ctl.c revision 273312
1/*-
2 * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3 * Copyright (c) 2012 The FreeBSD Foundation
4 * All rights reserved.
5 *
6 * Portions of this software were developed by Edward Tomasz Napierala
7 * under sponsorship from the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions, and the following disclaimer,
14 *    without modification.
15 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16 *    substantially similar to the "NO WARRANTY" disclaimer below
17 *    ("Disclaimer") and any redistribution must be conditioned upon
18 *    including a substantially similar Disclaimer requirement for further
19 *    binary redistribution.
20 *
21 * NO WARRANTY
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGES.
33 *
34 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
35 */
36/*
37 * CAM Target Layer, a SCSI device emulation subsystem.
38 *
39 * Author: Ken Merry <ken@FreeBSD.org>
40 */
41
42#define _CTL_C
43
44#include <sys/cdefs.h>
45__FBSDID("$FreeBSD: stable/10/sys/cam/ctl/ctl.c 273312 2014-10-20 07:34:37Z mav $");
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/kernel.h>
50#include <sys/types.h>
51#include <sys/kthread.h>
52#include <sys/bio.h>
53#include <sys/fcntl.h>
54#include <sys/lock.h>
55#include <sys/module.h>
56#include <sys/mutex.h>
57#include <sys/condvar.h>
58#include <sys/malloc.h>
59#include <sys/conf.h>
60#include <sys/ioccom.h>
61#include <sys/queue.h>
62#include <sys/sbuf.h>
63#include <sys/smp.h>
64#include <sys/endian.h>
65#include <sys/sysctl.h>
66
67#include <cam/cam.h>
68#include <cam/scsi/scsi_all.h>
69#include <cam/scsi/scsi_da.h>
70#include <cam/ctl/ctl_io.h>
71#include <cam/ctl/ctl.h>
72#include <cam/ctl/ctl_frontend.h>
73#include <cam/ctl/ctl_frontend_internal.h>
74#include <cam/ctl/ctl_util.h>
75#include <cam/ctl/ctl_backend.h>
76#include <cam/ctl/ctl_ioctl.h>
77#include <cam/ctl/ctl_ha.h>
78#include <cam/ctl/ctl_private.h>
79#include <cam/ctl/ctl_debug.h>
80#include <cam/ctl/ctl_scsi_all.h>
81#include <cam/ctl/ctl_error.h>
82
83struct ctl_softc *control_softc = NULL;
84
85/*
86 * Size and alignment macros needed for Copan-specific HA hardware.  These
87 * can go away when the HA code is re-written, and uses busdma for any
88 * hardware.
89 */
90#define	CTL_ALIGN_8B(target, source, type)				\
91	if (((uint32_t)source & 0x7) != 0)				\
92		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
93	else								\
94		target = (type)source;
95
96#define	CTL_SIZE_8B(target, size)					\
97	if ((size & 0x7) != 0)						\
98		target = size + (0x8 - (size & 0x7));			\
99	else								\
100		target = size;
101
102#define CTL_ALIGN_8B_MARGIN	16
103
104/*
105 * Template mode pages.
106 */
107
108/*
109 * Note that these are default values only.  The actual values will be
110 * filled in when the user does a mode sense.
111 */
112static struct copan_power_subpage power_page_default = {
113	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
114	/*subpage*/ PWR_SUBPAGE_CODE,
115	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
116			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
117	/*page_version*/ PWR_VERSION,
118	/* total_luns */ 26,
119	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
120	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
121		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
122		      0, 0, 0, 0, 0, 0}
123};
124
125static struct copan_power_subpage power_page_changeable = {
126	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
127	/*subpage*/ PWR_SUBPAGE_CODE,
128	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
129			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
130	/*page_version*/ 0,
131	/* total_luns */ 0,
132	/* max_active_luns*/ 0,
133	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
134		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
135		      0, 0, 0, 0, 0, 0}
136};
137
138static struct copan_aps_subpage aps_page_default = {
139	APS_PAGE_CODE | SMPH_SPF, //page_code
140	APS_SUBPAGE_CODE, //subpage
141	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
142	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
143	APS_VERSION, //page_version
144	0, //lock_active
145	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
146	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
147	0, 0, 0, 0, 0} //reserved
148};
149
150static struct copan_aps_subpage aps_page_changeable = {
151	APS_PAGE_CODE | SMPH_SPF, //page_code
152	APS_SUBPAGE_CODE, //subpage
153	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
154	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
155	0, //page_version
156	0, //lock_active
157	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
158	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
159	0, 0, 0, 0, 0} //reserved
160};
161
162static struct copan_debugconf_subpage debugconf_page_default = {
163	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
164	DBGCNF_SUBPAGE_CODE,		/* subpage */
165	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
166	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
167	DBGCNF_VERSION,			/* page_version */
168	{CTL_TIME_IO_DEFAULT_SECS>>8,
169	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
170};
171
172static struct copan_debugconf_subpage debugconf_page_changeable = {
173	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
174	DBGCNF_SUBPAGE_CODE,		/* subpage */
175	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
176	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
177	0,				/* page_version */
178	{0xff,0xff},			/* ctl_time_io_secs */
179};
180
181static struct scsi_format_page format_page_default = {
182	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
183	/*page_length*/sizeof(struct scsi_format_page) - 2,
184	/*tracks_per_zone*/ {0, 0},
185	/*alt_sectors_per_zone*/ {0, 0},
186	/*alt_tracks_per_zone*/ {0, 0},
187	/*alt_tracks_per_lun*/ {0, 0},
188	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
189			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
190	/*bytes_per_sector*/ {0, 0},
191	/*interleave*/ {0, 0},
192	/*track_skew*/ {0, 0},
193	/*cylinder_skew*/ {0, 0},
194	/*flags*/ SFP_HSEC,
195	/*reserved*/ {0, 0, 0}
196};
197
198static struct scsi_format_page format_page_changeable = {
199	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
200	/*page_length*/sizeof(struct scsi_format_page) - 2,
201	/*tracks_per_zone*/ {0, 0},
202	/*alt_sectors_per_zone*/ {0, 0},
203	/*alt_tracks_per_zone*/ {0, 0},
204	/*alt_tracks_per_lun*/ {0, 0},
205	/*sectors_per_track*/ {0, 0},
206	/*bytes_per_sector*/ {0, 0},
207	/*interleave*/ {0, 0},
208	/*track_skew*/ {0, 0},
209	/*cylinder_skew*/ {0, 0},
210	/*flags*/ 0,
211	/*reserved*/ {0, 0, 0}
212};
213
214static struct scsi_rigid_disk_page rigid_disk_page_default = {
215	/*page_code*/SMS_RIGID_DISK_PAGE,
216	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
217	/*cylinders*/ {0, 0, 0},
218	/*heads*/ CTL_DEFAULT_HEADS,
219	/*start_write_precomp*/ {0, 0, 0},
220	/*start_reduced_current*/ {0, 0, 0},
221	/*step_rate*/ {0, 0},
222	/*landing_zone_cylinder*/ {0, 0, 0},
223	/*rpl*/ SRDP_RPL_DISABLED,
224	/*rotational_offset*/ 0,
225	/*reserved1*/ 0,
226	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
227			   CTL_DEFAULT_ROTATION_RATE & 0xff},
228	/*reserved2*/ {0, 0}
229};
230
231static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
232	/*page_code*/SMS_RIGID_DISK_PAGE,
233	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
234	/*cylinders*/ {0, 0, 0},
235	/*heads*/ 0,
236	/*start_write_precomp*/ {0, 0, 0},
237	/*start_reduced_current*/ {0, 0, 0},
238	/*step_rate*/ {0, 0},
239	/*landing_zone_cylinder*/ {0, 0, 0},
240	/*rpl*/ 0,
241	/*rotational_offset*/ 0,
242	/*reserved1*/ 0,
243	/*rotation_rate*/ {0, 0},
244	/*reserved2*/ {0, 0}
245};
246
247static struct scsi_caching_page caching_page_default = {
248	/*page_code*/SMS_CACHING_PAGE,
249	/*page_length*/sizeof(struct scsi_caching_page) - 2,
250	/*flags1*/ SCP_DISC | SCP_WCE,
251	/*ret_priority*/ 0,
252	/*disable_pf_transfer_len*/ {0xff, 0xff},
253	/*min_prefetch*/ {0, 0},
254	/*max_prefetch*/ {0xff, 0xff},
255	/*max_pf_ceiling*/ {0xff, 0xff},
256	/*flags2*/ 0,
257	/*cache_segments*/ 0,
258	/*cache_seg_size*/ {0, 0},
259	/*reserved*/ 0,
260	/*non_cache_seg_size*/ {0, 0, 0}
261};
262
263static struct scsi_caching_page caching_page_changeable = {
264	/*page_code*/SMS_CACHING_PAGE,
265	/*page_length*/sizeof(struct scsi_caching_page) - 2,
266	/*flags1*/ SCP_WCE | SCP_RCD,
267	/*ret_priority*/ 0,
268	/*disable_pf_transfer_len*/ {0, 0},
269	/*min_prefetch*/ {0, 0},
270	/*max_prefetch*/ {0, 0},
271	/*max_pf_ceiling*/ {0, 0},
272	/*flags2*/ 0,
273	/*cache_segments*/ 0,
274	/*cache_seg_size*/ {0, 0},
275	/*reserved*/ 0,
276	/*non_cache_seg_size*/ {0, 0, 0}
277};
278
279static struct scsi_control_page control_page_default = {
280	/*page_code*/SMS_CONTROL_MODE_PAGE,
281	/*page_length*/sizeof(struct scsi_control_page) - 2,
282	/*rlec*/0,
283	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
284	/*eca_and_aen*/0,
285	/*flags4*/SCP_TAS,
286	/*aen_holdoff_period*/{0, 0},
287	/*busy_timeout_period*/{0, 0},
288	/*extended_selftest_completion_time*/{0, 0}
289};
290
291static struct scsi_control_page control_page_changeable = {
292	/*page_code*/SMS_CONTROL_MODE_PAGE,
293	/*page_length*/sizeof(struct scsi_control_page) - 2,
294	/*rlec*/SCP_DSENSE,
295	/*queue_flags*/SCP_QUEUE_ALG_MASK,
296	/*eca_and_aen*/SCP_SWP,
297	/*flags4*/0,
298	/*aen_holdoff_period*/{0, 0},
299	/*busy_timeout_period*/{0, 0},
300	/*extended_selftest_completion_time*/{0, 0}
301};
302
303
304/*
305 * XXX KDM move these into the softc.
306 */
307static int rcv_sync_msg;
308static int persis_offset;
309static uint8_t ctl_pause_rtr;
310static int     ctl_is_single = 1;
311static int     index_to_aps_page;
312
313SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
314static int worker_threads = -1;
315TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
316SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
317    &worker_threads, 1, "Number of worker threads");
318static int verbose = 0;
319TUNABLE_INT("kern.cam.ctl.verbose", &verbose);
320SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
321    &verbose, 0, "Show SCSI errors returned to initiator");
322
323/*
324 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
325 * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
326 * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
327 * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
328 */
329#define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
330
331static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
332				  int param);
333static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
334static int ctl_init(void);
335void ctl_shutdown(void);
336static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
337static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
338static void ctl_ioctl_online(void *arg);
339static void ctl_ioctl_offline(void *arg);
340static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
341static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
342static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
343static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
344static int ctl_ioctl_submit_wait(union ctl_io *io);
345static void ctl_ioctl_datamove(union ctl_io *io);
346static void ctl_ioctl_done(union ctl_io *io);
347static void ctl_ioctl_hard_startstop_callback(void *arg,
348					      struct cfi_metatask *metatask);
349static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
350static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
351			      struct ctl_ooa *ooa_hdr,
352			      struct ctl_ooa_entry *kern_entries);
353static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
354		     struct thread *td);
355static uint32_t ctl_map_lun(int port_num, uint32_t lun);
356static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
357#ifdef unused
358static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
359				   uint32_t targ_target, uint32_t targ_lun,
360				   int can_wait);
361static void ctl_kfree_io(union ctl_io *io);
362#endif /* unused */
363static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
364			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
365static int ctl_free_lun(struct ctl_lun *lun);
366static void ctl_create_lun(struct ctl_be_lun *be_lun);
367/**
368static void ctl_failover_change_pages(struct ctl_softc *softc,
369				      struct ctl_scsiio *ctsio, int master);
370**/
371
372static int ctl_do_mode_select(union ctl_io *io);
373static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
374			   uint64_t res_key, uint64_t sa_res_key,
375			   uint8_t type, uint32_t residx,
376			   struct ctl_scsiio *ctsio,
377			   struct scsi_per_res_out *cdb,
378			   struct scsi_per_res_out_parms* param);
379static void ctl_pro_preempt_other(struct ctl_lun *lun,
380				  union ctl_ha_msg *msg);
381static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
382static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
383static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
384static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
385static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
386static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
387static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
388					 int alloc_len);
389static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
390					 int alloc_len);
391static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
392static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
393static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
394static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
395static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
396static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
397static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
398    union ctl_io *pending_io, union ctl_io *ooa_io);
399static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
400				union ctl_io *starting_io);
401static int ctl_check_blocked(struct ctl_lun *lun);
402static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
403				struct ctl_lun *lun,
404				const struct ctl_cmd_entry *entry,
405				struct ctl_scsiio *ctsio);
406//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
407static void ctl_failover(void);
408static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
409			       struct ctl_scsiio *ctsio);
410static int ctl_scsiio(struct ctl_scsiio *ctsio);
411
412static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
413static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
414			    ctl_ua_type ua_type);
415static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
416			 ctl_ua_type ua_type);
417static int ctl_abort_task(union ctl_io *io);
418static int ctl_abort_task_set(union ctl_io *io);
419static int ctl_i_t_nexus_reset(union ctl_io *io);
420static void ctl_run_task(union ctl_io *io);
421#ifdef CTL_IO_DELAY
422static void ctl_datamove_timer_wakeup(void *arg);
423static void ctl_done_timer_wakeup(void *arg);
424#endif /* CTL_IO_DELAY */
425
426static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
427static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
428static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
429static void ctl_datamove_remote_write(union ctl_io *io);
430static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
431static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
432static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
433static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
434				    ctl_ha_dt_cb callback);
435static void ctl_datamove_remote_read(union ctl_io *io);
436static void ctl_datamove_remote(union ctl_io *io);
437static int ctl_process_done(union ctl_io *io);
438static void ctl_lun_thread(void *arg);
439static void ctl_work_thread(void *arg);
440static void ctl_enqueue_incoming(union ctl_io *io);
441static void ctl_enqueue_rtr(union ctl_io *io);
442static void ctl_enqueue_done(union ctl_io *io);
443static void ctl_enqueue_isc(union ctl_io *io);
444static const struct ctl_cmd_entry *
445    ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
446static const struct ctl_cmd_entry *
447    ctl_validate_command(struct ctl_scsiio *ctsio);
448static int ctl_cmd_applicable(uint8_t lun_type,
449    const struct ctl_cmd_entry *entry);
450
451/*
452 * Load the serialization table.  This isn't very pretty, but is probably
453 * the easiest way to do it.
454 */
455#include "ctl_ser_table.c"
456
457/*
458 * We only need to define open, close and ioctl routines for this driver.
459 */
460static struct cdevsw ctl_cdevsw = {
461	.d_version =	D_VERSION,
462	.d_flags =	0,
463	.d_open =	ctl_open,
464	.d_close =	ctl_close,
465	.d_ioctl =	ctl_ioctl,
466	.d_name =	"ctl",
467};
468
469
470MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
471MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
472
473static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
474
475static moduledata_t ctl_moduledata = {
476	"ctl",
477	ctl_module_event_handler,
478	NULL
479};
480
481DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
482MODULE_VERSION(ctl, 1);
483
484static struct ctl_frontend ioctl_frontend =
485{
486	.name = "ioctl",
487};
488
489static void
490ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
491			    union ctl_ha_msg *msg_info)
492{
493	struct ctl_scsiio *ctsio;
494
495	if (msg_info->hdr.original_sc == NULL) {
496		printf("%s: original_sc == NULL!\n", __func__);
497		/* XXX KDM now what? */
498		return;
499	}
500
501	ctsio = &msg_info->hdr.original_sc->scsiio;
502	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
503	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
504	ctsio->io_hdr.status = msg_info->hdr.status;
505	ctsio->scsi_status = msg_info->scsi.scsi_status;
506	ctsio->sense_len = msg_info->scsi.sense_len;
507	ctsio->sense_residual = msg_info->scsi.sense_residual;
508	ctsio->residual = msg_info->scsi.residual;
509	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
510	       sizeof(ctsio->sense_data));
511	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
512	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
513	ctl_enqueue_isc((union ctl_io *)ctsio);
514}
515
516static void
517ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
518				union ctl_ha_msg *msg_info)
519{
520	struct ctl_scsiio *ctsio;
521
522	if (msg_info->hdr.serializing_sc == NULL) {
523		printf("%s: serializing_sc == NULL!\n", __func__);
524		/* XXX KDM now what? */
525		return;
526	}
527
528	ctsio = &msg_info->hdr.serializing_sc->scsiio;
529#if 0
530	/*
531	 * Attempt to catch the situation where an I/O has
532	 * been freed, and we're using it again.
533	 */
534	if (ctsio->io_hdr.io_type == 0xff) {
535		union ctl_io *tmp_io;
536		tmp_io = (union ctl_io *)ctsio;
537		printf("%s: %p use after free!\n", __func__,
538		       ctsio);
539		printf("%s: type %d msg %d cdb %x iptl: "
540		       "%d:%d:%d:%d tag 0x%04x "
541		       "flag %#x status %x\n",
542			__func__,
543			tmp_io->io_hdr.io_type,
544			tmp_io->io_hdr.msg_type,
545			tmp_io->scsiio.cdb[0],
546			tmp_io->io_hdr.nexus.initid.id,
547			tmp_io->io_hdr.nexus.targ_port,
548			tmp_io->io_hdr.nexus.targ_target.id,
549			tmp_io->io_hdr.nexus.targ_lun,
550			(tmp_io->io_hdr.io_type ==
551			CTL_IO_TASK) ?
552			tmp_io->taskio.tag_num :
553			tmp_io->scsiio.tag_num,
554		        tmp_io->io_hdr.flags,
555			tmp_io->io_hdr.status);
556	}
557#endif
558	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
559	ctl_enqueue_isc((union ctl_io *)ctsio);
560}
561
562/*
563 * ISC (Inter Shelf Communication) event handler.  Events from the HA
564 * subsystem come in here.
565 */
566static void
567ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
568{
569	struct ctl_softc *ctl_softc;
570	union ctl_io *io;
571	struct ctl_prio *presio;
572	ctl_ha_status isc_status;
573
574	ctl_softc = control_softc;
575	io = NULL;
576
577
578#if 0
579	printf("CTL: Isc Msg event %d\n", event);
580#endif
581	if (event == CTL_HA_EVT_MSG_RECV) {
582		union ctl_ha_msg msg_info;
583
584		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
585					     sizeof(msg_info), /*wait*/ 0);
586#if 0
587		printf("CTL: msg_type %d\n", msg_info.msg_type);
588#endif
589		if (isc_status != 0) {
590			printf("Error receiving message, status = %d\n",
591			       isc_status);
592			return;
593		}
594
595		switch (msg_info.hdr.msg_type) {
596		case CTL_MSG_SERIALIZE:
597#if 0
598			printf("Serialize\n");
599#endif
600			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
601			if (io == NULL) {
602				printf("ctl_isc_event_handler: can't allocate "
603				       "ctl_io!\n");
604				/* Bad Juju */
605				/* Need to set busy and send msg back */
606				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
607				msg_info.hdr.status = CTL_SCSI_ERROR;
608				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
609				msg_info.scsi.sense_len = 0;
610			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
611				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
612				}
613				goto bailout;
614			}
615			ctl_zero_io(io);
616			// populate ctsio from msg_info
617			io->io_hdr.io_type = CTL_IO_SCSI;
618			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
619			io->io_hdr.original_sc = msg_info.hdr.original_sc;
620#if 0
621			printf("pOrig %x\n", (int)msg_info.original_sc);
622#endif
623			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
624					    CTL_FLAG_IO_ACTIVE;
625			/*
626			 * If we're in serialization-only mode, we don't
627			 * want to go through full done processing.  Thus
628			 * the COPY flag.
629			 *
630			 * XXX KDM add another flag that is more specific.
631			 */
632			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
633				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
634			io->io_hdr.nexus = msg_info.hdr.nexus;
635#if 0
636			printf("targ %d, port %d, iid %d, lun %d\n",
637			       io->io_hdr.nexus.targ_target.id,
638			       io->io_hdr.nexus.targ_port,
639			       io->io_hdr.nexus.initid.id,
640			       io->io_hdr.nexus.targ_lun);
641#endif
642			io->scsiio.tag_num = msg_info.scsi.tag_num;
643			io->scsiio.tag_type = msg_info.scsi.tag_type;
644			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
645			       CTL_MAX_CDBLEN);
646			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
647				const struct ctl_cmd_entry *entry;
648
649				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
650				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
651				io->io_hdr.flags |=
652					entry->flags & CTL_FLAG_DATA_MASK;
653			}
654			ctl_enqueue_isc(io);
655			break;
656
657		/* Performed on the Originating SC, XFER mode only */
658		case CTL_MSG_DATAMOVE: {
659			struct ctl_sg_entry *sgl;
660			int i, j;
661
662			io = msg_info.hdr.original_sc;
663			if (io == NULL) {
664				printf("%s: original_sc == NULL!\n", __func__);
665				/* XXX KDM do something here */
666				break;
667			}
668			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
669			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
670			/*
671			 * Keep track of this, we need to send it back over
672			 * when the datamove is complete.
673			 */
674			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
675
676			if (msg_info.dt.sg_sequence == 0) {
677				/*
678				 * XXX KDM we use the preallocated S/G list
679				 * here, but we'll need to change this to
680				 * dynamic allocation if we need larger S/G
681				 * lists.
682				 */
683				if (msg_info.dt.kern_sg_entries >
684				    sizeof(io->io_hdr.remote_sglist) /
685				    sizeof(io->io_hdr.remote_sglist[0])) {
686					printf("%s: number of S/G entries "
687					    "needed %u > allocated num %zd\n",
688					    __func__,
689					    msg_info.dt.kern_sg_entries,
690					    sizeof(io->io_hdr.remote_sglist)/
691					    sizeof(io->io_hdr.remote_sglist[0]));
692
693					/*
694					 * XXX KDM send a message back to
695					 * the other side to shut down the
696					 * DMA.  The error will come back
697					 * through via the normal channel.
698					 */
699					break;
700				}
701				sgl = io->io_hdr.remote_sglist;
702				memset(sgl, 0,
703				       sizeof(io->io_hdr.remote_sglist));
704
705				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
706
707				io->scsiio.kern_sg_entries =
708					msg_info.dt.kern_sg_entries;
709				io->scsiio.rem_sg_entries =
710					msg_info.dt.kern_sg_entries;
711				io->scsiio.kern_data_len =
712					msg_info.dt.kern_data_len;
713				io->scsiio.kern_total_len =
714					msg_info.dt.kern_total_len;
715				io->scsiio.kern_data_resid =
716					msg_info.dt.kern_data_resid;
717				io->scsiio.kern_rel_offset =
718					msg_info.dt.kern_rel_offset;
719				/*
720				 * Clear out per-DMA flags.
721				 */
722				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
723				/*
724				 * Add per-DMA flags that are set for this
725				 * particular DMA request.
726				 */
727				io->io_hdr.flags |= msg_info.dt.flags &
728						    CTL_FLAG_RDMA_MASK;
729			} else
730				sgl = (struct ctl_sg_entry *)
731					io->scsiio.kern_data_ptr;
732
733			for (i = msg_info.dt.sent_sg_entries, j = 0;
734			     i < (msg_info.dt.sent_sg_entries +
735			     msg_info.dt.cur_sg_entries); i++, j++) {
736				sgl[i].addr = msg_info.dt.sg_list[j].addr;
737				sgl[i].len = msg_info.dt.sg_list[j].len;
738
739#if 0
740				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
741				       __func__,
742				       msg_info.dt.sg_list[j].addr,
743				       msg_info.dt.sg_list[j].len,
744				       sgl[i].addr, sgl[i].len, j, i);
745#endif
746			}
747#if 0
748			memcpy(&sgl[msg_info.dt.sent_sg_entries],
749			       msg_info.dt.sg_list,
750			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
751#endif
752
753			/*
754			 * If this is the last piece of the I/O, we've got
755			 * the full S/G list.  Queue processing in the thread.
756			 * Otherwise wait for the next piece.
757			 */
758			if (msg_info.dt.sg_last != 0)
759				ctl_enqueue_isc(io);
760			break;
761		}
762		/* Performed on the Serializing (primary) SC, XFER mode only */
763		case CTL_MSG_DATAMOVE_DONE: {
764			if (msg_info.hdr.serializing_sc == NULL) {
765				printf("%s: serializing_sc == NULL!\n",
766				       __func__);
767				/* XXX KDM now what? */
768				break;
769			}
770			/*
771			 * We grab the sense information here in case
772			 * there was a failure, so we can return status
773			 * back to the initiator.
774			 */
775			io = msg_info.hdr.serializing_sc;
776			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
777			io->io_hdr.status = msg_info.hdr.status;
778			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
779			io->scsiio.sense_len = msg_info.scsi.sense_len;
780			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
781			io->io_hdr.port_status = msg_info.scsi.fetd_status;
782			io->scsiio.residual = msg_info.scsi.residual;
783			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
784			       sizeof(io->scsiio.sense_data));
785			ctl_enqueue_isc(io);
786			break;
787		}
788
789		/* Preformed on Originating SC, SER_ONLY mode */
790		case CTL_MSG_R2R:
791			io = msg_info.hdr.original_sc;
792			if (io == NULL) {
793				printf("%s: Major Bummer\n", __func__);
794				return;
795			} else {
796#if 0
797				printf("pOrig %x\n",(int) ctsio);
798#endif
799			}
800			io->io_hdr.msg_type = CTL_MSG_R2R;
801			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
802			ctl_enqueue_isc(io);
803			break;
804
805		/*
806		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
807		 * mode.
808		 * Performed on the Originating (i.e. secondary) SC in XFER
809		 * mode
810		 */
811		case CTL_MSG_FINISH_IO:
812			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
813				ctl_isc_handler_finish_xfer(ctl_softc,
814							    &msg_info);
815			else
816				ctl_isc_handler_finish_ser_only(ctl_softc,
817								&msg_info);
818			break;
819
820		/* Preformed on Originating SC */
821		case CTL_MSG_BAD_JUJU:
822			io = msg_info.hdr.original_sc;
823			if (io == NULL) {
824				printf("%s: Bad JUJU!, original_sc is NULL!\n",
825				       __func__);
826				break;
827			}
828			ctl_copy_sense_data(&msg_info, io);
829			/*
830			 * IO should have already been cleaned up on other
831			 * SC so clear this flag so we won't send a message
832			 * back to finish the IO there.
833			 */
834			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
835			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
836
837			/* io = msg_info.hdr.serializing_sc; */
838			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
839			ctl_enqueue_isc(io);
840			break;
841
842		/* Handle resets sent from the other side */
843		case CTL_MSG_MANAGE_TASKS: {
844			struct ctl_taskio *taskio;
845			taskio = (struct ctl_taskio *)ctl_alloc_io(
846				(void *)ctl_softc->othersc_pool);
847			if (taskio == NULL) {
848				printf("ctl_isc_event_handler: can't allocate "
849				       "ctl_io!\n");
850				/* Bad Juju */
851				/* should I just call the proper reset func
852				   here??? */
853				goto bailout;
854			}
855			ctl_zero_io((union ctl_io *)taskio);
856			taskio->io_hdr.io_type = CTL_IO_TASK;
857			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
858			taskio->io_hdr.nexus = msg_info.hdr.nexus;
859			taskio->task_action = msg_info.task.task_action;
860			taskio->tag_num = msg_info.task.tag_num;
861			taskio->tag_type = msg_info.task.tag_type;
862#ifdef CTL_TIME_IO
863			taskio->io_hdr.start_time = time_uptime;
864			getbintime(&taskio->io_hdr.start_bt);
865#if 0
866			cs_prof_gettime(&taskio->io_hdr.start_ticks);
867#endif
868#endif /* CTL_TIME_IO */
869			ctl_run_task((union ctl_io *)taskio);
870			break;
871		}
872		/* Persistent Reserve action which needs attention */
873		case CTL_MSG_PERS_ACTION:
874			presio = (struct ctl_prio *)ctl_alloc_io(
875				(void *)ctl_softc->othersc_pool);
876			if (presio == NULL) {
877				printf("ctl_isc_event_handler: can't allocate "
878				       "ctl_io!\n");
879				/* Bad Juju */
880				/* Need to set busy and send msg back */
881				goto bailout;
882			}
883			ctl_zero_io((union ctl_io *)presio);
884			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
885			presio->pr_msg = msg_info.pr;
886			ctl_enqueue_isc((union ctl_io *)presio);
887			break;
888		case CTL_MSG_SYNC_FE:
889			rcv_sync_msg = 1;
890			break;
891		case CTL_MSG_APS_LOCK: {
892			// It's quicker to execute this then to
893			// queue it.
894			struct ctl_lun *lun;
895			struct ctl_page_index *page_index;
896			struct copan_aps_subpage *current_sp;
897			uint32_t targ_lun;
898
899			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
900			lun = ctl_softc->ctl_luns[targ_lun];
901			mtx_lock(&lun->lun_lock);
902			page_index = &lun->mode_pages.index[index_to_aps_page];
903			current_sp = (struct copan_aps_subpage *)
904				     (page_index->page_data +
905				     (page_index->page_len * CTL_PAGE_CURRENT));
906
907			current_sp->lock_active = msg_info.aps.lock_flag;
908			mtx_unlock(&lun->lun_lock);
909		        break;
910		}
911		default:
912		        printf("How did I get here?\n");
913		}
914	} else if (event == CTL_HA_EVT_MSG_SENT) {
915		if (param != CTL_HA_STATUS_SUCCESS) {
916			printf("Bad status from ctl_ha_msg_send status %d\n",
917			       param);
918		}
919		return;
920	} else if (event == CTL_HA_EVT_DISCONNECT) {
921		printf("CTL: Got a disconnect from Isc\n");
922		return;
923	} else {
924		printf("ctl_isc_event_handler: Unknown event %d\n", event);
925		return;
926	}
927
928bailout:
929	return;
930}
931
932static void
933ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
934{
935	struct scsi_sense_data *sense;
936
937	sense = &dest->scsiio.sense_data;
938	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
939	dest->scsiio.scsi_status = src->scsi.scsi_status;
940	dest->scsiio.sense_len = src->scsi.sense_len;
941	dest->io_hdr.status = src->hdr.status;
942}
943
944static int
945ctl_init(void)
946{
947	struct ctl_softc *softc;
948	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
949	struct ctl_port *port;
950        uint8_t sc_id =0;
951	int i, error, retval;
952	//int isc_retval;
953
954	retval = 0;
955	ctl_pause_rtr = 0;
956        rcv_sync_msg = 0;
957
958	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
959			       M_WAITOK | M_ZERO);
960	softc = control_softc;
961
962	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
963			      "cam/ctl");
964
965	softc->dev->si_drv1 = softc;
966
967	/*
968	 * By default, return a "bad LUN" peripheral qualifier for unknown
969	 * LUNs.  The user can override this default using the tunable or
970	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
971	 */
972	softc->inquiry_pq_no_lun = 1;
973	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
974			  &softc->inquiry_pq_no_lun);
975	sysctl_ctx_init(&softc->sysctl_ctx);
976	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
977		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
978		CTLFLAG_RD, 0, "CAM Target Layer");
979
980	if (softc->sysctl_tree == NULL) {
981		printf("%s: unable to allocate sysctl tree\n", __func__);
982		destroy_dev(softc->dev);
983		free(control_softc, M_DEVBUF);
984		control_softc = NULL;
985		return (ENOMEM);
986	}
987
988	SYSCTL_ADD_INT(&softc->sysctl_ctx,
989		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
990		       "inquiry_pq_no_lun", CTLFLAG_RW,
991		       &softc->inquiry_pq_no_lun, 0,
992		       "Report no lun possible for invalid LUNs");
993
994	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
995	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
996	softc->open_count = 0;
997
998	/*
999	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1000	 * the drive.
1001	 */
1002	softc->flags = CTL_FLAG_REAL_SYNC;
1003
1004	/*
1005	 * In Copan's HA scheme, the "master" and "slave" roles are
1006	 * figured out through the slot the controller is in.  Although it
1007	 * is an active/active system, someone has to be in charge.
1008 	 */
1009#ifdef NEEDTOPORT
1010        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1011#endif
1012
1013        if (sc_id == 0) {
1014		softc->flags |= CTL_FLAG_MASTER_SHELF;
1015		persis_offset = 0;
1016	} else
1017		persis_offset = CTL_MAX_INITIATORS;
1018
1019	/*
1020	 * XXX KDM need to figure out where we want to get our target ID
1021	 * and WWID.  Is it different on each port?
1022	 */
1023	softc->target.id = 0;
1024	softc->target.wwid[0] = 0x12345678;
1025	softc->target.wwid[1] = 0x87654321;
1026	STAILQ_INIT(&softc->lun_list);
1027	STAILQ_INIT(&softc->pending_lun_queue);
1028	STAILQ_INIT(&softc->fe_list);
1029	STAILQ_INIT(&softc->port_list);
1030	STAILQ_INIT(&softc->be_list);
1031	STAILQ_INIT(&softc->io_pools);
1032	ctl_tpc_init(softc);
1033
1034	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1035			    &internal_pool)!= 0){
1036		printf("ctl: can't allocate %d entry internal pool, "
1037		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1038		return (ENOMEM);
1039	}
1040
1041	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1042			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1043		printf("ctl: can't allocate %d entry emergency pool, "
1044		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1045		ctl_pool_free(internal_pool);
1046		return (ENOMEM);
1047	}
1048
1049	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1050	                    &other_pool) != 0)
1051	{
1052		printf("ctl: can't allocate %d entry other SC pool, "
1053		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1054		ctl_pool_free(internal_pool);
1055		ctl_pool_free(emergency_pool);
1056		return (ENOMEM);
1057	}
1058
1059	softc->internal_pool = internal_pool;
1060	softc->emergency_pool = emergency_pool;
1061	softc->othersc_pool = other_pool;
1062
1063	if (worker_threads <= 0)
1064		worker_threads = max(1, mp_ncpus / 4);
1065	if (worker_threads > CTL_MAX_THREADS)
1066		worker_threads = CTL_MAX_THREADS;
1067
1068	for (i = 0; i < worker_threads; i++) {
1069		struct ctl_thread *thr = &softc->threads[i];
1070
1071		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1072		thr->ctl_softc = softc;
1073		STAILQ_INIT(&thr->incoming_queue);
1074		STAILQ_INIT(&thr->rtr_queue);
1075		STAILQ_INIT(&thr->done_queue);
1076		STAILQ_INIT(&thr->isc_queue);
1077
1078		error = kproc_kthread_add(ctl_work_thread, thr,
1079		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1080		if (error != 0) {
1081			printf("error creating CTL work thread!\n");
1082			ctl_pool_free(internal_pool);
1083			ctl_pool_free(emergency_pool);
1084			ctl_pool_free(other_pool);
1085			return (error);
1086		}
1087	}
1088	error = kproc_kthread_add(ctl_lun_thread, softc,
1089	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1090	if (error != 0) {
1091		printf("error creating CTL lun thread!\n");
1092		ctl_pool_free(internal_pool);
1093		ctl_pool_free(emergency_pool);
1094		ctl_pool_free(other_pool);
1095		return (error);
1096	}
1097	if (bootverbose)
1098		printf("ctl: CAM Target Layer loaded\n");
1099
1100	/*
1101	 * Initialize the ioctl front end.
1102	 */
1103	ctl_frontend_register(&ioctl_frontend);
1104	port = &softc->ioctl_info.port;
1105	port->frontend = &ioctl_frontend;
1106	sprintf(softc->ioctl_info.port_name, "ioctl");
1107	port->port_type = CTL_PORT_IOCTL;
1108	port->num_requested_ctl_io = 100;
1109	port->port_name = softc->ioctl_info.port_name;
1110	port->port_online = ctl_ioctl_online;
1111	port->port_offline = ctl_ioctl_offline;
1112	port->onoff_arg = &softc->ioctl_info;
1113	port->lun_enable = ctl_ioctl_lun_enable;
1114	port->lun_disable = ctl_ioctl_lun_disable;
1115	port->targ_lun_arg = &softc->ioctl_info;
1116	port->fe_datamove = ctl_ioctl_datamove;
1117	port->fe_done = ctl_ioctl_done;
1118	port->max_targets = 15;
1119	port->max_target_id = 15;
1120
1121	if (ctl_port_register(&softc->ioctl_info.port,
1122	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1123		printf("ctl: ioctl front end registration failed, will "
1124		       "continue anyway\n");
1125	}
1126
1127#ifdef CTL_IO_DELAY
1128	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1129		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1130		       sizeof(struct callout), CTL_TIMER_BYTES);
1131		return (EINVAL);
1132	}
1133#endif /* CTL_IO_DELAY */
1134
1135	return (0);
1136}
1137
1138void
1139ctl_shutdown(void)
1140{
1141	struct ctl_softc *softc;
1142	struct ctl_lun *lun, *next_lun;
1143	struct ctl_io_pool *pool;
1144
1145	softc = (struct ctl_softc *)control_softc;
1146
1147	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1148		printf("ctl: ioctl front end deregistration failed\n");
1149
1150	mtx_lock(&softc->ctl_lock);
1151
1152	/*
1153	 * Free up each LUN.
1154	 */
1155	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1156		next_lun = STAILQ_NEXT(lun, links);
1157		ctl_free_lun(lun);
1158	}
1159
1160	mtx_unlock(&softc->ctl_lock);
1161
1162	ctl_frontend_deregister(&ioctl_frontend);
1163
1164	/*
1165	 * This will rip the rug out from under any FETDs or anyone else
1166	 * that has a pool allocated.  Since we increment our module
1167	 * refcount any time someone outside the main CTL module allocates
1168	 * a pool, we shouldn't have any problems here.  The user won't be
1169	 * able to unload the CTL module until client modules have
1170	 * successfully unloaded.
1171	 */
1172	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1173		ctl_pool_free(pool);
1174
1175#if 0
1176	ctl_shutdown_thread(softc->work_thread);
1177	mtx_destroy(&softc->queue_lock);
1178#endif
1179
1180	ctl_tpc_shutdown(softc);
1181	mtx_destroy(&softc->pool_lock);
1182	mtx_destroy(&softc->ctl_lock);
1183
1184	destroy_dev(softc->dev);
1185
1186	sysctl_ctx_free(&softc->sysctl_ctx);
1187
1188	free(control_softc, M_DEVBUF);
1189	control_softc = NULL;
1190
1191	if (bootverbose)
1192		printf("ctl: CAM Target Layer unloaded\n");
1193}
1194
1195static int
1196ctl_module_event_handler(module_t mod, int what, void *arg)
1197{
1198
1199	switch (what) {
1200	case MOD_LOAD:
1201		return (ctl_init());
1202	case MOD_UNLOAD:
1203		return (EBUSY);
1204	default:
1205		return (EOPNOTSUPP);
1206	}
1207}
1208
1209/*
1210 * XXX KDM should we do some access checks here?  Bump a reference count to
1211 * prevent a CTL module from being unloaded while someone has it open?
1212 */
1213static int
1214ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1215{
1216	return (0);
1217}
1218
1219static int
1220ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1221{
1222	return (0);
1223}
1224
1225int
1226ctl_port_enable(ctl_port_type port_type)
1227{
1228	struct ctl_softc *softc;
1229	struct ctl_port *port;
1230
1231	if (ctl_is_single == 0) {
1232		union ctl_ha_msg msg_info;
1233		int isc_retval;
1234
1235#if 0
1236		printf("%s: HA mode, synchronizing frontend enable\n",
1237		        __func__);
1238#endif
1239		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1240	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1241		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1242			printf("Sync msg send error retval %d\n", isc_retval);
1243		}
1244		if (!rcv_sync_msg) {
1245			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1246			        sizeof(msg_info), 1);
1247		}
1248#if 0
1249        	printf("CTL:Frontend Enable\n");
1250	} else {
1251		printf("%s: single mode, skipping frontend synchronization\n",
1252		        __func__);
1253#endif
1254	}
1255
1256	softc = control_softc;
1257
1258	STAILQ_FOREACH(port, &softc->port_list, links) {
1259		if (port_type & port->port_type)
1260		{
1261#if 0
1262			printf("port %d\n", port->targ_port);
1263#endif
1264			ctl_port_online(port);
1265		}
1266	}
1267
1268	return (0);
1269}
1270
1271int
1272ctl_port_disable(ctl_port_type port_type)
1273{
1274	struct ctl_softc *softc;
1275	struct ctl_port *port;
1276
1277	softc = control_softc;
1278
1279	STAILQ_FOREACH(port, &softc->port_list, links) {
1280		if (port_type & port->port_type)
1281			ctl_port_offline(port);
1282	}
1283
1284	return (0);
1285}
1286
1287/*
1288 * Returns 0 for success, 1 for failure.
1289 * Currently the only failure mode is if there aren't enough entries
1290 * allocated.  So, in case of a failure, look at num_entries_dropped,
1291 * reallocate and try again.
1292 */
1293int
1294ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1295	      int *num_entries_filled, int *num_entries_dropped,
1296	      ctl_port_type port_type, int no_virtual)
1297{
1298	struct ctl_softc *softc;
1299	struct ctl_port *port;
1300	int entries_dropped, entries_filled;
1301	int retval;
1302	int i;
1303
1304	softc = control_softc;
1305
1306	retval = 0;
1307	entries_filled = 0;
1308	entries_dropped = 0;
1309
1310	i = 0;
1311	mtx_lock(&softc->ctl_lock);
1312	STAILQ_FOREACH(port, &softc->port_list, links) {
1313		struct ctl_port_entry *entry;
1314
1315		if ((port->port_type & port_type) == 0)
1316			continue;
1317
1318		if ((no_virtual != 0)
1319		 && (port->virtual_port != 0))
1320			continue;
1321
1322		if (entries_filled >= num_entries_alloced) {
1323			entries_dropped++;
1324			continue;
1325		}
1326		entry = &entries[i];
1327
1328		entry->port_type = port->port_type;
1329		strlcpy(entry->port_name, port->port_name,
1330			sizeof(entry->port_name));
1331		entry->physical_port = port->physical_port;
1332		entry->virtual_port = port->virtual_port;
1333		entry->wwnn = port->wwnn;
1334		entry->wwpn = port->wwpn;
1335
1336		i++;
1337		entries_filled++;
1338	}
1339
1340	mtx_unlock(&softc->ctl_lock);
1341
1342	if (entries_dropped > 0)
1343		retval = 1;
1344
1345	*num_entries_dropped = entries_dropped;
1346	*num_entries_filled = entries_filled;
1347
1348	return (retval);
1349}
1350
1351static void
1352ctl_ioctl_online(void *arg)
1353{
1354	struct ctl_ioctl_info *ioctl_info;
1355
1356	ioctl_info = (struct ctl_ioctl_info *)arg;
1357
1358	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1359}
1360
1361static void
1362ctl_ioctl_offline(void *arg)
1363{
1364	struct ctl_ioctl_info *ioctl_info;
1365
1366	ioctl_info = (struct ctl_ioctl_info *)arg;
1367
1368	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1369}
1370
1371/*
1372 * Remove an initiator by port number and initiator ID.
1373 * Returns 0 for success, -1 for failure.
1374 */
1375int
1376ctl_remove_initiator(struct ctl_port *port, int iid)
1377{
1378	struct ctl_softc *softc = control_softc;
1379
1380	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1381
1382	if (iid > CTL_MAX_INIT_PER_PORT) {
1383		printf("%s: initiator ID %u > maximun %u!\n",
1384		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1385		return (-1);
1386	}
1387
1388	mtx_lock(&softc->ctl_lock);
1389	port->wwpn_iid[iid].in_use--;
1390	port->wwpn_iid[iid].last_use = time_uptime;
1391	mtx_unlock(&softc->ctl_lock);
1392
1393	return (0);
1394}
1395
1396/*
1397 * Add an initiator to the initiator map.
1398 * Returns iid for success, < 0 for failure.
1399 */
1400int
1401ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1402{
1403	struct ctl_softc *softc = control_softc;
1404	time_t best_time;
1405	int i, best;
1406
1407	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1408
1409	if (iid >= CTL_MAX_INIT_PER_PORT) {
1410		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1411		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1412		free(name, M_CTL);
1413		return (-1);
1414	}
1415
1416	mtx_lock(&softc->ctl_lock);
1417
1418	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1419		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1420			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1421				iid = i;
1422				break;
1423			}
1424			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1425			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1426				iid = i;
1427				break;
1428			}
1429		}
1430	}
1431
1432	if (iid < 0) {
1433		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1434			if (port->wwpn_iid[i].in_use == 0 &&
1435			    port->wwpn_iid[i].wwpn == 0 &&
1436			    port->wwpn_iid[i].name == NULL) {
1437				iid = i;
1438				break;
1439			}
1440		}
1441	}
1442
1443	if (iid < 0) {
1444		best = -1;
1445		best_time = INT32_MAX;
1446		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1447			if (port->wwpn_iid[i].in_use == 0) {
1448				if (port->wwpn_iid[i].last_use < best_time) {
1449					best = i;
1450					best_time = port->wwpn_iid[i].last_use;
1451				}
1452			}
1453		}
1454		iid = best;
1455	}
1456
1457	if (iid < 0) {
1458		mtx_unlock(&softc->ctl_lock);
1459		free(name, M_CTL);
1460		return (-2);
1461	}
1462
1463	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1464		/*
1465		 * This is not an error yet.
1466		 */
1467		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1468#if 0
1469			printf("%s: port %d iid %u WWPN %#jx arrived"
1470			    " again\n", __func__, port->targ_port,
1471			    iid, (uintmax_t)wwpn);
1472#endif
1473			goto take;
1474		}
1475		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1476		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1477#if 0
1478			printf("%s: port %d iid %u name '%s' arrived"
1479			    " again\n", __func__, port->targ_port,
1480			    iid, name);
1481#endif
1482			goto take;
1483		}
1484
1485		/*
1486		 * This is an error, but what do we do about it?  The
1487		 * driver is telling us we have a new WWPN for this
1488		 * initiator ID, so we pretty much need to use it.
1489		 */
1490		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1491		    " but WWPN %#jx '%s' is still at that address\n",
1492		    __func__, port->targ_port, iid, wwpn, name,
1493		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1494		    port->wwpn_iid[iid].name);
1495
1496		/*
1497		 * XXX KDM clear have_ca and ua_pending on each LUN for
1498		 * this initiator.
1499		 */
1500	}
1501take:
1502	free(port->wwpn_iid[iid].name, M_CTL);
1503	port->wwpn_iid[iid].name = name;
1504	port->wwpn_iid[iid].wwpn = wwpn;
1505	port->wwpn_iid[iid].in_use++;
1506	mtx_unlock(&softc->ctl_lock);
1507
1508	return (iid);
1509}
1510
1511static int
1512ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1513{
1514	int len;
1515
1516	switch (port->port_type) {
1517	case CTL_PORT_FC:
1518	{
1519		struct scsi_transportid_fcp *id =
1520		    (struct scsi_transportid_fcp *)buf;
1521		if (port->wwpn_iid[iid].wwpn == 0)
1522			return (0);
1523		memset(id, 0, sizeof(*id));
1524		id->format_protocol = SCSI_PROTO_FC;
1525		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1526		return (sizeof(*id));
1527	}
1528	case CTL_PORT_ISCSI:
1529	{
1530		struct scsi_transportid_iscsi_port *id =
1531		    (struct scsi_transportid_iscsi_port *)buf;
1532		if (port->wwpn_iid[iid].name == NULL)
1533			return (0);
1534		memset(id, 0, 256);
1535		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1536		    SCSI_PROTO_ISCSI;
1537		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1538		len = roundup2(min(len, 252), 4);
1539		scsi_ulto2b(len, id->additional_length);
1540		return (sizeof(*id) + len);
1541	}
1542	case CTL_PORT_SAS:
1543	{
1544		struct scsi_transportid_sas *id =
1545		    (struct scsi_transportid_sas *)buf;
1546		if (port->wwpn_iid[iid].wwpn == 0)
1547			return (0);
1548		memset(id, 0, sizeof(*id));
1549		id->format_protocol = SCSI_PROTO_SAS;
1550		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1551		return (sizeof(*id));
1552	}
1553	default:
1554	{
1555		struct scsi_transportid_spi *id =
1556		    (struct scsi_transportid_spi *)buf;
1557		memset(id, 0, sizeof(*id));
1558		id->format_protocol = SCSI_PROTO_SPI;
1559		scsi_ulto2b(iid, id->scsi_addr);
1560		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1561		return (sizeof(*id));
1562	}
1563	}
1564}
1565
1566static int
1567ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1568{
1569	return (0);
1570}
1571
1572static int
1573ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1574{
1575	return (0);
1576}
1577
1578/*
1579 * Data movement routine for the CTL ioctl frontend port.
1580 */
1581static int
1582ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1583{
1584	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1585	struct ctl_sg_entry ext_entry, kern_entry;
1586	int ext_sglen, ext_sg_entries, kern_sg_entries;
1587	int ext_sg_start, ext_offset;
1588	int len_to_copy, len_copied;
1589	int kern_watermark, ext_watermark;
1590	int ext_sglist_malloced;
1591	int i, j;
1592
1593	ext_sglist_malloced = 0;
1594	ext_sg_start = 0;
1595	ext_offset = 0;
1596
1597	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1598
1599	/*
1600	 * If this flag is set, fake the data transfer.
1601	 */
1602	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1603		ctsio->ext_data_filled = ctsio->ext_data_len;
1604		goto bailout;
1605	}
1606
1607	/*
1608	 * To simplify things here, if we have a single buffer, stick it in
1609	 * a S/G entry and just make it a single entry S/G list.
1610	 */
1611	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1612		int len_seen;
1613
1614		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1615
1616		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1617							   M_WAITOK);
1618		ext_sglist_malloced = 1;
1619		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1620				   ext_sglen) != 0) {
1621			ctl_set_internal_failure(ctsio,
1622						 /*sks_valid*/ 0,
1623						 /*retry_count*/ 0);
1624			goto bailout;
1625		}
1626		ext_sg_entries = ctsio->ext_sg_entries;
1627		len_seen = 0;
1628		for (i = 0; i < ext_sg_entries; i++) {
1629			if ((len_seen + ext_sglist[i].len) >=
1630			     ctsio->ext_data_filled) {
1631				ext_sg_start = i;
1632				ext_offset = ctsio->ext_data_filled - len_seen;
1633				break;
1634			}
1635			len_seen += ext_sglist[i].len;
1636		}
1637	} else {
1638		ext_sglist = &ext_entry;
1639		ext_sglist->addr = ctsio->ext_data_ptr;
1640		ext_sglist->len = ctsio->ext_data_len;
1641		ext_sg_entries = 1;
1642		ext_sg_start = 0;
1643		ext_offset = ctsio->ext_data_filled;
1644	}
1645
1646	if (ctsio->kern_sg_entries > 0) {
1647		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1648		kern_sg_entries = ctsio->kern_sg_entries;
1649	} else {
1650		kern_sglist = &kern_entry;
1651		kern_sglist->addr = ctsio->kern_data_ptr;
1652		kern_sglist->len = ctsio->kern_data_len;
1653		kern_sg_entries = 1;
1654	}
1655
1656
1657	kern_watermark = 0;
1658	ext_watermark = ext_offset;
1659	len_copied = 0;
1660	for (i = ext_sg_start, j = 0;
1661	     i < ext_sg_entries && j < kern_sg_entries;) {
1662		uint8_t *ext_ptr, *kern_ptr;
1663
1664		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1665				      kern_sglist[j].len - kern_watermark);
1666
1667		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1668		ext_ptr = ext_ptr + ext_watermark;
1669		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1670			/*
1671			 * XXX KDM fix this!
1672			 */
1673			panic("need to implement bus address support");
1674#if 0
1675			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1676#endif
1677		} else
1678			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1679		kern_ptr = kern_ptr + kern_watermark;
1680
1681		kern_watermark += len_to_copy;
1682		ext_watermark += len_to_copy;
1683
1684		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1685		     CTL_FLAG_DATA_IN) {
1686			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1687					 "bytes to user\n", len_to_copy));
1688			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1689					 "to %p\n", kern_ptr, ext_ptr));
1690			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1691				ctl_set_internal_failure(ctsio,
1692							 /*sks_valid*/ 0,
1693							 /*retry_count*/ 0);
1694				goto bailout;
1695			}
1696		} else {
1697			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1698					 "bytes from user\n", len_to_copy));
1699			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1700					 "to %p\n", ext_ptr, kern_ptr));
1701			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1702				ctl_set_internal_failure(ctsio,
1703							 /*sks_valid*/ 0,
1704							 /*retry_count*/0);
1705				goto bailout;
1706			}
1707		}
1708
1709		len_copied += len_to_copy;
1710
1711		if (ext_sglist[i].len == ext_watermark) {
1712			i++;
1713			ext_watermark = 0;
1714		}
1715
1716		if (kern_sglist[j].len == kern_watermark) {
1717			j++;
1718			kern_watermark = 0;
1719		}
1720	}
1721
1722	ctsio->ext_data_filled += len_copied;
1723
1724	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1725			 "kern_sg_entries: %d\n", ext_sg_entries,
1726			 kern_sg_entries));
1727	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1728			 "kern_data_len = %d\n", ctsio->ext_data_len,
1729			 ctsio->kern_data_len));
1730
1731
1732	/* XXX KDM set residual?? */
1733bailout:
1734
1735	if (ext_sglist_malloced != 0)
1736		free(ext_sglist, M_CTL);
1737
1738	return (CTL_RETVAL_COMPLETE);
1739}
1740
1741/*
1742 * Serialize a command that went down the "wrong" side, and so was sent to
1743 * this controller for execution.  The logic is a little different than the
1744 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1745 * sent back to the other side, but in the success case, we execute the
1746 * command on this side (XFER mode) or tell the other side to execute it
1747 * (SER_ONLY mode).
1748 */
1749static int
1750ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1751{
1752	struct ctl_softc *ctl_softc;
1753	union ctl_ha_msg msg_info;
1754	struct ctl_lun *lun;
1755	int retval = 0;
1756	uint32_t targ_lun;
1757
1758	ctl_softc = control_softc;
1759
1760	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1761	lun = ctl_softc->ctl_luns[targ_lun];
1762	if (lun==NULL)
1763	{
1764		/*
1765		 * Why isn't LUN defined? The other side wouldn't
1766		 * send a cmd if the LUN is undefined.
1767		 */
1768		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1769
1770		/* "Logical unit not supported" */
1771		ctl_set_sense_data(&msg_info.scsi.sense_data,
1772				   lun,
1773				   /*sense_format*/SSD_TYPE_NONE,
1774				   /*current_error*/ 1,
1775				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1776				   /*asc*/ 0x25,
1777				   /*ascq*/ 0x00,
1778				   SSD_ELEM_NONE);
1779
1780		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1781		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1782		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1783		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1784		msg_info.hdr.serializing_sc = NULL;
1785		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1786	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1787				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1788		}
1789		return(1);
1790
1791	}
1792
1793	mtx_lock(&lun->lun_lock);
1794    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1795
1796	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1797		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1798		 ooa_links))) {
1799	case CTL_ACTION_BLOCK:
1800		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1801		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1802				  blocked_links);
1803		break;
1804	case CTL_ACTION_PASS:
1805	case CTL_ACTION_SKIP:
1806		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1807			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1808			ctl_enqueue_rtr((union ctl_io *)ctsio);
1809		} else {
1810
1811			/* send msg back to other side */
1812			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1813			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1814			msg_info.hdr.msg_type = CTL_MSG_R2R;
1815#if 0
1816			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1817#endif
1818		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1819			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1820			}
1821		}
1822		break;
1823	case CTL_ACTION_OVERLAP:
1824		/* OVERLAPPED COMMANDS ATTEMPTED */
1825		ctl_set_sense_data(&msg_info.scsi.sense_data,
1826				   lun,
1827				   /*sense_format*/SSD_TYPE_NONE,
1828				   /*current_error*/ 1,
1829				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1830				   /*asc*/ 0x4E,
1831				   /*ascq*/ 0x00,
1832				   SSD_ELEM_NONE);
1833
1834		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1835		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1836		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1837		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1838		msg_info.hdr.serializing_sc = NULL;
1839		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1840#if 0
1841		printf("BAD JUJU:Major Bummer Overlap\n");
1842#endif
1843		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1844		retval = 1;
1845		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1846		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1847		}
1848		break;
1849	case CTL_ACTION_OVERLAP_TAG:
1850		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1851		ctl_set_sense_data(&msg_info.scsi.sense_data,
1852				   lun,
1853				   /*sense_format*/SSD_TYPE_NONE,
1854				   /*current_error*/ 1,
1855				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1856				   /*asc*/ 0x4D,
1857				   /*ascq*/ ctsio->tag_num & 0xff,
1858				   SSD_ELEM_NONE);
1859
1860		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1861		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1862		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1863		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1864		msg_info.hdr.serializing_sc = NULL;
1865		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1866#if 0
1867		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1868#endif
1869		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1870		retval = 1;
1871		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1872		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1873		}
1874		break;
1875	case CTL_ACTION_ERROR:
1876	default:
1877		/* "Internal target failure" */
1878		ctl_set_sense_data(&msg_info.scsi.sense_data,
1879				   lun,
1880				   /*sense_format*/SSD_TYPE_NONE,
1881				   /*current_error*/ 1,
1882				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1883				   /*asc*/ 0x44,
1884				   /*ascq*/ 0x00,
1885				   SSD_ELEM_NONE);
1886
1887		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1888		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1889		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1890		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1891		msg_info.hdr.serializing_sc = NULL;
1892		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1893#if 0
1894		printf("BAD JUJU:Major Bummer HW Error\n");
1895#endif
1896		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1897		retval = 1;
1898		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1899		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1900		}
1901		break;
1902	}
1903	mtx_unlock(&lun->lun_lock);
1904	return (retval);
1905}
1906
1907static int
1908ctl_ioctl_submit_wait(union ctl_io *io)
1909{
1910	struct ctl_fe_ioctl_params params;
1911	ctl_fe_ioctl_state last_state;
1912	int done, retval;
1913
1914	retval = 0;
1915
1916	bzero(&params, sizeof(params));
1917
1918	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1919	cv_init(&params.sem, "ctlioccv");
1920	params.state = CTL_IOCTL_INPROG;
1921	last_state = params.state;
1922
1923	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1924
1925	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1926
1927	/* This shouldn't happen */
1928	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1929		return (retval);
1930
1931	done = 0;
1932
1933	do {
1934		mtx_lock(&params.ioctl_mtx);
1935		/*
1936		 * Check the state here, and don't sleep if the state has
1937		 * already changed (i.e. wakeup has already occured, but we
1938		 * weren't waiting yet).
1939		 */
1940		if (params.state == last_state) {
1941			/* XXX KDM cv_wait_sig instead? */
1942			cv_wait(&params.sem, &params.ioctl_mtx);
1943		}
1944		last_state = params.state;
1945
1946		switch (params.state) {
1947		case CTL_IOCTL_INPROG:
1948			/* Why did we wake up? */
1949			/* XXX KDM error here? */
1950			mtx_unlock(&params.ioctl_mtx);
1951			break;
1952		case CTL_IOCTL_DATAMOVE:
1953			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1954
1955			/*
1956			 * change last_state back to INPROG to avoid
1957			 * deadlock on subsequent data moves.
1958			 */
1959			params.state = last_state = CTL_IOCTL_INPROG;
1960
1961			mtx_unlock(&params.ioctl_mtx);
1962			ctl_ioctl_do_datamove(&io->scsiio);
1963			/*
1964			 * Note that in some cases, most notably writes,
1965			 * this will queue the I/O and call us back later.
1966			 * In other cases, generally reads, this routine
1967			 * will immediately call back and wake us up,
1968			 * probably using our own context.
1969			 */
1970			io->scsiio.be_move_done(io);
1971			break;
1972		case CTL_IOCTL_DONE:
1973			mtx_unlock(&params.ioctl_mtx);
1974			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1975			done = 1;
1976			break;
1977		default:
1978			mtx_unlock(&params.ioctl_mtx);
1979			/* XXX KDM error here? */
1980			break;
1981		}
1982	} while (done == 0);
1983
1984	mtx_destroy(&params.ioctl_mtx);
1985	cv_destroy(&params.sem);
1986
1987	return (CTL_RETVAL_COMPLETE);
1988}
1989
1990static void
1991ctl_ioctl_datamove(union ctl_io *io)
1992{
1993	struct ctl_fe_ioctl_params *params;
1994
1995	params = (struct ctl_fe_ioctl_params *)
1996		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1997
1998	mtx_lock(&params->ioctl_mtx);
1999	params->state = CTL_IOCTL_DATAMOVE;
2000	cv_broadcast(&params->sem);
2001	mtx_unlock(&params->ioctl_mtx);
2002}
2003
2004static void
2005ctl_ioctl_done(union ctl_io *io)
2006{
2007	struct ctl_fe_ioctl_params *params;
2008
2009	params = (struct ctl_fe_ioctl_params *)
2010		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2011
2012	mtx_lock(&params->ioctl_mtx);
2013	params->state = CTL_IOCTL_DONE;
2014	cv_broadcast(&params->sem);
2015	mtx_unlock(&params->ioctl_mtx);
2016}
2017
2018static void
2019ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2020{
2021	struct ctl_fe_ioctl_startstop_info *sd_info;
2022
2023	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2024
2025	sd_info->hs_info.status = metatask->status;
2026	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2027	sd_info->hs_info.luns_complete =
2028		metatask->taskinfo.startstop.luns_complete;
2029	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2030
2031	cv_broadcast(&sd_info->sem);
2032}
2033
2034static void
2035ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2036{
2037	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2038
2039	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2040
2041	mtx_lock(fe_bbr_info->lock);
2042	fe_bbr_info->bbr_info->status = metatask->status;
2043	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2044	fe_bbr_info->wakeup_done = 1;
2045	mtx_unlock(fe_bbr_info->lock);
2046
2047	cv_broadcast(&fe_bbr_info->sem);
2048}
2049
2050/*
2051 * Returns 0 for success, errno for failure.
2052 */
2053static int
2054ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2055		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2056{
2057	union ctl_io *io;
2058	int retval;
2059
2060	retval = 0;
2061
2062	mtx_lock(&lun->lun_lock);
2063	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2064	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2065	     ooa_links)) {
2066		struct ctl_ooa_entry *entry;
2067
2068		/*
2069		 * If we've got more than we can fit, just count the
2070		 * remaining entries.
2071		 */
2072		if (*cur_fill_num >= ooa_hdr->alloc_num)
2073			continue;
2074
2075		entry = &kern_entries[*cur_fill_num];
2076
2077		entry->tag_num = io->scsiio.tag_num;
2078		entry->lun_num = lun->lun;
2079#ifdef CTL_TIME_IO
2080		entry->start_bt = io->io_hdr.start_bt;
2081#endif
2082		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2083		entry->cdb_len = io->scsiio.cdb_len;
2084		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2085			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2086
2087		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2088			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2089
2090		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2091			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2092
2093		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2094			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2095
2096		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2097			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2098	}
2099	mtx_unlock(&lun->lun_lock);
2100
2101	return (retval);
2102}
2103
2104static void *
2105ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2106		 size_t error_str_len)
2107{
2108	void *kptr;
2109
2110	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2111
2112	if (copyin(user_addr, kptr, len) != 0) {
2113		snprintf(error_str, error_str_len, "Error copying %d bytes "
2114			 "from user address %p to kernel address %p", len,
2115			 user_addr, kptr);
2116		free(kptr, M_CTL);
2117		return (NULL);
2118	}
2119
2120	return (kptr);
2121}
2122
2123static void
2124ctl_free_args(int num_args, struct ctl_be_arg *args)
2125{
2126	int i;
2127
2128	if (args == NULL)
2129		return;
2130
2131	for (i = 0; i < num_args; i++) {
2132		free(args[i].kname, M_CTL);
2133		free(args[i].kvalue, M_CTL);
2134	}
2135
2136	free(args, M_CTL);
2137}
2138
2139static struct ctl_be_arg *
2140ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2141		char *error_str, size_t error_str_len)
2142{
2143	struct ctl_be_arg *args;
2144	int i;
2145
2146	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2147				error_str, error_str_len);
2148
2149	if (args == NULL)
2150		goto bailout;
2151
2152	for (i = 0; i < num_args; i++) {
2153		args[i].kname = NULL;
2154		args[i].kvalue = NULL;
2155	}
2156
2157	for (i = 0; i < num_args; i++) {
2158		uint8_t *tmpptr;
2159
2160		args[i].kname = ctl_copyin_alloc(args[i].name,
2161			args[i].namelen, error_str, error_str_len);
2162		if (args[i].kname == NULL)
2163			goto bailout;
2164
2165		if (args[i].kname[args[i].namelen - 1] != '\0') {
2166			snprintf(error_str, error_str_len, "Argument %d "
2167				 "name is not NUL-terminated", i);
2168			goto bailout;
2169		}
2170
2171		if (args[i].flags & CTL_BEARG_RD) {
2172			tmpptr = ctl_copyin_alloc(args[i].value,
2173				args[i].vallen, error_str, error_str_len);
2174			if (tmpptr == NULL)
2175				goto bailout;
2176			if ((args[i].flags & CTL_BEARG_ASCII)
2177			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2178				snprintf(error_str, error_str_len, "Argument "
2179				    "%d value is not NUL-terminated", i);
2180				goto bailout;
2181			}
2182			args[i].kvalue = tmpptr;
2183		} else {
2184			args[i].kvalue = malloc(args[i].vallen,
2185			    M_CTL, M_WAITOK | M_ZERO);
2186		}
2187	}
2188
2189	return (args);
2190bailout:
2191
2192	ctl_free_args(num_args, args);
2193
2194	return (NULL);
2195}
2196
2197static void
2198ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2199{
2200	int i;
2201
2202	for (i = 0; i < num_args; i++) {
2203		if (args[i].flags & CTL_BEARG_WR)
2204			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2205	}
2206}
2207
2208/*
2209 * Escape characters that are illegal or not recommended in XML.
2210 */
2211int
2212ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2213{
2214	int retval;
2215
2216	retval = 0;
2217
2218	for (; *str; str++) {
2219		switch (*str) {
2220		case '&':
2221			retval = sbuf_printf(sb, "&amp;");
2222			break;
2223		case '>':
2224			retval = sbuf_printf(sb, "&gt;");
2225			break;
2226		case '<':
2227			retval = sbuf_printf(sb, "&lt;");
2228			break;
2229		default:
2230			retval = sbuf_putc(sb, *str);
2231			break;
2232		}
2233
2234		if (retval != 0)
2235			break;
2236
2237	}
2238
2239	return (retval);
2240}
2241
2242static int
2243ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2244	  struct thread *td)
2245{
2246	struct ctl_softc *softc;
2247	int retval;
2248
2249	softc = control_softc;
2250
2251	retval = 0;
2252
2253	switch (cmd) {
2254	case CTL_IO: {
2255		union ctl_io *io;
2256		void *pool_tmp;
2257
2258		/*
2259		 * If we haven't been "enabled", don't allow any SCSI I/O
2260		 * to this FETD.
2261		 */
2262		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2263			retval = EPERM;
2264			break;
2265		}
2266
2267		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2268		if (io == NULL) {
2269			printf("ctl_ioctl: can't allocate ctl_io!\n");
2270			retval = ENOSPC;
2271			break;
2272		}
2273
2274		/*
2275		 * Need to save the pool reference so it doesn't get
2276		 * spammed by the user's ctl_io.
2277		 */
2278		pool_tmp = io->io_hdr.pool;
2279
2280		memcpy(io, (void *)addr, sizeof(*io));
2281
2282		io->io_hdr.pool = pool_tmp;
2283		/*
2284		 * No status yet, so make sure the status is set properly.
2285		 */
2286		io->io_hdr.status = CTL_STATUS_NONE;
2287
2288		/*
2289		 * The user sets the initiator ID, target and LUN IDs.
2290		 */
2291		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2292		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2293		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2294		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2295			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2296
2297		retval = ctl_ioctl_submit_wait(io);
2298
2299		if (retval != 0) {
2300			ctl_free_io(io);
2301			break;
2302		}
2303
2304		memcpy((void *)addr, io, sizeof(*io));
2305
2306		/* return this to our pool */
2307		ctl_free_io(io);
2308
2309		break;
2310	}
2311	case CTL_ENABLE_PORT:
2312	case CTL_DISABLE_PORT:
2313	case CTL_SET_PORT_WWNS: {
2314		struct ctl_port *port;
2315		struct ctl_port_entry *entry;
2316
2317		entry = (struct ctl_port_entry *)addr;
2318
2319		mtx_lock(&softc->ctl_lock);
2320		STAILQ_FOREACH(port, &softc->port_list, links) {
2321			int action, done;
2322
2323			action = 0;
2324			done = 0;
2325
2326			if ((entry->port_type == CTL_PORT_NONE)
2327			 && (entry->targ_port == port->targ_port)) {
2328				/*
2329				 * If the user only wants to enable or
2330				 * disable or set WWNs on a specific port,
2331				 * do the operation and we're done.
2332				 */
2333				action = 1;
2334				done = 1;
2335			} else if (entry->port_type & port->port_type) {
2336				/*
2337				 * Compare the user's type mask with the
2338				 * particular frontend type to see if we
2339				 * have a match.
2340				 */
2341				action = 1;
2342				done = 0;
2343
2344				/*
2345				 * Make sure the user isn't trying to set
2346				 * WWNs on multiple ports at the same time.
2347				 */
2348				if (cmd == CTL_SET_PORT_WWNS) {
2349					printf("%s: Can't set WWNs on "
2350					       "multiple ports\n", __func__);
2351					retval = EINVAL;
2352					break;
2353				}
2354			}
2355			if (action != 0) {
2356				/*
2357				 * XXX KDM we have to drop the lock here,
2358				 * because the online/offline operations
2359				 * can potentially block.  We need to
2360				 * reference count the frontends so they
2361				 * can't go away,
2362				 */
2363				mtx_unlock(&softc->ctl_lock);
2364
2365				if (cmd == CTL_ENABLE_PORT) {
2366					struct ctl_lun *lun;
2367
2368					STAILQ_FOREACH(lun, &softc->lun_list,
2369						       links) {
2370						port->lun_enable(port->targ_lun_arg,
2371						    lun->target,
2372						    lun->lun);
2373					}
2374
2375					ctl_port_online(port);
2376				} else if (cmd == CTL_DISABLE_PORT) {
2377					struct ctl_lun *lun;
2378
2379					ctl_port_offline(port);
2380
2381					STAILQ_FOREACH(lun, &softc->lun_list,
2382						       links) {
2383						port->lun_disable(
2384						    port->targ_lun_arg,
2385						    lun->target,
2386						    lun->lun);
2387					}
2388				}
2389
2390				mtx_lock(&softc->ctl_lock);
2391
2392				if (cmd == CTL_SET_PORT_WWNS)
2393					ctl_port_set_wwns(port,
2394					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2395					    1 : 0, entry->wwnn,
2396					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2397					    1 : 0, entry->wwpn);
2398			}
2399			if (done != 0)
2400				break;
2401		}
2402		mtx_unlock(&softc->ctl_lock);
2403		break;
2404	}
2405	case CTL_GET_PORT_LIST: {
2406		struct ctl_port *port;
2407		struct ctl_port_list *list;
2408		int i;
2409
2410		list = (struct ctl_port_list *)addr;
2411
2412		if (list->alloc_len != (list->alloc_num *
2413		    sizeof(struct ctl_port_entry))) {
2414			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2415			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2416			       "%zu\n", __func__, list->alloc_len,
2417			       list->alloc_num, sizeof(struct ctl_port_entry));
2418			retval = EINVAL;
2419			break;
2420		}
2421		list->fill_len = 0;
2422		list->fill_num = 0;
2423		list->dropped_num = 0;
2424		i = 0;
2425		mtx_lock(&softc->ctl_lock);
2426		STAILQ_FOREACH(port, &softc->port_list, links) {
2427			struct ctl_port_entry entry, *list_entry;
2428
2429			if (list->fill_num >= list->alloc_num) {
2430				list->dropped_num++;
2431				continue;
2432			}
2433
2434			entry.port_type = port->port_type;
2435			strlcpy(entry.port_name, port->port_name,
2436				sizeof(entry.port_name));
2437			entry.targ_port = port->targ_port;
2438			entry.physical_port = port->physical_port;
2439			entry.virtual_port = port->virtual_port;
2440			entry.wwnn = port->wwnn;
2441			entry.wwpn = port->wwpn;
2442			if (port->status & CTL_PORT_STATUS_ONLINE)
2443				entry.online = 1;
2444			else
2445				entry.online = 0;
2446
2447			list_entry = &list->entries[i];
2448
2449			retval = copyout(&entry, list_entry, sizeof(entry));
2450			if (retval != 0) {
2451				printf("%s: CTL_GET_PORT_LIST: copyout "
2452				       "returned %d\n", __func__, retval);
2453				break;
2454			}
2455			i++;
2456			list->fill_num++;
2457			list->fill_len += sizeof(entry);
2458		}
2459		mtx_unlock(&softc->ctl_lock);
2460
2461		/*
2462		 * If this is non-zero, we had a copyout fault, so there's
2463		 * probably no point in attempting to set the status inside
2464		 * the structure.
2465		 */
2466		if (retval != 0)
2467			break;
2468
2469		if (list->dropped_num > 0)
2470			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2471		else
2472			list->status = CTL_PORT_LIST_OK;
2473		break;
2474	}
2475	case CTL_DUMP_OOA: {
2476		struct ctl_lun *lun;
2477		union ctl_io *io;
2478		char printbuf[128];
2479		struct sbuf sb;
2480
2481		mtx_lock(&softc->ctl_lock);
2482		printf("Dumping OOA queues:\n");
2483		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2484			mtx_lock(&lun->lun_lock);
2485			for (io = (union ctl_io *)TAILQ_FIRST(
2486			     &lun->ooa_queue); io != NULL;
2487			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2488			     ooa_links)) {
2489				sbuf_new(&sb, printbuf, sizeof(printbuf),
2490					 SBUF_FIXEDLEN);
2491				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2492					    (intmax_t)lun->lun,
2493					    io->scsiio.tag_num,
2494					    (io->io_hdr.flags &
2495					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2496					    (io->io_hdr.flags &
2497					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2498					    (io->io_hdr.flags &
2499					    CTL_FLAG_ABORT) ? " ABORT" : "",
2500			                    (io->io_hdr.flags &
2501		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2502				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2503				sbuf_finish(&sb);
2504				printf("%s\n", sbuf_data(&sb));
2505			}
2506			mtx_unlock(&lun->lun_lock);
2507		}
2508		printf("OOA queues dump done\n");
2509		mtx_unlock(&softc->ctl_lock);
2510		break;
2511	}
2512	case CTL_GET_OOA: {
2513		struct ctl_lun *lun;
2514		struct ctl_ooa *ooa_hdr;
2515		struct ctl_ooa_entry *entries;
2516		uint32_t cur_fill_num;
2517
2518		ooa_hdr = (struct ctl_ooa *)addr;
2519
2520		if ((ooa_hdr->alloc_len == 0)
2521		 || (ooa_hdr->alloc_num == 0)) {
2522			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2523			       "must be non-zero\n", __func__,
2524			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2525			retval = EINVAL;
2526			break;
2527		}
2528
2529		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2530		    sizeof(struct ctl_ooa_entry))) {
2531			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2532			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2533			       __func__, ooa_hdr->alloc_len,
2534			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2535			retval = EINVAL;
2536			break;
2537		}
2538
2539		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2540		if (entries == NULL) {
2541			printf("%s: could not allocate %d bytes for OOA "
2542			       "dump\n", __func__, ooa_hdr->alloc_len);
2543			retval = ENOMEM;
2544			break;
2545		}
2546
2547		mtx_lock(&softc->ctl_lock);
2548		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2549		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2550		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2551			mtx_unlock(&softc->ctl_lock);
2552			free(entries, M_CTL);
2553			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2554			       __func__, (uintmax_t)ooa_hdr->lun_num);
2555			retval = EINVAL;
2556			break;
2557		}
2558
2559		cur_fill_num = 0;
2560
2561		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2562			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2563				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2564					ooa_hdr, entries);
2565				if (retval != 0)
2566					break;
2567			}
2568			if (retval != 0) {
2569				mtx_unlock(&softc->ctl_lock);
2570				free(entries, M_CTL);
2571				break;
2572			}
2573		} else {
2574			lun = softc->ctl_luns[ooa_hdr->lun_num];
2575
2576			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2577						    entries);
2578		}
2579		mtx_unlock(&softc->ctl_lock);
2580
2581		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2582		ooa_hdr->fill_len = ooa_hdr->fill_num *
2583			sizeof(struct ctl_ooa_entry);
2584		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2585		if (retval != 0) {
2586			printf("%s: error copying out %d bytes for OOA dump\n",
2587			       __func__, ooa_hdr->fill_len);
2588		}
2589
2590		getbintime(&ooa_hdr->cur_bt);
2591
2592		if (cur_fill_num > ooa_hdr->alloc_num) {
2593			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2594			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2595		} else {
2596			ooa_hdr->dropped_num = 0;
2597			ooa_hdr->status = CTL_OOA_OK;
2598		}
2599
2600		free(entries, M_CTL);
2601		break;
2602	}
2603	case CTL_CHECK_OOA: {
2604		union ctl_io *io;
2605		struct ctl_lun *lun;
2606		struct ctl_ooa_info *ooa_info;
2607
2608
2609		ooa_info = (struct ctl_ooa_info *)addr;
2610
2611		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2612			ooa_info->status = CTL_OOA_INVALID_LUN;
2613			break;
2614		}
2615		mtx_lock(&softc->ctl_lock);
2616		lun = softc->ctl_luns[ooa_info->lun_id];
2617		if (lun == NULL) {
2618			mtx_unlock(&softc->ctl_lock);
2619			ooa_info->status = CTL_OOA_INVALID_LUN;
2620			break;
2621		}
2622		mtx_lock(&lun->lun_lock);
2623		mtx_unlock(&softc->ctl_lock);
2624		ooa_info->num_entries = 0;
2625		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2626		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2627		     &io->io_hdr, ooa_links)) {
2628			ooa_info->num_entries++;
2629		}
2630		mtx_unlock(&lun->lun_lock);
2631
2632		ooa_info->status = CTL_OOA_SUCCESS;
2633
2634		break;
2635	}
2636	case CTL_HARD_START:
2637	case CTL_HARD_STOP: {
2638		struct ctl_fe_ioctl_startstop_info ss_info;
2639		struct cfi_metatask *metatask;
2640		struct mtx hs_mtx;
2641
2642		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2643
2644		cv_init(&ss_info.sem, "hard start/stop cv" );
2645
2646		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2647		if (metatask == NULL) {
2648			retval = ENOMEM;
2649			mtx_destroy(&hs_mtx);
2650			break;
2651		}
2652
2653		if (cmd == CTL_HARD_START)
2654			metatask->tasktype = CFI_TASK_STARTUP;
2655		else
2656			metatask->tasktype = CFI_TASK_SHUTDOWN;
2657
2658		metatask->callback = ctl_ioctl_hard_startstop_callback;
2659		metatask->callback_arg = &ss_info;
2660
2661		cfi_action(metatask);
2662
2663		/* Wait for the callback */
2664		mtx_lock(&hs_mtx);
2665		cv_wait_sig(&ss_info.sem, &hs_mtx);
2666		mtx_unlock(&hs_mtx);
2667
2668		/*
2669		 * All information has been copied from the metatask by the
2670		 * time cv_broadcast() is called, so we free the metatask here.
2671		 */
2672		cfi_free_metatask(metatask);
2673
2674		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2675
2676		mtx_destroy(&hs_mtx);
2677		break;
2678	}
2679	case CTL_BBRREAD: {
2680		struct ctl_bbrread_info *bbr_info;
2681		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2682		struct mtx bbr_mtx;
2683		struct cfi_metatask *metatask;
2684
2685		bbr_info = (struct ctl_bbrread_info *)addr;
2686
2687		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2688
2689		bzero(&bbr_mtx, sizeof(bbr_mtx));
2690		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2691
2692		fe_bbr_info.bbr_info = bbr_info;
2693		fe_bbr_info.lock = &bbr_mtx;
2694
2695		cv_init(&fe_bbr_info.sem, "BBR read cv");
2696		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2697
2698		if (metatask == NULL) {
2699			mtx_destroy(&bbr_mtx);
2700			cv_destroy(&fe_bbr_info.sem);
2701			retval = ENOMEM;
2702			break;
2703		}
2704		metatask->tasktype = CFI_TASK_BBRREAD;
2705		metatask->callback = ctl_ioctl_bbrread_callback;
2706		metatask->callback_arg = &fe_bbr_info;
2707		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2708		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2709		metatask->taskinfo.bbrread.len = bbr_info->len;
2710
2711		cfi_action(metatask);
2712
2713		mtx_lock(&bbr_mtx);
2714		while (fe_bbr_info.wakeup_done == 0)
2715			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2716		mtx_unlock(&bbr_mtx);
2717
2718		bbr_info->status = metatask->status;
2719		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2720		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2721		memcpy(&bbr_info->sense_data,
2722		       &metatask->taskinfo.bbrread.sense_data,
2723		       ctl_min(sizeof(bbr_info->sense_data),
2724			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2725
2726		cfi_free_metatask(metatask);
2727
2728		mtx_destroy(&bbr_mtx);
2729		cv_destroy(&fe_bbr_info.sem);
2730
2731		break;
2732	}
2733	case CTL_DELAY_IO: {
2734		struct ctl_io_delay_info *delay_info;
2735#ifdef CTL_IO_DELAY
2736		struct ctl_lun *lun;
2737#endif /* CTL_IO_DELAY */
2738
2739		delay_info = (struct ctl_io_delay_info *)addr;
2740
2741#ifdef CTL_IO_DELAY
2742		mtx_lock(&softc->ctl_lock);
2743
2744		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2745		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2746			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2747		} else {
2748			lun = softc->ctl_luns[delay_info->lun_id];
2749			mtx_lock(&lun->lun_lock);
2750
2751			delay_info->status = CTL_DELAY_STATUS_OK;
2752
2753			switch (delay_info->delay_type) {
2754			case CTL_DELAY_TYPE_CONT:
2755				break;
2756			case CTL_DELAY_TYPE_ONESHOT:
2757				break;
2758			default:
2759				delay_info->status =
2760					CTL_DELAY_STATUS_INVALID_TYPE;
2761				break;
2762			}
2763
2764			switch (delay_info->delay_loc) {
2765			case CTL_DELAY_LOC_DATAMOVE:
2766				lun->delay_info.datamove_type =
2767					delay_info->delay_type;
2768				lun->delay_info.datamove_delay =
2769					delay_info->delay_secs;
2770				break;
2771			case CTL_DELAY_LOC_DONE:
2772				lun->delay_info.done_type =
2773					delay_info->delay_type;
2774				lun->delay_info.done_delay =
2775					delay_info->delay_secs;
2776				break;
2777			default:
2778				delay_info->status =
2779					CTL_DELAY_STATUS_INVALID_LOC;
2780				break;
2781			}
2782			mtx_unlock(&lun->lun_lock);
2783		}
2784
2785		mtx_unlock(&softc->ctl_lock);
2786#else
2787		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2788#endif /* CTL_IO_DELAY */
2789		break;
2790	}
2791	case CTL_REALSYNC_SET: {
2792		int *syncstate;
2793
2794		syncstate = (int *)addr;
2795
2796		mtx_lock(&softc->ctl_lock);
2797		switch (*syncstate) {
2798		case 0:
2799			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2800			break;
2801		case 1:
2802			softc->flags |= CTL_FLAG_REAL_SYNC;
2803			break;
2804		default:
2805			retval = EINVAL;
2806			break;
2807		}
2808		mtx_unlock(&softc->ctl_lock);
2809		break;
2810	}
2811	case CTL_REALSYNC_GET: {
2812		int *syncstate;
2813
2814		syncstate = (int*)addr;
2815
2816		mtx_lock(&softc->ctl_lock);
2817		if (softc->flags & CTL_FLAG_REAL_SYNC)
2818			*syncstate = 1;
2819		else
2820			*syncstate = 0;
2821		mtx_unlock(&softc->ctl_lock);
2822
2823		break;
2824	}
2825	case CTL_SETSYNC:
2826	case CTL_GETSYNC: {
2827		struct ctl_sync_info *sync_info;
2828		struct ctl_lun *lun;
2829
2830		sync_info = (struct ctl_sync_info *)addr;
2831
2832		mtx_lock(&softc->ctl_lock);
2833		lun = softc->ctl_luns[sync_info->lun_id];
2834		if (lun == NULL) {
2835			mtx_unlock(&softc->ctl_lock);
2836			sync_info->status = CTL_GS_SYNC_NO_LUN;
2837		}
2838		/*
2839		 * Get or set the sync interval.  We're not bounds checking
2840		 * in the set case, hopefully the user won't do something
2841		 * silly.
2842		 */
2843		mtx_lock(&lun->lun_lock);
2844		mtx_unlock(&softc->ctl_lock);
2845		if (cmd == CTL_GETSYNC)
2846			sync_info->sync_interval = lun->sync_interval;
2847		else
2848			lun->sync_interval = sync_info->sync_interval;
2849		mtx_unlock(&lun->lun_lock);
2850
2851		sync_info->status = CTL_GS_SYNC_OK;
2852
2853		break;
2854	}
2855	case CTL_GETSTATS: {
2856		struct ctl_stats *stats;
2857		struct ctl_lun *lun;
2858		int i;
2859
2860		stats = (struct ctl_stats *)addr;
2861
2862		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2863		     stats->alloc_len) {
2864			stats->status = CTL_SS_NEED_MORE_SPACE;
2865			stats->num_luns = softc->num_luns;
2866			break;
2867		}
2868		/*
2869		 * XXX KDM no locking here.  If the LUN list changes,
2870		 * things can blow up.
2871		 */
2872		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2873		     i++, lun = STAILQ_NEXT(lun, links)) {
2874			retval = copyout(&lun->stats, &stats->lun_stats[i],
2875					 sizeof(lun->stats));
2876			if (retval != 0)
2877				break;
2878		}
2879		stats->num_luns = softc->num_luns;
2880		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2881				 softc->num_luns;
2882		stats->status = CTL_SS_OK;
2883#ifdef CTL_TIME_IO
2884		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2885#else
2886		stats->flags = CTL_STATS_FLAG_NONE;
2887#endif
2888		getnanouptime(&stats->timestamp);
2889		break;
2890	}
2891	case CTL_ERROR_INJECT: {
2892		struct ctl_error_desc *err_desc, *new_err_desc;
2893		struct ctl_lun *lun;
2894
2895		err_desc = (struct ctl_error_desc *)addr;
2896
2897		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2898				      M_WAITOK | M_ZERO);
2899		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2900
2901		mtx_lock(&softc->ctl_lock);
2902		lun = softc->ctl_luns[err_desc->lun_id];
2903		if (lun == NULL) {
2904			mtx_unlock(&softc->ctl_lock);
2905			free(new_err_desc, M_CTL);
2906			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2907			       __func__, (uintmax_t)err_desc->lun_id);
2908			retval = EINVAL;
2909			break;
2910		}
2911		mtx_lock(&lun->lun_lock);
2912		mtx_unlock(&softc->ctl_lock);
2913
2914		/*
2915		 * We could do some checking here to verify the validity
2916		 * of the request, but given the complexity of error
2917		 * injection requests, the checking logic would be fairly
2918		 * complex.
2919		 *
2920		 * For now, if the request is invalid, it just won't get
2921		 * executed and might get deleted.
2922		 */
2923		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2924
2925		/*
2926		 * XXX KDM check to make sure the serial number is unique,
2927		 * in case we somehow manage to wrap.  That shouldn't
2928		 * happen for a very long time, but it's the right thing to
2929		 * do.
2930		 */
2931		new_err_desc->serial = lun->error_serial;
2932		err_desc->serial = lun->error_serial;
2933		lun->error_serial++;
2934
2935		mtx_unlock(&lun->lun_lock);
2936		break;
2937	}
2938	case CTL_ERROR_INJECT_DELETE: {
2939		struct ctl_error_desc *delete_desc, *desc, *desc2;
2940		struct ctl_lun *lun;
2941		int delete_done;
2942
2943		delete_desc = (struct ctl_error_desc *)addr;
2944		delete_done = 0;
2945
2946		mtx_lock(&softc->ctl_lock);
2947		lun = softc->ctl_luns[delete_desc->lun_id];
2948		if (lun == NULL) {
2949			mtx_unlock(&softc->ctl_lock);
2950			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2951			       __func__, (uintmax_t)delete_desc->lun_id);
2952			retval = EINVAL;
2953			break;
2954		}
2955		mtx_lock(&lun->lun_lock);
2956		mtx_unlock(&softc->ctl_lock);
2957		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2958			if (desc->serial != delete_desc->serial)
2959				continue;
2960
2961			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2962				      links);
2963			free(desc, M_CTL);
2964			delete_done = 1;
2965		}
2966		mtx_unlock(&lun->lun_lock);
2967		if (delete_done == 0) {
2968			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2969			       "error serial %ju on LUN %u\n", __func__,
2970			       delete_desc->serial, delete_desc->lun_id);
2971			retval = EINVAL;
2972			break;
2973		}
2974		break;
2975	}
2976	case CTL_DUMP_STRUCTS: {
2977		int i, j, k, idx;
2978		struct ctl_port *port;
2979		struct ctl_frontend *fe;
2980
2981		mtx_lock(&softc->ctl_lock);
2982		printf("CTL Persistent Reservation information start:\n");
2983		for (i = 0; i < CTL_MAX_LUNS; i++) {
2984			struct ctl_lun *lun;
2985
2986			lun = softc->ctl_luns[i];
2987
2988			if ((lun == NULL)
2989			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2990				continue;
2991
2992			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2993				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2994					idx = j * CTL_MAX_INIT_PER_PORT + k;
2995					if (lun->per_res[idx].registered == 0)
2996						continue;
2997					printf("  LUN %d port %d iid %d key "
2998					       "%#jx\n", i, j, k,
2999					       (uintmax_t)scsi_8btou64(
3000					       lun->per_res[idx].res_key.key));
3001				}
3002			}
3003		}
3004		printf("CTL Persistent Reservation information end\n");
3005		printf("CTL Ports:\n");
3006		STAILQ_FOREACH(port, &softc->port_list, links) {
3007			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3008			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3009			       port->frontend->name, port->port_type,
3010			       port->physical_port, port->virtual_port,
3011			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3012			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3013				if (port->wwpn_iid[j].in_use == 0 &&
3014				    port->wwpn_iid[j].wwpn == 0 &&
3015				    port->wwpn_iid[j].name == NULL)
3016					continue;
3017
3018				printf("    iid %u use %d WWPN %#jx '%s'\n",
3019				    j, port->wwpn_iid[j].in_use,
3020				    (uintmax_t)port->wwpn_iid[j].wwpn,
3021				    port->wwpn_iid[j].name);
3022			}
3023		}
3024		printf("CTL Port information end\n");
3025		mtx_unlock(&softc->ctl_lock);
3026		/*
3027		 * XXX KDM calling this without a lock.  We'd likely want
3028		 * to drop the lock before calling the frontend's dump
3029		 * routine anyway.
3030		 */
3031		printf("CTL Frontends:\n");
3032		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3033			printf("  Frontend '%s'\n", fe->name);
3034			if (fe->fe_dump != NULL)
3035				fe->fe_dump();
3036		}
3037		printf("CTL Frontend information end\n");
3038		break;
3039	}
3040	case CTL_LUN_REQ: {
3041		struct ctl_lun_req *lun_req;
3042		struct ctl_backend_driver *backend;
3043
3044		lun_req = (struct ctl_lun_req *)addr;
3045
3046		backend = ctl_backend_find(lun_req->backend);
3047		if (backend == NULL) {
3048			lun_req->status = CTL_LUN_ERROR;
3049			snprintf(lun_req->error_str,
3050				 sizeof(lun_req->error_str),
3051				 "Backend \"%s\" not found.",
3052				 lun_req->backend);
3053			break;
3054		}
3055		if (lun_req->num_be_args > 0) {
3056			lun_req->kern_be_args = ctl_copyin_args(
3057				lun_req->num_be_args,
3058				lun_req->be_args,
3059				lun_req->error_str,
3060				sizeof(lun_req->error_str));
3061			if (lun_req->kern_be_args == NULL) {
3062				lun_req->status = CTL_LUN_ERROR;
3063				break;
3064			}
3065		}
3066
3067		retval = backend->ioctl(dev, cmd, addr, flag, td);
3068
3069		if (lun_req->num_be_args > 0) {
3070			ctl_copyout_args(lun_req->num_be_args,
3071				      lun_req->kern_be_args);
3072			ctl_free_args(lun_req->num_be_args,
3073				      lun_req->kern_be_args);
3074		}
3075		break;
3076	}
3077	case CTL_LUN_LIST: {
3078		struct sbuf *sb;
3079		struct ctl_lun *lun;
3080		struct ctl_lun_list *list;
3081		struct ctl_option *opt;
3082
3083		list = (struct ctl_lun_list *)addr;
3084
3085		/*
3086		 * Allocate a fixed length sbuf here, based on the length
3087		 * of the user's buffer.  We could allocate an auto-extending
3088		 * buffer, and then tell the user how much larger our
3089		 * amount of data is than his buffer, but that presents
3090		 * some problems:
3091		 *
3092		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3093		 *     we can't hold a lock while calling them with an
3094		 *     auto-extending buffer.
3095 		 *
3096		 * 2.  There is not currently a LUN reference counting
3097		 *     mechanism, outside of outstanding transactions on
3098		 *     the LUN's OOA queue.  So a LUN could go away on us
3099		 *     while we're getting the LUN number, backend-specific
3100		 *     information, etc.  Thus, given the way things
3101		 *     currently work, we need to hold the CTL lock while
3102		 *     grabbing LUN information.
3103		 *
3104		 * So, from the user's standpoint, the best thing to do is
3105		 * allocate what he thinks is a reasonable buffer length,
3106		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3107		 * double the buffer length and try again.  (And repeat
3108		 * that until he succeeds.)
3109		 */
3110		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3111		if (sb == NULL) {
3112			list->status = CTL_LUN_LIST_ERROR;
3113			snprintf(list->error_str, sizeof(list->error_str),
3114				 "Unable to allocate %d bytes for LUN list",
3115				 list->alloc_len);
3116			break;
3117		}
3118
3119		sbuf_printf(sb, "<ctllunlist>\n");
3120
3121		mtx_lock(&softc->ctl_lock);
3122		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3123			mtx_lock(&lun->lun_lock);
3124			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3125					     (uintmax_t)lun->lun);
3126
3127			/*
3128			 * Bail out as soon as we see that we've overfilled
3129			 * the buffer.
3130			 */
3131			if (retval != 0)
3132				break;
3133
3134			retval = sbuf_printf(sb, "\t<backend_type>%s"
3135					     "</backend_type>\n",
3136					     (lun->backend == NULL) ?  "none" :
3137					     lun->backend->name);
3138
3139			if (retval != 0)
3140				break;
3141
3142			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3143					     lun->be_lun->lun_type);
3144
3145			if (retval != 0)
3146				break;
3147
3148			if (lun->backend == NULL) {
3149				retval = sbuf_printf(sb, "</lun>\n");
3150				if (retval != 0)
3151					break;
3152				continue;
3153			}
3154
3155			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3156					     (lun->be_lun->maxlba > 0) ?
3157					     lun->be_lun->maxlba + 1 : 0);
3158
3159			if (retval != 0)
3160				break;
3161
3162			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3163					     lun->be_lun->blocksize);
3164
3165			if (retval != 0)
3166				break;
3167
3168			retval = sbuf_printf(sb, "\t<serial_number>");
3169
3170			if (retval != 0)
3171				break;
3172
3173			retval = ctl_sbuf_printf_esc(sb,
3174						     lun->be_lun->serial_num);
3175
3176			if (retval != 0)
3177				break;
3178
3179			retval = sbuf_printf(sb, "</serial_number>\n");
3180
3181			if (retval != 0)
3182				break;
3183
3184			retval = sbuf_printf(sb, "\t<device_id>");
3185
3186			if (retval != 0)
3187				break;
3188
3189			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3190
3191			if (retval != 0)
3192				break;
3193
3194			retval = sbuf_printf(sb, "</device_id>\n");
3195
3196			if (retval != 0)
3197				break;
3198
3199			if (lun->backend->lun_info != NULL) {
3200				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3201				if (retval != 0)
3202					break;
3203			}
3204			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3205				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3206				    opt->name, opt->value, opt->name);
3207				if (retval != 0)
3208					break;
3209			}
3210
3211			retval = sbuf_printf(sb, "</lun>\n");
3212
3213			if (retval != 0)
3214				break;
3215			mtx_unlock(&lun->lun_lock);
3216		}
3217		if (lun != NULL)
3218			mtx_unlock(&lun->lun_lock);
3219		mtx_unlock(&softc->ctl_lock);
3220
3221		if ((retval != 0)
3222		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3223			retval = 0;
3224			sbuf_delete(sb);
3225			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3226			snprintf(list->error_str, sizeof(list->error_str),
3227				 "Out of space, %d bytes is too small",
3228				 list->alloc_len);
3229			break;
3230		}
3231
3232		sbuf_finish(sb);
3233
3234		retval = copyout(sbuf_data(sb), list->lun_xml,
3235				 sbuf_len(sb) + 1);
3236
3237		list->fill_len = sbuf_len(sb) + 1;
3238		list->status = CTL_LUN_LIST_OK;
3239		sbuf_delete(sb);
3240		break;
3241	}
3242	case CTL_ISCSI: {
3243		struct ctl_iscsi *ci;
3244		struct ctl_frontend *fe;
3245
3246		ci = (struct ctl_iscsi *)addr;
3247
3248		fe = ctl_frontend_find("iscsi");
3249		if (fe == NULL) {
3250			ci->status = CTL_ISCSI_ERROR;
3251			snprintf(ci->error_str, sizeof(ci->error_str),
3252			    "Frontend \"iscsi\" not found.");
3253			break;
3254		}
3255
3256		retval = fe->ioctl(dev, cmd, addr, flag, td);
3257		break;
3258	}
3259	case CTL_PORT_REQ: {
3260		struct ctl_req *req;
3261		struct ctl_frontend *fe;
3262
3263		req = (struct ctl_req *)addr;
3264
3265		fe = ctl_frontend_find(req->driver);
3266		if (fe == NULL) {
3267			req->status = CTL_LUN_ERROR;
3268			snprintf(req->error_str, sizeof(req->error_str),
3269			    "Frontend \"%s\" not found.", req->driver);
3270			break;
3271		}
3272		if (req->num_args > 0) {
3273			req->kern_args = ctl_copyin_args(req->num_args,
3274			    req->args, req->error_str, sizeof(req->error_str));
3275			if (req->kern_args == NULL) {
3276				req->status = CTL_LUN_ERROR;
3277				break;
3278			}
3279		}
3280
3281		retval = fe->ioctl(dev, cmd, addr, flag, td);
3282
3283		if (req->num_args > 0) {
3284			ctl_copyout_args(req->num_args, req->kern_args);
3285			ctl_free_args(req->num_args, req->kern_args);
3286		}
3287		break;
3288	}
3289	case CTL_PORT_LIST: {
3290		struct sbuf *sb;
3291		struct ctl_port *port;
3292		struct ctl_lun_list *list;
3293		struct ctl_option *opt;
3294
3295		list = (struct ctl_lun_list *)addr;
3296
3297		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3298		if (sb == NULL) {
3299			list->status = CTL_LUN_LIST_ERROR;
3300			snprintf(list->error_str, sizeof(list->error_str),
3301				 "Unable to allocate %d bytes for LUN list",
3302				 list->alloc_len);
3303			break;
3304		}
3305
3306		sbuf_printf(sb, "<ctlportlist>\n");
3307
3308		mtx_lock(&softc->ctl_lock);
3309		STAILQ_FOREACH(port, &softc->port_list, links) {
3310			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3311					     (uintmax_t)port->targ_port);
3312
3313			/*
3314			 * Bail out as soon as we see that we've overfilled
3315			 * the buffer.
3316			 */
3317			if (retval != 0)
3318				break;
3319
3320			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3321			    "</frontend_type>\n", port->frontend->name);
3322			if (retval != 0)
3323				break;
3324
3325			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3326					     port->port_type);
3327			if (retval != 0)
3328				break;
3329
3330			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3331			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3332			if (retval != 0)
3333				break;
3334
3335			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3336			    port->port_name);
3337			if (retval != 0)
3338				break;
3339
3340			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3341			    port->physical_port);
3342			if (retval != 0)
3343				break;
3344
3345			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3346			    port->virtual_port);
3347			if (retval != 0)
3348				break;
3349
3350			retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3351			    (uintmax_t)port->wwnn);
3352			if (retval != 0)
3353				break;
3354
3355			retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3356			    (uintmax_t)port->wwpn);
3357			if (retval != 0)
3358				break;
3359
3360			if (port->port_info != NULL) {
3361				retval = port->port_info(port->onoff_arg, sb);
3362				if (retval != 0)
3363					break;
3364			}
3365			STAILQ_FOREACH(opt, &port->options, links) {
3366				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3367				    opt->name, opt->value, opt->name);
3368				if (retval != 0)
3369					break;
3370			}
3371
3372			retval = sbuf_printf(sb, "</targ_port>\n");
3373			if (retval != 0)
3374				break;
3375		}
3376		mtx_unlock(&softc->ctl_lock);
3377
3378		if ((retval != 0)
3379		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3380			retval = 0;
3381			sbuf_delete(sb);
3382			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3383			snprintf(list->error_str, sizeof(list->error_str),
3384				 "Out of space, %d bytes is too small",
3385				 list->alloc_len);
3386			break;
3387		}
3388
3389		sbuf_finish(sb);
3390
3391		retval = copyout(sbuf_data(sb), list->lun_xml,
3392				 sbuf_len(sb) + 1);
3393
3394		list->fill_len = sbuf_len(sb) + 1;
3395		list->status = CTL_LUN_LIST_OK;
3396		sbuf_delete(sb);
3397		break;
3398	}
3399	default: {
3400		/* XXX KDM should we fix this? */
3401#if 0
3402		struct ctl_backend_driver *backend;
3403		unsigned int type;
3404		int found;
3405
3406		found = 0;
3407
3408		/*
3409		 * We encode the backend type as the ioctl type for backend
3410		 * ioctls.  So parse it out here, and then search for a
3411		 * backend of this type.
3412		 */
3413		type = _IOC_TYPE(cmd);
3414
3415		STAILQ_FOREACH(backend, &softc->be_list, links) {
3416			if (backend->type == type) {
3417				found = 1;
3418				break;
3419			}
3420		}
3421		if (found == 0) {
3422			printf("ctl: unknown ioctl command %#lx or backend "
3423			       "%d\n", cmd, type);
3424			retval = EINVAL;
3425			break;
3426		}
3427		retval = backend->ioctl(dev, cmd, addr, flag, td);
3428#endif
3429		retval = ENOTTY;
3430		break;
3431	}
3432	}
3433	return (retval);
3434}
3435
3436uint32_t
3437ctl_get_initindex(struct ctl_nexus *nexus)
3438{
3439	if (nexus->targ_port < CTL_MAX_PORTS)
3440		return (nexus->initid.id +
3441			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3442	else
3443		return (nexus->initid.id +
3444		       ((nexus->targ_port - CTL_MAX_PORTS) *
3445			CTL_MAX_INIT_PER_PORT));
3446}
3447
3448uint32_t
3449ctl_get_resindex(struct ctl_nexus *nexus)
3450{
3451	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3452}
3453
3454uint32_t
3455ctl_port_idx(int port_num)
3456{
3457	if (port_num < CTL_MAX_PORTS)
3458		return(port_num);
3459	else
3460		return(port_num - CTL_MAX_PORTS);
3461}
3462
3463static uint32_t
3464ctl_map_lun(int port_num, uint32_t lun_id)
3465{
3466	struct ctl_port *port;
3467
3468	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3469	if (port == NULL)
3470		return (UINT32_MAX);
3471	if (port->lun_map == NULL)
3472		return (lun_id);
3473	return (port->lun_map(port->targ_lun_arg, lun_id));
3474}
3475
3476static uint32_t
3477ctl_map_lun_back(int port_num, uint32_t lun_id)
3478{
3479	struct ctl_port *port;
3480	uint32_t i;
3481
3482	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3483	if (port->lun_map == NULL)
3484		return (lun_id);
3485	for (i = 0; i < CTL_MAX_LUNS; i++) {
3486		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3487			return (i);
3488	}
3489	return (UINT32_MAX);
3490}
3491
3492/*
3493 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3494 * that are a power of 2.
3495 */
3496int
3497ctl_ffz(uint32_t *mask, uint32_t size)
3498{
3499	uint32_t num_chunks, num_pieces;
3500	int i, j;
3501
3502	num_chunks = (size >> 5);
3503	if (num_chunks == 0)
3504		num_chunks++;
3505	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3506
3507	for (i = 0; i < num_chunks; i++) {
3508		for (j = 0; j < num_pieces; j++) {
3509			if ((mask[i] & (1 << j)) == 0)
3510				return ((i << 5) + j);
3511		}
3512	}
3513
3514	return (-1);
3515}
3516
3517int
3518ctl_set_mask(uint32_t *mask, uint32_t bit)
3519{
3520	uint32_t chunk, piece;
3521
3522	chunk = bit >> 5;
3523	piece = bit % (sizeof(uint32_t) * 8);
3524
3525	if ((mask[chunk] & (1 << piece)) != 0)
3526		return (-1);
3527	else
3528		mask[chunk] |= (1 << piece);
3529
3530	return (0);
3531}
3532
3533int
3534ctl_clear_mask(uint32_t *mask, uint32_t bit)
3535{
3536	uint32_t chunk, piece;
3537
3538	chunk = bit >> 5;
3539	piece = bit % (sizeof(uint32_t) * 8);
3540
3541	if ((mask[chunk] & (1 << piece)) == 0)
3542		return (-1);
3543	else
3544		mask[chunk] &= ~(1 << piece);
3545
3546	return (0);
3547}
3548
3549int
3550ctl_is_set(uint32_t *mask, uint32_t bit)
3551{
3552	uint32_t chunk, piece;
3553
3554	chunk = bit >> 5;
3555	piece = bit % (sizeof(uint32_t) * 8);
3556
3557	if ((mask[chunk] & (1 << piece)) == 0)
3558		return (0);
3559	else
3560		return (1);
3561}
3562
3563#ifdef unused
3564/*
3565 * The bus, target and lun are optional, they can be filled in later.
3566 * can_wait is used to determine whether we can wait on the malloc or not.
3567 */
3568union ctl_io*
3569ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3570	      uint32_t targ_lun, int can_wait)
3571{
3572	union ctl_io *io;
3573
3574	if (can_wait)
3575		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3576	else
3577		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3578
3579	if (io != NULL) {
3580		io->io_hdr.io_type = io_type;
3581		io->io_hdr.targ_port = targ_port;
3582		/*
3583		 * XXX KDM this needs to change/go away.  We need to move
3584		 * to a preallocated pool of ctl_scsiio structures.
3585		 */
3586		io->io_hdr.nexus.targ_target.id = targ_target;
3587		io->io_hdr.nexus.targ_lun = targ_lun;
3588	}
3589
3590	return (io);
3591}
3592
3593void
3594ctl_kfree_io(union ctl_io *io)
3595{
3596	free(io, M_CTL);
3597}
3598#endif /* unused */
3599
3600/*
3601 * ctl_softc, pool_type, total_ctl_io are passed in.
3602 * npool is passed out.
3603 */
3604int
3605ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3606		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3607{
3608	uint32_t i;
3609	union ctl_io *cur_io, *next_io;
3610	struct ctl_io_pool *pool;
3611	int retval;
3612
3613	retval = 0;
3614
3615	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3616					    M_NOWAIT | M_ZERO);
3617	if (pool == NULL) {
3618		retval = ENOMEM;
3619		goto bailout;
3620	}
3621
3622	pool->type = pool_type;
3623	pool->ctl_softc = ctl_softc;
3624
3625	mtx_lock(&ctl_softc->pool_lock);
3626	pool->id = ctl_softc->cur_pool_id++;
3627	mtx_unlock(&ctl_softc->pool_lock);
3628
3629	pool->flags = CTL_POOL_FLAG_NONE;
3630	pool->refcount = 1;		/* Reference for validity. */
3631	STAILQ_INIT(&pool->free_queue);
3632
3633	/*
3634	 * XXX KDM other options here:
3635	 * - allocate a page at a time
3636	 * - allocate one big chunk of memory.
3637	 * Page allocation might work well, but would take a little more
3638	 * tracking.
3639	 */
3640	for (i = 0; i < total_ctl_io; i++) {
3641		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3642						M_NOWAIT);
3643		if (cur_io == NULL) {
3644			retval = ENOMEM;
3645			break;
3646		}
3647		cur_io->io_hdr.pool = pool;
3648		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3649		pool->total_ctl_io++;
3650		pool->free_ctl_io++;
3651	}
3652
3653	if (retval != 0) {
3654		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3655		     cur_io != NULL; cur_io = next_io) {
3656			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3657							      links);
3658			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3659				      ctl_io_hdr, links);
3660			free(cur_io, M_CTLIO);
3661		}
3662
3663		free(pool, M_CTL);
3664		goto bailout;
3665	}
3666	mtx_lock(&ctl_softc->pool_lock);
3667	ctl_softc->num_pools++;
3668	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3669	/*
3670	 * Increment our usage count if this is an external consumer, so we
3671	 * can't get unloaded until the external consumer (most likely a
3672	 * FETD) unloads and frees his pool.
3673	 *
3674	 * XXX KDM will this increment the caller's module use count, or
3675	 * mine?
3676	 */
3677#if 0
3678	if ((pool_type != CTL_POOL_EMERGENCY)
3679	 && (pool_type != CTL_POOL_INTERNAL)
3680	 && (pool_type != CTL_POOL_4OTHERSC))
3681		MOD_INC_USE_COUNT;
3682#endif
3683
3684	mtx_unlock(&ctl_softc->pool_lock);
3685
3686	*npool = pool;
3687
3688bailout:
3689
3690	return (retval);
3691}
3692
3693static int
3694ctl_pool_acquire(struct ctl_io_pool *pool)
3695{
3696
3697	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3698
3699	if (pool->flags & CTL_POOL_FLAG_INVALID)
3700		return (EINVAL);
3701
3702	pool->refcount++;
3703
3704	return (0);
3705}
3706
3707static void
3708ctl_pool_release(struct ctl_io_pool *pool)
3709{
3710	struct ctl_softc *ctl_softc = pool->ctl_softc;
3711	union ctl_io *io;
3712
3713	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3714
3715	if (--pool->refcount != 0)
3716		return;
3717
3718	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3719		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3720			      links);
3721		free(io, M_CTLIO);
3722	}
3723
3724	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3725	ctl_softc->num_pools--;
3726
3727	/*
3728	 * XXX KDM will this decrement the caller's usage count or mine?
3729	 */
3730#if 0
3731	if ((pool->type != CTL_POOL_EMERGENCY)
3732	 && (pool->type != CTL_POOL_INTERNAL)
3733	 && (pool->type != CTL_POOL_4OTHERSC))
3734		MOD_DEC_USE_COUNT;
3735#endif
3736
3737	free(pool, M_CTL);
3738}
3739
3740void
3741ctl_pool_free(struct ctl_io_pool *pool)
3742{
3743	struct ctl_softc *ctl_softc;
3744
3745	if (pool == NULL)
3746		return;
3747
3748	ctl_softc = pool->ctl_softc;
3749	mtx_lock(&ctl_softc->pool_lock);
3750	pool->flags |= CTL_POOL_FLAG_INVALID;
3751	ctl_pool_release(pool);
3752	mtx_unlock(&ctl_softc->pool_lock);
3753}
3754
3755/*
3756 * This routine does not block (except for spinlocks of course).
3757 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3758 * possible.
3759 */
3760union ctl_io *
3761ctl_alloc_io(void *pool_ref)
3762{
3763	union ctl_io *io;
3764	struct ctl_softc *ctl_softc;
3765	struct ctl_io_pool *pool, *npool;
3766	struct ctl_io_pool *emergency_pool;
3767
3768	pool = (struct ctl_io_pool *)pool_ref;
3769
3770	if (pool == NULL) {
3771		printf("%s: pool is NULL\n", __func__);
3772		return (NULL);
3773	}
3774
3775	emergency_pool = NULL;
3776
3777	ctl_softc = pool->ctl_softc;
3778
3779	mtx_lock(&ctl_softc->pool_lock);
3780	/*
3781	 * First, try to get the io structure from the user's pool.
3782	 */
3783	if (ctl_pool_acquire(pool) == 0) {
3784		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3785		if (io != NULL) {
3786			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3787			pool->total_allocated++;
3788			pool->free_ctl_io--;
3789			mtx_unlock(&ctl_softc->pool_lock);
3790			return (io);
3791		} else
3792			ctl_pool_release(pool);
3793	}
3794	/*
3795	 * If he doesn't have any io structures left, search for an
3796	 * emergency pool and grab one from there.
3797	 */
3798	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3799		if (npool->type != CTL_POOL_EMERGENCY)
3800			continue;
3801
3802		if (ctl_pool_acquire(npool) != 0)
3803			continue;
3804
3805		emergency_pool = npool;
3806
3807		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3808		if (io != NULL) {
3809			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3810			npool->total_allocated++;
3811			npool->free_ctl_io--;
3812			mtx_unlock(&ctl_softc->pool_lock);
3813			return (io);
3814		} else
3815			ctl_pool_release(npool);
3816	}
3817
3818	/* Drop the spinlock before we malloc */
3819	mtx_unlock(&ctl_softc->pool_lock);
3820
3821	/*
3822	 * The emergency pool (if it exists) didn't have one, so try an
3823	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3824	 */
3825	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3826	if (io != NULL) {
3827		/*
3828		 * If the emergency pool exists but is empty, add this
3829		 * ctl_io to its list when it gets freed.
3830		 */
3831		if (emergency_pool != NULL) {
3832			mtx_lock(&ctl_softc->pool_lock);
3833			if (ctl_pool_acquire(emergency_pool) == 0) {
3834				io->io_hdr.pool = emergency_pool;
3835				emergency_pool->total_ctl_io++;
3836				/*
3837				 * Need to bump this, otherwise
3838				 * total_allocated and total_freed won't
3839				 * match when we no longer have anything
3840				 * outstanding.
3841				 */
3842				emergency_pool->total_allocated++;
3843			}
3844			mtx_unlock(&ctl_softc->pool_lock);
3845		} else
3846			io->io_hdr.pool = NULL;
3847	}
3848
3849	return (io);
3850}
3851
3852void
3853ctl_free_io(union ctl_io *io)
3854{
3855	if (io == NULL)
3856		return;
3857
3858	/*
3859	 * If this ctl_io has a pool, return it to that pool.
3860	 */
3861	if (io->io_hdr.pool != NULL) {
3862		struct ctl_io_pool *pool;
3863
3864		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3865		mtx_lock(&pool->ctl_softc->pool_lock);
3866		io->io_hdr.io_type = 0xff;
3867		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3868		pool->total_freed++;
3869		pool->free_ctl_io++;
3870		ctl_pool_release(pool);
3871		mtx_unlock(&pool->ctl_softc->pool_lock);
3872	} else {
3873		/*
3874		 * Otherwise, just free it.  We probably malloced it and
3875		 * the emergency pool wasn't available.
3876		 */
3877		free(io, M_CTLIO);
3878	}
3879
3880}
3881
3882void
3883ctl_zero_io(union ctl_io *io)
3884{
3885	void *pool_ref;
3886
3887	if (io == NULL)
3888		return;
3889
3890	/*
3891	 * May need to preserve linked list pointers at some point too.
3892	 */
3893	pool_ref = io->io_hdr.pool;
3894
3895	memset(io, 0, sizeof(*io));
3896
3897	io->io_hdr.pool = pool_ref;
3898}
3899
3900/*
3901 * This routine is currently used for internal copies of ctl_ios that need
3902 * to persist for some reason after we've already returned status to the
3903 * FETD.  (Thus the flag set.)
3904 *
3905 * XXX XXX
3906 * Note that this makes a blind copy of all fields in the ctl_io, except
3907 * for the pool reference.  This includes any memory that has been
3908 * allocated!  That memory will no longer be valid after done has been
3909 * called, so this would be VERY DANGEROUS for command that actually does
3910 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3911 * start and stop commands, which don't transfer any data, so this is not a
3912 * problem.  If it is used for anything else, the caller would also need to
3913 * allocate data buffer space and this routine would need to be modified to
3914 * copy the data buffer(s) as well.
3915 */
3916void
3917ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3918{
3919	void *pool_ref;
3920
3921	if ((src == NULL)
3922	 || (dest == NULL))
3923		return;
3924
3925	/*
3926	 * May need to preserve linked list pointers at some point too.
3927	 */
3928	pool_ref = dest->io_hdr.pool;
3929
3930	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3931
3932	dest->io_hdr.pool = pool_ref;
3933	/*
3934	 * We need to know that this is an internal copy, and doesn't need
3935	 * to get passed back to the FETD that allocated it.
3936	 */
3937	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3938}
3939
3940#ifdef NEEDTOPORT
3941static void
3942ctl_update_power_subpage(struct copan_power_subpage *page)
3943{
3944	int num_luns, num_partitions, config_type;
3945	struct ctl_softc *softc;
3946	cs_BOOL_t aor_present, shelf_50pct_power;
3947	cs_raidset_personality_t rs_type;
3948	int max_active_luns;
3949
3950	softc = control_softc;
3951
3952	/* subtract out the processor LUN */
3953	num_luns = softc->num_luns - 1;
3954	/*
3955	 * Default to 7 LUNs active, which was the only number we allowed
3956	 * in the past.
3957	 */
3958	max_active_luns = 7;
3959
3960	num_partitions = config_GetRsPartitionInfo();
3961	config_type = config_GetConfigType();
3962	shelf_50pct_power = config_GetShelfPowerMode();
3963	aor_present = config_IsAorRsPresent();
3964
3965	rs_type = ddb_GetRsRaidType(1);
3966	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3967	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3968		EPRINT(0, "Unsupported RS type %d!", rs_type);
3969	}
3970
3971
3972	page->total_luns = num_luns;
3973
3974	switch (config_type) {
3975	case 40:
3976		/*
3977		 * In a 40 drive configuration, it doesn't matter what DC
3978		 * cards we have, whether we have AOR enabled or not,
3979		 * partitioning or not, or what type of RAIDset we have.
3980		 * In that scenario, we can power up every LUN we present
3981		 * to the user.
3982		 */
3983		max_active_luns = num_luns;
3984
3985		break;
3986	case 64:
3987		if (shelf_50pct_power == CS_FALSE) {
3988			/* 25% power */
3989			if (aor_present == CS_TRUE) {
3990				if (rs_type ==
3991				     CS_RAIDSET_PERSONALITY_RAID5) {
3992					max_active_luns = 7;
3993				} else if (rs_type ==
3994					 CS_RAIDSET_PERSONALITY_RAID1){
3995					max_active_luns = 14;
3996				} else {
3997					/* XXX KDM now what?? */
3998				}
3999			} else {
4000				if (rs_type ==
4001				     CS_RAIDSET_PERSONALITY_RAID5) {
4002					max_active_luns = 8;
4003				} else if (rs_type ==
4004					 CS_RAIDSET_PERSONALITY_RAID1){
4005					max_active_luns = 16;
4006				} else {
4007					/* XXX KDM now what?? */
4008				}
4009			}
4010		} else {
4011			/* 50% power */
4012			/*
4013			 * With 50% power in a 64 drive configuration, we
4014			 * can power all LUNs we present.
4015			 */
4016			max_active_luns = num_luns;
4017		}
4018		break;
4019	case 112:
4020		if (shelf_50pct_power == CS_FALSE) {
4021			/* 25% power */
4022			if (aor_present == CS_TRUE) {
4023				if (rs_type ==
4024				     CS_RAIDSET_PERSONALITY_RAID5) {
4025					max_active_luns = 7;
4026				} else if (rs_type ==
4027					 CS_RAIDSET_PERSONALITY_RAID1){
4028					max_active_luns = 14;
4029				} else {
4030					/* XXX KDM now what?? */
4031				}
4032			} else {
4033				if (rs_type ==
4034				     CS_RAIDSET_PERSONALITY_RAID5) {
4035					max_active_luns = 8;
4036				} else if (rs_type ==
4037					 CS_RAIDSET_PERSONALITY_RAID1){
4038					max_active_luns = 16;
4039				} else {
4040					/* XXX KDM now what?? */
4041				}
4042			}
4043		} else {
4044			/* 50% power */
4045			if (aor_present == CS_TRUE) {
4046				if (rs_type ==
4047				     CS_RAIDSET_PERSONALITY_RAID5) {
4048					max_active_luns = 14;
4049				} else if (rs_type ==
4050					 CS_RAIDSET_PERSONALITY_RAID1){
4051					/*
4052					 * We're assuming here that disk
4053					 * caching is enabled, and so we're
4054					 * able to power up half of each
4055					 * LUN, and cache all writes.
4056					 */
4057					max_active_luns = num_luns;
4058				} else {
4059					/* XXX KDM now what?? */
4060				}
4061			} else {
4062				if (rs_type ==
4063				     CS_RAIDSET_PERSONALITY_RAID5) {
4064					max_active_luns = 15;
4065				} else if (rs_type ==
4066					 CS_RAIDSET_PERSONALITY_RAID1){
4067					max_active_luns = 30;
4068				} else {
4069					/* XXX KDM now what?? */
4070				}
4071			}
4072		}
4073		break;
4074	default:
4075		/*
4076		 * In this case, we have an unknown configuration, so we
4077		 * just use the default from above.
4078		 */
4079		break;
4080	}
4081
4082	page->max_active_luns = max_active_luns;
4083#if 0
4084	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4085	       page->total_luns, page->max_active_luns);
4086#endif
4087}
4088#endif /* NEEDTOPORT */
4089
4090/*
4091 * This routine could be used in the future to load default and/or saved
4092 * mode page parameters for a particuar lun.
4093 */
4094static int
4095ctl_init_page_index(struct ctl_lun *lun)
4096{
4097	int i;
4098	struct ctl_page_index *page_index;
4099	struct ctl_softc *softc;
4100	const char *value;
4101
4102	memcpy(&lun->mode_pages.index, page_index_template,
4103	       sizeof(page_index_template));
4104
4105	softc = lun->ctl_softc;
4106
4107	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4108
4109		page_index = &lun->mode_pages.index[i];
4110		/*
4111		 * If this is a disk-only mode page, there's no point in
4112		 * setting it up.  For some pages, we have to have some
4113		 * basic information about the disk in order to calculate the
4114		 * mode page data.
4115		 */
4116		if ((lun->be_lun->lun_type != T_DIRECT)
4117		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4118			continue;
4119
4120		switch (page_index->page_code & SMPH_PC_MASK) {
4121		case SMS_FORMAT_DEVICE_PAGE: {
4122			struct scsi_format_page *format_page;
4123
4124			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4125				panic("subpage is incorrect!");
4126
4127			/*
4128			 * Sectors per track are set above.  Bytes per
4129			 * sector need to be set here on a per-LUN basis.
4130			 */
4131			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4132			       &format_page_default,
4133			       sizeof(format_page_default));
4134			memcpy(&lun->mode_pages.format_page[
4135			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4136			       sizeof(format_page_changeable));
4137			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4138			       &format_page_default,
4139			       sizeof(format_page_default));
4140			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4141			       &format_page_default,
4142			       sizeof(format_page_default));
4143
4144			format_page = &lun->mode_pages.format_page[
4145				CTL_PAGE_CURRENT];
4146			scsi_ulto2b(lun->be_lun->blocksize,
4147				    format_page->bytes_per_sector);
4148
4149			format_page = &lun->mode_pages.format_page[
4150				CTL_PAGE_DEFAULT];
4151			scsi_ulto2b(lun->be_lun->blocksize,
4152				    format_page->bytes_per_sector);
4153
4154			format_page = &lun->mode_pages.format_page[
4155				CTL_PAGE_SAVED];
4156			scsi_ulto2b(lun->be_lun->blocksize,
4157				    format_page->bytes_per_sector);
4158
4159			page_index->page_data =
4160				(uint8_t *)lun->mode_pages.format_page;
4161			break;
4162		}
4163		case SMS_RIGID_DISK_PAGE: {
4164			struct scsi_rigid_disk_page *rigid_disk_page;
4165			uint32_t sectors_per_cylinder;
4166			uint64_t cylinders;
4167#ifndef	__XSCALE__
4168			int shift;
4169#endif /* !__XSCALE__ */
4170
4171			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4172				panic("invalid subpage value %d",
4173				      page_index->subpage);
4174
4175			/*
4176			 * Rotation rate and sectors per track are set
4177			 * above.  We calculate the cylinders here based on
4178			 * capacity.  Due to the number of heads and
4179			 * sectors per track we're using, smaller arrays
4180			 * may turn out to have 0 cylinders.  Linux and
4181			 * FreeBSD don't pay attention to these mode pages
4182			 * to figure out capacity, but Solaris does.  It
4183			 * seems to deal with 0 cylinders just fine, and
4184			 * works out a fake geometry based on the capacity.
4185			 */
4186			memcpy(&lun->mode_pages.rigid_disk_page[
4187			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4188			       sizeof(rigid_disk_page_default));
4189			memcpy(&lun->mode_pages.rigid_disk_page[
4190			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4191			       sizeof(rigid_disk_page_changeable));
4192			memcpy(&lun->mode_pages.rigid_disk_page[
4193			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4194			       sizeof(rigid_disk_page_default));
4195			memcpy(&lun->mode_pages.rigid_disk_page[
4196			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4197			       sizeof(rigid_disk_page_default));
4198
4199			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4200				CTL_DEFAULT_HEADS;
4201
4202			/*
4203			 * The divide method here will be more accurate,
4204			 * probably, but results in floating point being
4205			 * used in the kernel on i386 (__udivdi3()).  On the
4206			 * XScale, though, __udivdi3() is implemented in
4207			 * software.
4208			 *
4209			 * The shift method for cylinder calculation is
4210			 * accurate if sectors_per_cylinder is a power of
4211			 * 2.  Otherwise it might be slightly off -- you
4212			 * might have a bit of a truncation problem.
4213			 */
4214#ifdef	__XSCALE__
4215			cylinders = (lun->be_lun->maxlba + 1) /
4216				sectors_per_cylinder;
4217#else
4218			for (shift = 31; shift > 0; shift--) {
4219				if (sectors_per_cylinder & (1 << shift))
4220					break;
4221			}
4222			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4223#endif
4224
4225			/*
4226			 * We've basically got 3 bytes, or 24 bits for the
4227			 * cylinder size in the mode page.  If we're over,
4228			 * just round down to 2^24.
4229			 */
4230			if (cylinders > 0xffffff)
4231				cylinders = 0xffffff;
4232
4233			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4234				CTL_PAGE_CURRENT];
4235			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4236
4237			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4238				CTL_PAGE_DEFAULT];
4239			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4240
4241			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4242				CTL_PAGE_SAVED];
4243			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4244
4245			page_index->page_data =
4246				(uint8_t *)lun->mode_pages.rigid_disk_page;
4247			break;
4248		}
4249		case SMS_CACHING_PAGE: {
4250			struct scsi_caching_page *caching_page;
4251
4252			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4253				panic("invalid subpage value %d",
4254				      page_index->subpage);
4255			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4256			       &caching_page_default,
4257			       sizeof(caching_page_default));
4258			memcpy(&lun->mode_pages.caching_page[
4259			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4260			       sizeof(caching_page_changeable));
4261			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4262			       &caching_page_default,
4263			       sizeof(caching_page_default));
4264			caching_page = &lun->mode_pages.caching_page[
4265			    CTL_PAGE_SAVED];
4266			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4267			if (value != NULL && strcmp(value, "off") == 0)
4268				caching_page->flags1 &= ~SCP_WCE;
4269			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4270			if (value != NULL && strcmp(value, "off") == 0)
4271				caching_page->flags1 |= SCP_RCD;
4272			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4273			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4274			       sizeof(caching_page_default));
4275			page_index->page_data =
4276				(uint8_t *)lun->mode_pages.caching_page;
4277			break;
4278		}
4279		case SMS_CONTROL_MODE_PAGE: {
4280			struct scsi_control_page *control_page;
4281
4282			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4283				panic("invalid subpage value %d",
4284				      page_index->subpage);
4285
4286			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4287			       &control_page_default,
4288			       sizeof(control_page_default));
4289			memcpy(&lun->mode_pages.control_page[
4290			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4291			       sizeof(control_page_changeable));
4292			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4293			       &control_page_default,
4294			       sizeof(control_page_default));
4295			control_page = &lun->mode_pages.control_page[
4296			    CTL_PAGE_SAVED];
4297			value = ctl_get_opt(&lun->be_lun->options, "reordering");
4298			if (value != NULL && strcmp(value, "unrestricted") == 0) {
4299				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4300				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4301			}
4302			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4303			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4304			       sizeof(control_page_default));
4305			page_index->page_data =
4306				(uint8_t *)lun->mode_pages.control_page;
4307			break;
4308
4309		}
4310		case SMS_VENDOR_SPECIFIC_PAGE:{
4311			switch (page_index->subpage) {
4312			case PWR_SUBPAGE_CODE: {
4313				struct copan_power_subpage *current_page,
4314							   *saved_page;
4315
4316				memcpy(&lun->mode_pages.power_subpage[
4317				       CTL_PAGE_CURRENT],
4318				       &power_page_default,
4319				       sizeof(power_page_default));
4320				memcpy(&lun->mode_pages.power_subpage[
4321				       CTL_PAGE_CHANGEABLE],
4322				       &power_page_changeable,
4323				       sizeof(power_page_changeable));
4324				memcpy(&lun->mode_pages.power_subpage[
4325				       CTL_PAGE_DEFAULT],
4326				       &power_page_default,
4327				       sizeof(power_page_default));
4328				memcpy(&lun->mode_pages.power_subpage[
4329				       CTL_PAGE_SAVED],
4330				       &power_page_default,
4331				       sizeof(power_page_default));
4332				page_index->page_data =
4333				    (uint8_t *)lun->mode_pages.power_subpage;
4334
4335				current_page = (struct copan_power_subpage *)
4336					(page_index->page_data +
4337					 (page_index->page_len *
4338					  CTL_PAGE_CURRENT));
4339			        saved_page = (struct copan_power_subpage *)
4340				        (page_index->page_data +
4341					 (page_index->page_len *
4342					  CTL_PAGE_SAVED));
4343				break;
4344			}
4345			case APS_SUBPAGE_CODE: {
4346				struct copan_aps_subpage *current_page,
4347							 *saved_page;
4348
4349				// This gets set multiple times but
4350				// it should always be the same. It's
4351				// only done during init so who cares.
4352				index_to_aps_page = i;
4353
4354				memcpy(&lun->mode_pages.aps_subpage[
4355				       CTL_PAGE_CURRENT],
4356				       &aps_page_default,
4357				       sizeof(aps_page_default));
4358				memcpy(&lun->mode_pages.aps_subpage[
4359				       CTL_PAGE_CHANGEABLE],
4360				       &aps_page_changeable,
4361				       sizeof(aps_page_changeable));
4362				memcpy(&lun->mode_pages.aps_subpage[
4363				       CTL_PAGE_DEFAULT],
4364				       &aps_page_default,
4365				       sizeof(aps_page_default));
4366				memcpy(&lun->mode_pages.aps_subpage[
4367				       CTL_PAGE_SAVED],
4368				       &aps_page_default,
4369				       sizeof(aps_page_default));
4370				page_index->page_data =
4371					(uint8_t *)lun->mode_pages.aps_subpage;
4372
4373				current_page = (struct copan_aps_subpage *)
4374					(page_index->page_data +
4375					 (page_index->page_len *
4376					  CTL_PAGE_CURRENT));
4377				saved_page = (struct copan_aps_subpage *)
4378					(page_index->page_data +
4379					 (page_index->page_len *
4380					  CTL_PAGE_SAVED));
4381				break;
4382			}
4383			case DBGCNF_SUBPAGE_CODE: {
4384				struct copan_debugconf_subpage *current_page,
4385							       *saved_page;
4386
4387				memcpy(&lun->mode_pages.debugconf_subpage[
4388				       CTL_PAGE_CURRENT],
4389				       &debugconf_page_default,
4390				       sizeof(debugconf_page_default));
4391				memcpy(&lun->mode_pages.debugconf_subpage[
4392				       CTL_PAGE_CHANGEABLE],
4393				       &debugconf_page_changeable,
4394				       sizeof(debugconf_page_changeable));
4395				memcpy(&lun->mode_pages.debugconf_subpage[
4396				       CTL_PAGE_DEFAULT],
4397				       &debugconf_page_default,
4398				       sizeof(debugconf_page_default));
4399				memcpy(&lun->mode_pages.debugconf_subpage[
4400				       CTL_PAGE_SAVED],
4401				       &debugconf_page_default,
4402				       sizeof(debugconf_page_default));
4403				page_index->page_data =
4404					(uint8_t *)lun->mode_pages.debugconf_subpage;
4405
4406				current_page = (struct copan_debugconf_subpage *)
4407					(page_index->page_data +
4408					 (page_index->page_len *
4409					  CTL_PAGE_CURRENT));
4410				saved_page = (struct copan_debugconf_subpage *)
4411					(page_index->page_data +
4412					 (page_index->page_len *
4413					  CTL_PAGE_SAVED));
4414				break;
4415			}
4416			default:
4417				panic("invalid subpage value %d",
4418				      page_index->subpage);
4419				break;
4420			}
4421   			break;
4422		}
4423		default:
4424			panic("invalid page value %d",
4425			      page_index->page_code & SMPH_PC_MASK);
4426			break;
4427    	}
4428	}
4429
4430	return (CTL_RETVAL_COMPLETE);
4431}
4432
4433/*
4434 * LUN allocation.
4435 *
4436 * Requirements:
4437 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4438 *   wants us to allocate the LUN and he can block.
4439 * - ctl_softc is always set
4440 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4441 *
4442 * Returns 0 for success, non-zero (errno) for failure.
4443 */
4444static int
4445ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4446	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4447{
4448	struct ctl_lun *nlun, *lun;
4449	struct ctl_port *port;
4450	struct scsi_vpd_id_descriptor *desc;
4451	struct scsi_vpd_id_t10 *t10id;
4452	const char *eui, *naa, *scsiname, *vendor, *value;
4453	int lun_number, i, lun_malloced;
4454	int devidlen, idlen1, idlen2 = 0, len;
4455
4456	if (be_lun == NULL)
4457		return (EINVAL);
4458
4459	/*
4460	 * We currently only support Direct Access or Processor LUN types.
4461	 */
4462	switch (be_lun->lun_type) {
4463	case T_DIRECT:
4464		break;
4465	case T_PROCESSOR:
4466		break;
4467	case T_SEQUENTIAL:
4468	case T_CHANGER:
4469	default:
4470		be_lun->lun_config_status(be_lun->be_lun,
4471					  CTL_LUN_CONFIG_FAILURE);
4472		break;
4473	}
4474	if (ctl_lun == NULL) {
4475		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4476		lun_malloced = 1;
4477	} else {
4478		lun_malloced = 0;
4479		lun = ctl_lun;
4480	}
4481
4482	memset(lun, 0, sizeof(*lun));
4483	if (lun_malloced)
4484		lun->flags = CTL_LUN_MALLOCED;
4485
4486	/* Generate LUN ID. */
4487	devidlen = max(CTL_DEVID_MIN_LEN,
4488	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4489	idlen1 = sizeof(*t10id) + devidlen;
4490	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4491	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4492	if (scsiname != NULL) {
4493		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4494		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4495	}
4496	eui = ctl_get_opt(&be_lun->options, "eui");
4497	if (eui != NULL) {
4498		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4499	}
4500	naa = ctl_get_opt(&be_lun->options, "naa");
4501	if (naa != NULL) {
4502		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4503	}
4504	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4505	    M_CTL, M_WAITOK | M_ZERO);
4506	lun->lun_devid->len = len;
4507	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4508	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4509	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4510	desc->length = idlen1;
4511	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4512	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4513	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4514		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4515	} else {
4516		strncpy(t10id->vendor, vendor,
4517		    min(sizeof(t10id->vendor), strlen(vendor)));
4518	}
4519	strncpy((char *)t10id->vendor_spec_id,
4520	    (char *)be_lun->device_id, devidlen);
4521	if (scsiname != NULL) {
4522		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4523		    desc->length);
4524		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4525		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4526		    SVPD_ID_TYPE_SCSI_NAME;
4527		desc->length = idlen2;
4528		strlcpy(desc->identifier, scsiname, idlen2);
4529	}
4530	if (eui != NULL) {
4531		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4532		    desc->length);
4533		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4534		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4535		    SVPD_ID_TYPE_EUI64;
4536		desc->length = 8;
4537		scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4538	}
4539	if (naa != NULL) {
4540		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4541		    desc->length);
4542		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4543		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4544		    SVPD_ID_TYPE_NAA;
4545		desc->length = 8;
4546		scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4547	}
4548
4549	mtx_lock(&ctl_softc->ctl_lock);
4550	/*
4551	 * See if the caller requested a particular LUN number.  If so, see
4552	 * if it is available.  Otherwise, allocate the first available LUN.
4553	 */
4554	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4555		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4556		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4557			mtx_unlock(&ctl_softc->ctl_lock);
4558			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4559				printf("ctl: requested LUN ID %d is higher "
4560				       "than CTL_MAX_LUNS - 1 (%d)\n",
4561				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4562			} else {
4563				/*
4564				 * XXX KDM return an error, or just assign
4565				 * another LUN ID in this case??
4566				 */
4567				printf("ctl: requested LUN ID %d is already "
4568				       "in use\n", be_lun->req_lun_id);
4569			}
4570			if (lun->flags & CTL_LUN_MALLOCED)
4571				free(lun, M_CTL);
4572			be_lun->lun_config_status(be_lun->be_lun,
4573						  CTL_LUN_CONFIG_FAILURE);
4574			return (ENOSPC);
4575		}
4576		lun_number = be_lun->req_lun_id;
4577	} else {
4578		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4579		if (lun_number == -1) {
4580			mtx_unlock(&ctl_softc->ctl_lock);
4581			printf("ctl: can't allocate LUN on target %ju, out of "
4582			       "LUNs\n", (uintmax_t)target_id.id);
4583			if (lun->flags & CTL_LUN_MALLOCED)
4584				free(lun, M_CTL);
4585			be_lun->lun_config_status(be_lun->be_lun,
4586						  CTL_LUN_CONFIG_FAILURE);
4587			return (ENOSPC);
4588		}
4589	}
4590	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4591
4592	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4593	lun->target = target_id;
4594	lun->lun = lun_number;
4595	lun->be_lun = be_lun;
4596	/*
4597	 * The processor LUN is always enabled.  Disk LUNs come on line
4598	 * disabled, and must be enabled by the backend.
4599	 */
4600	lun->flags |= CTL_LUN_DISABLED;
4601	lun->backend = be_lun->be;
4602	be_lun->ctl_lun = lun;
4603	be_lun->lun_id = lun_number;
4604	atomic_add_int(&be_lun->be->num_luns, 1);
4605	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4606		lun->flags |= CTL_LUN_STOPPED;
4607
4608	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4609		lun->flags |= CTL_LUN_INOPERABLE;
4610
4611	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4612		lun->flags |= CTL_LUN_PRIMARY_SC;
4613
4614	value = ctl_get_opt(&be_lun->options, "readonly");
4615	if (value != NULL && strcmp(value, "on") == 0)
4616		lun->flags |= CTL_LUN_READONLY;
4617
4618	lun->ctl_softc = ctl_softc;
4619	TAILQ_INIT(&lun->ooa_queue);
4620	TAILQ_INIT(&lun->blocked_queue);
4621	STAILQ_INIT(&lun->error_list);
4622	ctl_tpc_lun_init(lun);
4623
4624	/*
4625	 * Initialize the mode page index.
4626	 */
4627	ctl_init_page_index(lun);
4628
4629	/*
4630	 * Set the poweron UA for all initiators on this LUN only.
4631	 */
4632	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4633		lun->pending_ua[i] = CTL_UA_POWERON;
4634
4635	/*
4636	 * Now, before we insert this lun on the lun list, set the lun
4637	 * inventory changed UA for all other luns.
4638	 */
4639	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4640		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4641			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4642		}
4643	}
4644
4645	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4646
4647	ctl_softc->ctl_luns[lun_number] = lun;
4648
4649	ctl_softc->num_luns++;
4650
4651	/* Setup statistics gathering */
4652	lun->stats.device_type = be_lun->lun_type;
4653	lun->stats.lun_number = lun_number;
4654	if (lun->stats.device_type == T_DIRECT)
4655		lun->stats.blocksize = be_lun->blocksize;
4656	else
4657		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4658	for (i = 0;i < CTL_MAX_PORTS;i++)
4659		lun->stats.ports[i].targ_port = i;
4660
4661	mtx_unlock(&ctl_softc->ctl_lock);
4662
4663	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4664
4665	/*
4666	 * Run through each registered FETD and bring it online if it isn't
4667	 * already.  Enable the target ID if it hasn't been enabled, and
4668	 * enable this particular LUN.
4669	 */
4670	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4671		int retval;
4672
4673		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4674		if (retval != 0) {
4675			printf("ctl_alloc_lun: FETD %s port %d returned error "
4676			       "%d for lun_enable on target %ju lun %d\n",
4677			       port->port_name, port->targ_port, retval,
4678			       (uintmax_t)target_id.id, lun_number);
4679		} else
4680			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4681	}
4682	return (0);
4683}
4684
4685/*
4686 * Delete a LUN.
4687 * Assumptions:
4688 * - LUN has already been marked invalid and any pending I/O has been taken
4689 *   care of.
4690 */
4691static int
4692ctl_free_lun(struct ctl_lun *lun)
4693{
4694	struct ctl_softc *softc;
4695#if 0
4696	struct ctl_port *port;
4697#endif
4698	struct ctl_lun *nlun;
4699	int i;
4700
4701	softc = lun->ctl_softc;
4702
4703	mtx_assert(&softc->ctl_lock, MA_OWNED);
4704
4705	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4706
4707	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4708
4709	softc->ctl_luns[lun->lun] = NULL;
4710
4711	if (!TAILQ_EMPTY(&lun->ooa_queue))
4712		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4713
4714	softc->num_luns--;
4715
4716	/*
4717	 * XXX KDM this scheme only works for a single target/multiple LUN
4718	 * setup.  It needs to be revamped for a multiple target scheme.
4719	 *
4720	 * XXX KDM this results in port->lun_disable() getting called twice,
4721	 * once when ctl_disable_lun() is called, and a second time here.
4722	 * We really need to re-think the LUN disable semantics.  There
4723	 * should probably be several steps/levels to LUN removal:
4724	 *  - disable
4725	 *  - invalidate
4726	 *  - free
4727 	 *
4728	 * Right now we only have a disable method when communicating to
4729	 * the front end ports, at least for individual LUNs.
4730	 */
4731#if 0
4732	STAILQ_FOREACH(port, &softc->port_list, links) {
4733		int retval;
4734
4735		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4736					 lun->lun);
4737		if (retval != 0) {
4738			printf("ctl_free_lun: FETD %s port %d returned error "
4739			       "%d for lun_disable on target %ju lun %jd\n",
4740			       port->port_name, port->targ_port, retval,
4741			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4742		}
4743
4744		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4745			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4746
4747			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4748			if (retval != 0) {
4749				printf("ctl_free_lun: FETD %s port %d "
4750				       "returned error %d for targ_disable on "
4751				       "target %ju\n", port->port_name,
4752				       port->targ_port, retval,
4753				       (uintmax_t)lun->target.id);
4754			} else
4755				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4756
4757			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4758				continue;
4759
4760#if 0
4761			port->port_offline(port->onoff_arg);
4762			port->status &= ~CTL_PORT_STATUS_ONLINE;
4763#endif
4764		}
4765	}
4766#endif
4767
4768	/*
4769	 * Tell the backend to free resources, if this LUN has a backend.
4770	 */
4771	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4772	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4773
4774	ctl_tpc_lun_shutdown(lun);
4775	mtx_destroy(&lun->lun_lock);
4776	free(lun->lun_devid, M_CTL);
4777	if (lun->flags & CTL_LUN_MALLOCED)
4778		free(lun, M_CTL);
4779
4780	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4781		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4782			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4783		}
4784	}
4785
4786	return (0);
4787}
4788
4789static void
4790ctl_create_lun(struct ctl_be_lun *be_lun)
4791{
4792	struct ctl_softc *ctl_softc;
4793
4794	ctl_softc = control_softc;
4795
4796	/*
4797	 * ctl_alloc_lun() should handle all potential failure cases.
4798	 */
4799	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4800}
4801
4802int
4803ctl_add_lun(struct ctl_be_lun *be_lun)
4804{
4805	struct ctl_softc *ctl_softc = control_softc;
4806
4807	mtx_lock(&ctl_softc->ctl_lock);
4808	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4809	mtx_unlock(&ctl_softc->ctl_lock);
4810	wakeup(&ctl_softc->pending_lun_queue);
4811
4812	return (0);
4813}
4814
4815int
4816ctl_enable_lun(struct ctl_be_lun *be_lun)
4817{
4818	struct ctl_softc *ctl_softc;
4819	struct ctl_port *port, *nport;
4820	struct ctl_lun *lun;
4821	int retval;
4822
4823	ctl_softc = control_softc;
4824
4825	lun = (struct ctl_lun *)be_lun->ctl_lun;
4826
4827	mtx_lock(&ctl_softc->ctl_lock);
4828	mtx_lock(&lun->lun_lock);
4829	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4830		/*
4831		 * eh?  Why did we get called if the LUN is already
4832		 * enabled?
4833		 */
4834		mtx_unlock(&lun->lun_lock);
4835		mtx_unlock(&ctl_softc->ctl_lock);
4836		return (0);
4837	}
4838	lun->flags &= ~CTL_LUN_DISABLED;
4839	mtx_unlock(&lun->lun_lock);
4840
4841	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4842		nport = STAILQ_NEXT(port, links);
4843
4844		/*
4845		 * Drop the lock while we call the FETD's enable routine.
4846		 * This can lead to a callback into CTL (at least in the
4847		 * case of the internal initiator frontend.
4848		 */
4849		mtx_unlock(&ctl_softc->ctl_lock);
4850		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4851		mtx_lock(&ctl_softc->ctl_lock);
4852		if (retval != 0) {
4853			printf("%s: FETD %s port %d returned error "
4854			       "%d for lun_enable on target %ju lun %jd\n",
4855			       __func__, port->port_name, port->targ_port, retval,
4856			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4857		}
4858#if 0
4859		 else {
4860            /* NOTE:  TODO:  why does lun enable affect port status? */
4861			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4862		}
4863#endif
4864	}
4865
4866	mtx_unlock(&ctl_softc->ctl_lock);
4867
4868	return (0);
4869}
4870
4871int
4872ctl_disable_lun(struct ctl_be_lun *be_lun)
4873{
4874	struct ctl_softc *ctl_softc;
4875	struct ctl_port *port;
4876	struct ctl_lun *lun;
4877	int retval;
4878
4879	ctl_softc = control_softc;
4880
4881	lun = (struct ctl_lun *)be_lun->ctl_lun;
4882
4883	mtx_lock(&ctl_softc->ctl_lock);
4884	mtx_lock(&lun->lun_lock);
4885	if (lun->flags & CTL_LUN_DISABLED) {
4886		mtx_unlock(&lun->lun_lock);
4887		mtx_unlock(&ctl_softc->ctl_lock);
4888		return (0);
4889	}
4890	lun->flags |= CTL_LUN_DISABLED;
4891	mtx_unlock(&lun->lun_lock);
4892
4893	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4894		mtx_unlock(&ctl_softc->ctl_lock);
4895		/*
4896		 * Drop the lock before we call the frontend's disable
4897		 * routine, to avoid lock order reversals.
4898		 *
4899		 * XXX KDM what happens if the frontend list changes while
4900		 * we're traversing it?  It's unlikely, but should be handled.
4901		 */
4902		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4903					 lun->lun);
4904		mtx_lock(&ctl_softc->ctl_lock);
4905		if (retval != 0) {
4906			printf("ctl_alloc_lun: FETD %s port %d returned error "
4907			       "%d for lun_disable on target %ju lun %jd\n",
4908			       port->port_name, port->targ_port, retval,
4909			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4910		}
4911	}
4912
4913	mtx_unlock(&ctl_softc->ctl_lock);
4914
4915	return (0);
4916}
4917
4918int
4919ctl_start_lun(struct ctl_be_lun *be_lun)
4920{
4921	struct ctl_softc *ctl_softc;
4922	struct ctl_lun *lun;
4923
4924	ctl_softc = control_softc;
4925
4926	lun = (struct ctl_lun *)be_lun->ctl_lun;
4927
4928	mtx_lock(&lun->lun_lock);
4929	lun->flags &= ~CTL_LUN_STOPPED;
4930	mtx_unlock(&lun->lun_lock);
4931
4932	return (0);
4933}
4934
4935int
4936ctl_stop_lun(struct ctl_be_lun *be_lun)
4937{
4938	struct ctl_softc *ctl_softc;
4939	struct ctl_lun *lun;
4940
4941	ctl_softc = control_softc;
4942
4943	lun = (struct ctl_lun *)be_lun->ctl_lun;
4944
4945	mtx_lock(&lun->lun_lock);
4946	lun->flags |= CTL_LUN_STOPPED;
4947	mtx_unlock(&lun->lun_lock);
4948
4949	return (0);
4950}
4951
4952int
4953ctl_lun_offline(struct ctl_be_lun *be_lun)
4954{
4955	struct ctl_softc *ctl_softc;
4956	struct ctl_lun *lun;
4957
4958	ctl_softc = control_softc;
4959
4960	lun = (struct ctl_lun *)be_lun->ctl_lun;
4961
4962	mtx_lock(&lun->lun_lock);
4963	lun->flags |= CTL_LUN_OFFLINE;
4964	mtx_unlock(&lun->lun_lock);
4965
4966	return (0);
4967}
4968
4969int
4970ctl_lun_online(struct ctl_be_lun *be_lun)
4971{
4972	struct ctl_softc *ctl_softc;
4973	struct ctl_lun *lun;
4974
4975	ctl_softc = control_softc;
4976
4977	lun = (struct ctl_lun *)be_lun->ctl_lun;
4978
4979	mtx_lock(&lun->lun_lock);
4980	lun->flags &= ~CTL_LUN_OFFLINE;
4981	mtx_unlock(&lun->lun_lock);
4982
4983	return (0);
4984}
4985
4986int
4987ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4988{
4989	struct ctl_softc *ctl_softc;
4990	struct ctl_lun *lun;
4991
4992	ctl_softc = control_softc;
4993
4994	lun = (struct ctl_lun *)be_lun->ctl_lun;
4995
4996	mtx_lock(&lun->lun_lock);
4997
4998	/*
4999	 * The LUN needs to be disabled before it can be marked invalid.
5000	 */
5001	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
5002		mtx_unlock(&lun->lun_lock);
5003		return (-1);
5004	}
5005	/*
5006	 * Mark the LUN invalid.
5007	 */
5008	lun->flags |= CTL_LUN_INVALID;
5009
5010	/*
5011	 * If there is nothing in the OOA queue, go ahead and free the LUN.
5012	 * If we have something in the OOA queue, we'll free it when the
5013	 * last I/O completes.
5014	 */
5015	if (TAILQ_EMPTY(&lun->ooa_queue)) {
5016		mtx_unlock(&lun->lun_lock);
5017		mtx_lock(&ctl_softc->ctl_lock);
5018		ctl_free_lun(lun);
5019		mtx_unlock(&ctl_softc->ctl_lock);
5020	} else
5021		mtx_unlock(&lun->lun_lock);
5022
5023	return (0);
5024}
5025
5026int
5027ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5028{
5029	struct ctl_softc *ctl_softc;
5030	struct ctl_lun *lun;
5031
5032	ctl_softc = control_softc;
5033	lun = (struct ctl_lun *)be_lun->ctl_lun;
5034
5035	mtx_lock(&lun->lun_lock);
5036	lun->flags |= CTL_LUN_INOPERABLE;
5037	mtx_unlock(&lun->lun_lock);
5038
5039	return (0);
5040}
5041
5042int
5043ctl_lun_operable(struct ctl_be_lun *be_lun)
5044{
5045	struct ctl_softc *ctl_softc;
5046	struct ctl_lun *lun;
5047
5048	ctl_softc = control_softc;
5049	lun = (struct ctl_lun *)be_lun->ctl_lun;
5050
5051	mtx_lock(&lun->lun_lock);
5052	lun->flags &= ~CTL_LUN_INOPERABLE;
5053	mtx_unlock(&lun->lun_lock);
5054
5055	return (0);
5056}
5057
5058int
5059ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5060		   int lock)
5061{
5062	struct ctl_softc *softc;
5063	struct ctl_lun *lun;
5064	struct copan_aps_subpage *current_sp;
5065	struct ctl_page_index *page_index;
5066	int i;
5067
5068	softc = control_softc;
5069
5070	mtx_lock(&softc->ctl_lock);
5071
5072	lun = (struct ctl_lun *)be_lun->ctl_lun;
5073	mtx_lock(&lun->lun_lock);
5074
5075	page_index = NULL;
5076	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5077		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5078		     APS_PAGE_CODE)
5079			continue;
5080
5081		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5082			continue;
5083		page_index = &lun->mode_pages.index[i];
5084	}
5085
5086	if (page_index == NULL) {
5087		mtx_unlock(&lun->lun_lock);
5088		mtx_unlock(&softc->ctl_lock);
5089		printf("%s: APS subpage not found for lun %ju!\n", __func__,
5090		       (uintmax_t)lun->lun);
5091		return (1);
5092	}
5093#if 0
5094	if ((softc->aps_locked_lun != 0)
5095	 && (softc->aps_locked_lun != lun->lun)) {
5096		printf("%s: attempt to lock LUN %llu when %llu is already "
5097		       "locked\n");
5098		mtx_unlock(&lun->lun_lock);
5099		mtx_unlock(&softc->ctl_lock);
5100		return (1);
5101	}
5102#endif
5103
5104	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5105		(page_index->page_len * CTL_PAGE_CURRENT));
5106
5107	if (lock != 0) {
5108		current_sp->lock_active = APS_LOCK_ACTIVE;
5109		softc->aps_locked_lun = lun->lun;
5110	} else {
5111		current_sp->lock_active = 0;
5112		softc->aps_locked_lun = 0;
5113	}
5114
5115
5116	/*
5117	 * If we're in HA mode, try to send the lock message to the other
5118	 * side.
5119	 */
5120	if (ctl_is_single == 0) {
5121		int isc_retval;
5122		union ctl_ha_msg lock_msg;
5123
5124		lock_msg.hdr.nexus = *nexus;
5125		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5126		if (lock != 0)
5127			lock_msg.aps.lock_flag = 1;
5128		else
5129			lock_msg.aps.lock_flag = 0;
5130		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5131					 sizeof(lock_msg), 0);
5132		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5133			printf("%s: APS (lock=%d) error returned from "
5134			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5135			mtx_unlock(&lun->lun_lock);
5136			mtx_unlock(&softc->ctl_lock);
5137			return (1);
5138		}
5139	}
5140
5141	mtx_unlock(&lun->lun_lock);
5142	mtx_unlock(&softc->ctl_lock);
5143
5144	return (0);
5145}
5146
5147void
5148ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5149{
5150	struct ctl_lun *lun;
5151	struct ctl_softc *softc;
5152	int i;
5153
5154	softc = control_softc;
5155
5156	lun = (struct ctl_lun *)be_lun->ctl_lun;
5157
5158	mtx_lock(&lun->lun_lock);
5159
5160	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5161		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5162
5163	mtx_unlock(&lun->lun_lock);
5164}
5165
5166/*
5167 * Backend "memory move is complete" callback for requests that never
5168 * make it down to say RAIDCore's configuration code.
5169 */
5170int
5171ctl_config_move_done(union ctl_io *io)
5172{
5173	int retval;
5174
5175	retval = CTL_RETVAL_COMPLETE;
5176
5177
5178	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5179	/*
5180	 * XXX KDM this shouldn't happen, but what if it does?
5181	 */
5182	if (io->io_hdr.io_type != CTL_IO_SCSI)
5183		panic("I/O type isn't CTL_IO_SCSI!");
5184
5185	if ((io->io_hdr.port_status == 0)
5186	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5187	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5188		io->io_hdr.status = CTL_SUCCESS;
5189	else if ((io->io_hdr.port_status != 0)
5190	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5191	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5192		/*
5193		 * For hardware error sense keys, the sense key
5194		 * specific value is defined to be a retry count,
5195		 * but we use it to pass back an internal FETD
5196		 * error code.  XXX KDM  Hopefully the FETD is only
5197		 * using 16 bits for an error code, since that's
5198		 * all the space we have in the sks field.
5199		 */
5200		ctl_set_internal_failure(&io->scsiio,
5201					 /*sks_valid*/ 1,
5202					 /*retry_count*/
5203					 io->io_hdr.port_status);
5204		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5205			free(io->scsiio.kern_data_ptr, M_CTL);
5206		ctl_done(io);
5207		goto bailout;
5208	}
5209
5210	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5211	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5212	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5213		/*
5214		 * XXX KDM just assuming a single pointer here, and not a
5215		 * S/G list.  If we start using S/G lists for config data,
5216		 * we'll need to know how to clean them up here as well.
5217		 */
5218		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5219			free(io->scsiio.kern_data_ptr, M_CTL);
5220		/* Hopefully the user has already set the status... */
5221		ctl_done(io);
5222	} else {
5223		/*
5224		 * XXX KDM now we need to continue data movement.  Some
5225		 * options:
5226		 * - call ctl_scsiio() again?  We don't do this for data
5227		 *   writes, because for those at least we know ahead of
5228		 *   time where the write will go and how long it is.  For
5229		 *   config writes, though, that information is largely
5230		 *   contained within the write itself, thus we need to
5231		 *   parse out the data again.
5232		 *
5233		 * - Call some other function once the data is in?
5234		 */
5235
5236		/*
5237		 * XXX KDM call ctl_scsiio() again for now, and check flag
5238		 * bits to see whether we're allocated or not.
5239		 */
5240		retval = ctl_scsiio(&io->scsiio);
5241	}
5242bailout:
5243	return (retval);
5244}
5245
5246/*
5247 * This gets called by a backend driver when it is done with a
5248 * data_submit method.
5249 */
5250void
5251ctl_data_submit_done(union ctl_io *io)
5252{
5253	/*
5254	 * If the IO_CONT flag is set, we need to call the supplied
5255	 * function to continue processing the I/O, instead of completing
5256	 * the I/O just yet.
5257	 *
5258	 * If there is an error, though, we don't want to keep processing.
5259	 * Instead, just send status back to the initiator.
5260	 */
5261	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5262	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5263	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5264	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5265		io->scsiio.io_cont(io);
5266		return;
5267	}
5268	ctl_done(io);
5269}
5270
5271/*
5272 * This gets called by a backend driver when it is done with a
5273 * configuration write.
5274 */
5275void
5276ctl_config_write_done(union ctl_io *io)
5277{
5278	uint8_t *buf;
5279
5280	/*
5281	 * If the IO_CONT flag is set, we need to call the supplied
5282	 * function to continue processing the I/O, instead of completing
5283	 * the I/O just yet.
5284	 *
5285	 * If there is an error, though, we don't want to keep processing.
5286	 * Instead, just send status back to the initiator.
5287	 */
5288	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5289	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5290	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5291	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5292		io->scsiio.io_cont(io);
5293		return;
5294	}
5295	/*
5296	 * Since a configuration write can be done for commands that actually
5297	 * have data allocated, like write buffer, and commands that have
5298	 * no data, like start/stop unit, we need to check here.
5299	 */
5300	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5301		buf = io->scsiio.kern_data_ptr;
5302	else
5303		buf = NULL;
5304	ctl_done(io);
5305	if (buf)
5306		free(buf, M_CTL);
5307}
5308
5309/*
5310 * SCSI release command.
5311 */
5312int
5313ctl_scsi_release(struct ctl_scsiio *ctsio)
5314{
5315	int length, longid, thirdparty_id, resv_id;
5316	struct ctl_softc *ctl_softc;
5317	struct ctl_lun *lun;
5318	uint32_t residx;
5319
5320	length = 0;
5321	resv_id = 0;
5322
5323	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5324
5325	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5326	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5327	ctl_softc = control_softc;
5328
5329	switch (ctsio->cdb[0]) {
5330	case RELEASE_10: {
5331		struct scsi_release_10 *cdb;
5332
5333		cdb = (struct scsi_release_10 *)ctsio->cdb;
5334
5335		if (cdb->byte2 & SR10_LONGID)
5336			longid = 1;
5337		else
5338			thirdparty_id = cdb->thirdparty_id;
5339
5340		resv_id = cdb->resv_id;
5341		length = scsi_2btoul(cdb->length);
5342		break;
5343	}
5344	}
5345
5346
5347	/*
5348	 * XXX KDM right now, we only support LUN reservation.  We don't
5349	 * support 3rd party reservations, or extent reservations, which
5350	 * might actually need the parameter list.  If we've gotten this
5351	 * far, we've got a LUN reservation.  Anything else got kicked out
5352	 * above.  So, according to SPC, ignore the length.
5353	 */
5354	length = 0;
5355
5356	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5357	 && (length > 0)) {
5358		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5359		ctsio->kern_data_len = length;
5360		ctsio->kern_total_len = length;
5361		ctsio->kern_data_resid = 0;
5362		ctsio->kern_rel_offset = 0;
5363		ctsio->kern_sg_entries = 0;
5364		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5365		ctsio->be_move_done = ctl_config_move_done;
5366		ctl_datamove((union ctl_io *)ctsio);
5367
5368		return (CTL_RETVAL_COMPLETE);
5369	}
5370
5371	if (length > 0)
5372		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5373
5374	mtx_lock(&lun->lun_lock);
5375
5376	/*
5377	 * According to SPC, it is not an error for an intiator to attempt
5378	 * to release a reservation on a LUN that isn't reserved, or that
5379	 * is reserved by another initiator.  The reservation can only be
5380	 * released, though, by the initiator who made it or by one of
5381	 * several reset type events.
5382	 */
5383	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5384			lun->flags &= ~CTL_LUN_RESERVED;
5385
5386	mtx_unlock(&lun->lun_lock);
5387
5388	ctsio->scsi_status = SCSI_STATUS_OK;
5389	ctsio->io_hdr.status = CTL_SUCCESS;
5390
5391	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5392		free(ctsio->kern_data_ptr, M_CTL);
5393		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5394	}
5395
5396	ctl_done((union ctl_io *)ctsio);
5397	return (CTL_RETVAL_COMPLETE);
5398}
5399
5400int
5401ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5402{
5403	int extent, thirdparty, longid;
5404	int resv_id, length;
5405	uint64_t thirdparty_id;
5406	struct ctl_softc *ctl_softc;
5407	struct ctl_lun *lun;
5408	uint32_t residx;
5409
5410	extent = 0;
5411	thirdparty = 0;
5412	longid = 0;
5413	resv_id = 0;
5414	length = 0;
5415	thirdparty_id = 0;
5416
5417	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5418
5419	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5420	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5421	ctl_softc = control_softc;
5422
5423	switch (ctsio->cdb[0]) {
5424	case RESERVE_10: {
5425		struct scsi_reserve_10 *cdb;
5426
5427		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5428
5429		if (cdb->byte2 & SR10_LONGID)
5430			longid = 1;
5431		else
5432			thirdparty_id = cdb->thirdparty_id;
5433
5434		resv_id = cdb->resv_id;
5435		length = scsi_2btoul(cdb->length);
5436		break;
5437	}
5438	}
5439
5440	/*
5441	 * XXX KDM right now, we only support LUN reservation.  We don't
5442	 * support 3rd party reservations, or extent reservations, which
5443	 * might actually need the parameter list.  If we've gotten this
5444	 * far, we've got a LUN reservation.  Anything else got kicked out
5445	 * above.  So, according to SPC, ignore the length.
5446	 */
5447	length = 0;
5448
5449	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5450	 && (length > 0)) {
5451		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5452		ctsio->kern_data_len = length;
5453		ctsio->kern_total_len = length;
5454		ctsio->kern_data_resid = 0;
5455		ctsio->kern_rel_offset = 0;
5456		ctsio->kern_sg_entries = 0;
5457		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5458		ctsio->be_move_done = ctl_config_move_done;
5459		ctl_datamove((union ctl_io *)ctsio);
5460
5461		return (CTL_RETVAL_COMPLETE);
5462	}
5463
5464	if (length > 0)
5465		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5466
5467	mtx_lock(&lun->lun_lock);
5468	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5469		ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5470		ctsio->io_hdr.status = CTL_SCSI_ERROR;
5471		goto bailout;
5472	}
5473
5474	lun->flags |= CTL_LUN_RESERVED;
5475	lun->res_idx = residx;
5476
5477	ctsio->scsi_status = SCSI_STATUS_OK;
5478	ctsio->io_hdr.status = CTL_SUCCESS;
5479
5480bailout:
5481	mtx_unlock(&lun->lun_lock);
5482
5483	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5484		free(ctsio->kern_data_ptr, M_CTL);
5485		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5486	}
5487
5488	ctl_done((union ctl_io *)ctsio);
5489	return (CTL_RETVAL_COMPLETE);
5490}
5491
5492int
5493ctl_start_stop(struct ctl_scsiio *ctsio)
5494{
5495	struct scsi_start_stop_unit *cdb;
5496	struct ctl_lun *lun;
5497	struct ctl_softc *ctl_softc;
5498	int retval;
5499
5500	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5501
5502	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5503	ctl_softc = control_softc;
5504	retval = 0;
5505
5506	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5507
5508	/*
5509	 * XXX KDM
5510	 * We don't support the immediate bit on a stop unit.  In order to
5511	 * do that, we would need to code up a way to know that a stop is
5512	 * pending, and hold off any new commands until it completes, one
5513	 * way or another.  Then we could accept or reject those commands
5514	 * depending on its status.  We would almost need to do the reverse
5515	 * of what we do below for an immediate start -- return the copy of
5516	 * the ctl_io to the FETD with status to send to the host (and to
5517	 * free the copy!) and then free the original I/O once the stop
5518	 * actually completes.  That way, the OOA queue mechanism can work
5519	 * to block commands that shouldn't proceed.  Another alternative
5520	 * would be to put the copy in the queue in place of the original,
5521	 * and return the original back to the caller.  That could be
5522	 * slightly safer..
5523	 */
5524	if ((cdb->byte2 & SSS_IMMED)
5525	 && ((cdb->how & SSS_START) == 0)) {
5526		ctl_set_invalid_field(ctsio,
5527				      /*sks_valid*/ 1,
5528				      /*command*/ 1,
5529				      /*field*/ 1,
5530				      /*bit_valid*/ 1,
5531				      /*bit*/ 0);
5532		ctl_done((union ctl_io *)ctsio);
5533		return (CTL_RETVAL_COMPLETE);
5534	}
5535
5536	if ((lun->flags & CTL_LUN_PR_RESERVED)
5537	 && ((cdb->how & SSS_START)==0)) {
5538		uint32_t residx;
5539
5540		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5541		if (!lun->per_res[residx].registered
5542		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5543
5544			ctl_set_reservation_conflict(ctsio);
5545			ctl_done((union ctl_io *)ctsio);
5546			return (CTL_RETVAL_COMPLETE);
5547		}
5548	}
5549
5550	/*
5551	 * If there is no backend on this device, we can't start or stop
5552	 * it.  In theory we shouldn't get any start/stop commands in the
5553	 * first place at this level if the LUN doesn't have a backend.
5554	 * That should get stopped by the command decode code.
5555	 */
5556	if (lun->backend == NULL) {
5557		ctl_set_invalid_opcode(ctsio);
5558		ctl_done((union ctl_io *)ctsio);
5559		return (CTL_RETVAL_COMPLETE);
5560	}
5561
5562	/*
5563	 * XXX KDM Copan-specific offline behavior.
5564	 * Figure out a reasonable way to port this?
5565	 */
5566#ifdef NEEDTOPORT
5567	mtx_lock(&lun->lun_lock);
5568
5569	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5570	 && (lun->flags & CTL_LUN_OFFLINE)) {
5571		/*
5572		 * If the LUN is offline, and the on/offline bit isn't set,
5573		 * reject the start or stop.  Otherwise, let it through.
5574		 */
5575		mtx_unlock(&lun->lun_lock);
5576		ctl_set_lun_not_ready(ctsio);
5577		ctl_done((union ctl_io *)ctsio);
5578	} else {
5579		mtx_unlock(&lun->lun_lock);
5580#endif /* NEEDTOPORT */
5581		/*
5582		 * This could be a start or a stop when we're online,
5583		 * or a stop/offline or start/online.  A start or stop when
5584		 * we're offline is covered in the case above.
5585		 */
5586		/*
5587		 * In the non-immediate case, we send the request to
5588		 * the backend and return status to the user when
5589		 * it is done.
5590		 *
5591		 * In the immediate case, we allocate a new ctl_io
5592		 * to hold a copy of the request, and send that to
5593		 * the backend.  We then set good status on the
5594		 * user's request and return it immediately.
5595		 */
5596		if (cdb->byte2 & SSS_IMMED) {
5597			union ctl_io *new_io;
5598
5599			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5600			if (new_io == NULL) {
5601				ctl_set_busy(ctsio);
5602				ctl_done((union ctl_io *)ctsio);
5603			} else {
5604				ctl_copy_io((union ctl_io *)ctsio,
5605					    new_io);
5606				retval = lun->backend->config_write(new_io);
5607				ctl_set_success(ctsio);
5608				ctl_done((union ctl_io *)ctsio);
5609			}
5610		} else {
5611			retval = lun->backend->config_write(
5612				(union ctl_io *)ctsio);
5613		}
5614#ifdef NEEDTOPORT
5615	}
5616#endif
5617	return (retval);
5618}
5619
5620/*
5621 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5622 * we don't really do anything with the LBA and length fields if the user
5623 * passes them in.  Instead we'll just flush out the cache for the entire
5624 * LUN.
5625 */
5626int
5627ctl_sync_cache(struct ctl_scsiio *ctsio)
5628{
5629	struct ctl_lun *lun;
5630	struct ctl_softc *ctl_softc;
5631	uint64_t starting_lba;
5632	uint32_t block_count;
5633	int retval;
5634
5635	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5636
5637	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5638	ctl_softc = control_softc;
5639	retval = 0;
5640
5641	switch (ctsio->cdb[0]) {
5642	case SYNCHRONIZE_CACHE: {
5643		struct scsi_sync_cache *cdb;
5644		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5645
5646		starting_lba = scsi_4btoul(cdb->begin_lba);
5647		block_count = scsi_2btoul(cdb->lb_count);
5648		break;
5649	}
5650	case SYNCHRONIZE_CACHE_16: {
5651		struct scsi_sync_cache_16 *cdb;
5652		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5653
5654		starting_lba = scsi_8btou64(cdb->begin_lba);
5655		block_count = scsi_4btoul(cdb->lb_count);
5656		break;
5657	}
5658	default:
5659		ctl_set_invalid_opcode(ctsio);
5660		ctl_done((union ctl_io *)ctsio);
5661		goto bailout;
5662		break; /* NOTREACHED */
5663	}
5664
5665	/*
5666	 * We check the LBA and length, but don't do anything with them.
5667	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5668	 * get flushed.  This check will just help satisfy anyone who wants
5669	 * to see an error for an out of range LBA.
5670	 */
5671	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5672		ctl_set_lba_out_of_range(ctsio);
5673		ctl_done((union ctl_io *)ctsio);
5674		goto bailout;
5675	}
5676
5677	/*
5678	 * If this LUN has no backend, we can't flush the cache anyway.
5679	 */
5680	if (lun->backend == NULL) {
5681		ctl_set_invalid_opcode(ctsio);
5682		ctl_done((union ctl_io *)ctsio);
5683		goto bailout;
5684	}
5685
5686	/*
5687	 * Check to see whether we're configured to send the SYNCHRONIZE
5688	 * CACHE command directly to the back end.
5689	 */
5690	mtx_lock(&lun->lun_lock);
5691	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5692	 && (++(lun->sync_count) >= lun->sync_interval)) {
5693		lun->sync_count = 0;
5694		mtx_unlock(&lun->lun_lock);
5695		retval = lun->backend->config_write((union ctl_io *)ctsio);
5696	} else {
5697		mtx_unlock(&lun->lun_lock);
5698		ctl_set_success(ctsio);
5699		ctl_done((union ctl_io *)ctsio);
5700	}
5701
5702bailout:
5703
5704	return (retval);
5705}
5706
5707int
5708ctl_format(struct ctl_scsiio *ctsio)
5709{
5710	struct scsi_format *cdb;
5711	struct ctl_lun *lun;
5712	struct ctl_softc *ctl_softc;
5713	int length, defect_list_len;
5714
5715	CTL_DEBUG_PRINT(("ctl_format\n"));
5716
5717	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5718	ctl_softc = control_softc;
5719
5720	cdb = (struct scsi_format *)ctsio->cdb;
5721
5722	length = 0;
5723	if (cdb->byte2 & SF_FMTDATA) {
5724		if (cdb->byte2 & SF_LONGLIST)
5725			length = sizeof(struct scsi_format_header_long);
5726		else
5727			length = sizeof(struct scsi_format_header_short);
5728	}
5729
5730	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5731	 && (length > 0)) {
5732		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5733		ctsio->kern_data_len = length;
5734		ctsio->kern_total_len = length;
5735		ctsio->kern_data_resid = 0;
5736		ctsio->kern_rel_offset = 0;
5737		ctsio->kern_sg_entries = 0;
5738		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5739		ctsio->be_move_done = ctl_config_move_done;
5740		ctl_datamove((union ctl_io *)ctsio);
5741
5742		return (CTL_RETVAL_COMPLETE);
5743	}
5744
5745	defect_list_len = 0;
5746
5747	if (cdb->byte2 & SF_FMTDATA) {
5748		if (cdb->byte2 & SF_LONGLIST) {
5749			struct scsi_format_header_long *header;
5750
5751			header = (struct scsi_format_header_long *)
5752				ctsio->kern_data_ptr;
5753
5754			defect_list_len = scsi_4btoul(header->defect_list_len);
5755			if (defect_list_len != 0) {
5756				ctl_set_invalid_field(ctsio,
5757						      /*sks_valid*/ 1,
5758						      /*command*/ 0,
5759						      /*field*/ 2,
5760						      /*bit_valid*/ 0,
5761						      /*bit*/ 0);
5762				goto bailout;
5763			}
5764		} else {
5765			struct scsi_format_header_short *header;
5766
5767			header = (struct scsi_format_header_short *)
5768				ctsio->kern_data_ptr;
5769
5770			defect_list_len = scsi_2btoul(header->defect_list_len);
5771			if (defect_list_len != 0) {
5772				ctl_set_invalid_field(ctsio,
5773						      /*sks_valid*/ 1,
5774						      /*command*/ 0,
5775						      /*field*/ 2,
5776						      /*bit_valid*/ 0,
5777						      /*bit*/ 0);
5778				goto bailout;
5779			}
5780		}
5781	}
5782
5783	/*
5784	 * The format command will clear out the "Medium format corrupted"
5785	 * status if set by the configuration code.  That status is really
5786	 * just a way to notify the host that we have lost the media, and
5787	 * get them to issue a command that will basically make them think
5788	 * they're blowing away the media.
5789	 */
5790	mtx_lock(&lun->lun_lock);
5791	lun->flags &= ~CTL_LUN_INOPERABLE;
5792	mtx_unlock(&lun->lun_lock);
5793
5794	ctsio->scsi_status = SCSI_STATUS_OK;
5795	ctsio->io_hdr.status = CTL_SUCCESS;
5796bailout:
5797
5798	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5799		free(ctsio->kern_data_ptr, M_CTL);
5800		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5801	}
5802
5803	ctl_done((union ctl_io *)ctsio);
5804	return (CTL_RETVAL_COMPLETE);
5805}
5806
5807int
5808ctl_read_buffer(struct ctl_scsiio *ctsio)
5809{
5810	struct scsi_read_buffer *cdb;
5811	struct ctl_lun *lun;
5812	int buffer_offset, len;
5813	static uint8_t descr[4];
5814	static uint8_t echo_descr[4] = { 0 };
5815
5816	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5817
5818	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5819	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5820
5821	if (lun->flags & CTL_LUN_PR_RESERVED) {
5822		uint32_t residx;
5823
5824		/*
5825		 * XXX KDM need a lock here.
5826		 */
5827		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5828		if ((lun->res_type == SPR_TYPE_EX_AC
5829		  && residx != lun->pr_res_idx)
5830		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5831		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5832		  && !lun->per_res[residx].registered)) {
5833			ctl_set_reservation_conflict(ctsio);
5834			ctl_done((union ctl_io *)ctsio);
5835			return (CTL_RETVAL_COMPLETE);
5836	        }
5837	}
5838
5839	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5840	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5841	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5842		ctl_set_invalid_field(ctsio,
5843				      /*sks_valid*/ 1,
5844				      /*command*/ 1,
5845				      /*field*/ 1,
5846				      /*bit_valid*/ 1,
5847				      /*bit*/ 4);
5848		ctl_done((union ctl_io *)ctsio);
5849		return (CTL_RETVAL_COMPLETE);
5850	}
5851
5852	len = scsi_3btoul(cdb->length);
5853	buffer_offset = scsi_3btoul(cdb->offset);
5854
5855	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5856		ctl_set_invalid_field(ctsio,
5857				      /*sks_valid*/ 1,
5858				      /*command*/ 1,
5859				      /*field*/ 6,
5860				      /*bit_valid*/ 0,
5861				      /*bit*/ 0);
5862		ctl_done((union ctl_io *)ctsio);
5863		return (CTL_RETVAL_COMPLETE);
5864	}
5865
5866	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5867		descr[0] = 0;
5868		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5869		ctsio->kern_data_ptr = descr;
5870		len = min(len, sizeof(descr));
5871	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5872		ctsio->kern_data_ptr = echo_descr;
5873		len = min(len, sizeof(echo_descr));
5874	} else
5875		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5876	ctsio->kern_data_len = len;
5877	ctsio->kern_total_len = len;
5878	ctsio->kern_data_resid = 0;
5879	ctsio->kern_rel_offset = 0;
5880	ctsio->kern_sg_entries = 0;
5881	ctsio->be_move_done = ctl_config_move_done;
5882	ctl_datamove((union ctl_io *)ctsio);
5883
5884	return (CTL_RETVAL_COMPLETE);
5885}
5886
5887int
5888ctl_write_buffer(struct ctl_scsiio *ctsio)
5889{
5890	struct scsi_write_buffer *cdb;
5891	struct ctl_lun *lun;
5892	int buffer_offset, len;
5893
5894	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5895
5896	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5897	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5898
5899	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5900		ctl_set_invalid_field(ctsio,
5901				      /*sks_valid*/ 1,
5902				      /*command*/ 1,
5903				      /*field*/ 1,
5904				      /*bit_valid*/ 1,
5905				      /*bit*/ 4);
5906		ctl_done((union ctl_io *)ctsio);
5907		return (CTL_RETVAL_COMPLETE);
5908	}
5909
5910	len = scsi_3btoul(cdb->length);
5911	buffer_offset = scsi_3btoul(cdb->offset);
5912
5913	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5914		ctl_set_invalid_field(ctsio,
5915				      /*sks_valid*/ 1,
5916				      /*command*/ 1,
5917				      /*field*/ 6,
5918				      /*bit_valid*/ 0,
5919				      /*bit*/ 0);
5920		ctl_done((union ctl_io *)ctsio);
5921		return (CTL_RETVAL_COMPLETE);
5922	}
5923
5924	/*
5925	 * If we've got a kernel request that hasn't been malloced yet,
5926	 * malloc it and tell the caller the data buffer is here.
5927	 */
5928	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5929		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5930		ctsio->kern_data_len = len;
5931		ctsio->kern_total_len = len;
5932		ctsio->kern_data_resid = 0;
5933		ctsio->kern_rel_offset = 0;
5934		ctsio->kern_sg_entries = 0;
5935		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5936		ctsio->be_move_done = ctl_config_move_done;
5937		ctl_datamove((union ctl_io *)ctsio);
5938
5939		return (CTL_RETVAL_COMPLETE);
5940	}
5941
5942	ctl_done((union ctl_io *)ctsio);
5943
5944	return (CTL_RETVAL_COMPLETE);
5945}
5946
5947int
5948ctl_write_same(struct ctl_scsiio *ctsio)
5949{
5950	struct ctl_lun *lun;
5951	struct ctl_lba_len_flags *lbalen;
5952	uint64_t lba;
5953	uint32_t num_blocks;
5954	int len, retval;
5955	uint8_t byte2;
5956
5957	retval = CTL_RETVAL_COMPLETE;
5958
5959	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5960
5961	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5962
5963	switch (ctsio->cdb[0]) {
5964	case WRITE_SAME_10: {
5965		struct scsi_write_same_10 *cdb;
5966
5967		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5968
5969		lba = scsi_4btoul(cdb->addr);
5970		num_blocks = scsi_2btoul(cdb->length);
5971		byte2 = cdb->byte2;
5972		break;
5973	}
5974	case WRITE_SAME_16: {
5975		struct scsi_write_same_16 *cdb;
5976
5977		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5978
5979		lba = scsi_8btou64(cdb->addr);
5980		num_blocks = scsi_4btoul(cdb->length);
5981		byte2 = cdb->byte2;
5982		break;
5983	}
5984	default:
5985		/*
5986		 * We got a command we don't support.  This shouldn't
5987		 * happen, commands should be filtered out above us.
5988		 */
5989		ctl_set_invalid_opcode(ctsio);
5990		ctl_done((union ctl_io *)ctsio);
5991
5992		return (CTL_RETVAL_COMPLETE);
5993		break; /* NOTREACHED */
5994	}
5995
5996	/* NDOB and ANCHOR flags can be used only together with UNMAP */
5997	if ((byte2 & SWS_UNMAP) == 0 &&
5998	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5999		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
6000		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
6001		ctl_done((union ctl_io *)ctsio);
6002		return (CTL_RETVAL_COMPLETE);
6003	}
6004
6005	/*
6006	 * The first check is to make sure we're in bounds, the second
6007	 * check is to catch wrap-around problems.  If the lba + num blocks
6008	 * is less than the lba, then we've wrapped around and the block
6009	 * range is invalid anyway.
6010	 */
6011	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6012	 || ((lba + num_blocks) < lba)) {
6013		ctl_set_lba_out_of_range(ctsio);
6014		ctl_done((union ctl_io *)ctsio);
6015		return (CTL_RETVAL_COMPLETE);
6016	}
6017
6018	/* Zero number of blocks means "to the last logical block" */
6019	if (num_blocks == 0) {
6020		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
6021			ctl_set_invalid_field(ctsio,
6022					      /*sks_valid*/ 0,
6023					      /*command*/ 1,
6024					      /*field*/ 0,
6025					      /*bit_valid*/ 0,
6026					      /*bit*/ 0);
6027			ctl_done((union ctl_io *)ctsio);
6028			return (CTL_RETVAL_COMPLETE);
6029		}
6030		num_blocks = (lun->be_lun->maxlba + 1) - lba;
6031	}
6032
6033	len = lun->be_lun->blocksize;
6034
6035	/*
6036	 * If we've got a kernel request that hasn't been malloced yet,
6037	 * malloc it and tell the caller the data buffer is here.
6038	 */
6039	if ((byte2 & SWS_NDOB) == 0 &&
6040	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6041		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6042		ctsio->kern_data_len = len;
6043		ctsio->kern_total_len = len;
6044		ctsio->kern_data_resid = 0;
6045		ctsio->kern_rel_offset = 0;
6046		ctsio->kern_sg_entries = 0;
6047		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6048		ctsio->be_move_done = ctl_config_move_done;
6049		ctl_datamove((union ctl_io *)ctsio);
6050
6051		return (CTL_RETVAL_COMPLETE);
6052	}
6053
6054	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6055	lbalen->lba = lba;
6056	lbalen->len = num_blocks;
6057	lbalen->flags = byte2;
6058	retval = lun->backend->config_write((union ctl_io *)ctsio);
6059
6060	return (retval);
6061}
6062
6063int
6064ctl_unmap(struct ctl_scsiio *ctsio)
6065{
6066	struct ctl_lun *lun;
6067	struct scsi_unmap *cdb;
6068	struct ctl_ptr_len_flags *ptrlen;
6069	struct scsi_unmap_header *hdr;
6070	struct scsi_unmap_desc *buf, *end, *endnz, *range;
6071	uint64_t lba;
6072	uint32_t num_blocks;
6073	int len, retval;
6074	uint8_t byte2;
6075
6076	retval = CTL_RETVAL_COMPLETE;
6077
6078	CTL_DEBUG_PRINT(("ctl_unmap\n"));
6079
6080	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6081	cdb = (struct scsi_unmap *)ctsio->cdb;
6082
6083	len = scsi_2btoul(cdb->length);
6084	byte2 = cdb->byte2;
6085
6086	/*
6087	 * If we've got a kernel request that hasn't been malloced yet,
6088	 * malloc it and tell the caller the data buffer is here.
6089	 */
6090	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6091		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6092		ctsio->kern_data_len = len;
6093		ctsio->kern_total_len = len;
6094		ctsio->kern_data_resid = 0;
6095		ctsio->kern_rel_offset = 0;
6096		ctsio->kern_sg_entries = 0;
6097		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6098		ctsio->be_move_done = ctl_config_move_done;
6099		ctl_datamove((union ctl_io *)ctsio);
6100
6101		return (CTL_RETVAL_COMPLETE);
6102	}
6103
6104	len = ctsio->kern_total_len - ctsio->kern_data_resid;
6105	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6106	if (len < sizeof (*hdr) ||
6107	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6108	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6109	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6110		ctl_set_invalid_field(ctsio,
6111				      /*sks_valid*/ 0,
6112				      /*command*/ 0,
6113				      /*field*/ 0,
6114				      /*bit_valid*/ 0,
6115				      /*bit*/ 0);
6116		ctl_done((union ctl_io *)ctsio);
6117		return (CTL_RETVAL_COMPLETE);
6118	}
6119	len = scsi_2btoul(hdr->desc_length);
6120	buf = (struct scsi_unmap_desc *)(hdr + 1);
6121	end = buf + len / sizeof(*buf);
6122
6123	endnz = buf;
6124	for (range = buf; range < end; range++) {
6125		lba = scsi_8btou64(range->lba);
6126		num_blocks = scsi_4btoul(range->length);
6127		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6128		 || ((lba + num_blocks) < lba)) {
6129			ctl_set_lba_out_of_range(ctsio);
6130			ctl_done((union ctl_io *)ctsio);
6131			return (CTL_RETVAL_COMPLETE);
6132		}
6133		if (num_blocks != 0)
6134			endnz = range + 1;
6135	}
6136
6137	/*
6138	 * Block backend can not handle zero last range.
6139	 * Filter it out and return if there is nothing left.
6140	 */
6141	len = (uint8_t *)endnz - (uint8_t *)buf;
6142	if (len == 0) {
6143		ctl_set_success(ctsio);
6144		ctl_done((union ctl_io *)ctsio);
6145		return (CTL_RETVAL_COMPLETE);
6146	}
6147
6148	mtx_lock(&lun->lun_lock);
6149	ptrlen = (struct ctl_ptr_len_flags *)
6150	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6151	ptrlen->ptr = (void *)buf;
6152	ptrlen->len = len;
6153	ptrlen->flags = byte2;
6154	ctl_check_blocked(lun);
6155	mtx_unlock(&lun->lun_lock);
6156
6157	retval = lun->backend->config_write((union ctl_io *)ctsio);
6158	return (retval);
6159}
6160
6161/*
6162 * Note that this function currently doesn't actually do anything inside
6163 * CTL to enforce things if the DQue bit is turned on.
6164 *
6165 * Also note that this function can't be used in the default case, because
6166 * the DQue bit isn't set in the changeable mask for the control mode page
6167 * anyway.  This is just here as an example for how to implement a page
6168 * handler, and a placeholder in case we want to allow the user to turn
6169 * tagged queueing on and off.
6170 *
6171 * The D_SENSE bit handling is functional, however, and will turn
6172 * descriptor sense on and off for a given LUN.
6173 */
6174int
6175ctl_control_page_handler(struct ctl_scsiio *ctsio,
6176			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6177{
6178	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6179	struct ctl_lun *lun;
6180	struct ctl_softc *softc;
6181	int set_ua;
6182	uint32_t initidx;
6183
6184	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6185	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6186	set_ua = 0;
6187
6188	user_cp = (struct scsi_control_page *)page_ptr;
6189	current_cp = (struct scsi_control_page *)
6190		(page_index->page_data + (page_index->page_len *
6191		CTL_PAGE_CURRENT));
6192	saved_cp = (struct scsi_control_page *)
6193		(page_index->page_data + (page_index->page_len *
6194		CTL_PAGE_SAVED));
6195
6196	softc = control_softc;
6197
6198	mtx_lock(&lun->lun_lock);
6199	if (((current_cp->rlec & SCP_DSENSE) == 0)
6200	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6201		/*
6202		 * Descriptor sense is currently turned off and the user
6203		 * wants to turn it on.
6204		 */
6205		current_cp->rlec |= SCP_DSENSE;
6206		saved_cp->rlec |= SCP_DSENSE;
6207		lun->flags |= CTL_LUN_SENSE_DESC;
6208		set_ua = 1;
6209	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6210		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6211		/*
6212		 * Descriptor sense is currently turned on, and the user
6213		 * wants to turn it off.
6214		 */
6215		current_cp->rlec &= ~SCP_DSENSE;
6216		saved_cp->rlec &= ~SCP_DSENSE;
6217		lun->flags &= ~CTL_LUN_SENSE_DESC;
6218		set_ua = 1;
6219	}
6220	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6221	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6222		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6223		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6224		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6225		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6226		set_ua = 1;
6227	}
6228	if ((current_cp->eca_and_aen & SCP_SWP) !=
6229	    (user_cp->eca_and_aen & SCP_SWP)) {
6230		current_cp->eca_and_aen &= ~SCP_SWP;
6231		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6232		saved_cp->eca_and_aen &= ~SCP_SWP;
6233		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6234		set_ua = 1;
6235	}
6236	if (set_ua != 0) {
6237		int i;
6238		/*
6239		 * Let other initiators know that the mode
6240		 * parameters for this LUN have changed.
6241		 */
6242		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6243			if (i == initidx)
6244				continue;
6245
6246			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6247		}
6248	}
6249	mtx_unlock(&lun->lun_lock);
6250
6251	return (0);
6252}
6253
6254int
6255ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6256		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6257{
6258	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6259	struct ctl_lun *lun;
6260	int set_ua;
6261	uint32_t initidx;
6262
6263	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6264	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6265	set_ua = 0;
6266
6267	user_cp = (struct scsi_caching_page *)page_ptr;
6268	current_cp = (struct scsi_caching_page *)
6269		(page_index->page_data + (page_index->page_len *
6270		CTL_PAGE_CURRENT));
6271	saved_cp = (struct scsi_caching_page *)
6272		(page_index->page_data + (page_index->page_len *
6273		CTL_PAGE_SAVED));
6274
6275	mtx_lock(&lun->lun_lock);
6276	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6277	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6278		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6279		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6280		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6281		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6282		set_ua = 1;
6283	}
6284	if (set_ua != 0) {
6285		int i;
6286		/*
6287		 * Let other initiators know that the mode
6288		 * parameters for this LUN have changed.
6289		 */
6290		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6291			if (i == initidx)
6292				continue;
6293
6294			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6295		}
6296	}
6297	mtx_unlock(&lun->lun_lock);
6298
6299	return (0);
6300}
6301
6302int
6303ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6304		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6305{
6306	return (0);
6307}
6308
6309int
6310ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6311			   struct ctl_page_index *page_index, int pc)
6312{
6313	struct copan_power_subpage *page;
6314
6315	page = (struct copan_power_subpage *)page_index->page_data +
6316		(page_index->page_len * pc);
6317
6318	switch (pc) {
6319	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6320		/*
6321		 * We don't update the changable bits for this page.
6322		 */
6323		break;
6324	case SMS_PAGE_CTRL_CURRENT >> 6:
6325	case SMS_PAGE_CTRL_DEFAULT >> 6:
6326	case SMS_PAGE_CTRL_SAVED >> 6:
6327#ifdef NEEDTOPORT
6328		ctl_update_power_subpage(page);
6329#endif
6330		break;
6331	default:
6332#ifdef NEEDTOPORT
6333		EPRINT(0, "Invalid PC %d!!", pc);
6334#endif
6335		break;
6336	}
6337	return (0);
6338}
6339
6340
6341int
6342ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6343		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6344{
6345	struct copan_aps_subpage *user_sp;
6346	struct copan_aps_subpage *current_sp;
6347	union ctl_modepage_info *modepage_info;
6348	struct ctl_softc *softc;
6349	struct ctl_lun *lun;
6350	int retval;
6351
6352	retval = CTL_RETVAL_COMPLETE;
6353	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6354		     (page_index->page_len * CTL_PAGE_CURRENT));
6355	softc = control_softc;
6356	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6357
6358	user_sp = (struct copan_aps_subpage *)page_ptr;
6359
6360	modepage_info = (union ctl_modepage_info *)
6361		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6362
6363	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6364	modepage_info->header.subpage = page_index->subpage;
6365	modepage_info->aps.lock_active = user_sp->lock_active;
6366
6367	mtx_lock(&softc->ctl_lock);
6368
6369	/*
6370	 * If there is a request to lock the LUN and another LUN is locked
6371	 * this is an error. If the requested LUN is already locked ignore
6372	 * the request. If no LUN is locked attempt to lock it.
6373	 * if there is a request to unlock the LUN and the LUN is currently
6374	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6375	 * if another LUN is locked or no LUN is locked.
6376	 */
6377	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6378		if (softc->aps_locked_lun == lun->lun) {
6379			/*
6380			 * This LUN is already locked, so we're done.
6381			 */
6382			retval = CTL_RETVAL_COMPLETE;
6383		} else if (softc->aps_locked_lun == 0) {
6384			/*
6385			 * No one has the lock, pass the request to the
6386			 * backend.
6387			 */
6388			retval = lun->backend->config_write(
6389				(union ctl_io *)ctsio);
6390		} else {
6391			/*
6392			 * Someone else has the lock, throw out the request.
6393			 */
6394			ctl_set_already_locked(ctsio);
6395			free(ctsio->kern_data_ptr, M_CTL);
6396			ctl_done((union ctl_io *)ctsio);
6397
6398			/*
6399			 * Set the return value so that ctl_do_mode_select()
6400			 * won't try to complete the command.  We already
6401			 * completed it here.
6402			 */
6403			retval = CTL_RETVAL_ERROR;
6404		}
6405	} else if (softc->aps_locked_lun == lun->lun) {
6406		/*
6407		 * This LUN is locked, so pass the unlock request to the
6408		 * backend.
6409		 */
6410		retval = lun->backend->config_write((union ctl_io *)ctsio);
6411	}
6412	mtx_unlock(&softc->ctl_lock);
6413
6414	return (retval);
6415}
6416
6417int
6418ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6419				struct ctl_page_index *page_index,
6420				uint8_t *page_ptr)
6421{
6422	uint8_t *c;
6423	int i;
6424
6425	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6426	ctl_time_io_secs =
6427		(c[0] << 8) |
6428		(c[1] << 0) |
6429		0;
6430	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6431	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6432	printf("page data:");
6433	for (i=0; i<8; i++)
6434		printf(" %.2x",page_ptr[i]);
6435	printf("\n");
6436	return (0);
6437}
6438
6439int
6440ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6441			       struct ctl_page_index *page_index,
6442			       int pc)
6443{
6444	struct copan_debugconf_subpage *page;
6445
6446	page = (struct copan_debugconf_subpage *)page_index->page_data +
6447		(page_index->page_len * pc);
6448
6449	switch (pc) {
6450	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6451	case SMS_PAGE_CTRL_DEFAULT >> 6:
6452	case SMS_PAGE_CTRL_SAVED >> 6:
6453		/*
6454		 * We don't update the changable or default bits for this page.
6455		 */
6456		break;
6457	case SMS_PAGE_CTRL_CURRENT >> 6:
6458		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6459		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6460		break;
6461	default:
6462#ifdef NEEDTOPORT
6463		EPRINT(0, "Invalid PC %d!!", pc);
6464#endif /* NEEDTOPORT */
6465		break;
6466	}
6467	return (0);
6468}
6469
6470
6471static int
6472ctl_do_mode_select(union ctl_io *io)
6473{
6474	struct scsi_mode_page_header *page_header;
6475	struct ctl_page_index *page_index;
6476	struct ctl_scsiio *ctsio;
6477	int control_dev, page_len;
6478	int page_len_offset, page_len_size;
6479	union ctl_modepage_info *modepage_info;
6480	struct ctl_lun *lun;
6481	int *len_left, *len_used;
6482	int retval, i;
6483
6484	ctsio = &io->scsiio;
6485	page_index = NULL;
6486	page_len = 0;
6487	retval = CTL_RETVAL_COMPLETE;
6488
6489	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6490
6491	if (lun->be_lun->lun_type != T_DIRECT)
6492		control_dev = 1;
6493	else
6494		control_dev = 0;
6495
6496	modepage_info = (union ctl_modepage_info *)
6497		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6498	len_left = &modepage_info->header.len_left;
6499	len_used = &modepage_info->header.len_used;
6500
6501do_next_page:
6502
6503	page_header = (struct scsi_mode_page_header *)
6504		(ctsio->kern_data_ptr + *len_used);
6505
6506	if (*len_left == 0) {
6507		free(ctsio->kern_data_ptr, M_CTL);
6508		ctl_set_success(ctsio);
6509		ctl_done((union ctl_io *)ctsio);
6510		return (CTL_RETVAL_COMPLETE);
6511	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6512
6513		free(ctsio->kern_data_ptr, M_CTL);
6514		ctl_set_param_len_error(ctsio);
6515		ctl_done((union ctl_io *)ctsio);
6516		return (CTL_RETVAL_COMPLETE);
6517
6518	} else if ((page_header->page_code & SMPH_SPF)
6519		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6520
6521		free(ctsio->kern_data_ptr, M_CTL);
6522		ctl_set_param_len_error(ctsio);
6523		ctl_done((union ctl_io *)ctsio);
6524		return (CTL_RETVAL_COMPLETE);
6525	}
6526
6527
6528	/*
6529	 * XXX KDM should we do something with the block descriptor?
6530	 */
6531	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6532
6533		if ((control_dev != 0)
6534		 && (lun->mode_pages.index[i].page_flags &
6535		     CTL_PAGE_FLAG_DISK_ONLY))
6536			continue;
6537
6538		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6539		    (page_header->page_code & SMPH_PC_MASK))
6540			continue;
6541
6542		/*
6543		 * If neither page has a subpage code, then we've got a
6544		 * match.
6545		 */
6546		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6547		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6548			page_index = &lun->mode_pages.index[i];
6549			page_len = page_header->page_length;
6550			break;
6551		}
6552
6553		/*
6554		 * If both pages have subpages, then the subpage numbers
6555		 * have to match.
6556		 */
6557		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6558		  && (page_header->page_code & SMPH_SPF)) {
6559			struct scsi_mode_page_header_sp *sph;
6560
6561			sph = (struct scsi_mode_page_header_sp *)page_header;
6562
6563			if (lun->mode_pages.index[i].subpage ==
6564			    sph->subpage) {
6565				page_index = &lun->mode_pages.index[i];
6566				page_len = scsi_2btoul(sph->page_length);
6567				break;
6568			}
6569		}
6570	}
6571
6572	/*
6573	 * If we couldn't find the page, or if we don't have a mode select
6574	 * handler for it, send back an error to the user.
6575	 */
6576	if ((page_index == NULL)
6577	 || (page_index->select_handler == NULL)) {
6578		ctl_set_invalid_field(ctsio,
6579				      /*sks_valid*/ 1,
6580				      /*command*/ 0,
6581				      /*field*/ *len_used,
6582				      /*bit_valid*/ 0,
6583				      /*bit*/ 0);
6584		free(ctsio->kern_data_ptr, M_CTL);
6585		ctl_done((union ctl_io *)ctsio);
6586		return (CTL_RETVAL_COMPLETE);
6587	}
6588
6589	if (page_index->page_code & SMPH_SPF) {
6590		page_len_offset = 2;
6591		page_len_size = 2;
6592	} else {
6593		page_len_size = 1;
6594		page_len_offset = 1;
6595	}
6596
6597	/*
6598	 * If the length the initiator gives us isn't the one we specify in
6599	 * the mode page header, or if they didn't specify enough data in
6600	 * the CDB to avoid truncating this page, kick out the request.
6601	 */
6602	if ((page_len != (page_index->page_len - page_len_offset -
6603			  page_len_size))
6604	 || (*len_left < page_index->page_len)) {
6605
6606
6607		ctl_set_invalid_field(ctsio,
6608				      /*sks_valid*/ 1,
6609				      /*command*/ 0,
6610				      /*field*/ *len_used + page_len_offset,
6611				      /*bit_valid*/ 0,
6612				      /*bit*/ 0);
6613		free(ctsio->kern_data_ptr, M_CTL);
6614		ctl_done((union ctl_io *)ctsio);
6615		return (CTL_RETVAL_COMPLETE);
6616	}
6617
6618	/*
6619	 * Run through the mode page, checking to make sure that the bits
6620	 * the user changed are actually legal for him to change.
6621	 */
6622	for (i = 0; i < page_index->page_len; i++) {
6623		uint8_t *user_byte, *change_mask, *current_byte;
6624		int bad_bit;
6625		int j;
6626
6627		user_byte = (uint8_t *)page_header + i;
6628		change_mask = page_index->page_data +
6629			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6630		current_byte = page_index->page_data +
6631			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6632
6633		/*
6634		 * Check to see whether the user set any bits in this byte
6635		 * that he is not allowed to set.
6636		 */
6637		if ((*user_byte & ~(*change_mask)) ==
6638		    (*current_byte & ~(*change_mask)))
6639			continue;
6640
6641		/*
6642		 * Go through bit by bit to determine which one is illegal.
6643		 */
6644		bad_bit = 0;
6645		for (j = 7; j >= 0; j--) {
6646			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6647			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6648				bad_bit = i;
6649				break;
6650			}
6651		}
6652		ctl_set_invalid_field(ctsio,
6653				      /*sks_valid*/ 1,
6654				      /*command*/ 0,
6655				      /*field*/ *len_used + i,
6656				      /*bit_valid*/ 1,
6657				      /*bit*/ bad_bit);
6658		free(ctsio->kern_data_ptr, M_CTL);
6659		ctl_done((union ctl_io *)ctsio);
6660		return (CTL_RETVAL_COMPLETE);
6661	}
6662
6663	/*
6664	 * Decrement these before we call the page handler, since we may
6665	 * end up getting called back one way or another before the handler
6666	 * returns to this context.
6667	 */
6668	*len_left -= page_index->page_len;
6669	*len_used += page_index->page_len;
6670
6671	retval = page_index->select_handler(ctsio, page_index,
6672					    (uint8_t *)page_header);
6673
6674	/*
6675	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6676	 * wait until this queued command completes to finish processing
6677	 * the mode page.  If it returns anything other than
6678	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6679	 * already set the sense information, freed the data pointer, and
6680	 * completed the io for us.
6681	 */
6682	if (retval != CTL_RETVAL_COMPLETE)
6683		goto bailout_no_done;
6684
6685	/*
6686	 * If the initiator sent us more than one page, parse the next one.
6687	 */
6688	if (*len_left > 0)
6689		goto do_next_page;
6690
6691	ctl_set_success(ctsio);
6692	free(ctsio->kern_data_ptr, M_CTL);
6693	ctl_done((union ctl_io *)ctsio);
6694
6695bailout_no_done:
6696
6697	return (CTL_RETVAL_COMPLETE);
6698
6699}
6700
6701int
6702ctl_mode_select(struct ctl_scsiio *ctsio)
6703{
6704	int param_len, pf, sp;
6705	int header_size, bd_len;
6706	int len_left, len_used;
6707	struct ctl_page_index *page_index;
6708	struct ctl_lun *lun;
6709	int control_dev, page_len;
6710	union ctl_modepage_info *modepage_info;
6711	int retval;
6712
6713	pf = 0;
6714	sp = 0;
6715	page_len = 0;
6716	len_used = 0;
6717	len_left = 0;
6718	retval = 0;
6719	bd_len = 0;
6720	page_index = NULL;
6721
6722	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6723
6724	if (lun->be_lun->lun_type != T_DIRECT)
6725		control_dev = 1;
6726	else
6727		control_dev = 0;
6728
6729	switch (ctsio->cdb[0]) {
6730	case MODE_SELECT_6: {
6731		struct scsi_mode_select_6 *cdb;
6732
6733		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6734
6735		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6736		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6737
6738		param_len = cdb->length;
6739		header_size = sizeof(struct scsi_mode_header_6);
6740		break;
6741	}
6742	case MODE_SELECT_10: {
6743		struct scsi_mode_select_10 *cdb;
6744
6745		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6746
6747		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6748		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6749
6750		param_len = scsi_2btoul(cdb->length);
6751		header_size = sizeof(struct scsi_mode_header_10);
6752		break;
6753	}
6754	default:
6755		ctl_set_invalid_opcode(ctsio);
6756		ctl_done((union ctl_io *)ctsio);
6757		return (CTL_RETVAL_COMPLETE);
6758		break; /* NOTREACHED */
6759	}
6760
6761	/*
6762	 * From SPC-3:
6763	 * "A parameter list length of zero indicates that the Data-Out Buffer
6764	 * shall be empty. This condition shall not be considered as an error."
6765	 */
6766	if (param_len == 0) {
6767		ctl_set_success(ctsio);
6768		ctl_done((union ctl_io *)ctsio);
6769		return (CTL_RETVAL_COMPLETE);
6770	}
6771
6772	/*
6773	 * Since we'll hit this the first time through, prior to
6774	 * allocation, we don't need to free a data buffer here.
6775	 */
6776	if (param_len < header_size) {
6777		ctl_set_param_len_error(ctsio);
6778		ctl_done((union ctl_io *)ctsio);
6779		return (CTL_RETVAL_COMPLETE);
6780	}
6781
6782	/*
6783	 * Allocate the data buffer and grab the user's data.  In theory,
6784	 * we shouldn't have to sanity check the parameter list length here
6785	 * because the maximum size is 64K.  We should be able to malloc
6786	 * that much without too many problems.
6787	 */
6788	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6789		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6790		ctsio->kern_data_len = param_len;
6791		ctsio->kern_total_len = param_len;
6792		ctsio->kern_data_resid = 0;
6793		ctsio->kern_rel_offset = 0;
6794		ctsio->kern_sg_entries = 0;
6795		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6796		ctsio->be_move_done = ctl_config_move_done;
6797		ctl_datamove((union ctl_io *)ctsio);
6798
6799		return (CTL_RETVAL_COMPLETE);
6800	}
6801
6802	switch (ctsio->cdb[0]) {
6803	case MODE_SELECT_6: {
6804		struct scsi_mode_header_6 *mh6;
6805
6806		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6807		bd_len = mh6->blk_desc_len;
6808		break;
6809	}
6810	case MODE_SELECT_10: {
6811		struct scsi_mode_header_10 *mh10;
6812
6813		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6814		bd_len = scsi_2btoul(mh10->blk_desc_len);
6815		break;
6816	}
6817	default:
6818		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6819		break;
6820	}
6821
6822	if (param_len < (header_size + bd_len)) {
6823		free(ctsio->kern_data_ptr, M_CTL);
6824		ctl_set_param_len_error(ctsio);
6825		ctl_done((union ctl_io *)ctsio);
6826		return (CTL_RETVAL_COMPLETE);
6827	}
6828
6829	/*
6830	 * Set the IO_CONT flag, so that if this I/O gets passed to
6831	 * ctl_config_write_done(), it'll get passed back to
6832	 * ctl_do_mode_select() for further processing, or completion if
6833	 * we're all done.
6834	 */
6835	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6836	ctsio->io_cont = ctl_do_mode_select;
6837
6838	modepage_info = (union ctl_modepage_info *)
6839		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6840
6841	memset(modepage_info, 0, sizeof(*modepage_info));
6842
6843	len_left = param_len - header_size - bd_len;
6844	len_used = header_size + bd_len;
6845
6846	modepage_info->header.len_left = len_left;
6847	modepage_info->header.len_used = len_used;
6848
6849	return (ctl_do_mode_select((union ctl_io *)ctsio));
6850}
6851
6852int
6853ctl_mode_sense(struct ctl_scsiio *ctsio)
6854{
6855	struct ctl_lun *lun;
6856	int pc, page_code, dbd, llba, subpage;
6857	int alloc_len, page_len, header_len, total_len;
6858	struct scsi_mode_block_descr *block_desc;
6859	struct ctl_page_index *page_index;
6860	int control_dev;
6861
6862	dbd = 0;
6863	llba = 0;
6864	block_desc = NULL;
6865	page_index = NULL;
6866
6867	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6868
6869	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6870
6871	if (lun->be_lun->lun_type != T_DIRECT)
6872		control_dev = 1;
6873	else
6874		control_dev = 0;
6875
6876	if (lun->flags & CTL_LUN_PR_RESERVED) {
6877		uint32_t residx;
6878
6879		/*
6880		 * XXX KDM need a lock here.
6881		 */
6882		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6883		if ((lun->res_type == SPR_TYPE_EX_AC
6884		  && residx != lun->pr_res_idx)
6885		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6886		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6887		  && !lun->per_res[residx].registered)) {
6888			ctl_set_reservation_conflict(ctsio);
6889			ctl_done((union ctl_io *)ctsio);
6890			return (CTL_RETVAL_COMPLETE);
6891		}
6892	}
6893
6894	switch (ctsio->cdb[0]) {
6895	case MODE_SENSE_6: {
6896		struct scsi_mode_sense_6 *cdb;
6897
6898		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6899
6900		header_len = sizeof(struct scsi_mode_hdr_6);
6901		if (cdb->byte2 & SMS_DBD)
6902			dbd = 1;
6903		else
6904			header_len += sizeof(struct scsi_mode_block_descr);
6905
6906		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6907		page_code = cdb->page & SMS_PAGE_CODE;
6908		subpage = cdb->subpage;
6909		alloc_len = cdb->length;
6910		break;
6911	}
6912	case MODE_SENSE_10: {
6913		struct scsi_mode_sense_10 *cdb;
6914
6915		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6916
6917		header_len = sizeof(struct scsi_mode_hdr_10);
6918
6919		if (cdb->byte2 & SMS_DBD)
6920			dbd = 1;
6921		else
6922			header_len += sizeof(struct scsi_mode_block_descr);
6923		if (cdb->byte2 & SMS10_LLBAA)
6924			llba = 1;
6925		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6926		page_code = cdb->page & SMS_PAGE_CODE;
6927		subpage = cdb->subpage;
6928		alloc_len = scsi_2btoul(cdb->length);
6929		break;
6930	}
6931	default:
6932		ctl_set_invalid_opcode(ctsio);
6933		ctl_done((union ctl_io *)ctsio);
6934		return (CTL_RETVAL_COMPLETE);
6935		break; /* NOTREACHED */
6936	}
6937
6938	/*
6939	 * We have to make a first pass through to calculate the size of
6940	 * the pages that match the user's query.  Then we allocate enough
6941	 * memory to hold it, and actually copy the data into the buffer.
6942	 */
6943	switch (page_code) {
6944	case SMS_ALL_PAGES_PAGE: {
6945		int i;
6946
6947		page_len = 0;
6948
6949		/*
6950		 * At the moment, values other than 0 and 0xff here are
6951		 * reserved according to SPC-3.
6952		 */
6953		if ((subpage != SMS_SUBPAGE_PAGE_0)
6954		 && (subpage != SMS_SUBPAGE_ALL)) {
6955			ctl_set_invalid_field(ctsio,
6956					      /*sks_valid*/ 1,
6957					      /*command*/ 1,
6958					      /*field*/ 3,
6959					      /*bit_valid*/ 0,
6960					      /*bit*/ 0);
6961			ctl_done((union ctl_io *)ctsio);
6962			return (CTL_RETVAL_COMPLETE);
6963		}
6964
6965		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6966			if ((control_dev != 0)
6967			 && (lun->mode_pages.index[i].page_flags &
6968			     CTL_PAGE_FLAG_DISK_ONLY))
6969				continue;
6970
6971			/*
6972			 * We don't use this subpage if the user didn't
6973			 * request all subpages.
6974			 */
6975			if ((lun->mode_pages.index[i].subpage != 0)
6976			 && (subpage == SMS_SUBPAGE_PAGE_0))
6977				continue;
6978
6979#if 0
6980			printf("found page %#x len %d\n",
6981			       lun->mode_pages.index[i].page_code &
6982			       SMPH_PC_MASK,
6983			       lun->mode_pages.index[i].page_len);
6984#endif
6985			page_len += lun->mode_pages.index[i].page_len;
6986		}
6987		break;
6988	}
6989	default: {
6990		int i;
6991
6992		page_len = 0;
6993
6994		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6995			/* Look for the right page code */
6996			if ((lun->mode_pages.index[i].page_code &
6997			     SMPH_PC_MASK) != page_code)
6998				continue;
6999
7000			/* Look for the right subpage or the subpage wildcard*/
7001			if ((lun->mode_pages.index[i].subpage != subpage)
7002			 && (subpage != SMS_SUBPAGE_ALL))
7003				continue;
7004
7005			/* Make sure the page is supported for this dev type */
7006			if ((control_dev != 0)
7007			 && (lun->mode_pages.index[i].page_flags &
7008			     CTL_PAGE_FLAG_DISK_ONLY))
7009				continue;
7010
7011#if 0
7012			printf("found page %#x len %d\n",
7013			       lun->mode_pages.index[i].page_code &
7014			       SMPH_PC_MASK,
7015			       lun->mode_pages.index[i].page_len);
7016#endif
7017
7018			page_len += lun->mode_pages.index[i].page_len;
7019		}
7020
7021		if (page_len == 0) {
7022			ctl_set_invalid_field(ctsio,
7023					      /*sks_valid*/ 1,
7024					      /*command*/ 1,
7025					      /*field*/ 2,
7026					      /*bit_valid*/ 1,
7027					      /*bit*/ 5);
7028			ctl_done((union ctl_io *)ctsio);
7029			return (CTL_RETVAL_COMPLETE);
7030		}
7031		break;
7032	}
7033	}
7034
7035	total_len = header_len + page_len;
7036#if 0
7037	printf("header_len = %d, page_len = %d, total_len = %d\n",
7038	       header_len, page_len, total_len);
7039#endif
7040
7041	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7042	ctsio->kern_sg_entries = 0;
7043	ctsio->kern_data_resid = 0;
7044	ctsio->kern_rel_offset = 0;
7045	if (total_len < alloc_len) {
7046		ctsio->residual = alloc_len - total_len;
7047		ctsio->kern_data_len = total_len;
7048		ctsio->kern_total_len = total_len;
7049	} else {
7050		ctsio->residual = 0;
7051		ctsio->kern_data_len = alloc_len;
7052		ctsio->kern_total_len = alloc_len;
7053	}
7054
7055	switch (ctsio->cdb[0]) {
7056	case MODE_SENSE_6: {
7057		struct scsi_mode_hdr_6 *header;
7058
7059		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7060
7061		header->datalen = ctl_min(total_len - 1, 254);
7062		if (control_dev == 0) {
7063			header->dev_specific = 0x10; /* DPOFUA */
7064			if ((lun->flags & CTL_LUN_READONLY) ||
7065			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
7066			    .eca_and_aen & SCP_SWP) != 0)
7067				    header->dev_specific |= 0x80; /* WP */
7068		}
7069		if (dbd)
7070			header->block_descr_len = 0;
7071		else
7072			header->block_descr_len =
7073				sizeof(struct scsi_mode_block_descr);
7074		block_desc = (struct scsi_mode_block_descr *)&header[1];
7075		break;
7076	}
7077	case MODE_SENSE_10: {
7078		struct scsi_mode_hdr_10 *header;
7079		int datalen;
7080
7081		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7082
7083		datalen = ctl_min(total_len - 2, 65533);
7084		scsi_ulto2b(datalen, header->datalen);
7085		if (control_dev == 0) {
7086			header->dev_specific = 0x10; /* DPOFUA */
7087			if ((lun->flags & CTL_LUN_READONLY) ||
7088			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
7089			    .eca_and_aen & SCP_SWP) != 0)
7090				    header->dev_specific |= 0x80; /* WP */
7091		}
7092		if (dbd)
7093			scsi_ulto2b(0, header->block_descr_len);
7094		else
7095			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7096				    header->block_descr_len);
7097		block_desc = (struct scsi_mode_block_descr *)&header[1];
7098		break;
7099	}
7100	default:
7101		panic("invalid CDB type %#x", ctsio->cdb[0]);
7102		break; /* NOTREACHED */
7103	}
7104
7105	/*
7106	 * If we've got a disk, use its blocksize in the block
7107	 * descriptor.  Otherwise, just set it to 0.
7108	 */
7109	if (dbd == 0) {
7110		if (control_dev == 0)
7111			scsi_ulto3b(lun->be_lun->blocksize,
7112				    block_desc->block_len);
7113		else
7114			scsi_ulto3b(0, block_desc->block_len);
7115	}
7116
7117	switch (page_code) {
7118	case SMS_ALL_PAGES_PAGE: {
7119		int i, data_used;
7120
7121		data_used = header_len;
7122		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7123			struct ctl_page_index *page_index;
7124
7125			page_index = &lun->mode_pages.index[i];
7126
7127			if ((control_dev != 0)
7128			 && (page_index->page_flags &
7129			    CTL_PAGE_FLAG_DISK_ONLY))
7130				continue;
7131
7132			/*
7133			 * We don't use this subpage if the user didn't
7134			 * request all subpages.  We already checked (above)
7135			 * to make sure the user only specified a subpage
7136			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7137			 */
7138			if ((page_index->subpage != 0)
7139			 && (subpage == SMS_SUBPAGE_PAGE_0))
7140				continue;
7141
7142			/*
7143			 * Call the handler, if it exists, to update the
7144			 * page to the latest values.
7145			 */
7146			if (page_index->sense_handler != NULL)
7147				page_index->sense_handler(ctsio, page_index,pc);
7148
7149			memcpy(ctsio->kern_data_ptr + data_used,
7150			       page_index->page_data +
7151			       (page_index->page_len * pc),
7152			       page_index->page_len);
7153			data_used += page_index->page_len;
7154		}
7155		break;
7156	}
7157	default: {
7158		int i, data_used;
7159
7160		data_used = header_len;
7161
7162		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7163			struct ctl_page_index *page_index;
7164
7165			page_index = &lun->mode_pages.index[i];
7166
7167			/* Look for the right page code */
7168			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7169				continue;
7170
7171			/* Look for the right subpage or the subpage wildcard*/
7172			if ((page_index->subpage != subpage)
7173			 && (subpage != SMS_SUBPAGE_ALL))
7174				continue;
7175
7176			/* Make sure the page is supported for this dev type */
7177			if ((control_dev != 0)
7178			 && (page_index->page_flags &
7179			     CTL_PAGE_FLAG_DISK_ONLY))
7180				continue;
7181
7182			/*
7183			 * Call the handler, if it exists, to update the
7184			 * page to the latest values.
7185			 */
7186			if (page_index->sense_handler != NULL)
7187				page_index->sense_handler(ctsio, page_index,pc);
7188
7189			memcpy(ctsio->kern_data_ptr + data_used,
7190			       page_index->page_data +
7191			       (page_index->page_len * pc),
7192			       page_index->page_len);
7193			data_used += page_index->page_len;
7194		}
7195		break;
7196	}
7197	}
7198
7199	ctsio->scsi_status = SCSI_STATUS_OK;
7200
7201	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7202	ctsio->be_move_done = ctl_config_move_done;
7203	ctl_datamove((union ctl_io *)ctsio);
7204
7205	return (CTL_RETVAL_COMPLETE);
7206}
7207
7208int
7209ctl_read_capacity(struct ctl_scsiio *ctsio)
7210{
7211	struct scsi_read_capacity *cdb;
7212	struct scsi_read_capacity_data *data;
7213	struct ctl_lun *lun;
7214	uint32_t lba;
7215
7216	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7217
7218	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7219
7220	lba = scsi_4btoul(cdb->addr);
7221	if (((cdb->pmi & SRC_PMI) == 0)
7222	 && (lba != 0)) {
7223		ctl_set_invalid_field(/*ctsio*/ ctsio,
7224				      /*sks_valid*/ 1,
7225				      /*command*/ 1,
7226				      /*field*/ 2,
7227				      /*bit_valid*/ 0,
7228				      /*bit*/ 0);
7229		ctl_done((union ctl_io *)ctsio);
7230		return (CTL_RETVAL_COMPLETE);
7231	}
7232
7233	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7234
7235	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7236	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7237	ctsio->residual = 0;
7238	ctsio->kern_data_len = sizeof(*data);
7239	ctsio->kern_total_len = sizeof(*data);
7240	ctsio->kern_data_resid = 0;
7241	ctsio->kern_rel_offset = 0;
7242	ctsio->kern_sg_entries = 0;
7243
7244	/*
7245	 * If the maximum LBA is greater than 0xfffffffe, the user must
7246	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7247	 * serivce action set.
7248	 */
7249	if (lun->be_lun->maxlba > 0xfffffffe)
7250		scsi_ulto4b(0xffffffff, data->addr);
7251	else
7252		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7253
7254	/*
7255	 * XXX KDM this may not be 512 bytes...
7256	 */
7257	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7258
7259	ctsio->scsi_status = SCSI_STATUS_OK;
7260
7261	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7262	ctsio->be_move_done = ctl_config_move_done;
7263	ctl_datamove((union ctl_io *)ctsio);
7264
7265	return (CTL_RETVAL_COMPLETE);
7266}
7267
7268int
7269ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7270{
7271	struct scsi_read_capacity_16 *cdb;
7272	struct scsi_read_capacity_data_long *data;
7273	struct ctl_lun *lun;
7274	uint64_t lba;
7275	uint32_t alloc_len;
7276
7277	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7278
7279	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7280
7281	alloc_len = scsi_4btoul(cdb->alloc_len);
7282	lba = scsi_8btou64(cdb->addr);
7283
7284	if ((cdb->reladr & SRC16_PMI)
7285	 && (lba != 0)) {
7286		ctl_set_invalid_field(/*ctsio*/ ctsio,
7287				      /*sks_valid*/ 1,
7288				      /*command*/ 1,
7289				      /*field*/ 2,
7290				      /*bit_valid*/ 0,
7291				      /*bit*/ 0);
7292		ctl_done((union ctl_io *)ctsio);
7293		return (CTL_RETVAL_COMPLETE);
7294	}
7295
7296	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7297
7298	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7299	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7300
7301	if (sizeof(*data) < alloc_len) {
7302		ctsio->residual = alloc_len - sizeof(*data);
7303		ctsio->kern_data_len = sizeof(*data);
7304		ctsio->kern_total_len = sizeof(*data);
7305	} else {
7306		ctsio->residual = 0;
7307		ctsio->kern_data_len = alloc_len;
7308		ctsio->kern_total_len = alloc_len;
7309	}
7310	ctsio->kern_data_resid = 0;
7311	ctsio->kern_rel_offset = 0;
7312	ctsio->kern_sg_entries = 0;
7313
7314	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7315	/* XXX KDM this may not be 512 bytes... */
7316	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7317	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7318	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7319	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7320		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7321
7322	ctsio->scsi_status = SCSI_STATUS_OK;
7323
7324	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7325	ctsio->be_move_done = ctl_config_move_done;
7326	ctl_datamove((union ctl_io *)ctsio);
7327
7328	return (CTL_RETVAL_COMPLETE);
7329}
7330
7331int
7332ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7333{
7334	struct scsi_maintenance_in *cdb;
7335	int retval;
7336	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7337	int num_target_port_groups, num_target_ports, single;
7338	struct ctl_lun *lun;
7339	struct ctl_softc *softc;
7340	struct ctl_port *port;
7341	struct scsi_target_group_data *rtg_ptr;
7342	struct scsi_target_group_data_extended *rtg_ext_ptr;
7343	struct scsi_target_port_group_descriptor *tpg_desc;
7344
7345	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7346
7347	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7348	softc = control_softc;
7349	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7350
7351	retval = CTL_RETVAL_COMPLETE;
7352
7353	switch (cdb->byte2 & STG_PDF_MASK) {
7354	case STG_PDF_LENGTH:
7355		ext = 0;
7356		break;
7357	case STG_PDF_EXTENDED:
7358		ext = 1;
7359		break;
7360	default:
7361		ctl_set_invalid_field(/*ctsio*/ ctsio,
7362				      /*sks_valid*/ 1,
7363				      /*command*/ 1,
7364				      /*field*/ 2,
7365				      /*bit_valid*/ 1,
7366				      /*bit*/ 5);
7367		ctl_done((union ctl_io *)ctsio);
7368		return(retval);
7369	}
7370
7371	single = ctl_is_single;
7372	if (single)
7373		num_target_port_groups = 1;
7374	else
7375		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7376	num_target_ports = 0;
7377	mtx_lock(&softc->ctl_lock);
7378	STAILQ_FOREACH(port, &softc->port_list, links) {
7379		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7380			continue;
7381		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7382			continue;
7383		num_target_ports++;
7384	}
7385	mtx_unlock(&softc->ctl_lock);
7386
7387	if (ext)
7388		total_len = sizeof(struct scsi_target_group_data_extended);
7389	else
7390		total_len = sizeof(struct scsi_target_group_data);
7391	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7392		num_target_port_groups +
7393	    sizeof(struct scsi_target_port_descriptor) *
7394		num_target_ports * num_target_port_groups;
7395
7396	alloc_len = scsi_4btoul(cdb->length);
7397
7398	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7399
7400	ctsio->kern_sg_entries = 0;
7401
7402	if (total_len < alloc_len) {
7403		ctsio->residual = alloc_len - total_len;
7404		ctsio->kern_data_len = total_len;
7405		ctsio->kern_total_len = total_len;
7406	} else {
7407		ctsio->residual = 0;
7408		ctsio->kern_data_len = alloc_len;
7409		ctsio->kern_total_len = alloc_len;
7410	}
7411	ctsio->kern_data_resid = 0;
7412	ctsio->kern_rel_offset = 0;
7413
7414	if (ext) {
7415		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7416		    ctsio->kern_data_ptr;
7417		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7418		rtg_ext_ptr->format_type = 0x10;
7419		rtg_ext_ptr->implicit_transition_time = 0;
7420		tpg_desc = &rtg_ext_ptr->groups[0];
7421	} else {
7422		rtg_ptr = (struct scsi_target_group_data *)
7423		    ctsio->kern_data_ptr;
7424		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7425		tpg_desc = &rtg_ptr->groups[0];
7426	}
7427
7428	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7429	mtx_lock(&softc->ctl_lock);
7430	for (g = 0; g < num_target_port_groups; g++) {
7431		if (g == pg)
7432			tpg_desc->pref_state = TPG_PRIMARY |
7433			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7434		else
7435			tpg_desc->pref_state =
7436			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7437		tpg_desc->support = TPG_AO_SUP;
7438		if (!single)
7439			tpg_desc->support |= TPG_AN_SUP;
7440		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7441		tpg_desc->status = TPG_IMPLICIT;
7442		pc = 0;
7443		STAILQ_FOREACH(port, &softc->port_list, links) {
7444			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7445				continue;
7446			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7447			    CTL_MAX_LUNS)
7448				continue;
7449			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7450			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7451			    relative_target_port_identifier);
7452			pc++;
7453		}
7454		tpg_desc->target_port_count = pc;
7455		tpg_desc = (struct scsi_target_port_group_descriptor *)
7456		    &tpg_desc->descriptors[pc];
7457	}
7458	mtx_unlock(&softc->ctl_lock);
7459
7460	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7461	ctsio->be_move_done = ctl_config_move_done;
7462
7463	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7464			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7465			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7466			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7467			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7468
7469	ctl_datamove((union ctl_io *)ctsio);
7470	return(retval);
7471}
7472
7473int
7474ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7475{
7476	struct ctl_lun *lun;
7477	struct scsi_report_supported_opcodes *cdb;
7478	const struct ctl_cmd_entry *entry, *sentry;
7479	struct scsi_report_supported_opcodes_all *all;
7480	struct scsi_report_supported_opcodes_descr *descr;
7481	struct scsi_report_supported_opcodes_one *one;
7482	int retval;
7483	int alloc_len, total_len;
7484	int opcode, service_action, i, j, num;
7485
7486	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7487
7488	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7489	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7490
7491	retval = CTL_RETVAL_COMPLETE;
7492
7493	opcode = cdb->requested_opcode;
7494	service_action = scsi_2btoul(cdb->requested_service_action);
7495	switch (cdb->options & RSO_OPTIONS_MASK) {
7496	case RSO_OPTIONS_ALL:
7497		num = 0;
7498		for (i = 0; i < 256; i++) {
7499			entry = &ctl_cmd_table[i];
7500			if (entry->flags & CTL_CMD_FLAG_SA5) {
7501				for (j = 0; j < 32; j++) {
7502					sentry = &((const struct ctl_cmd_entry *)
7503					    entry->execute)[j];
7504					if (ctl_cmd_applicable(
7505					    lun->be_lun->lun_type, sentry))
7506						num++;
7507				}
7508			} else {
7509				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7510				    entry))
7511					num++;
7512			}
7513		}
7514		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7515		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7516		break;
7517	case RSO_OPTIONS_OC:
7518		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7519			ctl_set_invalid_field(/*ctsio*/ ctsio,
7520					      /*sks_valid*/ 1,
7521					      /*command*/ 1,
7522					      /*field*/ 2,
7523					      /*bit_valid*/ 1,
7524					      /*bit*/ 2);
7525			ctl_done((union ctl_io *)ctsio);
7526			return (CTL_RETVAL_COMPLETE);
7527		}
7528		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7529		break;
7530	case RSO_OPTIONS_OC_SA:
7531		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7532		    service_action >= 32) {
7533			ctl_set_invalid_field(/*ctsio*/ ctsio,
7534					      /*sks_valid*/ 1,
7535					      /*command*/ 1,
7536					      /*field*/ 2,
7537					      /*bit_valid*/ 1,
7538					      /*bit*/ 2);
7539			ctl_done((union ctl_io *)ctsio);
7540			return (CTL_RETVAL_COMPLETE);
7541		}
7542		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7543		break;
7544	default:
7545		ctl_set_invalid_field(/*ctsio*/ ctsio,
7546				      /*sks_valid*/ 1,
7547				      /*command*/ 1,
7548				      /*field*/ 2,
7549				      /*bit_valid*/ 1,
7550				      /*bit*/ 2);
7551		ctl_done((union ctl_io *)ctsio);
7552		return (CTL_RETVAL_COMPLETE);
7553	}
7554
7555	alloc_len = scsi_4btoul(cdb->length);
7556
7557	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7558
7559	ctsio->kern_sg_entries = 0;
7560
7561	if (total_len < alloc_len) {
7562		ctsio->residual = alloc_len - total_len;
7563		ctsio->kern_data_len = total_len;
7564		ctsio->kern_total_len = total_len;
7565	} else {
7566		ctsio->residual = 0;
7567		ctsio->kern_data_len = alloc_len;
7568		ctsio->kern_total_len = alloc_len;
7569	}
7570	ctsio->kern_data_resid = 0;
7571	ctsio->kern_rel_offset = 0;
7572
7573	switch (cdb->options & RSO_OPTIONS_MASK) {
7574	case RSO_OPTIONS_ALL:
7575		all = (struct scsi_report_supported_opcodes_all *)
7576		    ctsio->kern_data_ptr;
7577		num = 0;
7578		for (i = 0; i < 256; i++) {
7579			entry = &ctl_cmd_table[i];
7580			if (entry->flags & CTL_CMD_FLAG_SA5) {
7581				for (j = 0; j < 32; j++) {
7582					sentry = &((const struct ctl_cmd_entry *)
7583					    entry->execute)[j];
7584					if (!ctl_cmd_applicable(
7585					    lun->be_lun->lun_type, sentry))
7586						continue;
7587					descr = &all->descr[num++];
7588					descr->opcode = i;
7589					scsi_ulto2b(j, descr->service_action);
7590					descr->flags = RSO_SERVACTV;
7591					scsi_ulto2b(sentry->length,
7592					    descr->cdb_length);
7593				}
7594			} else {
7595				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7596				    entry))
7597					continue;
7598				descr = &all->descr[num++];
7599				descr->opcode = i;
7600				scsi_ulto2b(0, descr->service_action);
7601				descr->flags = 0;
7602				scsi_ulto2b(entry->length, descr->cdb_length);
7603			}
7604		}
7605		scsi_ulto4b(
7606		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7607		    all->length);
7608		break;
7609	case RSO_OPTIONS_OC:
7610		one = (struct scsi_report_supported_opcodes_one *)
7611		    ctsio->kern_data_ptr;
7612		entry = &ctl_cmd_table[opcode];
7613		goto fill_one;
7614	case RSO_OPTIONS_OC_SA:
7615		one = (struct scsi_report_supported_opcodes_one *)
7616		    ctsio->kern_data_ptr;
7617		entry = &ctl_cmd_table[opcode];
7618		entry = &((const struct ctl_cmd_entry *)
7619		    entry->execute)[service_action];
7620fill_one:
7621		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7622			one->support = 3;
7623			scsi_ulto2b(entry->length, one->cdb_length);
7624			one->cdb_usage[0] = opcode;
7625			memcpy(&one->cdb_usage[1], entry->usage,
7626			    entry->length - 1);
7627		} else
7628			one->support = 1;
7629		break;
7630	}
7631
7632	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7633	ctsio->be_move_done = ctl_config_move_done;
7634
7635	ctl_datamove((union ctl_io *)ctsio);
7636	return(retval);
7637}
7638
7639int
7640ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7641{
7642	struct ctl_lun *lun;
7643	struct scsi_report_supported_tmf *cdb;
7644	struct scsi_report_supported_tmf_data *data;
7645	int retval;
7646	int alloc_len, total_len;
7647
7648	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7649
7650	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7651	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7652
7653	retval = CTL_RETVAL_COMPLETE;
7654
7655	total_len = sizeof(struct scsi_report_supported_tmf_data);
7656	alloc_len = scsi_4btoul(cdb->length);
7657
7658	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7659
7660	ctsio->kern_sg_entries = 0;
7661
7662	if (total_len < alloc_len) {
7663		ctsio->residual = alloc_len - total_len;
7664		ctsio->kern_data_len = total_len;
7665		ctsio->kern_total_len = total_len;
7666	} else {
7667		ctsio->residual = 0;
7668		ctsio->kern_data_len = alloc_len;
7669		ctsio->kern_total_len = alloc_len;
7670	}
7671	ctsio->kern_data_resid = 0;
7672	ctsio->kern_rel_offset = 0;
7673
7674	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7675	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7676	data->byte2 |= RST_ITNRS;
7677
7678	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7679	ctsio->be_move_done = ctl_config_move_done;
7680
7681	ctl_datamove((union ctl_io *)ctsio);
7682	return (retval);
7683}
7684
7685int
7686ctl_report_timestamp(struct ctl_scsiio *ctsio)
7687{
7688	struct ctl_lun *lun;
7689	struct scsi_report_timestamp *cdb;
7690	struct scsi_report_timestamp_data *data;
7691	struct timeval tv;
7692	int64_t timestamp;
7693	int retval;
7694	int alloc_len, total_len;
7695
7696	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7697
7698	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7699	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7700
7701	retval = CTL_RETVAL_COMPLETE;
7702
7703	total_len = sizeof(struct scsi_report_timestamp_data);
7704	alloc_len = scsi_4btoul(cdb->length);
7705
7706	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7707
7708	ctsio->kern_sg_entries = 0;
7709
7710	if (total_len < alloc_len) {
7711		ctsio->residual = alloc_len - total_len;
7712		ctsio->kern_data_len = total_len;
7713		ctsio->kern_total_len = total_len;
7714	} else {
7715		ctsio->residual = 0;
7716		ctsio->kern_data_len = alloc_len;
7717		ctsio->kern_total_len = alloc_len;
7718	}
7719	ctsio->kern_data_resid = 0;
7720	ctsio->kern_rel_offset = 0;
7721
7722	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7723	scsi_ulto2b(sizeof(*data) - 2, data->length);
7724	data->origin = RTS_ORIG_OUTSIDE;
7725	getmicrotime(&tv);
7726	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7727	scsi_ulto4b(timestamp >> 16, data->timestamp);
7728	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7729
7730	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7731	ctsio->be_move_done = ctl_config_move_done;
7732
7733	ctl_datamove((union ctl_io *)ctsio);
7734	return (retval);
7735}
7736
7737int
7738ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7739{
7740	struct scsi_per_res_in *cdb;
7741	int alloc_len, total_len = 0;
7742	/* struct scsi_per_res_in_rsrv in_data; */
7743	struct ctl_lun *lun;
7744	struct ctl_softc *softc;
7745
7746	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7747
7748	softc = control_softc;
7749
7750	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7751
7752	alloc_len = scsi_2btoul(cdb->length);
7753
7754	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7755
7756retry:
7757	mtx_lock(&lun->lun_lock);
7758	switch (cdb->action) {
7759	case SPRI_RK: /* read keys */
7760		total_len = sizeof(struct scsi_per_res_in_keys) +
7761			lun->pr_key_count *
7762			sizeof(struct scsi_per_res_key);
7763		break;
7764	case SPRI_RR: /* read reservation */
7765		if (lun->flags & CTL_LUN_PR_RESERVED)
7766			total_len = sizeof(struct scsi_per_res_in_rsrv);
7767		else
7768			total_len = sizeof(struct scsi_per_res_in_header);
7769		break;
7770	case SPRI_RC: /* report capabilities */
7771		total_len = sizeof(struct scsi_per_res_cap);
7772		break;
7773	case SPRI_RS: /* read full status */
7774		total_len = sizeof(struct scsi_per_res_in_header) +
7775		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7776		    lun->pr_key_count;
7777		break;
7778	default:
7779		panic("Invalid PR type %x", cdb->action);
7780	}
7781	mtx_unlock(&lun->lun_lock);
7782
7783	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7784
7785	if (total_len < alloc_len) {
7786		ctsio->residual = alloc_len - total_len;
7787		ctsio->kern_data_len = total_len;
7788		ctsio->kern_total_len = total_len;
7789	} else {
7790		ctsio->residual = 0;
7791		ctsio->kern_data_len = alloc_len;
7792		ctsio->kern_total_len = alloc_len;
7793	}
7794
7795	ctsio->kern_data_resid = 0;
7796	ctsio->kern_rel_offset = 0;
7797	ctsio->kern_sg_entries = 0;
7798
7799	mtx_lock(&lun->lun_lock);
7800	switch (cdb->action) {
7801	case SPRI_RK: { // read keys
7802        struct scsi_per_res_in_keys *res_keys;
7803		int i, key_count;
7804
7805		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7806
7807		/*
7808		 * We had to drop the lock to allocate our buffer, which
7809		 * leaves time for someone to come in with another
7810		 * persistent reservation.  (That is unlikely, though,
7811		 * since this should be the only persistent reservation
7812		 * command active right now.)
7813		 */
7814		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7815		    (lun->pr_key_count *
7816		     sizeof(struct scsi_per_res_key)))){
7817			mtx_unlock(&lun->lun_lock);
7818			free(ctsio->kern_data_ptr, M_CTL);
7819			printf("%s: reservation length changed, retrying\n",
7820			       __func__);
7821			goto retry;
7822		}
7823
7824		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7825
7826		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7827			     lun->pr_key_count, res_keys->header.length);
7828
7829		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7830			if (!lun->per_res[i].registered)
7831				continue;
7832
7833			/*
7834			 * We used lun->pr_key_count to calculate the
7835			 * size to allocate.  If it turns out the number of
7836			 * initiators with the registered flag set is
7837			 * larger than that (i.e. they haven't been kept in
7838			 * sync), we've got a problem.
7839			 */
7840			if (key_count >= lun->pr_key_count) {
7841#ifdef NEEDTOPORT
7842				csevent_log(CSC_CTL | CSC_SHELF_SW |
7843					    CTL_PR_ERROR,
7844					    csevent_LogType_Fault,
7845					    csevent_AlertLevel_Yellow,
7846					    csevent_FRU_ShelfController,
7847					    csevent_FRU_Firmware,
7848				        csevent_FRU_Unknown,
7849					    "registered keys %d >= key "
7850					    "count %d", key_count,
7851					    lun->pr_key_count);
7852#endif
7853				key_count++;
7854				continue;
7855			}
7856			memcpy(res_keys->keys[key_count].key,
7857			       lun->per_res[i].res_key.key,
7858			       ctl_min(sizeof(res_keys->keys[key_count].key),
7859			       sizeof(lun->per_res[i].res_key)));
7860			key_count++;
7861		}
7862		break;
7863	}
7864	case SPRI_RR: { // read reservation
7865		struct scsi_per_res_in_rsrv *res;
7866		int tmp_len, header_only;
7867
7868		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7869
7870		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7871
7872		if (lun->flags & CTL_LUN_PR_RESERVED)
7873		{
7874			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7875			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7876				    res->header.length);
7877			header_only = 0;
7878		} else {
7879			tmp_len = sizeof(struct scsi_per_res_in_header);
7880			scsi_ulto4b(0, res->header.length);
7881			header_only = 1;
7882		}
7883
7884		/*
7885		 * We had to drop the lock to allocate our buffer, which
7886		 * leaves time for someone to come in with another
7887		 * persistent reservation.  (That is unlikely, though,
7888		 * since this should be the only persistent reservation
7889		 * command active right now.)
7890		 */
7891		if (tmp_len != total_len) {
7892			mtx_unlock(&lun->lun_lock);
7893			free(ctsio->kern_data_ptr, M_CTL);
7894			printf("%s: reservation status changed, retrying\n",
7895			       __func__);
7896			goto retry;
7897		}
7898
7899		/*
7900		 * No reservation held, so we're done.
7901		 */
7902		if (header_only != 0)
7903			break;
7904
7905		/*
7906		 * If the registration is an All Registrants type, the key
7907		 * is 0, since it doesn't really matter.
7908		 */
7909		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7910			memcpy(res->data.reservation,
7911			       &lun->per_res[lun->pr_res_idx].res_key,
7912			       sizeof(struct scsi_per_res_key));
7913		}
7914		res->data.scopetype = lun->res_type;
7915		break;
7916	}
7917	case SPRI_RC:     //report capabilities
7918	{
7919		struct scsi_per_res_cap *res_cap;
7920		uint16_t type_mask;
7921
7922		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7923		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7924		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7925		type_mask = SPRI_TM_WR_EX_AR |
7926			    SPRI_TM_EX_AC_RO |
7927			    SPRI_TM_WR_EX_RO |
7928			    SPRI_TM_EX_AC |
7929			    SPRI_TM_WR_EX |
7930			    SPRI_TM_EX_AC_AR;
7931		scsi_ulto2b(type_mask, res_cap->type_mask);
7932		break;
7933	}
7934	case SPRI_RS: { // read full status
7935		struct scsi_per_res_in_full *res_status;
7936		struct scsi_per_res_in_full_desc *res_desc;
7937		struct ctl_port *port;
7938		int i, len;
7939
7940		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7941
7942		/*
7943		 * We had to drop the lock to allocate our buffer, which
7944		 * leaves time for someone to come in with another
7945		 * persistent reservation.  (That is unlikely, though,
7946		 * since this should be the only persistent reservation
7947		 * command active right now.)
7948		 */
7949		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7950		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7951		     lun->pr_key_count)){
7952			mtx_unlock(&lun->lun_lock);
7953			free(ctsio->kern_data_ptr, M_CTL);
7954			printf("%s: reservation length changed, retrying\n",
7955			       __func__);
7956			goto retry;
7957		}
7958
7959		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7960
7961		res_desc = &res_status->desc[0];
7962		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7963			if (!lun->per_res[i].registered)
7964				continue;
7965
7966			memcpy(&res_desc->res_key, &lun->per_res[i].res_key.key,
7967			    sizeof(res_desc->res_key));
7968			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7969			    (lun->pr_res_idx == i ||
7970			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7971				res_desc->flags = SPRI_FULL_R_HOLDER;
7972				res_desc->scopetype = lun->res_type;
7973			}
7974			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7975			    res_desc->rel_trgt_port_id);
7976			len = 0;
7977			port = softc->ctl_ports[
7978			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7979			if (port != NULL)
7980				len = ctl_create_iid(port,
7981				    i % CTL_MAX_INIT_PER_PORT,
7982				    res_desc->transport_id);
7983			scsi_ulto4b(len, res_desc->additional_length);
7984			res_desc = (struct scsi_per_res_in_full_desc *)
7985			    &res_desc->transport_id[len];
7986		}
7987		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7988		    res_status->header.length);
7989		break;
7990	}
7991	default:
7992		/*
7993		 * This is a bug, because we just checked for this above,
7994		 * and should have returned an error.
7995		 */
7996		panic("Invalid PR type %x", cdb->action);
7997		break; /* NOTREACHED */
7998	}
7999	mtx_unlock(&lun->lun_lock);
8000
8001	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8002	ctsio->be_move_done = ctl_config_move_done;
8003
8004	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
8005			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
8006			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
8007			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
8008			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
8009
8010	ctl_datamove((union ctl_io *)ctsio);
8011
8012	return (CTL_RETVAL_COMPLETE);
8013}
8014
8015/*
8016 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
8017 * it should return.
8018 */
8019static int
8020ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
8021		uint64_t sa_res_key, uint8_t type, uint32_t residx,
8022		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
8023		struct scsi_per_res_out_parms* param)
8024{
8025	union ctl_ha_msg persis_io;
8026	int retval, i;
8027	int isc_retval;
8028
8029	retval = 0;
8030
8031	mtx_lock(&lun->lun_lock);
8032	if (sa_res_key == 0) {
8033		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8034			/* validate scope and type */
8035			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8036			     SPR_LU_SCOPE) {
8037				mtx_unlock(&lun->lun_lock);
8038				ctl_set_invalid_field(/*ctsio*/ ctsio,
8039						      /*sks_valid*/ 1,
8040						      /*command*/ 1,
8041						      /*field*/ 2,
8042						      /*bit_valid*/ 1,
8043						      /*bit*/ 4);
8044				ctl_done((union ctl_io *)ctsio);
8045				return (1);
8046			}
8047
8048		        if (type>8 || type==2 || type==4 || type==0) {
8049				mtx_unlock(&lun->lun_lock);
8050				ctl_set_invalid_field(/*ctsio*/ ctsio,
8051       	           				      /*sks_valid*/ 1,
8052						      /*command*/ 1,
8053						      /*field*/ 2,
8054						      /*bit_valid*/ 1,
8055						      /*bit*/ 0);
8056				ctl_done((union ctl_io *)ctsio);
8057				return (1);
8058		        }
8059
8060			/* temporarily unregister this nexus */
8061			lun->per_res[residx].registered = 0;
8062
8063			/*
8064			 * Unregister everybody else and build UA for
8065			 * them
8066			 */
8067			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8068				if (lun->per_res[i].registered == 0)
8069					continue;
8070
8071				if (!persis_offset
8072				 && i <CTL_MAX_INITIATORS)
8073					lun->pending_ua[i] |=
8074						CTL_UA_REG_PREEMPT;
8075				else if (persis_offset
8076				      && i >= persis_offset)
8077					lun->pending_ua[i-persis_offset] |=
8078						CTL_UA_REG_PREEMPT;
8079				lun->per_res[i].registered = 0;
8080				memset(&lun->per_res[i].res_key, 0,
8081				       sizeof(struct scsi_per_res_key));
8082			}
8083			lun->per_res[residx].registered = 1;
8084			lun->pr_key_count = 1;
8085			lun->res_type = type;
8086			if (lun->res_type != SPR_TYPE_WR_EX_AR
8087			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8088				lun->pr_res_idx = residx;
8089
8090			/* send msg to other side */
8091			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8092			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8093			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8094			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8095			persis_io.pr.pr_info.res_type = type;
8096			memcpy(persis_io.pr.pr_info.sa_res_key,
8097			       param->serv_act_res_key,
8098			       sizeof(param->serv_act_res_key));
8099			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8100			     &persis_io, sizeof(persis_io), 0)) >
8101			     CTL_HA_STATUS_SUCCESS) {
8102				printf("CTL:Persis Out error returned "
8103				       "from ctl_ha_msg_send %d\n",
8104				       isc_retval);
8105			}
8106		} else {
8107			/* not all registrants */
8108			mtx_unlock(&lun->lun_lock);
8109			free(ctsio->kern_data_ptr, M_CTL);
8110			ctl_set_invalid_field(ctsio,
8111					      /*sks_valid*/ 1,
8112					      /*command*/ 0,
8113					      /*field*/ 8,
8114					      /*bit_valid*/ 0,
8115					      /*bit*/ 0);
8116			ctl_done((union ctl_io *)ctsio);
8117			return (1);
8118		}
8119	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8120		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8121		int found = 0;
8122
8123		if (res_key == sa_res_key) {
8124			/* special case */
8125			/*
8126			 * The spec implies this is not good but doesn't
8127			 * say what to do. There are two choices either
8128			 * generate a res conflict or check condition
8129			 * with illegal field in parameter data. Since
8130			 * that is what is done when the sa_res_key is
8131			 * zero I'll take that approach since this has
8132			 * to do with the sa_res_key.
8133			 */
8134			mtx_unlock(&lun->lun_lock);
8135			free(ctsio->kern_data_ptr, M_CTL);
8136			ctl_set_invalid_field(ctsio,
8137					      /*sks_valid*/ 1,
8138					      /*command*/ 0,
8139					      /*field*/ 8,
8140					      /*bit_valid*/ 0,
8141					      /*bit*/ 0);
8142			ctl_done((union ctl_io *)ctsio);
8143			return (1);
8144		}
8145
8146		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8147			if (lun->per_res[i].registered
8148			 && memcmp(param->serv_act_res_key,
8149			    lun->per_res[i].res_key.key,
8150			    sizeof(struct scsi_per_res_key)) != 0)
8151				continue;
8152
8153			found = 1;
8154			lun->per_res[i].registered = 0;
8155			memset(&lun->per_res[i].res_key, 0,
8156			       sizeof(struct scsi_per_res_key));
8157			lun->pr_key_count--;
8158
8159			if (!persis_offset && i < CTL_MAX_INITIATORS)
8160				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8161			else if (persis_offset && i >= persis_offset)
8162				lun->pending_ua[i-persis_offset] |=
8163					CTL_UA_REG_PREEMPT;
8164		}
8165		if (!found) {
8166			mtx_unlock(&lun->lun_lock);
8167			free(ctsio->kern_data_ptr, M_CTL);
8168			ctl_set_reservation_conflict(ctsio);
8169			ctl_done((union ctl_io *)ctsio);
8170			return (CTL_RETVAL_COMPLETE);
8171		}
8172		/* send msg to other side */
8173		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8174		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8175		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8176		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8177		persis_io.pr.pr_info.res_type = type;
8178		memcpy(persis_io.pr.pr_info.sa_res_key,
8179		       param->serv_act_res_key,
8180		       sizeof(param->serv_act_res_key));
8181		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8182		     &persis_io, sizeof(persis_io), 0)) >
8183		     CTL_HA_STATUS_SUCCESS) {
8184			printf("CTL:Persis Out error returned from "
8185			       "ctl_ha_msg_send %d\n", isc_retval);
8186		}
8187	} else {
8188		/* Reserved but not all registrants */
8189		/* sa_res_key is res holder */
8190		if (memcmp(param->serv_act_res_key,
8191                   lun->per_res[lun->pr_res_idx].res_key.key,
8192                   sizeof(struct scsi_per_res_key)) == 0) {
8193			/* validate scope and type */
8194			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8195			     SPR_LU_SCOPE) {
8196				mtx_unlock(&lun->lun_lock);
8197				ctl_set_invalid_field(/*ctsio*/ ctsio,
8198						      /*sks_valid*/ 1,
8199						      /*command*/ 1,
8200						      /*field*/ 2,
8201						      /*bit_valid*/ 1,
8202						      /*bit*/ 4);
8203				ctl_done((union ctl_io *)ctsio);
8204				return (1);
8205			}
8206
8207			if (type>8 || type==2 || type==4 || type==0) {
8208				mtx_unlock(&lun->lun_lock);
8209				ctl_set_invalid_field(/*ctsio*/ ctsio,
8210						      /*sks_valid*/ 1,
8211						      /*command*/ 1,
8212						      /*field*/ 2,
8213						      /*bit_valid*/ 1,
8214						      /*bit*/ 0);
8215				ctl_done((union ctl_io *)ctsio);
8216				return (1);
8217			}
8218
8219			/*
8220			 * Do the following:
8221			 * if sa_res_key != res_key remove all
8222			 * registrants w/sa_res_key and generate UA
8223			 * for these registrants(Registrations
8224			 * Preempted) if it wasn't an exclusive
8225			 * reservation generate UA(Reservations
8226			 * Preempted) for all other registered nexuses
8227			 * if the type has changed. Establish the new
8228			 * reservation and holder. If res_key and
8229			 * sa_res_key are the same do the above
8230			 * except don't unregister the res holder.
8231			 */
8232
8233			/*
8234			 * Temporarily unregister so it won't get
8235			 * removed or UA generated
8236			 */
8237			lun->per_res[residx].registered = 0;
8238			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8239				if (lun->per_res[i].registered == 0)
8240					continue;
8241
8242				if (memcmp(param->serv_act_res_key,
8243				    lun->per_res[i].res_key.key,
8244				    sizeof(struct scsi_per_res_key)) == 0) {
8245					lun->per_res[i].registered = 0;
8246					memset(&lun->per_res[i].res_key,
8247					       0,
8248					       sizeof(struct scsi_per_res_key));
8249					lun->pr_key_count--;
8250
8251					if (!persis_offset
8252					 && i < CTL_MAX_INITIATORS)
8253						lun->pending_ua[i] |=
8254							CTL_UA_REG_PREEMPT;
8255					else if (persis_offset
8256					      && i >= persis_offset)
8257						lun->pending_ua[i-persis_offset] |=
8258						  CTL_UA_REG_PREEMPT;
8259				} else if (type != lun->res_type
8260					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8261					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8262						if (!persis_offset
8263						 && i < CTL_MAX_INITIATORS)
8264							lun->pending_ua[i] |=
8265							CTL_UA_RES_RELEASE;
8266						else if (persis_offset
8267						      && i >= persis_offset)
8268							lun->pending_ua[
8269							i-persis_offset] |=
8270							CTL_UA_RES_RELEASE;
8271				}
8272			}
8273			lun->per_res[residx].registered = 1;
8274			lun->res_type = type;
8275			if (lun->res_type != SPR_TYPE_WR_EX_AR
8276			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8277				lun->pr_res_idx = residx;
8278			else
8279				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8280
8281			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8282			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8283			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8284			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8285			persis_io.pr.pr_info.res_type = type;
8286			memcpy(persis_io.pr.pr_info.sa_res_key,
8287			       param->serv_act_res_key,
8288			       sizeof(param->serv_act_res_key));
8289			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8290			     &persis_io, sizeof(persis_io), 0)) >
8291			     CTL_HA_STATUS_SUCCESS) {
8292				printf("CTL:Persis Out error returned "
8293				       "from ctl_ha_msg_send %d\n",
8294				       isc_retval);
8295			}
8296		} else {
8297			/*
8298			 * sa_res_key is not the res holder just
8299			 * remove registrants
8300			 */
8301			int found=0;
8302
8303			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8304				if (memcmp(param->serv_act_res_key,
8305				    lun->per_res[i].res_key.key,
8306				    sizeof(struct scsi_per_res_key)) != 0)
8307					continue;
8308
8309				found = 1;
8310				lun->per_res[i].registered = 0;
8311				memset(&lun->per_res[i].res_key, 0,
8312				       sizeof(struct scsi_per_res_key));
8313				lun->pr_key_count--;
8314
8315				if (!persis_offset
8316				 && i < CTL_MAX_INITIATORS)
8317					lun->pending_ua[i] |=
8318						CTL_UA_REG_PREEMPT;
8319				else if (persis_offset
8320				      && i >= persis_offset)
8321					lun->pending_ua[i-persis_offset] |=
8322						CTL_UA_REG_PREEMPT;
8323			}
8324
8325			if (!found) {
8326				mtx_unlock(&lun->lun_lock);
8327				free(ctsio->kern_data_ptr, M_CTL);
8328				ctl_set_reservation_conflict(ctsio);
8329				ctl_done((union ctl_io *)ctsio);
8330		        	return (1);
8331			}
8332			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8333			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8334			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8335			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8336			persis_io.pr.pr_info.res_type = type;
8337			memcpy(persis_io.pr.pr_info.sa_res_key,
8338			       param->serv_act_res_key,
8339			       sizeof(param->serv_act_res_key));
8340			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8341			     &persis_io, sizeof(persis_io), 0)) >
8342			     CTL_HA_STATUS_SUCCESS) {
8343				printf("CTL:Persis Out error returned "
8344				       "from ctl_ha_msg_send %d\n",
8345				isc_retval);
8346			}
8347		}
8348	}
8349
8350	lun->PRGeneration++;
8351	mtx_unlock(&lun->lun_lock);
8352
8353	return (retval);
8354}
8355
8356static void
8357ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8358{
8359	int i;
8360
8361	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8362	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8363	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8364		   msg->pr.pr_info.sa_res_key,
8365		   sizeof(struct scsi_per_res_key)) != 0) {
8366		uint64_t sa_res_key;
8367		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8368
8369		if (sa_res_key == 0) {
8370			/* temporarily unregister this nexus */
8371			lun->per_res[msg->pr.pr_info.residx].registered = 0;
8372
8373			/*
8374			 * Unregister everybody else and build UA for
8375			 * them
8376			 */
8377			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8378				if (lun->per_res[i].registered == 0)
8379					continue;
8380
8381				if (!persis_offset
8382				 && i < CTL_MAX_INITIATORS)
8383					lun->pending_ua[i] |=
8384						CTL_UA_REG_PREEMPT;
8385				else if (persis_offset && i >= persis_offset)
8386					lun->pending_ua[i - persis_offset] |=
8387						CTL_UA_REG_PREEMPT;
8388				lun->per_res[i].registered = 0;
8389				memset(&lun->per_res[i].res_key, 0,
8390				       sizeof(struct scsi_per_res_key));
8391			}
8392
8393			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8394			lun->pr_key_count = 1;
8395			lun->res_type = msg->pr.pr_info.res_type;
8396			if (lun->res_type != SPR_TYPE_WR_EX_AR
8397			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8398				lun->pr_res_idx = msg->pr.pr_info.residx;
8399		} else {
8400		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8401				if (memcmp(msg->pr.pr_info.sa_res_key,
8402		                   lun->per_res[i].res_key.key,
8403		                   sizeof(struct scsi_per_res_key)) != 0)
8404					continue;
8405
8406				lun->per_res[i].registered = 0;
8407				memset(&lun->per_res[i].res_key, 0,
8408				       sizeof(struct scsi_per_res_key));
8409				lun->pr_key_count--;
8410
8411				if (!persis_offset
8412				 && i < persis_offset)
8413					lun->pending_ua[i] |=
8414						CTL_UA_REG_PREEMPT;
8415				else if (persis_offset
8416				      && i >= persis_offset)
8417					lun->pending_ua[i - persis_offset] |=
8418						CTL_UA_REG_PREEMPT;
8419			}
8420		}
8421	} else {
8422		/*
8423		 * Temporarily unregister so it won't get removed
8424		 * or UA generated
8425		 */
8426		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8427		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8428			if (lun->per_res[i].registered == 0)
8429				continue;
8430
8431			if (memcmp(msg->pr.pr_info.sa_res_key,
8432	                   lun->per_res[i].res_key.key,
8433	                   sizeof(struct scsi_per_res_key)) == 0) {
8434				lun->per_res[i].registered = 0;
8435				memset(&lun->per_res[i].res_key, 0,
8436				       sizeof(struct scsi_per_res_key));
8437				lun->pr_key_count--;
8438				if (!persis_offset
8439				 && i < CTL_MAX_INITIATORS)
8440					lun->pending_ua[i] |=
8441						CTL_UA_REG_PREEMPT;
8442				else if (persis_offset
8443				      && i >= persis_offset)
8444					lun->pending_ua[i - persis_offset] |=
8445						CTL_UA_REG_PREEMPT;
8446			} else if (msg->pr.pr_info.res_type != lun->res_type
8447				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8448				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8449					if (!persis_offset
8450					 && i < persis_offset)
8451						lun->pending_ua[i] |=
8452							CTL_UA_RES_RELEASE;
8453					else if (persis_offset
8454					      && i >= persis_offset)
8455					lun->pending_ua[i - persis_offset] |=
8456						CTL_UA_RES_RELEASE;
8457			}
8458		}
8459		lun->per_res[msg->pr.pr_info.residx].registered = 1;
8460		lun->res_type = msg->pr.pr_info.res_type;
8461		if (lun->res_type != SPR_TYPE_WR_EX_AR
8462		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8463			lun->pr_res_idx = msg->pr.pr_info.residx;
8464		else
8465			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8466	}
8467	lun->PRGeneration++;
8468
8469}
8470
8471
8472int
8473ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8474{
8475	int retval;
8476	int isc_retval;
8477	u_int32_t param_len;
8478	struct scsi_per_res_out *cdb;
8479	struct ctl_lun *lun;
8480	struct scsi_per_res_out_parms* param;
8481	struct ctl_softc *softc;
8482	uint32_t residx;
8483	uint64_t res_key, sa_res_key;
8484	uint8_t type;
8485	union ctl_ha_msg persis_io;
8486	int    i;
8487
8488	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8489
8490	retval = CTL_RETVAL_COMPLETE;
8491
8492	softc = control_softc;
8493
8494	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8495	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8496
8497	/*
8498	 * We only support whole-LUN scope.  The scope & type are ignored for
8499	 * register, register and ignore existing key and clear.
8500	 * We sometimes ignore scope and type on preempts too!!
8501	 * Verify reservation type here as well.
8502	 */
8503	type = cdb->scope_type & SPR_TYPE_MASK;
8504	if ((cdb->action == SPRO_RESERVE)
8505	 || (cdb->action == SPRO_RELEASE)) {
8506		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8507			ctl_set_invalid_field(/*ctsio*/ ctsio,
8508					      /*sks_valid*/ 1,
8509					      /*command*/ 1,
8510					      /*field*/ 2,
8511					      /*bit_valid*/ 1,
8512					      /*bit*/ 4);
8513			ctl_done((union ctl_io *)ctsio);
8514			return (CTL_RETVAL_COMPLETE);
8515		}
8516
8517		if (type>8 || type==2 || type==4 || type==0) {
8518			ctl_set_invalid_field(/*ctsio*/ ctsio,
8519					      /*sks_valid*/ 1,
8520					      /*command*/ 1,
8521					      /*field*/ 2,
8522					      /*bit_valid*/ 1,
8523					      /*bit*/ 0);
8524			ctl_done((union ctl_io *)ctsio);
8525			return (CTL_RETVAL_COMPLETE);
8526		}
8527	}
8528
8529	param_len = scsi_4btoul(cdb->length);
8530
8531	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8532		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8533		ctsio->kern_data_len = param_len;
8534		ctsio->kern_total_len = param_len;
8535		ctsio->kern_data_resid = 0;
8536		ctsio->kern_rel_offset = 0;
8537		ctsio->kern_sg_entries = 0;
8538		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8539		ctsio->be_move_done = ctl_config_move_done;
8540		ctl_datamove((union ctl_io *)ctsio);
8541
8542		return (CTL_RETVAL_COMPLETE);
8543	}
8544
8545	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8546
8547	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8548	res_key = scsi_8btou64(param->res_key.key);
8549	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8550
8551	/*
8552	 * Validate the reservation key here except for SPRO_REG_IGNO
8553	 * This must be done for all other service actions
8554	 */
8555	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8556		mtx_lock(&lun->lun_lock);
8557		if (lun->per_res[residx].registered) {
8558		    if (memcmp(param->res_key.key,
8559			       lun->per_res[residx].res_key.key,
8560			       ctl_min(sizeof(param->res_key),
8561			       sizeof(lun->per_res[residx].res_key))) != 0) {
8562				/*
8563				 * The current key passed in doesn't match
8564				 * the one the initiator previously
8565				 * registered.
8566				 */
8567				mtx_unlock(&lun->lun_lock);
8568				free(ctsio->kern_data_ptr, M_CTL);
8569				ctl_set_reservation_conflict(ctsio);
8570				ctl_done((union ctl_io *)ctsio);
8571				return (CTL_RETVAL_COMPLETE);
8572			}
8573		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8574			/*
8575			 * We are not registered
8576			 */
8577			mtx_unlock(&lun->lun_lock);
8578			free(ctsio->kern_data_ptr, M_CTL);
8579			ctl_set_reservation_conflict(ctsio);
8580			ctl_done((union ctl_io *)ctsio);
8581			return (CTL_RETVAL_COMPLETE);
8582		} else if (res_key != 0) {
8583			/*
8584			 * We are not registered and trying to register but
8585			 * the register key isn't zero.
8586			 */
8587			mtx_unlock(&lun->lun_lock);
8588			free(ctsio->kern_data_ptr, M_CTL);
8589			ctl_set_reservation_conflict(ctsio);
8590			ctl_done((union ctl_io *)ctsio);
8591			return (CTL_RETVAL_COMPLETE);
8592		}
8593		mtx_unlock(&lun->lun_lock);
8594	}
8595
8596	switch (cdb->action & SPRO_ACTION_MASK) {
8597	case SPRO_REGISTER:
8598	case SPRO_REG_IGNO: {
8599
8600#if 0
8601		printf("Registration received\n");
8602#endif
8603
8604		/*
8605		 * We don't support any of these options, as we report in
8606		 * the read capabilities request (see
8607		 * ctl_persistent_reserve_in(), above).
8608		 */
8609		if ((param->flags & SPR_SPEC_I_PT)
8610		 || (param->flags & SPR_ALL_TG_PT)
8611		 || (param->flags & SPR_APTPL)) {
8612			int bit_ptr;
8613
8614			if (param->flags & SPR_APTPL)
8615				bit_ptr = 0;
8616			else if (param->flags & SPR_ALL_TG_PT)
8617				bit_ptr = 2;
8618			else /* SPR_SPEC_I_PT */
8619				bit_ptr = 3;
8620
8621			free(ctsio->kern_data_ptr, M_CTL);
8622			ctl_set_invalid_field(ctsio,
8623					      /*sks_valid*/ 1,
8624					      /*command*/ 0,
8625					      /*field*/ 20,
8626					      /*bit_valid*/ 1,
8627					      /*bit*/ bit_ptr);
8628			ctl_done((union ctl_io *)ctsio);
8629			return (CTL_RETVAL_COMPLETE);
8630		}
8631
8632		mtx_lock(&lun->lun_lock);
8633
8634		/*
8635		 * The initiator wants to clear the
8636		 * key/unregister.
8637		 */
8638		if (sa_res_key == 0) {
8639			if ((res_key == 0
8640			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8641			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8642			  && !lun->per_res[residx].registered)) {
8643				mtx_unlock(&lun->lun_lock);
8644				goto done;
8645			}
8646
8647			lun->per_res[residx].registered = 0;
8648			memset(&lun->per_res[residx].res_key,
8649			       0, sizeof(lun->per_res[residx].res_key));
8650			lun->pr_key_count--;
8651
8652			if (residx == lun->pr_res_idx) {
8653				lun->flags &= ~CTL_LUN_PR_RESERVED;
8654				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8655
8656				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8657				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8658				 && lun->pr_key_count) {
8659					/*
8660					 * If the reservation is a registrants
8661					 * only type we need to generate a UA
8662					 * for other registered inits.  The
8663					 * sense code should be RESERVATIONS
8664					 * RELEASED
8665					 */
8666
8667					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8668						if (lun->per_res[
8669						    i+persis_offset].registered
8670						    == 0)
8671							continue;
8672						lun->pending_ua[i] |=
8673							CTL_UA_RES_RELEASE;
8674					}
8675				}
8676				lun->res_type = 0;
8677			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8678				if (lun->pr_key_count==0) {
8679					lun->flags &= ~CTL_LUN_PR_RESERVED;
8680					lun->res_type = 0;
8681					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8682				}
8683			}
8684			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8685			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8686			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8687			persis_io.pr.pr_info.residx = residx;
8688			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8689			     &persis_io, sizeof(persis_io), 0 )) >
8690			     CTL_HA_STATUS_SUCCESS) {
8691				printf("CTL:Persis Out error returned from "
8692				       "ctl_ha_msg_send %d\n", isc_retval);
8693			}
8694		} else /* sa_res_key != 0 */ {
8695
8696			/*
8697			 * If we aren't registered currently then increment
8698			 * the key count and set the registered flag.
8699			 */
8700			if (!lun->per_res[residx].registered) {
8701				lun->pr_key_count++;
8702				lun->per_res[residx].registered = 1;
8703			}
8704
8705			memcpy(&lun->per_res[residx].res_key,
8706			       param->serv_act_res_key,
8707			       ctl_min(sizeof(param->serv_act_res_key),
8708			       sizeof(lun->per_res[residx].res_key)));
8709
8710			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8711			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8712			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8713			persis_io.pr.pr_info.residx = residx;
8714			memcpy(persis_io.pr.pr_info.sa_res_key,
8715			       param->serv_act_res_key,
8716			       sizeof(param->serv_act_res_key));
8717			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8718			     &persis_io, sizeof(persis_io), 0)) >
8719			     CTL_HA_STATUS_SUCCESS) {
8720				printf("CTL:Persis Out error returned from "
8721				       "ctl_ha_msg_send %d\n", isc_retval);
8722			}
8723		}
8724		lun->PRGeneration++;
8725		mtx_unlock(&lun->lun_lock);
8726
8727		break;
8728	}
8729	case SPRO_RESERVE:
8730#if 0
8731                printf("Reserve executed type %d\n", type);
8732#endif
8733		mtx_lock(&lun->lun_lock);
8734		if (lun->flags & CTL_LUN_PR_RESERVED) {
8735			/*
8736			 * if this isn't the reservation holder and it's
8737			 * not a "all registrants" type or if the type is
8738			 * different then we have a conflict
8739			 */
8740			if ((lun->pr_res_idx != residx
8741			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8742			 || lun->res_type != type) {
8743				mtx_unlock(&lun->lun_lock);
8744				free(ctsio->kern_data_ptr, M_CTL);
8745				ctl_set_reservation_conflict(ctsio);
8746				ctl_done((union ctl_io *)ctsio);
8747				return (CTL_RETVAL_COMPLETE);
8748			}
8749			mtx_unlock(&lun->lun_lock);
8750		} else /* create a reservation */ {
8751			/*
8752			 * If it's not an "all registrants" type record
8753			 * reservation holder
8754			 */
8755			if (type != SPR_TYPE_WR_EX_AR
8756			 && type != SPR_TYPE_EX_AC_AR)
8757				lun->pr_res_idx = residx; /* Res holder */
8758			else
8759				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8760
8761			lun->flags |= CTL_LUN_PR_RESERVED;
8762			lun->res_type = type;
8763
8764			mtx_unlock(&lun->lun_lock);
8765
8766			/* send msg to other side */
8767			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8768			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8769			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8770			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8771			persis_io.pr.pr_info.res_type = type;
8772			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8773			     &persis_io, sizeof(persis_io), 0)) >
8774			     CTL_HA_STATUS_SUCCESS) {
8775				printf("CTL:Persis Out error returned from "
8776				       "ctl_ha_msg_send %d\n", isc_retval);
8777			}
8778		}
8779		break;
8780
8781	case SPRO_RELEASE:
8782		mtx_lock(&lun->lun_lock);
8783		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8784			/* No reservation exists return good status */
8785			mtx_unlock(&lun->lun_lock);
8786			goto done;
8787		}
8788		/*
8789		 * Is this nexus a reservation holder?
8790		 */
8791		if (lun->pr_res_idx != residx
8792		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8793			/*
8794			 * not a res holder return good status but
8795			 * do nothing
8796			 */
8797			mtx_unlock(&lun->lun_lock);
8798			goto done;
8799		}
8800
8801		if (lun->res_type != type) {
8802			mtx_unlock(&lun->lun_lock);
8803			free(ctsio->kern_data_ptr, M_CTL);
8804			ctl_set_illegal_pr_release(ctsio);
8805			ctl_done((union ctl_io *)ctsio);
8806			return (CTL_RETVAL_COMPLETE);
8807		}
8808
8809		/* okay to release */
8810		lun->flags &= ~CTL_LUN_PR_RESERVED;
8811		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8812		lun->res_type = 0;
8813
8814		/*
8815		 * if this isn't an exclusive access
8816		 * res generate UA for all other
8817		 * registrants.
8818		 */
8819		if (type != SPR_TYPE_EX_AC
8820		 && type != SPR_TYPE_WR_EX) {
8821			/*
8822			 * temporarily unregister so we don't generate UA
8823			 */
8824			lun->per_res[residx].registered = 0;
8825
8826			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8827				if (lun->per_res[i+persis_offset].registered
8828				    == 0)
8829					continue;
8830				lun->pending_ua[i] |=
8831					CTL_UA_RES_RELEASE;
8832			}
8833
8834			lun->per_res[residx].registered = 1;
8835		}
8836		mtx_unlock(&lun->lun_lock);
8837		/* Send msg to other side */
8838		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8839		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8840		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8841		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8842		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8843			printf("CTL:Persis Out error returned from "
8844			       "ctl_ha_msg_send %d\n", isc_retval);
8845		}
8846		break;
8847
8848	case SPRO_CLEAR:
8849		/* send msg to other side */
8850
8851		mtx_lock(&lun->lun_lock);
8852		lun->flags &= ~CTL_LUN_PR_RESERVED;
8853		lun->res_type = 0;
8854		lun->pr_key_count = 0;
8855		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8856
8857
8858		memset(&lun->per_res[residx].res_key,
8859		       0, sizeof(lun->per_res[residx].res_key));
8860		lun->per_res[residx].registered = 0;
8861
8862		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8863			if (lun->per_res[i].registered) {
8864				if (!persis_offset && i < CTL_MAX_INITIATORS)
8865					lun->pending_ua[i] |=
8866						CTL_UA_RES_PREEMPT;
8867				else if (persis_offset && i >= persis_offset)
8868					lun->pending_ua[i-persis_offset] |=
8869					    CTL_UA_RES_PREEMPT;
8870
8871				memset(&lun->per_res[i].res_key,
8872				       0, sizeof(struct scsi_per_res_key));
8873				lun->per_res[i].registered = 0;
8874			}
8875		lun->PRGeneration++;
8876		mtx_unlock(&lun->lun_lock);
8877		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8878		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8879		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8880		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8881		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8882			printf("CTL:Persis Out error returned from "
8883			       "ctl_ha_msg_send %d\n", isc_retval);
8884		}
8885		break;
8886
8887	case SPRO_PREEMPT: {
8888		int nretval;
8889
8890		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8891					  residx, ctsio, cdb, param);
8892		if (nretval != 0)
8893			return (CTL_RETVAL_COMPLETE);
8894		break;
8895	}
8896	default:
8897		panic("Invalid PR type %x", cdb->action);
8898	}
8899
8900done:
8901	free(ctsio->kern_data_ptr, M_CTL);
8902	ctl_set_success(ctsio);
8903	ctl_done((union ctl_io *)ctsio);
8904
8905	return (retval);
8906}
8907
8908/*
8909 * This routine is for handling a message from the other SC pertaining to
8910 * persistent reserve out. All the error checking will have been done
8911 * so only perorming the action need be done here to keep the two
8912 * in sync.
8913 */
8914static void
8915ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8916{
8917	struct ctl_lun *lun;
8918	struct ctl_softc *softc;
8919	int i;
8920	uint32_t targ_lun;
8921
8922	softc = control_softc;
8923
8924	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8925	lun = softc->ctl_luns[targ_lun];
8926	mtx_lock(&lun->lun_lock);
8927	switch(msg->pr.pr_info.action) {
8928	case CTL_PR_REG_KEY:
8929		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8930			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8931			lun->pr_key_count++;
8932		}
8933		lun->PRGeneration++;
8934		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8935		       msg->pr.pr_info.sa_res_key,
8936		       sizeof(struct scsi_per_res_key));
8937		break;
8938
8939	case CTL_PR_UNREG_KEY:
8940		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8941		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8942		       0, sizeof(struct scsi_per_res_key));
8943		lun->pr_key_count--;
8944
8945		/* XXX Need to see if the reservation has been released */
8946		/* if so do we need to generate UA? */
8947		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8948			lun->flags &= ~CTL_LUN_PR_RESERVED;
8949			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8950
8951			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8952			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8953			 && lun->pr_key_count) {
8954				/*
8955				 * If the reservation is a registrants
8956				 * only type we need to generate a UA
8957				 * for other registered inits.  The
8958				 * sense code should be RESERVATIONS
8959				 * RELEASED
8960				 */
8961
8962				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8963					if (lun->per_res[i+
8964					    persis_offset].registered == 0)
8965						continue;
8966
8967					lun->pending_ua[i] |=
8968						CTL_UA_RES_RELEASE;
8969				}
8970			}
8971			lun->res_type = 0;
8972		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8973			if (lun->pr_key_count==0) {
8974				lun->flags &= ~CTL_LUN_PR_RESERVED;
8975				lun->res_type = 0;
8976				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8977			}
8978		}
8979		lun->PRGeneration++;
8980		break;
8981
8982	case CTL_PR_RESERVE:
8983		lun->flags |= CTL_LUN_PR_RESERVED;
8984		lun->res_type = msg->pr.pr_info.res_type;
8985		lun->pr_res_idx = msg->pr.pr_info.residx;
8986
8987		break;
8988
8989	case CTL_PR_RELEASE:
8990		/*
8991		 * if this isn't an exclusive access res generate UA for all
8992		 * other registrants.
8993		 */
8994		if (lun->res_type != SPR_TYPE_EX_AC
8995		 && lun->res_type != SPR_TYPE_WR_EX) {
8996			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8997				if (lun->per_res[i+persis_offset].registered)
8998					lun->pending_ua[i] |=
8999						CTL_UA_RES_RELEASE;
9000		}
9001
9002		lun->flags &= ~CTL_LUN_PR_RESERVED;
9003		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
9004		lun->res_type = 0;
9005		break;
9006
9007	case CTL_PR_PREEMPT:
9008		ctl_pro_preempt_other(lun, msg);
9009		break;
9010	case CTL_PR_CLEAR:
9011		lun->flags &= ~CTL_LUN_PR_RESERVED;
9012		lun->res_type = 0;
9013		lun->pr_key_count = 0;
9014		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
9015
9016		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
9017			if (lun->per_res[i].registered == 0)
9018				continue;
9019			if (!persis_offset
9020			 && i < CTL_MAX_INITIATORS)
9021				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
9022			else if (persis_offset
9023			      && i >= persis_offset)
9024				lun->pending_ua[i-persis_offset] |=
9025					CTL_UA_RES_PREEMPT;
9026			memset(&lun->per_res[i].res_key, 0,
9027			       sizeof(struct scsi_per_res_key));
9028			lun->per_res[i].registered = 0;
9029		}
9030		lun->PRGeneration++;
9031		break;
9032	}
9033
9034	mtx_unlock(&lun->lun_lock);
9035}
9036
9037int
9038ctl_read_write(struct ctl_scsiio *ctsio)
9039{
9040	struct ctl_lun *lun;
9041	struct ctl_lba_len_flags *lbalen;
9042	uint64_t lba;
9043	uint32_t num_blocks;
9044	int flags, retval;
9045	int isread;
9046
9047	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9048
9049	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
9050
9051	flags = 0;
9052	retval = CTL_RETVAL_COMPLETE;
9053
9054	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
9055	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
9056	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
9057		uint32_t residx;
9058
9059		/*
9060		 * XXX KDM need a lock here.
9061		 */
9062		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9063		if ((lun->res_type == SPR_TYPE_EX_AC
9064		  && residx != lun->pr_res_idx)
9065		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
9066		   || lun->res_type == SPR_TYPE_EX_AC_AR)
9067		  && !lun->per_res[residx].registered)) {
9068			ctl_set_reservation_conflict(ctsio);
9069			ctl_done((union ctl_io *)ctsio);
9070			return (CTL_RETVAL_COMPLETE);
9071	        }
9072	}
9073
9074	switch (ctsio->cdb[0]) {
9075	case READ_6:
9076	case WRITE_6: {
9077		struct scsi_rw_6 *cdb;
9078
9079		cdb = (struct scsi_rw_6 *)ctsio->cdb;
9080
9081		lba = scsi_3btoul(cdb->addr);
9082		/* only 5 bits are valid in the most significant address byte */
9083		lba &= 0x1fffff;
9084		num_blocks = cdb->length;
9085		/*
9086		 * This is correct according to SBC-2.
9087		 */
9088		if (num_blocks == 0)
9089			num_blocks = 256;
9090		break;
9091	}
9092	case READ_10:
9093	case WRITE_10: {
9094		struct scsi_rw_10 *cdb;
9095
9096		cdb = (struct scsi_rw_10 *)ctsio->cdb;
9097		if (cdb->byte2 & SRW10_FUA)
9098			flags |= CTL_LLF_FUA;
9099		if (cdb->byte2 & SRW10_DPO)
9100			flags |= CTL_LLF_DPO;
9101		lba = scsi_4btoul(cdb->addr);
9102		num_blocks = scsi_2btoul(cdb->length);
9103		break;
9104	}
9105	case WRITE_VERIFY_10: {
9106		struct scsi_write_verify_10 *cdb;
9107
9108		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9109		flags |= CTL_LLF_FUA;
9110		if (cdb->byte2 & SWV_DPO)
9111			flags |= CTL_LLF_DPO;
9112		lba = scsi_4btoul(cdb->addr);
9113		num_blocks = scsi_2btoul(cdb->length);
9114		break;
9115	}
9116	case READ_12:
9117	case WRITE_12: {
9118		struct scsi_rw_12 *cdb;
9119
9120		cdb = (struct scsi_rw_12 *)ctsio->cdb;
9121		if (cdb->byte2 & SRW12_FUA)
9122			flags |= CTL_LLF_FUA;
9123		if (cdb->byte2 & SRW12_DPO)
9124			flags |= CTL_LLF_DPO;
9125		lba = scsi_4btoul(cdb->addr);
9126		num_blocks = scsi_4btoul(cdb->length);
9127		break;
9128	}
9129	case WRITE_VERIFY_12: {
9130		struct scsi_write_verify_12 *cdb;
9131
9132		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9133		flags |= CTL_LLF_FUA;
9134		if (cdb->byte2 & SWV_DPO)
9135			flags |= CTL_LLF_DPO;
9136		lba = scsi_4btoul(cdb->addr);
9137		num_blocks = scsi_4btoul(cdb->length);
9138		break;
9139	}
9140	case READ_16:
9141	case WRITE_16: {
9142		struct scsi_rw_16 *cdb;
9143
9144		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9145		if (cdb->byte2 & SRW12_FUA)
9146			flags |= CTL_LLF_FUA;
9147		if (cdb->byte2 & SRW12_DPO)
9148			flags |= CTL_LLF_DPO;
9149		lba = scsi_8btou64(cdb->addr);
9150		num_blocks = scsi_4btoul(cdb->length);
9151		break;
9152	}
9153	case WRITE_ATOMIC_16: {
9154		struct scsi_rw_16 *cdb;
9155
9156		if (lun->be_lun->atomicblock == 0) {
9157			ctl_set_invalid_opcode(ctsio);
9158			ctl_done((union ctl_io *)ctsio);
9159			return (CTL_RETVAL_COMPLETE);
9160		}
9161
9162		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9163		if (cdb->byte2 & SRW12_FUA)
9164			flags |= CTL_LLF_FUA;
9165		if (cdb->byte2 & SRW12_DPO)
9166			flags |= CTL_LLF_DPO;
9167		lba = scsi_8btou64(cdb->addr);
9168		num_blocks = scsi_4btoul(cdb->length);
9169		if (num_blocks > lun->be_lun->atomicblock) {
9170			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
9171			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
9172			    /*bit*/ 0);
9173			ctl_done((union ctl_io *)ctsio);
9174			return (CTL_RETVAL_COMPLETE);
9175		}
9176		break;
9177	}
9178	case WRITE_VERIFY_16: {
9179		struct scsi_write_verify_16 *cdb;
9180
9181		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9182		flags |= CTL_LLF_FUA;
9183		if (cdb->byte2 & SWV_DPO)
9184			flags |= CTL_LLF_DPO;
9185		lba = scsi_8btou64(cdb->addr);
9186		num_blocks = scsi_4btoul(cdb->length);
9187		break;
9188	}
9189	default:
9190		/*
9191		 * We got a command we don't support.  This shouldn't
9192		 * happen, commands should be filtered out above us.
9193		 */
9194		ctl_set_invalid_opcode(ctsio);
9195		ctl_done((union ctl_io *)ctsio);
9196
9197		return (CTL_RETVAL_COMPLETE);
9198		break; /* NOTREACHED */
9199	}
9200
9201	/*
9202	 * The first check is to make sure we're in bounds, the second
9203	 * check is to catch wrap-around problems.  If the lba + num blocks
9204	 * is less than the lba, then we've wrapped around and the block
9205	 * range is invalid anyway.
9206	 */
9207	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9208	 || ((lba + num_blocks) < lba)) {
9209		ctl_set_lba_out_of_range(ctsio);
9210		ctl_done((union ctl_io *)ctsio);
9211		return (CTL_RETVAL_COMPLETE);
9212	}
9213
9214	/*
9215	 * According to SBC-3, a transfer length of 0 is not an error.
9216	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9217	 * translates to 256 blocks for those commands.
9218	 */
9219	if (num_blocks == 0) {
9220		ctl_set_success(ctsio);
9221		ctl_done((union ctl_io *)ctsio);
9222		return (CTL_RETVAL_COMPLETE);
9223	}
9224
9225	/* Set FUA and/or DPO if caches are disabled. */
9226	if (isread) {
9227		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9228		    SCP_RCD) != 0)
9229			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9230	} else {
9231		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9232		    SCP_WCE) == 0)
9233			flags |= CTL_LLF_FUA;
9234	}
9235
9236	lbalen = (struct ctl_lba_len_flags *)
9237	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9238	lbalen->lba = lba;
9239	lbalen->len = num_blocks;
9240	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9241
9242	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9243	ctsio->kern_rel_offset = 0;
9244
9245	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9246
9247	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9248
9249	return (retval);
9250}
9251
9252static int
9253ctl_cnw_cont(union ctl_io *io)
9254{
9255	struct ctl_scsiio *ctsio;
9256	struct ctl_lun *lun;
9257	struct ctl_lba_len_flags *lbalen;
9258	int retval;
9259
9260	ctsio = &io->scsiio;
9261	ctsio->io_hdr.status = CTL_STATUS_NONE;
9262	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9263	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9264	lbalen = (struct ctl_lba_len_flags *)
9265	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9266	lbalen->flags &= ~CTL_LLF_COMPARE;
9267	lbalen->flags |= CTL_LLF_WRITE;
9268
9269	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9270	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9271	return (retval);
9272}
9273
9274int
9275ctl_cnw(struct ctl_scsiio *ctsio)
9276{
9277	struct ctl_lun *lun;
9278	struct ctl_lba_len_flags *lbalen;
9279	uint64_t lba;
9280	uint32_t num_blocks;
9281	int flags, retval;
9282
9283	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9284
9285	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9286
9287	flags = 0;
9288	retval = CTL_RETVAL_COMPLETE;
9289
9290	switch (ctsio->cdb[0]) {
9291	case COMPARE_AND_WRITE: {
9292		struct scsi_compare_and_write *cdb;
9293
9294		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9295		if (cdb->byte2 & SRW10_FUA)
9296			flags |= CTL_LLF_FUA;
9297		if (cdb->byte2 & SRW10_DPO)
9298			flags |= CTL_LLF_DPO;
9299		lba = scsi_8btou64(cdb->addr);
9300		num_blocks = cdb->length;
9301		break;
9302	}
9303	default:
9304		/*
9305		 * We got a command we don't support.  This shouldn't
9306		 * happen, commands should be filtered out above us.
9307		 */
9308		ctl_set_invalid_opcode(ctsio);
9309		ctl_done((union ctl_io *)ctsio);
9310
9311		return (CTL_RETVAL_COMPLETE);
9312		break; /* NOTREACHED */
9313	}
9314
9315	/*
9316	 * The first check is to make sure we're in bounds, the second
9317	 * check is to catch wrap-around problems.  If the lba + num blocks
9318	 * is less than the lba, then we've wrapped around and the block
9319	 * range is invalid anyway.
9320	 */
9321	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9322	 || ((lba + num_blocks) < lba)) {
9323		ctl_set_lba_out_of_range(ctsio);
9324		ctl_done((union ctl_io *)ctsio);
9325		return (CTL_RETVAL_COMPLETE);
9326	}
9327
9328	/*
9329	 * According to SBC-3, a transfer length of 0 is not an error.
9330	 */
9331	if (num_blocks == 0) {
9332		ctl_set_success(ctsio);
9333		ctl_done((union ctl_io *)ctsio);
9334		return (CTL_RETVAL_COMPLETE);
9335	}
9336
9337	/* Set FUA if write cache is disabled. */
9338	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9339	    SCP_WCE) == 0)
9340		flags |= CTL_LLF_FUA;
9341
9342	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9343	ctsio->kern_rel_offset = 0;
9344
9345	/*
9346	 * Set the IO_CONT flag, so that if this I/O gets passed to
9347	 * ctl_data_submit_done(), it'll get passed back to
9348	 * ctl_ctl_cnw_cont() for further processing.
9349	 */
9350	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9351	ctsio->io_cont = ctl_cnw_cont;
9352
9353	lbalen = (struct ctl_lba_len_flags *)
9354	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9355	lbalen->lba = lba;
9356	lbalen->len = num_blocks;
9357	lbalen->flags = CTL_LLF_COMPARE | flags;
9358
9359	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9360	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9361	return (retval);
9362}
9363
9364int
9365ctl_verify(struct ctl_scsiio *ctsio)
9366{
9367	struct ctl_lun *lun;
9368	struct ctl_lba_len_flags *lbalen;
9369	uint64_t lba;
9370	uint32_t num_blocks;
9371	int bytchk, flags;
9372	int retval;
9373
9374	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9375
9376	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9377
9378	bytchk = 0;
9379	flags = CTL_LLF_FUA;
9380	retval = CTL_RETVAL_COMPLETE;
9381
9382	switch (ctsio->cdb[0]) {
9383	case VERIFY_10: {
9384		struct scsi_verify_10 *cdb;
9385
9386		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9387		if (cdb->byte2 & SVFY_BYTCHK)
9388			bytchk = 1;
9389		if (cdb->byte2 & SVFY_DPO)
9390			flags |= CTL_LLF_DPO;
9391		lba = scsi_4btoul(cdb->addr);
9392		num_blocks = scsi_2btoul(cdb->length);
9393		break;
9394	}
9395	case VERIFY_12: {
9396		struct scsi_verify_12 *cdb;
9397
9398		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9399		if (cdb->byte2 & SVFY_BYTCHK)
9400			bytchk = 1;
9401		if (cdb->byte2 & SVFY_DPO)
9402			flags |= CTL_LLF_DPO;
9403		lba = scsi_4btoul(cdb->addr);
9404		num_blocks = scsi_4btoul(cdb->length);
9405		break;
9406	}
9407	case VERIFY_16: {
9408		struct scsi_rw_16 *cdb;
9409
9410		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9411		if (cdb->byte2 & SVFY_BYTCHK)
9412			bytchk = 1;
9413		if (cdb->byte2 & SVFY_DPO)
9414			flags |= CTL_LLF_DPO;
9415		lba = scsi_8btou64(cdb->addr);
9416		num_blocks = scsi_4btoul(cdb->length);
9417		break;
9418	}
9419	default:
9420		/*
9421		 * We got a command we don't support.  This shouldn't
9422		 * happen, commands should be filtered out above us.
9423		 */
9424		ctl_set_invalid_opcode(ctsio);
9425		ctl_done((union ctl_io *)ctsio);
9426		return (CTL_RETVAL_COMPLETE);
9427	}
9428
9429	/*
9430	 * The first check is to make sure we're in bounds, the second
9431	 * check is to catch wrap-around problems.  If the lba + num blocks
9432	 * is less than the lba, then we've wrapped around and the block
9433	 * range is invalid anyway.
9434	 */
9435	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9436	 || ((lba + num_blocks) < lba)) {
9437		ctl_set_lba_out_of_range(ctsio);
9438		ctl_done((union ctl_io *)ctsio);
9439		return (CTL_RETVAL_COMPLETE);
9440	}
9441
9442	/*
9443	 * According to SBC-3, a transfer length of 0 is not an error.
9444	 */
9445	if (num_blocks == 0) {
9446		ctl_set_success(ctsio);
9447		ctl_done((union ctl_io *)ctsio);
9448		return (CTL_RETVAL_COMPLETE);
9449	}
9450
9451	lbalen = (struct ctl_lba_len_flags *)
9452	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9453	lbalen->lba = lba;
9454	lbalen->len = num_blocks;
9455	if (bytchk) {
9456		lbalen->flags = CTL_LLF_COMPARE | flags;
9457		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9458	} else {
9459		lbalen->flags = CTL_LLF_VERIFY | flags;
9460		ctsio->kern_total_len = 0;
9461	}
9462	ctsio->kern_rel_offset = 0;
9463
9464	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9465	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9466	return (retval);
9467}
9468
9469int
9470ctl_report_luns(struct ctl_scsiio *ctsio)
9471{
9472	struct scsi_report_luns *cdb;
9473	struct scsi_report_luns_data *lun_data;
9474	struct ctl_lun *lun, *request_lun;
9475	int num_luns, retval;
9476	uint32_t alloc_len, lun_datalen;
9477	int num_filled, well_known;
9478	uint32_t initidx, targ_lun_id, lun_id;
9479
9480	retval = CTL_RETVAL_COMPLETE;
9481	well_known = 0;
9482
9483	cdb = (struct scsi_report_luns *)ctsio->cdb;
9484
9485	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9486
9487	mtx_lock(&control_softc->ctl_lock);
9488	num_luns = control_softc->num_luns;
9489	mtx_unlock(&control_softc->ctl_lock);
9490
9491	switch (cdb->select_report) {
9492	case RPL_REPORT_DEFAULT:
9493	case RPL_REPORT_ALL:
9494		break;
9495	case RPL_REPORT_WELLKNOWN:
9496		well_known = 1;
9497		num_luns = 0;
9498		break;
9499	default:
9500		ctl_set_invalid_field(ctsio,
9501				      /*sks_valid*/ 1,
9502				      /*command*/ 1,
9503				      /*field*/ 2,
9504				      /*bit_valid*/ 0,
9505				      /*bit*/ 0);
9506		ctl_done((union ctl_io *)ctsio);
9507		return (retval);
9508		break; /* NOTREACHED */
9509	}
9510
9511	alloc_len = scsi_4btoul(cdb->length);
9512	/*
9513	 * The initiator has to allocate at least 16 bytes for this request,
9514	 * so he can at least get the header and the first LUN.  Otherwise
9515	 * we reject the request (per SPC-3 rev 14, section 6.21).
9516	 */
9517	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9518	    sizeof(struct scsi_report_luns_lundata))) {
9519		ctl_set_invalid_field(ctsio,
9520				      /*sks_valid*/ 1,
9521				      /*command*/ 1,
9522				      /*field*/ 6,
9523				      /*bit_valid*/ 0,
9524				      /*bit*/ 0);
9525		ctl_done((union ctl_io *)ctsio);
9526		return (retval);
9527	}
9528
9529	request_lun = (struct ctl_lun *)
9530		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9531
9532	lun_datalen = sizeof(*lun_data) +
9533		(num_luns * sizeof(struct scsi_report_luns_lundata));
9534
9535	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9536	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9537	ctsio->kern_sg_entries = 0;
9538
9539	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9540
9541	mtx_lock(&control_softc->ctl_lock);
9542	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9543		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9544		if (lun_id >= CTL_MAX_LUNS)
9545			continue;
9546		lun = control_softc->ctl_luns[lun_id];
9547		if (lun == NULL)
9548			continue;
9549
9550		if (targ_lun_id <= 0xff) {
9551			/*
9552			 * Peripheral addressing method, bus number 0.
9553			 */
9554			lun_data->luns[num_filled].lundata[0] =
9555				RPL_LUNDATA_ATYP_PERIPH;
9556			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9557			num_filled++;
9558		} else if (targ_lun_id <= 0x3fff) {
9559			/*
9560			 * Flat addressing method.
9561			 */
9562			lun_data->luns[num_filled].lundata[0] =
9563				RPL_LUNDATA_ATYP_FLAT |
9564				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9565#ifdef OLDCTLHEADERS
9566				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9567				(targ_lun_id & SRLD_BUS_LUN_MASK);
9568#endif
9569			lun_data->luns[num_filled].lundata[1] =
9570#ifdef OLDCTLHEADERS
9571				targ_lun_id >> SRLD_BUS_LUN_BITS;
9572#endif
9573				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9574			num_filled++;
9575		} else {
9576			printf("ctl_report_luns: bogus LUN number %jd, "
9577			       "skipping\n", (intmax_t)targ_lun_id);
9578		}
9579		/*
9580		 * According to SPC-3, rev 14 section 6.21:
9581		 *
9582		 * "The execution of a REPORT LUNS command to any valid and
9583		 * installed logical unit shall clear the REPORTED LUNS DATA
9584		 * HAS CHANGED unit attention condition for all logical
9585		 * units of that target with respect to the requesting
9586		 * initiator. A valid and installed logical unit is one
9587		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9588		 * INQUIRY data (see 6.4.2)."
9589		 *
9590		 * If request_lun is NULL, the LUN this report luns command
9591		 * was issued to is either disabled or doesn't exist. In that
9592		 * case, we shouldn't clear any pending lun change unit
9593		 * attention.
9594		 */
9595		if (request_lun != NULL) {
9596			mtx_lock(&lun->lun_lock);
9597			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9598			mtx_unlock(&lun->lun_lock);
9599		}
9600	}
9601	mtx_unlock(&control_softc->ctl_lock);
9602
9603	/*
9604	 * It's quite possible that we've returned fewer LUNs than we allocated
9605	 * space for.  Trim it.
9606	 */
9607	lun_datalen = sizeof(*lun_data) +
9608		(num_filled * sizeof(struct scsi_report_luns_lundata));
9609
9610	if (lun_datalen < alloc_len) {
9611		ctsio->residual = alloc_len - lun_datalen;
9612		ctsio->kern_data_len = lun_datalen;
9613		ctsio->kern_total_len = lun_datalen;
9614	} else {
9615		ctsio->residual = 0;
9616		ctsio->kern_data_len = alloc_len;
9617		ctsio->kern_total_len = alloc_len;
9618	}
9619	ctsio->kern_data_resid = 0;
9620	ctsio->kern_rel_offset = 0;
9621	ctsio->kern_sg_entries = 0;
9622
9623	/*
9624	 * We set this to the actual data length, regardless of how much
9625	 * space we actually have to return results.  If the user looks at
9626	 * this value, he'll know whether or not he allocated enough space
9627	 * and reissue the command if necessary.  We don't support well
9628	 * known logical units, so if the user asks for that, return none.
9629	 */
9630	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9631
9632	/*
9633	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9634	 * this request.
9635	 */
9636	ctsio->scsi_status = SCSI_STATUS_OK;
9637
9638	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9639	ctsio->be_move_done = ctl_config_move_done;
9640	ctl_datamove((union ctl_io *)ctsio);
9641
9642	return (retval);
9643}
9644
9645int
9646ctl_request_sense(struct ctl_scsiio *ctsio)
9647{
9648	struct scsi_request_sense *cdb;
9649	struct scsi_sense_data *sense_ptr;
9650	struct ctl_lun *lun;
9651	uint32_t initidx;
9652	int have_error;
9653	scsi_sense_data_type sense_format;
9654
9655	cdb = (struct scsi_request_sense *)ctsio->cdb;
9656
9657	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9658
9659	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9660
9661	/*
9662	 * Determine which sense format the user wants.
9663	 */
9664	if (cdb->byte2 & SRS_DESC)
9665		sense_format = SSD_TYPE_DESC;
9666	else
9667		sense_format = SSD_TYPE_FIXED;
9668
9669	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9670	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9671	ctsio->kern_sg_entries = 0;
9672
9673	/*
9674	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9675	 * larger than the largest allowed value for the length field in the
9676	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9677	 */
9678	ctsio->residual = 0;
9679	ctsio->kern_data_len = cdb->length;
9680	ctsio->kern_total_len = cdb->length;
9681
9682	ctsio->kern_data_resid = 0;
9683	ctsio->kern_rel_offset = 0;
9684	ctsio->kern_sg_entries = 0;
9685
9686	/*
9687	 * If we don't have a LUN, we don't have any pending sense.
9688	 */
9689	if (lun == NULL)
9690		goto no_sense;
9691
9692	have_error = 0;
9693	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9694	/*
9695	 * Check for pending sense, and then for pending unit attentions.
9696	 * Pending sense gets returned first, then pending unit attentions.
9697	 */
9698	mtx_lock(&lun->lun_lock);
9699#ifdef CTL_WITH_CA
9700	if (ctl_is_set(lun->have_ca, initidx)) {
9701		scsi_sense_data_type stored_format;
9702
9703		/*
9704		 * Check to see which sense format was used for the stored
9705		 * sense data.
9706		 */
9707		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9708
9709		/*
9710		 * If the user requested a different sense format than the
9711		 * one we stored, then we need to convert it to the other
9712		 * format.  If we're going from descriptor to fixed format
9713		 * sense data, we may lose things in translation, depending
9714		 * on what options were used.
9715		 *
9716		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9717		 * for some reason we'll just copy it out as-is.
9718		 */
9719		if ((stored_format == SSD_TYPE_FIXED)
9720		 && (sense_format == SSD_TYPE_DESC))
9721			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9722			    &lun->pending_sense[initidx],
9723			    (struct scsi_sense_data_desc *)sense_ptr);
9724		else if ((stored_format == SSD_TYPE_DESC)
9725		      && (sense_format == SSD_TYPE_FIXED))
9726			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9727			    &lun->pending_sense[initidx],
9728			    (struct scsi_sense_data_fixed *)sense_ptr);
9729		else
9730			memcpy(sense_ptr, &lun->pending_sense[initidx],
9731			       ctl_min(sizeof(*sense_ptr),
9732			       sizeof(lun->pending_sense[initidx])));
9733
9734		ctl_clear_mask(lun->have_ca, initidx);
9735		have_error = 1;
9736	} else
9737#endif
9738	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9739		ctl_ua_type ua_type;
9740
9741		ua_type = ctl_build_ua(&lun->pending_ua[initidx],
9742				       sense_ptr, sense_format);
9743		if (ua_type != CTL_UA_NONE)
9744			have_error = 1;
9745	}
9746	mtx_unlock(&lun->lun_lock);
9747
9748	/*
9749	 * We already have a pending error, return it.
9750	 */
9751	if (have_error != 0) {
9752		/*
9753		 * We report the SCSI status as OK, since the status of the
9754		 * request sense command itself is OK.
9755		 */
9756		ctsio->scsi_status = SCSI_STATUS_OK;
9757
9758		/*
9759		 * We report 0 for the sense length, because we aren't doing
9760		 * autosense in this case.  We're reporting sense as
9761		 * parameter data.
9762		 */
9763		ctsio->sense_len = 0;
9764		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9765		ctsio->be_move_done = ctl_config_move_done;
9766		ctl_datamove((union ctl_io *)ctsio);
9767
9768		return (CTL_RETVAL_COMPLETE);
9769	}
9770
9771no_sense:
9772
9773	/*
9774	 * No sense information to report, so we report that everything is
9775	 * okay.
9776	 */
9777	ctl_set_sense_data(sense_ptr,
9778			   lun,
9779			   sense_format,
9780			   /*current_error*/ 1,
9781			   /*sense_key*/ SSD_KEY_NO_SENSE,
9782			   /*asc*/ 0x00,
9783			   /*ascq*/ 0x00,
9784			   SSD_ELEM_NONE);
9785
9786	ctsio->scsi_status = SCSI_STATUS_OK;
9787
9788	/*
9789	 * We report 0 for the sense length, because we aren't doing
9790	 * autosense in this case.  We're reporting sense as parameter data.
9791	 */
9792	ctsio->sense_len = 0;
9793	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9794	ctsio->be_move_done = ctl_config_move_done;
9795	ctl_datamove((union ctl_io *)ctsio);
9796
9797	return (CTL_RETVAL_COMPLETE);
9798}
9799
9800int
9801ctl_tur(struct ctl_scsiio *ctsio)
9802{
9803	struct ctl_lun *lun;
9804
9805	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9806
9807	CTL_DEBUG_PRINT(("ctl_tur\n"));
9808
9809	if (lun == NULL)
9810		return (EINVAL);
9811
9812	ctsio->scsi_status = SCSI_STATUS_OK;
9813	ctsio->io_hdr.status = CTL_SUCCESS;
9814
9815	ctl_done((union ctl_io *)ctsio);
9816
9817	return (CTL_RETVAL_COMPLETE);
9818}
9819
9820#ifdef notyet
9821static int
9822ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9823{
9824
9825}
9826#endif
9827
9828static int
9829ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9830{
9831	struct scsi_vpd_supported_pages *pages;
9832	int sup_page_size;
9833	struct ctl_lun *lun;
9834
9835	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9836
9837	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9838	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9839	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9840	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9841	ctsio->kern_sg_entries = 0;
9842
9843	if (sup_page_size < alloc_len) {
9844		ctsio->residual = alloc_len - sup_page_size;
9845		ctsio->kern_data_len = sup_page_size;
9846		ctsio->kern_total_len = sup_page_size;
9847	} else {
9848		ctsio->residual = 0;
9849		ctsio->kern_data_len = alloc_len;
9850		ctsio->kern_total_len = alloc_len;
9851	}
9852	ctsio->kern_data_resid = 0;
9853	ctsio->kern_rel_offset = 0;
9854	ctsio->kern_sg_entries = 0;
9855
9856	/*
9857	 * The control device is always connected.  The disk device, on the
9858	 * other hand, may not be online all the time.  Need to change this
9859	 * to figure out whether the disk device is actually online or not.
9860	 */
9861	if (lun != NULL)
9862		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9863				lun->be_lun->lun_type;
9864	else
9865		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9866
9867	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9868	/* Supported VPD pages */
9869	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9870	/* Serial Number */
9871	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9872	/* Device Identification */
9873	pages->page_list[2] = SVPD_DEVICE_ID;
9874	/* Extended INQUIRY Data */
9875	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9876	/* Mode Page Policy */
9877	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9878	/* SCSI Ports */
9879	pages->page_list[5] = SVPD_SCSI_PORTS;
9880	/* Third-party Copy */
9881	pages->page_list[6] = SVPD_SCSI_TPC;
9882	/* Block limits */
9883	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9884	/* Block Device Characteristics */
9885	pages->page_list[8] = SVPD_BDC;
9886	/* Logical Block Provisioning */
9887	pages->page_list[9] = SVPD_LBP;
9888
9889	ctsio->scsi_status = SCSI_STATUS_OK;
9890
9891	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9892	ctsio->be_move_done = ctl_config_move_done;
9893	ctl_datamove((union ctl_io *)ctsio);
9894
9895	return (CTL_RETVAL_COMPLETE);
9896}
9897
9898static int
9899ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9900{
9901	struct scsi_vpd_unit_serial_number *sn_ptr;
9902	struct ctl_lun *lun;
9903
9904	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9905
9906	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9907	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9908	ctsio->kern_sg_entries = 0;
9909
9910	if (sizeof(*sn_ptr) < alloc_len) {
9911		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9912		ctsio->kern_data_len = sizeof(*sn_ptr);
9913		ctsio->kern_total_len = sizeof(*sn_ptr);
9914	} else {
9915		ctsio->residual = 0;
9916		ctsio->kern_data_len = alloc_len;
9917		ctsio->kern_total_len = alloc_len;
9918	}
9919	ctsio->kern_data_resid = 0;
9920	ctsio->kern_rel_offset = 0;
9921	ctsio->kern_sg_entries = 0;
9922
9923	/*
9924	 * The control device is always connected.  The disk device, on the
9925	 * other hand, may not be online all the time.  Need to change this
9926	 * to figure out whether the disk device is actually online or not.
9927	 */
9928	if (lun != NULL)
9929		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9930				  lun->be_lun->lun_type;
9931	else
9932		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9933
9934	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9935	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9936	/*
9937	 * If we don't have a LUN, we just leave the serial number as
9938	 * all spaces.
9939	 */
9940	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9941	if (lun != NULL) {
9942		strncpy((char *)sn_ptr->serial_num,
9943			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9944	}
9945	ctsio->scsi_status = SCSI_STATUS_OK;
9946
9947	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9948	ctsio->be_move_done = ctl_config_move_done;
9949	ctl_datamove((union ctl_io *)ctsio);
9950
9951	return (CTL_RETVAL_COMPLETE);
9952}
9953
9954
9955static int
9956ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9957{
9958	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9959	struct ctl_lun *lun;
9960	int data_len;
9961
9962	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9963
9964	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9965	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9966	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9967	ctsio->kern_sg_entries = 0;
9968
9969	if (data_len < alloc_len) {
9970		ctsio->residual = alloc_len - data_len;
9971		ctsio->kern_data_len = data_len;
9972		ctsio->kern_total_len = data_len;
9973	} else {
9974		ctsio->residual = 0;
9975		ctsio->kern_data_len = alloc_len;
9976		ctsio->kern_total_len = alloc_len;
9977	}
9978	ctsio->kern_data_resid = 0;
9979	ctsio->kern_rel_offset = 0;
9980	ctsio->kern_sg_entries = 0;
9981
9982	/*
9983	 * The control device is always connected.  The disk device, on the
9984	 * other hand, may not be online all the time.
9985	 */
9986	if (lun != NULL)
9987		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9988				     lun->be_lun->lun_type;
9989	else
9990		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9991	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9992	eid_ptr->page_length = data_len - 4;
9993	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9994	eid_ptr->flags3 = SVPD_EID_V_SUP;
9995
9996	ctsio->scsi_status = SCSI_STATUS_OK;
9997	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9998	ctsio->be_move_done = ctl_config_move_done;
9999	ctl_datamove((union ctl_io *)ctsio);
10000
10001	return (CTL_RETVAL_COMPLETE);
10002}
10003
10004static int
10005ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
10006{
10007	struct scsi_vpd_mode_page_policy *mpp_ptr;
10008	struct ctl_lun *lun;
10009	int data_len;
10010
10011	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10012
10013	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
10014	    sizeof(struct scsi_vpd_mode_page_policy_descr);
10015
10016	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10017	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
10018	ctsio->kern_sg_entries = 0;
10019
10020	if (data_len < alloc_len) {
10021		ctsio->residual = alloc_len - data_len;
10022		ctsio->kern_data_len = data_len;
10023		ctsio->kern_total_len = data_len;
10024	} else {
10025		ctsio->residual = 0;
10026		ctsio->kern_data_len = alloc_len;
10027		ctsio->kern_total_len = alloc_len;
10028	}
10029	ctsio->kern_data_resid = 0;
10030	ctsio->kern_rel_offset = 0;
10031	ctsio->kern_sg_entries = 0;
10032
10033	/*
10034	 * The control device is always connected.  The disk device, on the
10035	 * other hand, may not be online all the time.
10036	 */
10037	if (lun != NULL)
10038		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10039				     lun->be_lun->lun_type;
10040	else
10041		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10042	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
10043	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
10044	mpp_ptr->descr[0].page_code = 0x3f;
10045	mpp_ptr->descr[0].subpage_code = 0xff;
10046	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
10047
10048	ctsio->scsi_status = SCSI_STATUS_OK;
10049	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10050	ctsio->be_move_done = ctl_config_move_done;
10051	ctl_datamove((union ctl_io *)ctsio);
10052
10053	return (CTL_RETVAL_COMPLETE);
10054}
10055
10056static int
10057ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
10058{
10059	struct scsi_vpd_device_id *devid_ptr;
10060	struct scsi_vpd_id_descriptor *desc;
10061	struct ctl_softc *ctl_softc;
10062	struct ctl_lun *lun;
10063	struct ctl_port *port;
10064	int data_len;
10065	uint8_t proto;
10066
10067	ctl_softc = control_softc;
10068
10069	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
10070	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10071
10072	data_len = sizeof(struct scsi_vpd_device_id) +
10073	    sizeof(struct scsi_vpd_id_descriptor) +
10074		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
10075	    sizeof(struct scsi_vpd_id_descriptor) +
10076		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
10077	if (lun && lun->lun_devid)
10078		data_len += lun->lun_devid->len;
10079	if (port->port_devid)
10080		data_len += port->port_devid->len;
10081	if (port->target_devid)
10082		data_len += port->target_devid->len;
10083
10084	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10085	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
10086	ctsio->kern_sg_entries = 0;
10087
10088	if (data_len < alloc_len) {
10089		ctsio->residual = alloc_len - data_len;
10090		ctsio->kern_data_len = data_len;
10091		ctsio->kern_total_len = data_len;
10092	} else {
10093		ctsio->residual = 0;
10094		ctsio->kern_data_len = alloc_len;
10095		ctsio->kern_total_len = alloc_len;
10096	}
10097	ctsio->kern_data_resid = 0;
10098	ctsio->kern_rel_offset = 0;
10099	ctsio->kern_sg_entries = 0;
10100
10101	/*
10102	 * The control device is always connected.  The disk device, on the
10103	 * other hand, may not be online all the time.
10104	 */
10105	if (lun != NULL)
10106		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10107				     lun->be_lun->lun_type;
10108	else
10109		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10110	devid_ptr->page_code = SVPD_DEVICE_ID;
10111	scsi_ulto2b(data_len - 4, devid_ptr->length);
10112
10113	if (port->port_type == CTL_PORT_FC)
10114		proto = SCSI_PROTO_FC << 4;
10115	else if (port->port_type == CTL_PORT_ISCSI)
10116		proto = SCSI_PROTO_ISCSI << 4;
10117	else
10118		proto = SCSI_PROTO_SPI << 4;
10119	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
10120
10121	/*
10122	 * We're using a LUN association here.  i.e., this device ID is a
10123	 * per-LUN identifier.
10124	 */
10125	if (lun && lun->lun_devid) {
10126		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
10127		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10128		    lun->lun_devid->len);
10129	}
10130
10131	/*
10132	 * This is for the WWPN which is a port association.
10133	 */
10134	if (port->port_devid) {
10135		memcpy(desc, port->port_devid->data, port->port_devid->len);
10136		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10137		    port->port_devid->len);
10138	}
10139
10140	/*
10141	 * This is for the Relative Target Port(type 4h) identifier
10142	 */
10143	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10144	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10145	    SVPD_ID_TYPE_RELTARG;
10146	desc->length = 4;
10147	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
10148	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10149	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
10150
10151	/*
10152	 * This is for the Target Port Group(type 5h) identifier
10153	 */
10154	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10155	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10156	    SVPD_ID_TYPE_TPORTGRP;
10157	desc->length = 4;
10158	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
10159	    &desc->identifier[2]);
10160	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10161	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10162
10163	/*
10164	 * This is for the Target identifier
10165	 */
10166	if (port->target_devid) {
10167		memcpy(desc, port->target_devid->data, port->target_devid->len);
10168	}
10169
10170	ctsio->scsi_status = SCSI_STATUS_OK;
10171	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10172	ctsio->be_move_done = ctl_config_move_done;
10173	ctl_datamove((union ctl_io *)ctsio);
10174
10175	return (CTL_RETVAL_COMPLETE);
10176}
10177
10178static int
10179ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10180{
10181	struct ctl_softc *softc = control_softc;
10182	struct scsi_vpd_scsi_ports *sp;
10183	struct scsi_vpd_port_designation *pd;
10184	struct scsi_vpd_port_designation_cont *pdc;
10185	struct ctl_lun *lun;
10186	struct ctl_port *port;
10187	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10188	int num_target_port_groups, single;
10189
10190	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10191
10192	single = ctl_is_single;
10193	if (single)
10194		num_target_port_groups = 1;
10195	else
10196		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10197	num_target_ports = 0;
10198	iid_len = 0;
10199	id_len = 0;
10200	mtx_lock(&softc->ctl_lock);
10201	STAILQ_FOREACH(port, &softc->port_list, links) {
10202		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10203			continue;
10204		if (lun != NULL &&
10205		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10206		    CTL_MAX_LUNS)
10207			continue;
10208		num_target_ports++;
10209		if (port->init_devid)
10210			iid_len += port->init_devid->len;
10211		if (port->port_devid)
10212			id_len += port->port_devid->len;
10213	}
10214	mtx_unlock(&softc->ctl_lock);
10215
10216	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10217	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10218	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10219	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10220	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10221	ctsio->kern_sg_entries = 0;
10222
10223	if (data_len < alloc_len) {
10224		ctsio->residual = alloc_len - data_len;
10225		ctsio->kern_data_len = data_len;
10226		ctsio->kern_total_len = data_len;
10227	} else {
10228		ctsio->residual = 0;
10229		ctsio->kern_data_len = alloc_len;
10230		ctsio->kern_total_len = alloc_len;
10231	}
10232	ctsio->kern_data_resid = 0;
10233	ctsio->kern_rel_offset = 0;
10234	ctsio->kern_sg_entries = 0;
10235
10236	/*
10237	 * The control device is always connected.  The disk device, on the
10238	 * other hand, may not be online all the time.  Need to change this
10239	 * to figure out whether the disk device is actually online or not.
10240	 */
10241	if (lun != NULL)
10242		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10243				  lun->be_lun->lun_type;
10244	else
10245		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10246
10247	sp->page_code = SVPD_SCSI_PORTS;
10248	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10249	    sp->page_length);
10250	pd = &sp->design[0];
10251
10252	mtx_lock(&softc->ctl_lock);
10253	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10254		pg = 0;
10255	else
10256		pg = 1;
10257	for (g = 0; g < num_target_port_groups; g++) {
10258		STAILQ_FOREACH(port, &softc->port_list, links) {
10259			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10260				continue;
10261			if (lun != NULL &&
10262			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10263			    CTL_MAX_LUNS)
10264				continue;
10265			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10266			scsi_ulto2b(p, pd->relative_port_id);
10267			if (port->init_devid && g == pg) {
10268				iid_len = port->init_devid->len;
10269				memcpy(pd->initiator_transportid,
10270				    port->init_devid->data, port->init_devid->len);
10271			} else
10272				iid_len = 0;
10273			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10274			pdc = (struct scsi_vpd_port_designation_cont *)
10275			    (&pd->initiator_transportid[iid_len]);
10276			if (port->port_devid && g == pg) {
10277				id_len = port->port_devid->len;
10278				memcpy(pdc->target_port_descriptors,
10279				    port->port_devid->data, port->port_devid->len);
10280			} else
10281				id_len = 0;
10282			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10283			pd = (struct scsi_vpd_port_designation *)
10284			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10285		}
10286	}
10287	mtx_unlock(&softc->ctl_lock);
10288
10289	ctsio->scsi_status = SCSI_STATUS_OK;
10290	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10291	ctsio->be_move_done = ctl_config_move_done;
10292	ctl_datamove((union ctl_io *)ctsio);
10293
10294	return (CTL_RETVAL_COMPLETE);
10295}
10296
10297static int
10298ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10299{
10300	struct scsi_vpd_block_limits *bl_ptr;
10301	struct ctl_lun *lun;
10302	int bs;
10303
10304	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10305
10306	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10307	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10308	ctsio->kern_sg_entries = 0;
10309
10310	if (sizeof(*bl_ptr) < alloc_len) {
10311		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10312		ctsio->kern_data_len = sizeof(*bl_ptr);
10313		ctsio->kern_total_len = sizeof(*bl_ptr);
10314	} else {
10315		ctsio->residual = 0;
10316		ctsio->kern_data_len = alloc_len;
10317		ctsio->kern_total_len = alloc_len;
10318	}
10319	ctsio->kern_data_resid = 0;
10320	ctsio->kern_rel_offset = 0;
10321	ctsio->kern_sg_entries = 0;
10322
10323	/*
10324	 * The control device is always connected.  The disk device, on the
10325	 * other hand, may not be online all the time.  Need to change this
10326	 * to figure out whether the disk device is actually online or not.
10327	 */
10328	if (lun != NULL)
10329		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10330				  lun->be_lun->lun_type;
10331	else
10332		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10333
10334	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10335	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
10336	bl_ptr->max_cmp_write_len = 0xff;
10337	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10338	if (lun != NULL) {
10339		bs = lun->be_lun->blocksize;
10340		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10341		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10342			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10343			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10344			if (lun->be_lun->pblockexp != 0) {
10345				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10346				    bl_ptr->opt_unmap_grain);
10347				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10348				    bl_ptr->unmap_grain_align);
10349			}
10350		}
10351		scsi_ulto4b(lun->be_lun->atomicblock,
10352		    bl_ptr->max_atomic_transfer_length);
10353		scsi_ulto4b(0, bl_ptr->atomic_alignment);
10354		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10355	}
10356	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10357
10358	ctsio->scsi_status = SCSI_STATUS_OK;
10359	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10360	ctsio->be_move_done = ctl_config_move_done;
10361	ctl_datamove((union ctl_io *)ctsio);
10362
10363	return (CTL_RETVAL_COMPLETE);
10364}
10365
10366static int
10367ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10368{
10369	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10370	struct ctl_lun *lun;
10371
10372	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10373
10374	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10375	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10376	ctsio->kern_sg_entries = 0;
10377
10378	if (sizeof(*bdc_ptr) < alloc_len) {
10379		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10380		ctsio->kern_data_len = sizeof(*bdc_ptr);
10381		ctsio->kern_total_len = sizeof(*bdc_ptr);
10382	} else {
10383		ctsio->residual = 0;
10384		ctsio->kern_data_len = alloc_len;
10385		ctsio->kern_total_len = alloc_len;
10386	}
10387	ctsio->kern_data_resid = 0;
10388	ctsio->kern_rel_offset = 0;
10389	ctsio->kern_sg_entries = 0;
10390
10391	/*
10392	 * The control device is always connected.  The disk device, on the
10393	 * other hand, may not be online all the time.  Need to change this
10394	 * to figure out whether the disk device is actually online or not.
10395	 */
10396	if (lun != NULL)
10397		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10398				  lun->be_lun->lun_type;
10399	else
10400		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10401	bdc_ptr->page_code = SVPD_BDC;
10402	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10403	scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10404	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10405
10406	ctsio->scsi_status = SCSI_STATUS_OK;
10407	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10408	ctsio->be_move_done = ctl_config_move_done;
10409	ctl_datamove((union ctl_io *)ctsio);
10410
10411	return (CTL_RETVAL_COMPLETE);
10412}
10413
10414static int
10415ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10416{
10417	struct scsi_vpd_logical_block_prov *lbp_ptr;
10418	struct ctl_lun *lun;
10419
10420	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10421
10422	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10423	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10424	ctsio->kern_sg_entries = 0;
10425
10426	if (sizeof(*lbp_ptr) < alloc_len) {
10427		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10428		ctsio->kern_data_len = sizeof(*lbp_ptr);
10429		ctsio->kern_total_len = sizeof(*lbp_ptr);
10430	} else {
10431		ctsio->residual = 0;
10432		ctsio->kern_data_len = alloc_len;
10433		ctsio->kern_total_len = alloc_len;
10434	}
10435	ctsio->kern_data_resid = 0;
10436	ctsio->kern_rel_offset = 0;
10437	ctsio->kern_sg_entries = 0;
10438
10439	/*
10440	 * The control device is always connected.  The disk device, on the
10441	 * other hand, may not be online all the time.  Need to change this
10442	 * to figure out whether the disk device is actually online or not.
10443	 */
10444	if (lun != NULL)
10445		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10446				  lun->be_lun->lun_type;
10447	else
10448		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10449
10450	lbp_ptr->page_code = SVPD_LBP;
10451	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10452	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10453		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10454		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10455		lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10456	}
10457
10458	ctsio->scsi_status = SCSI_STATUS_OK;
10459	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10460	ctsio->be_move_done = ctl_config_move_done;
10461	ctl_datamove((union ctl_io *)ctsio);
10462
10463	return (CTL_RETVAL_COMPLETE);
10464}
10465
10466static int
10467ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10468{
10469	struct scsi_inquiry *cdb;
10470	struct ctl_lun *lun;
10471	int alloc_len, retval;
10472
10473	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10474	cdb = (struct scsi_inquiry *)ctsio->cdb;
10475
10476	retval = CTL_RETVAL_COMPLETE;
10477
10478	alloc_len = scsi_2btoul(cdb->length);
10479
10480	switch (cdb->page_code) {
10481	case SVPD_SUPPORTED_PAGES:
10482		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10483		break;
10484	case SVPD_UNIT_SERIAL_NUMBER:
10485		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10486		break;
10487	case SVPD_DEVICE_ID:
10488		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10489		break;
10490	case SVPD_EXTENDED_INQUIRY_DATA:
10491		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10492		break;
10493	case SVPD_MODE_PAGE_POLICY:
10494		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10495		break;
10496	case SVPD_SCSI_PORTS:
10497		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10498		break;
10499	case SVPD_SCSI_TPC:
10500		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10501		break;
10502	case SVPD_BLOCK_LIMITS:
10503		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10504		break;
10505	case SVPD_BDC:
10506		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10507		break;
10508	case SVPD_LBP:
10509		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10510		break;
10511	default:
10512		ctl_set_invalid_field(ctsio,
10513				      /*sks_valid*/ 1,
10514				      /*command*/ 1,
10515				      /*field*/ 2,
10516				      /*bit_valid*/ 0,
10517				      /*bit*/ 0);
10518		ctl_done((union ctl_io *)ctsio);
10519		retval = CTL_RETVAL_COMPLETE;
10520		break;
10521	}
10522
10523	return (retval);
10524}
10525
10526static int
10527ctl_inquiry_std(struct ctl_scsiio *ctsio)
10528{
10529	struct scsi_inquiry_data *inq_ptr;
10530	struct scsi_inquiry *cdb;
10531	struct ctl_softc *ctl_softc;
10532	struct ctl_lun *lun;
10533	char *val;
10534	uint32_t alloc_len, data_len;
10535	ctl_port_type port_type;
10536
10537	ctl_softc = control_softc;
10538
10539	/*
10540	 * Figure out whether we're talking to a Fibre Channel port or not.
10541	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10542	 * SCSI front ends.
10543	 */
10544	port_type = ctl_softc->ctl_ports[
10545	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10546	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10547		port_type = CTL_PORT_SCSI;
10548
10549	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10550	cdb = (struct scsi_inquiry *)ctsio->cdb;
10551	alloc_len = scsi_2btoul(cdb->length);
10552
10553	/*
10554	 * We malloc the full inquiry data size here and fill it
10555	 * in.  If the user only asks for less, we'll give him
10556	 * that much.
10557	 */
10558	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10559	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10560	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10561	ctsio->kern_sg_entries = 0;
10562	ctsio->kern_data_resid = 0;
10563	ctsio->kern_rel_offset = 0;
10564
10565	if (data_len < alloc_len) {
10566		ctsio->residual = alloc_len - data_len;
10567		ctsio->kern_data_len = data_len;
10568		ctsio->kern_total_len = data_len;
10569	} else {
10570		ctsio->residual = 0;
10571		ctsio->kern_data_len = alloc_len;
10572		ctsio->kern_total_len = alloc_len;
10573	}
10574
10575	/*
10576	 * If we have a LUN configured, report it as connected.  Otherwise,
10577	 * report that it is offline or no device is supported, depending
10578	 * on the value of inquiry_pq_no_lun.
10579	 *
10580	 * According to the spec (SPC-4 r34), the peripheral qualifier
10581	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10582	 *
10583	 * "A peripheral device having the specified peripheral device type
10584	 * is not connected to this logical unit. However, the device
10585	 * server is capable of supporting the specified peripheral device
10586	 * type on this logical unit."
10587	 *
10588	 * According to the same spec, the peripheral qualifier
10589	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10590	 *
10591	 * "The device server is not capable of supporting a peripheral
10592	 * device on this logical unit. For this peripheral qualifier the
10593	 * peripheral device type shall be set to 1Fh. All other peripheral
10594	 * device type values are reserved for this peripheral qualifier."
10595	 *
10596	 * Given the text, it would seem that we probably want to report that
10597	 * the LUN is offline here.  There is no LUN connected, but we can
10598	 * support a LUN at the given LUN number.
10599	 *
10600	 * In the real world, though, it sounds like things are a little
10601	 * different:
10602	 *
10603	 * - Linux, when presented with a LUN with the offline peripheral
10604	 *   qualifier, will create an sg driver instance for it.  So when
10605	 *   you attach it to CTL, you wind up with a ton of sg driver
10606	 *   instances.  (One for every LUN that Linux bothered to probe.)
10607	 *   Linux does this despite the fact that it issues a REPORT LUNs
10608	 *   to LUN 0 to get the inventory of supported LUNs.
10609	 *
10610	 * - There is other anecdotal evidence (from Emulex folks) about
10611	 *   arrays that use the offline peripheral qualifier for LUNs that
10612	 *   are on the "passive" path in an active/passive array.
10613	 *
10614	 * So the solution is provide a hopefully reasonable default
10615	 * (return bad/no LUN) and allow the user to change the behavior
10616	 * with a tunable/sysctl variable.
10617	 */
10618	if (lun != NULL)
10619		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10620				  lun->be_lun->lun_type;
10621	else if (ctl_softc->inquiry_pq_no_lun == 0)
10622		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10623	else
10624		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10625
10626	/* RMB in byte 2 is 0 */
10627	inq_ptr->version = SCSI_REV_SPC4;
10628
10629	/*
10630	 * According to SAM-3, even if a device only supports a single
10631	 * level of LUN addressing, it should still set the HISUP bit:
10632	 *
10633	 * 4.9.1 Logical unit numbers overview
10634	 *
10635	 * All logical unit number formats described in this standard are
10636	 * hierarchical in structure even when only a single level in that
10637	 * hierarchy is used. The HISUP bit shall be set to one in the
10638	 * standard INQUIRY data (see SPC-2) when any logical unit number
10639	 * format described in this standard is used.  Non-hierarchical
10640	 * formats are outside the scope of this standard.
10641	 *
10642	 * Therefore we set the HiSup bit here.
10643	 *
10644	 * The reponse format is 2, per SPC-3.
10645	 */
10646	inq_ptr->response_format = SID_HiSup | 2;
10647
10648	inq_ptr->additional_length = data_len -
10649	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10650	CTL_DEBUG_PRINT(("additional_length = %d\n",
10651			 inq_ptr->additional_length));
10652
10653	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10654	/* 16 bit addressing */
10655	if (port_type == CTL_PORT_SCSI)
10656		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10657	/* XXX set the SID_MultiP bit here if we're actually going to
10658	   respond on multiple ports */
10659	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10660
10661	/* 16 bit data bus, synchronous transfers */
10662	if (port_type == CTL_PORT_SCSI)
10663		inq_ptr->flags = SID_WBus16 | SID_Sync;
10664	/*
10665	 * XXX KDM do we want to support tagged queueing on the control
10666	 * device at all?
10667	 */
10668	if ((lun == NULL)
10669	 || (lun->be_lun->lun_type != T_PROCESSOR))
10670		inq_ptr->flags |= SID_CmdQue;
10671	/*
10672	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10673	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10674	 * name and 4 bytes for the revision.
10675	 */
10676	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10677	    "vendor")) == NULL) {
10678		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10679	} else {
10680		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10681		strncpy(inq_ptr->vendor, val,
10682		    min(sizeof(inq_ptr->vendor), strlen(val)));
10683	}
10684	if (lun == NULL) {
10685		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10686		    sizeof(inq_ptr->product));
10687	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10688		switch (lun->be_lun->lun_type) {
10689		case T_DIRECT:
10690			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10691			    sizeof(inq_ptr->product));
10692			break;
10693		case T_PROCESSOR:
10694			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10695			    sizeof(inq_ptr->product));
10696			break;
10697		default:
10698			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10699			    sizeof(inq_ptr->product));
10700			break;
10701		}
10702	} else {
10703		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10704		strncpy(inq_ptr->product, val,
10705		    min(sizeof(inq_ptr->product), strlen(val)));
10706	}
10707
10708	/*
10709	 * XXX make this a macro somewhere so it automatically gets
10710	 * incremented when we make changes.
10711	 */
10712	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10713	    "revision")) == NULL) {
10714		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10715	} else {
10716		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10717		strncpy(inq_ptr->revision, val,
10718		    min(sizeof(inq_ptr->revision), strlen(val)));
10719	}
10720
10721	/*
10722	 * For parallel SCSI, we support double transition and single
10723	 * transition clocking.  We also support QAS (Quick Arbitration
10724	 * and Selection) and Information Unit transfers on both the
10725	 * control and array devices.
10726	 */
10727	if (port_type == CTL_PORT_SCSI)
10728		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10729				    SID_SPI_IUS;
10730
10731	/* SAM-5 (no version claimed) */
10732	scsi_ulto2b(0x00A0, inq_ptr->version1);
10733	/* SPC-4 (no version claimed) */
10734	scsi_ulto2b(0x0460, inq_ptr->version2);
10735	if (port_type == CTL_PORT_FC) {
10736		/* FCP-2 ANSI INCITS.350:2003 */
10737		scsi_ulto2b(0x0917, inq_ptr->version3);
10738	} else if (port_type == CTL_PORT_SCSI) {
10739		/* SPI-4 ANSI INCITS.362:200x */
10740		scsi_ulto2b(0x0B56, inq_ptr->version3);
10741	} else if (port_type == CTL_PORT_ISCSI) {
10742		/* iSCSI (no version claimed) */
10743		scsi_ulto2b(0x0960, inq_ptr->version3);
10744	} else if (port_type == CTL_PORT_SAS) {
10745		/* SAS (no version claimed) */
10746		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10747	}
10748
10749	if (lun == NULL) {
10750		/* SBC-4 (no version claimed) */
10751		scsi_ulto2b(0x0600, inq_ptr->version4);
10752	} else {
10753		switch (lun->be_lun->lun_type) {
10754		case T_DIRECT:
10755			/* SBC-4 (no version claimed) */
10756			scsi_ulto2b(0x0600, inq_ptr->version4);
10757			break;
10758		case T_PROCESSOR:
10759		default:
10760			break;
10761		}
10762	}
10763
10764	ctsio->scsi_status = SCSI_STATUS_OK;
10765	if (ctsio->kern_data_len > 0) {
10766		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10767		ctsio->be_move_done = ctl_config_move_done;
10768		ctl_datamove((union ctl_io *)ctsio);
10769	} else {
10770		ctsio->io_hdr.status = CTL_SUCCESS;
10771		ctl_done((union ctl_io *)ctsio);
10772	}
10773
10774	return (CTL_RETVAL_COMPLETE);
10775}
10776
10777int
10778ctl_inquiry(struct ctl_scsiio *ctsio)
10779{
10780	struct scsi_inquiry *cdb;
10781	int retval;
10782
10783	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10784
10785	cdb = (struct scsi_inquiry *)ctsio->cdb;
10786	if (cdb->byte2 & SI_EVPD)
10787		retval = ctl_inquiry_evpd(ctsio);
10788	else if (cdb->page_code == 0)
10789		retval = ctl_inquiry_std(ctsio);
10790	else {
10791		ctl_set_invalid_field(ctsio,
10792				      /*sks_valid*/ 1,
10793				      /*command*/ 1,
10794				      /*field*/ 2,
10795				      /*bit_valid*/ 0,
10796				      /*bit*/ 0);
10797		ctl_done((union ctl_io *)ctsio);
10798		return (CTL_RETVAL_COMPLETE);
10799	}
10800
10801	return (retval);
10802}
10803
10804/*
10805 * For known CDB types, parse the LBA and length.
10806 */
10807static int
10808ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10809{
10810	if (io->io_hdr.io_type != CTL_IO_SCSI)
10811		return (1);
10812
10813	switch (io->scsiio.cdb[0]) {
10814	case COMPARE_AND_WRITE: {
10815		struct scsi_compare_and_write *cdb;
10816
10817		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10818
10819		*lba = scsi_8btou64(cdb->addr);
10820		*len = cdb->length;
10821		break;
10822	}
10823	case READ_6:
10824	case WRITE_6: {
10825		struct scsi_rw_6 *cdb;
10826
10827		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10828
10829		*lba = scsi_3btoul(cdb->addr);
10830		/* only 5 bits are valid in the most significant address byte */
10831		*lba &= 0x1fffff;
10832		*len = cdb->length;
10833		break;
10834	}
10835	case READ_10:
10836	case WRITE_10: {
10837		struct scsi_rw_10 *cdb;
10838
10839		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10840
10841		*lba = scsi_4btoul(cdb->addr);
10842		*len = scsi_2btoul(cdb->length);
10843		break;
10844	}
10845	case WRITE_VERIFY_10: {
10846		struct scsi_write_verify_10 *cdb;
10847
10848		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10849
10850		*lba = scsi_4btoul(cdb->addr);
10851		*len = scsi_2btoul(cdb->length);
10852		break;
10853	}
10854	case READ_12:
10855	case WRITE_12: {
10856		struct scsi_rw_12 *cdb;
10857
10858		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10859
10860		*lba = scsi_4btoul(cdb->addr);
10861		*len = scsi_4btoul(cdb->length);
10862		break;
10863	}
10864	case WRITE_VERIFY_12: {
10865		struct scsi_write_verify_12 *cdb;
10866
10867		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10868
10869		*lba = scsi_4btoul(cdb->addr);
10870		*len = scsi_4btoul(cdb->length);
10871		break;
10872	}
10873	case READ_16:
10874	case WRITE_16:
10875	case WRITE_ATOMIC_16: {
10876		struct scsi_rw_16 *cdb;
10877
10878		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10879
10880		*lba = scsi_8btou64(cdb->addr);
10881		*len = scsi_4btoul(cdb->length);
10882		break;
10883	}
10884	case WRITE_VERIFY_16: {
10885		struct scsi_write_verify_16 *cdb;
10886
10887		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10888
10889		*lba = scsi_8btou64(cdb->addr);
10890		*len = scsi_4btoul(cdb->length);
10891		break;
10892	}
10893	case WRITE_SAME_10: {
10894		struct scsi_write_same_10 *cdb;
10895
10896		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10897
10898		*lba = scsi_4btoul(cdb->addr);
10899		*len = scsi_2btoul(cdb->length);
10900		break;
10901	}
10902	case WRITE_SAME_16: {
10903		struct scsi_write_same_16 *cdb;
10904
10905		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10906
10907		*lba = scsi_8btou64(cdb->addr);
10908		*len = scsi_4btoul(cdb->length);
10909		break;
10910	}
10911	case VERIFY_10: {
10912		struct scsi_verify_10 *cdb;
10913
10914		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10915
10916		*lba = scsi_4btoul(cdb->addr);
10917		*len = scsi_2btoul(cdb->length);
10918		break;
10919	}
10920	case VERIFY_12: {
10921		struct scsi_verify_12 *cdb;
10922
10923		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10924
10925		*lba = scsi_4btoul(cdb->addr);
10926		*len = scsi_4btoul(cdb->length);
10927		break;
10928	}
10929	case VERIFY_16: {
10930		struct scsi_verify_16 *cdb;
10931
10932		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10933
10934		*lba = scsi_8btou64(cdb->addr);
10935		*len = scsi_4btoul(cdb->length);
10936		break;
10937	}
10938	case UNMAP: {
10939		*lba = 0;
10940		*len = UINT64_MAX;
10941		break;
10942	}
10943	default:
10944		return (1);
10945		break; /* NOTREACHED */
10946	}
10947
10948	return (0);
10949}
10950
10951static ctl_action
10952ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10953{
10954	uint64_t endlba1, endlba2;
10955
10956	endlba1 = lba1 + len1 - 1;
10957	endlba2 = lba2 + len2 - 1;
10958
10959	if ((endlba1 < lba2)
10960	 || (endlba2 < lba1))
10961		return (CTL_ACTION_PASS);
10962	else
10963		return (CTL_ACTION_BLOCK);
10964}
10965
10966static int
10967ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10968{
10969	struct ctl_ptr_len_flags *ptrlen;
10970	struct scsi_unmap_desc *buf, *end, *range;
10971	uint64_t lba;
10972	uint32_t len;
10973
10974	/* If not UNMAP -- go other way. */
10975	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10976	    io->scsiio.cdb[0] != UNMAP)
10977		return (CTL_ACTION_ERROR);
10978
10979	/* If UNMAP without data -- block and wait for data. */
10980	ptrlen = (struct ctl_ptr_len_flags *)
10981	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10982	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10983	    ptrlen->ptr == NULL)
10984		return (CTL_ACTION_BLOCK);
10985
10986	/* UNMAP with data -- check for collision. */
10987	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10988	end = buf + ptrlen->len / sizeof(*buf);
10989	for (range = buf; range < end; range++) {
10990		lba = scsi_8btou64(range->lba);
10991		len = scsi_4btoul(range->length);
10992		if ((lba < lba2 + len2) && (lba + len > lba2))
10993			return (CTL_ACTION_BLOCK);
10994	}
10995	return (CTL_ACTION_PASS);
10996}
10997
10998static ctl_action
10999ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
11000{
11001	uint64_t lba1, lba2;
11002	uint64_t len1, len2;
11003	int retval;
11004
11005	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
11006		return (CTL_ACTION_ERROR);
11007
11008	retval = ctl_extent_check_unmap(io2, lba1, len1);
11009	if (retval != CTL_ACTION_ERROR)
11010		return (retval);
11011
11012	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
11013		return (CTL_ACTION_ERROR);
11014
11015	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
11016}
11017
11018static ctl_action
11019ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
11020    union ctl_io *ooa_io)
11021{
11022	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
11023	ctl_serialize_action *serialize_row;
11024
11025	/*
11026	 * The initiator attempted multiple untagged commands at the same
11027	 * time.  Can't do that.
11028	 */
11029	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11030	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11031	 && ((pending_io->io_hdr.nexus.targ_port ==
11032	      ooa_io->io_hdr.nexus.targ_port)
11033	  && (pending_io->io_hdr.nexus.initid.id ==
11034	      ooa_io->io_hdr.nexus.initid.id))
11035	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
11036		return (CTL_ACTION_OVERLAP);
11037
11038	/*
11039	 * The initiator attempted to send multiple tagged commands with
11040	 * the same ID.  (It's fine if different initiators have the same
11041	 * tag ID.)
11042	 *
11043	 * Even if all of those conditions are true, we don't kill the I/O
11044	 * if the command ahead of us has been aborted.  We won't end up
11045	 * sending it to the FETD, and it's perfectly legal to resend a
11046	 * command with the same tag number as long as the previous
11047	 * instance of this tag number has been aborted somehow.
11048	 */
11049	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11050	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11051	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11052	 && ((pending_io->io_hdr.nexus.targ_port ==
11053	      ooa_io->io_hdr.nexus.targ_port)
11054	  && (pending_io->io_hdr.nexus.initid.id ==
11055	      ooa_io->io_hdr.nexus.initid.id))
11056	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
11057		return (CTL_ACTION_OVERLAP_TAG);
11058
11059	/*
11060	 * If we get a head of queue tag, SAM-3 says that we should
11061	 * immediately execute it.
11062	 *
11063	 * What happens if this command would normally block for some other
11064	 * reason?  e.g. a request sense with a head of queue tag
11065	 * immediately after a write.  Normally that would block, but this
11066	 * will result in its getting executed immediately...
11067	 *
11068	 * We currently return "pass" instead of "skip", so we'll end up
11069	 * going through the rest of the queue to check for overlapped tags.
11070	 *
11071	 * XXX KDM check for other types of blockage first??
11072	 */
11073	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11074		return (CTL_ACTION_PASS);
11075
11076	/*
11077	 * Ordered tags have to block until all items ahead of them
11078	 * have completed.  If we get called with an ordered tag, we always
11079	 * block, if something else is ahead of us in the queue.
11080	 */
11081	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11082		return (CTL_ACTION_BLOCK);
11083
11084	/*
11085	 * Simple tags get blocked until all head of queue and ordered tags
11086	 * ahead of them have completed.  I'm lumping untagged commands in
11087	 * with simple tags here.  XXX KDM is that the right thing to do?
11088	 */
11089	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11090	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11091	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11092	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11093		return (CTL_ACTION_BLOCK);
11094
11095	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11096	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11097
11098	serialize_row = ctl_serialize_table[ooa_entry->seridx];
11099
11100	switch (serialize_row[pending_entry->seridx]) {
11101	case CTL_SER_BLOCK:
11102		return (CTL_ACTION_BLOCK);
11103	case CTL_SER_EXTENT:
11104		return (ctl_extent_check(pending_io, ooa_io));
11105	case CTL_SER_EXTENTOPT:
11106		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11107		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11108			return (ctl_extent_check(pending_io, ooa_io));
11109		/* FALLTHROUGH */
11110	case CTL_SER_PASS:
11111		return (CTL_ACTION_PASS);
11112	case CTL_SER_BLOCKOPT:
11113		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11114		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11115			return (CTL_ACTION_BLOCK);
11116		return (CTL_ACTION_PASS);
11117	case CTL_SER_SKIP:
11118		return (CTL_ACTION_SKIP);
11119	default:
11120		panic("invalid serialization value %d",
11121		      serialize_row[pending_entry->seridx]);
11122	}
11123
11124	return (CTL_ACTION_ERROR);
11125}
11126
11127/*
11128 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11129 * Assumptions:
11130 * - pending_io is generally either incoming, or on the blocked queue
11131 * - starting I/O is the I/O we want to start the check with.
11132 */
11133static ctl_action
11134ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11135	      union ctl_io *starting_io)
11136{
11137	union ctl_io *ooa_io;
11138	ctl_action action;
11139
11140	mtx_assert(&lun->lun_lock, MA_OWNED);
11141
11142	/*
11143	 * Run back along the OOA queue, starting with the current
11144	 * blocked I/O and going through every I/O before it on the
11145	 * queue.  If starting_io is NULL, we'll just end up returning
11146	 * CTL_ACTION_PASS.
11147	 */
11148	for (ooa_io = starting_io; ooa_io != NULL;
11149	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11150	     ooa_links)){
11151
11152		/*
11153		 * This routine just checks to see whether
11154		 * cur_blocked is blocked by ooa_io, which is ahead
11155		 * of it in the queue.  It doesn't queue/dequeue
11156		 * cur_blocked.
11157		 */
11158		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11159		switch (action) {
11160		case CTL_ACTION_BLOCK:
11161		case CTL_ACTION_OVERLAP:
11162		case CTL_ACTION_OVERLAP_TAG:
11163		case CTL_ACTION_SKIP:
11164		case CTL_ACTION_ERROR:
11165			return (action);
11166			break; /* NOTREACHED */
11167		case CTL_ACTION_PASS:
11168			break;
11169		default:
11170			panic("invalid action %d", action);
11171			break;  /* NOTREACHED */
11172		}
11173	}
11174
11175	return (CTL_ACTION_PASS);
11176}
11177
11178/*
11179 * Assumptions:
11180 * - An I/O has just completed, and has been removed from the per-LUN OOA
11181 *   queue, so some items on the blocked queue may now be unblocked.
11182 */
11183static int
11184ctl_check_blocked(struct ctl_lun *lun)
11185{
11186	union ctl_io *cur_blocked, *next_blocked;
11187
11188	mtx_assert(&lun->lun_lock, MA_OWNED);
11189
11190	/*
11191	 * Run forward from the head of the blocked queue, checking each
11192	 * entry against the I/Os prior to it on the OOA queue to see if
11193	 * there is still any blockage.
11194	 *
11195	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11196	 * with our removing a variable on it while it is traversing the
11197	 * list.
11198	 */
11199	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11200	     cur_blocked != NULL; cur_blocked = next_blocked) {
11201		union ctl_io *prev_ooa;
11202		ctl_action action;
11203
11204		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11205							  blocked_links);
11206
11207		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11208						      ctl_ooaq, ooa_links);
11209
11210		/*
11211		 * If cur_blocked happens to be the first item in the OOA
11212		 * queue now, prev_ooa will be NULL, and the action
11213		 * returned will just be CTL_ACTION_PASS.
11214		 */
11215		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11216
11217		switch (action) {
11218		case CTL_ACTION_BLOCK:
11219			/* Nothing to do here, still blocked */
11220			break;
11221		case CTL_ACTION_OVERLAP:
11222		case CTL_ACTION_OVERLAP_TAG:
11223			/*
11224			 * This shouldn't happen!  In theory we've already
11225			 * checked this command for overlap...
11226			 */
11227			break;
11228		case CTL_ACTION_PASS:
11229		case CTL_ACTION_SKIP: {
11230			struct ctl_softc *softc;
11231			const struct ctl_cmd_entry *entry;
11232			uint32_t initidx;
11233			int isc_retval;
11234
11235			/*
11236			 * The skip case shouldn't happen, this transaction
11237			 * should have never made it onto the blocked queue.
11238			 */
11239			/*
11240			 * This I/O is no longer blocked, we can remove it
11241			 * from the blocked queue.  Since this is a TAILQ
11242			 * (doubly linked list), we can do O(1) removals
11243			 * from any place on the list.
11244			 */
11245			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11246				     blocked_links);
11247			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11248
11249			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11250				/*
11251				 * Need to send IO back to original side to
11252				 * run
11253				 */
11254				union ctl_ha_msg msg_info;
11255
11256				msg_info.hdr.original_sc =
11257					cur_blocked->io_hdr.original_sc;
11258				msg_info.hdr.serializing_sc = cur_blocked;
11259				msg_info.hdr.msg_type = CTL_MSG_R2R;
11260				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11261				     &msg_info, sizeof(msg_info), 0)) >
11262				     CTL_HA_STATUS_SUCCESS) {
11263					printf("CTL:Check Blocked error from "
11264					       "ctl_ha_msg_send %d\n",
11265					       isc_retval);
11266				}
11267				break;
11268			}
11269			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11270			softc = control_softc;
11271
11272			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11273
11274			/*
11275			 * Check this I/O for LUN state changes that may
11276			 * have happened while this command was blocked.
11277			 * The LUN state may have been changed by a command
11278			 * ahead of us in the queue, so we need to re-check
11279			 * for any states that can be caused by SCSI
11280			 * commands.
11281			 */
11282			if (ctl_scsiio_lun_check(softc, lun, entry,
11283						 &cur_blocked->scsiio) == 0) {
11284				cur_blocked->io_hdr.flags |=
11285				                      CTL_FLAG_IS_WAS_ON_RTR;
11286				ctl_enqueue_rtr(cur_blocked);
11287			} else
11288				ctl_done(cur_blocked);
11289			break;
11290		}
11291		default:
11292			/*
11293			 * This probably shouldn't happen -- we shouldn't
11294			 * get CTL_ACTION_ERROR, or anything else.
11295			 */
11296			break;
11297		}
11298	}
11299
11300	return (CTL_RETVAL_COMPLETE);
11301}
11302
11303/*
11304 * This routine (with one exception) checks LUN flags that can be set by
11305 * commands ahead of us in the OOA queue.  These flags have to be checked
11306 * when a command initially comes in, and when we pull a command off the
11307 * blocked queue and are preparing to execute it.  The reason we have to
11308 * check these flags for commands on the blocked queue is that the LUN
11309 * state may have been changed by a command ahead of us while we're on the
11310 * blocked queue.
11311 *
11312 * Ordering is somewhat important with these checks, so please pay
11313 * careful attention to the placement of any new checks.
11314 */
11315static int
11316ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11317    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11318{
11319	int retval;
11320	uint32_t residx;
11321
11322	retval = 0;
11323
11324	mtx_assert(&lun->lun_lock, MA_OWNED);
11325
11326	/*
11327	 * If this shelf is a secondary shelf controller, we have to reject
11328	 * any media access commands.
11329	 */
11330#if 0
11331	/* No longer needed for HA */
11332	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11333	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11334		ctl_set_lun_standby(ctsio);
11335		retval = 1;
11336		goto bailout;
11337	}
11338#endif
11339
11340	if (entry->pattern & CTL_LUN_PAT_WRITE) {
11341		if (lun->flags & CTL_LUN_READONLY) {
11342			ctl_set_sense(ctsio, /*current_error*/ 1,
11343			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11344			    /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
11345			retval = 1;
11346			goto bailout;
11347		}
11348		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11349		    .eca_and_aen & SCP_SWP) != 0) {
11350			ctl_set_sense(ctsio, /*current_error*/ 1,
11351			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11352			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11353			retval = 1;
11354			goto bailout;
11355		}
11356	}
11357
11358	/*
11359	 * Check for a reservation conflict.  If this command isn't allowed
11360	 * even on reserved LUNs, and if this initiator isn't the one who
11361	 * reserved us, reject the command with a reservation conflict.
11362	 */
11363	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11364	if ((lun->flags & CTL_LUN_RESERVED)
11365	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11366		if (lun->res_idx != residx) {
11367			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11368			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11369			retval = 1;
11370			goto bailout;
11371		}
11372	}
11373
11374	if ((lun->flags & CTL_LUN_PR_RESERVED)
11375	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11376		/*
11377		 * if we aren't registered or it's a res holder type
11378		 * reservation and this isn't the res holder then set a
11379		 * conflict.
11380		 * NOTE: Commands which might be allowed on write exclusive
11381		 * type reservations are checked in the particular command
11382		 * for a conflict. Read and SSU are the only ones.
11383		 */
11384		if (!lun->per_res[residx].registered
11385		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11386			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11387			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11388			retval = 1;
11389			goto bailout;
11390		}
11391
11392	}
11393
11394	if ((lun->flags & CTL_LUN_OFFLINE)
11395	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11396		ctl_set_lun_not_ready(ctsio);
11397		retval = 1;
11398		goto bailout;
11399	}
11400
11401	/*
11402	 * If the LUN is stopped, see if this particular command is allowed
11403	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11404	 */
11405	if ((lun->flags & CTL_LUN_STOPPED)
11406	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11407		/* "Logical unit not ready, initializing cmd. required" */
11408		ctl_set_lun_stopped(ctsio);
11409		retval = 1;
11410		goto bailout;
11411	}
11412
11413	if ((lun->flags & CTL_LUN_INOPERABLE)
11414	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11415		/* "Medium format corrupted" */
11416		ctl_set_medium_format_corrupted(ctsio);
11417		retval = 1;
11418		goto bailout;
11419	}
11420
11421bailout:
11422	return (retval);
11423
11424}
11425
11426static void
11427ctl_failover_io(union ctl_io *io, int have_lock)
11428{
11429	ctl_set_busy(&io->scsiio);
11430	ctl_done(io);
11431}
11432
11433static void
11434ctl_failover(void)
11435{
11436	struct ctl_lun *lun;
11437	struct ctl_softc *ctl_softc;
11438	union ctl_io *next_io, *pending_io;
11439	union ctl_io *io;
11440	int lun_idx;
11441	int i;
11442
11443	ctl_softc = control_softc;
11444
11445	mtx_lock(&ctl_softc->ctl_lock);
11446	/*
11447	 * Remove any cmds from the other SC from the rtr queue.  These
11448	 * will obviously only be for LUNs for which we're the primary.
11449	 * We can't send status or get/send data for these commands.
11450	 * Since they haven't been executed yet, we can just remove them.
11451	 * We'll either abort them or delete them below, depending on
11452	 * which HA mode we're in.
11453	 */
11454#ifdef notyet
11455	mtx_lock(&ctl_softc->queue_lock);
11456	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11457	     io != NULL; io = next_io) {
11458		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11459		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11460			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11461				      ctl_io_hdr, links);
11462	}
11463	mtx_unlock(&ctl_softc->queue_lock);
11464#endif
11465
11466	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11467		lun = ctl_softc->ctl_luns[lun_idx];
11468		if (lun==NULL)
11469			continue;
11470
11471		/*
11472		 * Processor LUNs are primary on both sides.
11473		 * XXX will this always be true?
11474		 */
11475		if (lun->be_lun->lun_type == T_PROCESSOR)
11476			continue;
11477
11478		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11479		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11480			printf("FAILOVER: primary lun %d\n", lun_idx);
11481		        /*
11482			 * Remove all commands from the other SC. First from the
11483			 * blocked queue then from the ooa queue. Once we have
11484			 * removed them. Call ctl_check_blocked to see if there
11485			 * is anything that can run.
11486			 */
11487			for (io = (union ctl_io *)TAILQ_FIRST(
11488			     &lun->blocked_queue); io != NULL; io = next_io) {
11489
11490		        	next_io = (union ctl_io *)TAILQ_NEXT(
11491				    &io->io_hdr, blocked_links);
11492
11493				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11494					TAILQ_REMOVE(&lun->blocked_queue,
11495						     &io->io_hdr,blocked_links);
11496					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11497					TAILQ_REMOVE(&lun->ooa_queue,
11498						     &io->io_hdr, ooa_links);
11499
11500					ctl_free_io(io);
11501				}
11502			}
11503
11504			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11505	     		     io != NULL; io = next_io) {
11506
11507		        	next_io = (union ctl_io *)TAILQ_NEXT(
11508				    &io->io_hdr, ooa_links);
11509
11510				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11511
11512					TAILQ_REMOVE(&lun->ooa_queue,
11513						&io->io_hdr,
11514					     	ooa_links);
11515
11516					ctl_free_io(io);
11517				}
11518			}
11519			ctl_check_blocked(lun);
11520		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11521			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11522
11523			printf("FAILOVER: primary lun %d\n", lun_idx);
11524			/*
11525			 * Abort all commands from the other SC.  We can't
11526			 * send status back for them now.  These should get
11527			 * cleaned up when they are completed or come out
11528			 * for a datamove operation.
11529			 */
11530			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11531	     		     io != NULL; io = next_io) {
11532		        	next_io = (union ctl_io *)TAILQ_NEXT(
11533					&io->io_hdr, ooa_links);
11534
11535				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11536					io->io_hdr.flags |= CTL_FLAG_ABORT;
11537			}
11538		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11539			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11540
11541			printf("FAILOVER: secondary lun %d\n", lun_idx);
11542
11543			lun->flags |= CTL_LUN_PRIMARY_SC;
11544
11545			/*
11546			 * We send all I/O that was sent to this controller
11547			 * and redirected to the other side back with
11548			 * busy status, and have the initiator retry it.
11549			 * Figuring out how much data has been transferred,
11550			 * etc. and picking up where we left off would be
11551			 * very tricky.
11552			 *
11553			 * XXX KDM need to remove I/O from the blocked
11554			 * queue as well!
11555			 */
11556			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11557			     &lun->ooa_queue); pending_io != NULL;
11558			     pending_io = next_io) {
11559
11560				next_io =  (union ctl_io *)TAILQ_NEXT(
11561					&pending_io->io_hdr, ooa_links);
11562
11563				pending_io->io_hdr.flags &=
11564					~CTL_FLAG_SENT_2OTHER_SC;
11565
11566				if (pending_io->io_hdr.flags &
11567				    CTL_FLAG_IO_ACTIVE) {
11568					pending_io->io_hdr.flags |=
11569						CTL_FLAG_FAILOVER;
11570				} else {
11571					ctl_set_busy(&pending_io->scsiio);
11572					ctl_done(pending_io);
11573				}
11574			}
11575
11576			/*
11577			 * Build Unit Attention
11578			 */
11579			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11580				lun->pending_ua[i] |=
11581				                     CTL_UA_ASYM_ACC_CHANGE;
11582			}
11583		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11584			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11585			printf("FAILOVER: secondary lun %d\n", lun_idx);
11586			/*
11587			 * if the first io on the OOA is not on the RtR queue
11588			 * add it.
11589			 */
11590			lun->flags |= CTL_LUN_PRIMARY_SC;
11591
11592			pending_io = (union ctl_io *)TAILQ_FIRST(
11593			    &lun->ooa_queue);
11594			if (pending_io==NULL) {
11595				printf("Nothing on OOA queue\n");
11596				continue;
11597			}
11598
11599			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11600			if ((pending_io->io_hdr.flags &
11601			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11602				pending_io->io_hdr.flags |=
11603				    CTL_FLAG_IS_WAS_ON_RTR;
11604				ctl_enqueue_rtr(pending_io);
11605			}
11606#if 0
11607			else
11608			{
11609				printf("Tag 0x%04x is running\n",
11610				      pending_io->scsiio.tag_num);
11611			}
11612#endif
11613
11614			next_io = (union ctl_io *)TAILQ_NEXT(
11615			    &pending_io->io_hdr, ooa_links);
11616			for (pending_io=next_io; pending_io != NULL;
11617			     pending_io = next_io) {
11618				pending_io->io_hdr.flags &=
11619				    ~CTL_FLAG_SENT_2OTHER_SC;
11620				next_io = (union ctl_io *)TAILQ_NEXT(
11621					&pending_io->io_hdr, ooa_links);
11622				if (pending_io->io_hdr.flags &
11623				    CTL_FLAG_IS_WAS_ON_RTR) {
11624#if 0
11625				        printf("Tag 0x%04x is running\n",
11626				      		pending_io->scsiio.tag_num);
11627#endif
11628					continue;
11629				}
11630
11631				switch (ctl_check_ooa(lun, pending_io,
11632			            (union ctl_io *)TAILQ_PREV(
11633				    &pending_io->io_hdr, ctl_ooaq,
11634				    ooa_links))) {
11635
11636				case CTL_ACTION_BLOCK:
11637					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11638							  &pending_io->io_hdr,
11639							  blocked_links);
11640					pending_io->io_hdr.flags |=
11641					    CTL_FLAG_BLOCKED;
11642					break;
11643				case CTL_ACTION_PASS:
11644				case CTL_ACTION_SKIP:
11645					pending_io->io_hdr.flags |=
11646					    CTL_FLAG_IS_WAS_ON_RTR;
11647					ctl_enqueue_rtr(pending_io);
11648					break;
11649				case CTL_ACTION_OVERLAP:
11650					ctl_set_overlapped_cmd(
11651					    (struct ctl_scsiio *)pending_io);
11652					ctl_done(pending_io);
11653					break;
11654				case CTL_ACTION_OVERLAP_TAG:
11655					ctl_set_overlapped_tag(
11656					    (struct ctl_scsiio *)pending_io,
11657					    pending_io->scsiio.tag_num & 0xff);
11658					ctl_done(pending_io);
11659					break;
11660				case CTL_ACTION_ERROR:
11661				default:
11662					ctl_set_internal_failure(
11663						(struct ctl_scsiio *)pending_io,
11664						0,  // sks_valid
11665						0); //retry count
11666					ctl_done(pending_io);
11667					break;
11668				}
11669			}
11670
11671			/*
11672			 * Build Unit Attention
11673			 */
11674			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11675				lun->pending_ua[i] |=
11676				                     CTL_UA_ASYM_ACC_CHANGE;
11677			}
11678		} else {
11679			panic("Unhandled HA mode failover, LUN flags = %#x, "
11680			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11681		}
11682	}
11683	ctl_pause_rtr = 0;
11684	mtx_unlock(&ctl_softc->ctl_lock);
11685}
11686
11687static int
11688ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11689{
11690	struct ctl_lun *lun;
11691	const struct ctl_cmd_entry *entry;
11692	uint32_t initidx, targ_lun;
11693	int retval;
11694
11695	retval = 0;
11696
11697	lun = NULL;
11698
11699	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11700	if ((targ_lun < CTL_MAX_LUNS)
11701	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11702		lun = ctl_softc->ctl_luns[targ_lun];
11703		/*
11704		 * If the LUN is invalid, pretend that it doesn't exist.
11705		 * It will go away as soon as all pending I/O has been
11706		 * completed.
11707		 */
11708		if (lun->flags & CTL_LUN_DISABLED) {
11709			lun = NULL;
11710		} else {
11711			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11712			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11713				lun->be_lun;
11714			if (lun->be_lun->lun_type == T_PROCESSOR) {
11715				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11716			}
11717
11718			/*
11719			 * Every I/O goes into the OOA queue for a
11720			 * particular LUN, and stays there until completion.
11721			 */
11722			mtx_lock(&lun->lun_lock);
11723			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11724			    ooa_links);
11725		}
11726	} else {
11727		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11728		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11729	}
11730
11731	/* Get command entry and return error if it is unsuppotyed. */
11732	entry = ctl_validate_command(ctsio);
11733	if (entry == NULL) {
11734		if (lun)
11735			mtx_unlock(&lun->lun_lock);
11736		return (retval);
11737	}
11738
11739	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11740	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11741
11742	/*
11743	 * Check to see whether we can send this command to LUNs that don't
11744	 * exist.  This should pretty much only be the case for inquiry
11745	 * and request sense.  Further checks, below, really require having
11746	 * a LUN, so we can't really check the command anymore.  Just put
11747	 * it on the rtr queue.
11748	 */
11749	if (lun == NULL) {
11750		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11751			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11752			ctl_enqueue_rtr((union ctl_io *)ctsio);
11753			return (retval);
11754		}
11755
11756		ctl_set_unsupported_lun(ctsio);
11757		ctl_done((union ctl_io *)ctsio);
11758		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11759		return (retval);
11760	} else {
11761		/*
11762		 * Make sure we support this particular command on this LUN.
11763		 * e.g., we don't support writes to the control LUN.
11764		 */
11765		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11766			mtx_unlock(&lun->lun_lock);
11767			ctl_set_invalid_opcode(ctsio);
11768			ctl_done((union ctl_io *)ctsio);
11769			return (retval);
11770		}
11771	}
11772
11773	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11774
11775#ifdef CTL_WITH_CA
11776	/*
11777	 * If we've got a request sense, it'll clear the contingent
11778	 * allegiance condition.  Otherwise, if we have a CA condition for
11779	 * this initiator, clear it, because it sent down a command other
11780	 * than request sense.
11781	 */
11782	if ((ctsio->cdb[0] != REQUEST_SENSE)
11783	 && (ctl_is_set(lun->have_ca, initidx)))
11784		ctl_clear_mask(lun->have_ca, initidx);
11785#endif
11786
11787	/*
11788	 * If the command has this flag set, it handles its own unit
11789	 * attention reporting, we shouldn't do anything.  Otherwise we
11790	 * check for any pending unit attentions, and send them back to the
11791	 * initiator.  We only do this when a command initially comes in,
11792	 * not when we pull it off the blocked queue.
11793	 *
11794	 * According to SAM-3, section 5.3.2, the order that things get
11795	 * presented back to the host is basically unit attentions caused
11796	 * by some sort of reset event, busy status, reservation conflicts
11797	 * or task set full, and finally any other status.
11798	 *
11799	 * One issue here is that some of the unit attentions we report
11800	 * don't fall into the "reset" category (e.g. "reported luns data
11801	 * has changed").  So reporting it here, before the reservation
11802	 * check, may be technically wrong.  I guess the only thing to do
11803	 * would be to check for and report the reset events here, and then
11804	 * check for the other unit attention types after we check for a
11805	 * reservation conflict.
11806	 *
11807	 * XXX KDM need to fix this
11808	 */
11809	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11810		ctl_ua_type ua_type;
11811
11812		if (lun->pending_ua[initidx] != CTL_UA_NONE) {
11813			scsi_sense_data_type sense_format;
11814
11815			if (lun != NULL)
11816				sense_format = (lun->flags &
11817				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11818				    SSD_TYPE_FIXED;
11819			else
11820				sense_format = SSD_TYPE_FIXED;
11821
11822			ua_type = ctl_build_ua(&lun->pending_ua[initidx],
11823			    &ctsio->sense_data, sense_format);
11824			if (ua_type != CTL_UA_NONE) {
11825				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11826				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11827						       CTL_AUTOSENSE;
11828				ctsio->sense_len = SSD_FULL_SIZE;
11829				mtx_unlock(&lun->lun_lock);
11830				ctl_done((union ctl_io *)ctsio);
11831				return (retval);
11832			}
11833		}
11834	}
11835
11836
11837	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11838		mtx_unlock(&lun->lun_lock);
11839		ctl_done((union ctl_io *)ctsio);
11840		return (retval);
11841	}
11842
11843	/*
11844	 * XXX CHD this is where we want to send IO to other side if
11845	 * this LUN is secondary on this SC. We will need to make a copy
11846	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11847	 * the copy we send as FROM_OTHER.
11848	 * We also need to stuff the address of the original IO so we can
11849	 * find it easily. Something similar will need be done on the other
11850	 * side so when we are done we can find the copy.
11851	 */
11852	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11853		union ctl_ha_msg msg_info;
11854		int isc_retval;
11855
11856		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11857
11858		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11859		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11860#if 0
11861		printf("1. ctsio %p\n", ctsio);
11862#endif
11863		msg_info.hdr.serializing_sc = NULL;
11864		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11865		msg_info.scsi.tag_num = ctsio->tag_num;
11866		msg_info.scsi.tag_type = ctsio->tag_type;
11867		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11868
11869		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11870
11871		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11872		    (void *)&msg_info, sizeof(msg_info), 0)) >
11873		    CTL_HA_STATUS_SUCCESS) {
11874			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11875			       isc_retval);
11876			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11877		} else {
11878#if 0
11879			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11880#endif
11881		}
11882
11883		/*
11884		 * XXX KDM this I/O is off the incoming queue, but hasn't
11885		 * been inserted on any other queue.  We may need to come
11886		 * up with a holding queue while we wait for serialization
11887		 * so that we have an idea of what we're waiting for from
11888		 * the other side.
11889		 */
11890		mtx_unlock(&lun->lun_lock);
11891		return (retval);
11892	}
11893
11894	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11895			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11896			      ctl_ooaq, ooa_links))) {
11897	case CTL_ACTION_BLOCK:
11898		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11899		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11900				  blocked_links);
11901		mtx_unlock(&lun->lun_lock);
11902		return (retval);
11903	case CTL_ACTION_PASS:
11904	case CTL_ACTION_SKIP:
11905		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11906		mtx_unlock(&lun->lun_lock);
11907		ctl_enqueue_rtr((union ctl_io *)ctsio);
11908		break;
11909	case CTL_ACTION_OVERLAP:
11910		mtx_unlock(&lun->lun_lock);
11911		ctl_set_overlapped_cmd(ctsio);
11912		ctl_done((union ctl_io *)ctsio);
11913		break;
11914	case CTL_ACTION_OVERLAP_TAG:
11915		mtx_unlock(&lun->lun_lock);
11916		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11917		ctl_done((union ctl_io *)ctsio);
11918		break;
11919	case CTL_ACTION_ERROR:
11920	default:
11921		mtx_unlock(&lun->lun_lock);
11922		ctl_set_internal_failure(ctsio,
11923					 /*sks_valid*/ 0,
11924					 /*retry_count*/ 0);
11925		ctl_done((union ctl_io *)ctsio);
11926		break;
11927	}
11928	return (retval);
11929}
11930
11931const struct ctl_cmd_entry *
11932ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11933{
11934	const struct ctl_cmd_entry *entry;
11935	int service_action;
11936
11937	entry = &ctl_cmd_table[ctsio->cdb[0]];
11938	if (sa)
11939		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11940	if (entry->flags & CTL_CMD_FLAG_SA5) {
11941		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11942		entry = &((const struct ctl_cmd_entry *)
11943		    entry->execute)[service_action];
11944	}
11945	return (entry);
11946}
11947
11948const struct ctl_cmd_entry *
11949ctl_validate_command(struct ctl_scsiio *ctsio)
11950{
11951	const struct ctl_cmd_entry *entry;
11952	int i, sa;
11953	uint8_t diff;
11954
11955	entry = ctl_get_cmd_entry(ctsio, &sa);
11956	if (entry->execute == NULL) {
11957		if (sa)
11958			ctl_set_invalid_field(ctsio,
11959					      /*sks_valid*/ 1,
11960					      /*command*/ 1,
11961					      /*field*/ 1,
11962					      /*bit_valid*/ 1,
11963					      /*bit*/ 4);
11964		else
11965			ctl_set_invalid_opcode(ctsio);
11966		ctl_done((union ctl_io *)ctsio);
11967		return (NULL);
11968	}
11969	KASSERT(entry->length > 0,
11970	    ("Not defined length for command 0x%02x/0x%02x",
11971	     ctsio->cdb[0], ctsio->cdb[1]));
11972	for (i = 1; i < entry->length; i++) {
11973		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11974		if (diff == 0)
11975			continue;
11976		ctl_set_invalid_field(ctsio,
11977				      /*sks_valid*/ 1,
11978				      /*command*/ 1,
11979				      /*field*/ i,
11980				      /*bit_valid*/ 1,
11981				      /*bit*/ fls(diff) - 1);
11982		ctl_done((union ctl_io *)ctsio);
11983		return (NULL);
11984	}
11985	return (entry);
11986}
11987
11988static int
11989ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11990{
11991
11992	switch (lun_type) {
11993	case T_PROCESSOR:
11994		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11995		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11996			return (0);
11997		break;
11998	case T_DIRECT:
11999		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
12000		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
12001			return (0);
12002		break;
12003	default:
12004		return (0);
12005	}
12006	return (1);
12007}
12008
12009static int
12010ctl_scsiio(struct ctl_scsiio *ctsio)
12011{
12012	int retval;
12013	const struct ctl_cmd_entry *entry;
12014
12015	retval = CTL_RETVAL_COMPLETE;
12016
12017	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
12018
12019	entry = ctl_get_cmd_entry(ctsio, NULL);
12020
12021	/*
12022	 * If this I/O has been aborted, just send it straight to
12023	 * ctl_done() without executing it.
12024	 */
12025	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
12026		ctl_done((union ctl_io *)ctsio);
12027		goto bailout;
12028	}
12029
12030	/*
12031	 * All the checks should have been handled by ctl_scsiio_precheck().
12032	 * We should be clear now to just execute the I/O.
12033	 */
12034	retval = entry->execute(ctsio);
12035
12036bailout:
12037	return (retval);
12038}
12039
12040/*
12041 * Since we only implement one target right now, a bus reset simply resets
12042 * our single target.
12043 */
12044static int
12045ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
12046{
12047	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
12048}
12049
12050static int
12051ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
12052		 ctl_ua_type ua_type)
12053{
12054	struct ctl_lun *lun;
12055	int retval;
12056
12057	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12058		union ctl_ha_msg msg_info;
12059
12060		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
12061		msg_info.hdr.nexus = io->io_hdr.nexus;
12062		if (ua_type==CTL_UA_TARG_RESET)
12063			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
12064		else
12065			msg_info.task.task_action = CTL_TASK_BUS_RESET;
12066		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12067		msg_info.hdr.original_sc = NULL;
12068		msg_info.hdr.serializing_sc = NULL;
12069		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12070		    (void *)&msg_info, sizeof(msg_info), 0)) {
12071		}
12072	}
12073	retval = 0;
12074
12075	mtx_lock(&ctl_softc->ctl_lock);
12076	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
12077		retval += ctl_lun_reset(lun, io, ua_type);
12078	mtx_unlock(&ctl_softc->ctl_lock);
12079
12080	return (retval);
12081}
12082
12083/*
12084 * The LUN should always be set.  The I/O is optional, and is used to
12085 * distinguish between I/Os sent by this initiator, and by other
12086 * initiators.  We set unit attention for initiators other than this one.
12087 * SAM-3 is vague on this point.  It does say that a unit attention should
12088 * be established for other initiators when a LUN is reset (see section
12089 * 5.7.3), but it doesn't specifically say that the unit attention should
12090 * be established for this particular initiator when a LUN is reset.  Here
12091 * is the relevant text, from SAM-3 rev 8:
12092 *
12093 * 5.7.2 When a SCSI initiator port aborts its own tasks
12094 *
12095 * When a SCSI initiator port causes its own task(s) to be aborted, no
12096 * notification that the task(s) have been aborted shall be returned to
12097 * the SCSI initiator port other than the completion response for the
12098 * command or task management function action that caused the task(s) to
12099 * be aborted and notification(s) associated with related effects of the
12100 * action (e.g., a reset unit attention condition).
12101 *
12102 * XXX KDM for now, we're setting unit attention for all initiators.
12103 */
12104static int
12105ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
12106{
12107	union ctl_io *xio;
12108#if 0
12109	uint32_t initindex;
12110#endif
12111	int i;
12112
12113	mtx_lock(&lun->lun_lock);
12114	/*
12115	 * Run through the OOA queue and abort each I/O.
12116	 */
12117#if 0
12118	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12119#endif
12120	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12121	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12122		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
12123	}
12124
12125	/*
12126	 * This version sets unit attention for every
12127	 */
12128#if 0
12129	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12130	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12131		if (initindex == i)
12132			continue;
12133		lun->pending_ua[i] |= ua_type;
12134	}
12135#endif
12136
12137	/*
12138	 * A reset (any kind, really) clears reservations established with
12139	 * RESERVE/RELEASE.  It does not clear reservations established
12140	 * with PERSISTENT RESERVE OUT, but we don't support that at the
12141	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
12142	 * reservations made with the RESERVE/RELEASE commands, because
12143	 * those commands are obsolete in SPC-3.
12144	 */
12145	lun->flags &= ~CTL_LUN_RESERVED;
12146
12147	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12148#ifdef CTL_WITH_CA
12149		ctl_clear_mask(lun->have_ca, i);
12150#endif
12151		lun->pending_ua[i] |= ua_type;
12152	}
12153	mtx_unlock(&lun->lun_lock);
12154
12155	return (0);
12156}
12157
12158static void
12159ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
12160    int other_sc)
12161{
12162	union ctl_io *xio;
12163
12164	mtx_assert(&lun->lun_lock, MA_OWNED);
12165
12166	/*
12167	 * Run through the OOA queue and attempt to find the given I/O.
12168	 * The target port, initiator ID, tag type and tag number have to
12169	 * match the values that we got from the initiator.  If we have an
12170	 * untagged command to abort, simply abort the first untagged command
12171	 * we come to.  We only allow one untagged command at a time of course.
12172	 */
12173	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12174	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12175
12176		if ((targ_port == UINT32_MAX ||
12177		     targ_port == xio->io_hdr.nexus.targ_port) &&
12178		    (init_id == UINT32_MAX ||
12179		     init_id == xio->io_hdr.nexus.initid.id)) {
12180			if (targ_port != xio->io_hdr.nexus.targ_port ||
12181			    init_id != xio->io_hdr.nexus.initid.id)
12182				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12183			xio->io_hdr.flags |= CTL_FLAG_ABORT;
12184			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12185				union ctl_ha_msg msg_info;
12186
12187				msg_info.hdr.nexus = xio->io_hdr.nexus;
12188				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12189				msg_info.task.tag_num = xio->scsiio.tag_num;
12190				msg_info.task.tag_type = xio->scsiio.tag_type;
12191				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12192				msg_info.hdr.original_sc = NULL;
12193				msg_info.hdr.serializing_sc = NULL;
12194				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12195				    (void *)&msg_info, sizeof(msg_info), 0);
12196			}
12197		}
12198	}
12199}
12200
12201static int
12202ctl_abort_task_set(union ctl_io *io)
12203{
12204	struct ctl_softc *softc = control_softc;
12205	struct ctl_lun *lun;
12206	uint32_t targ_lun;
12207
12208	/*
12209	 * Look up the LUN.
12210	 */
12211	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12212	mtx_lock(&softc->ctl_lock);
12213	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12214		lun = softc->ctl_luns[targ_lun];
12215	else {
12216		mtx_unlock(&softc->ctl_lock);
12217		return (1);
12218	}
12219
12220	mtx_lock(&lun->lun_lock);
12221	mtx_unlock(&softc->ctl_lock);
12222	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12223		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12224		    io->io_hdr.nexus.initid.id,
12225		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12226	} else { /* CTL_TASK_CLEAR_TASK_SET */
12227		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12228		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12229	}
12230	mtx_unlock(&lun->lun_lock);
12231	return (0);
12232}
12233
12234static int
12235ctl_i_t_nexus_reset(union ctl_io *io)
12236{
12237	struct ctl_softc *softc = control_softc;
12238	struct ctl_lun *lun;
12239	uint32_t initindex, residx;
12240
12241	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12242	residx = ctl_get_resindex(&io->io_hdr.nexus);
12243	mtx_lock(&softc->ctl_lock);
12244	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12245		mtx_lock(&lun->lun_lock);
12246		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12247		    io->io_hdr.nexus.initid.id,
12248		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12249#ifdef CTL_WITH_CA
12250		ctl_clear_mask(lun->have_ca, initindex);
12251#endif
12252		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
12253			lun->flags &= ~CTL_LUN_RESERVED;
12254		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12255		mtx_unlock(&lun->lun_lock);
12256	}
12257	mtx_unlock(&softc->ctl_lock);
12258	return (0);
12259}
12260
12261static int
12262ctl_abort_task(union ctl_io *io)
12263{
12264	union ctl_io *xio;
12265	struct ctl_lun *lun;
12266	struct ctl_softc *ctl_softc;
12267#if 0
12268	struct sbuf sb;
12269	char printbuf[128];
12270#endif
12271	int found;
12272	uint32_t targ_lun;
12273
12274	ctl_softc = control_softc;
12275	found = 0;
12276
12277	/*
12278	 * Look up the LUN.
12279	 */
12280	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12281	mtx_lock(&ctl_softc->ctl_lock);
12282	if ((targ_lun < CTL_MAX_LUNS)
12283	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12284		lun = ctl_softc->ctl_luns[targ_lun];
12285	else {
12286		mtx_unlock(&ctl_softc->ctl_lock);
12287		return (1);
12288	}
12289
12290#if 0
12291	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12292	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12293#endif
12294
12295	mtx_lock(&lun->lun_lock);
12296	mtx_unlock(&ctl_softc->ctl_lock);
12297	/*
12298	 * Run through the OOA queue and attempt to find the given I/O.
12299	 * The target port, initiator ID, tag type and tag number have to
12300	 * match the values that we got from the initiator.  If we have an
12301	 * untagged command to abort, simply abort the first untagged command
12302	 * we come to.  We only allow one untagged command at a time of course.
12303	 */
12304#if 0
12305	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12306#endif
12307	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12308	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12309#if 0
12310		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12311
12312		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12313			    lun->lun, xio->scsiio.tag_num,
12314			    xio->scsiio.tag_type,
12315			    (xio->io_hdr.blocked_links.tqe_prev
12316			    == NULL) ? "" : " BLOCKED",
12317			    (xio->io_hdr.flags &
12318			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12319			    (xio->io_hdr.flags &
12320			    CTL_FLAG_ABORT) ? " ABORT" : "",
12321			    (xio->io_hdr.flags &
12322			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12323		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12324		sbuf_finish(&sb);
12325		printf("%s\n", sbuf_data(&sb));
12326#endif
12327
12328		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12329		 && (xio->io_hdr.nexus.initid.id ==
12330		     io->io_hdr.nexus.initid.id)) {
12331			/*
12332			 * If the abort says that the task is untagged, the
12333			 * task in the queue must be untagged.  Otherwise,
12334			 * we just check to see whether the tag numbers
12335			 * match.  This is because the QLogic firmware
12336			 * doesn't pass back the tag type in an abort
12337			 * request.
12338			 */
12339#if 0
12340			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12341			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12342			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12343#endif
12344			/*
12345			 * XXX KDM we've got problems with FC, because it
12346			 * doesn't send down a tag type with aborts.  So we
12347			 * can only really go by the tag number...
12348			 * This may cause problems with parallel SCSI.
12349			 * Need to figure that out!!
12350			 */
12351			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12352				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12353				found = 1;
12354				if ((io->io_hdr.flags &
12355				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12356				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12357					union ctl_ha_msg msg_info;
12358
12359					io->io_hdr.flags |=
12360					                CTL_FLAG_SENT_2OTHER_SC;
12361					msg_info.hdr.nexus = io->io_hdr.nexus;
12362					msg_info.task.task_action =
12363						CTL_TASK_ABORT_TASK;
12364					msg_info.task.tag_num =
12365						io->taskio.tag_num;
12366					msg_info.task.tag_type =
12367						io->taskio.tag_type;
12368					msg_info.hdr.msg_type =
12369						CTL_MSG_MANAGE_TASKS;
12370					msg_info.hdr.original_sc = NULL;
12371					msg_info.hdr.serializing_sc = NULL;
12372#if 0
12373					printf("Sent Abort to other side\n");
12374#endif
12375					if (CTL_HA_STATUS_SUCCESS !=
12376					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12377		    				(void *)&msg_info,
12378						sizeof(msg_info), 0)) {
12379					}
12380				}
12381#if 0
12382				printf("ctl_abort_task: found I/O to abort\n");
12383#endif
12384				break;
12385			}
12386		}
12387	}
12388	mtx_unlock(&lun->lun_lock);
12389
12390	if (found == 0) {
12391		/*
12392		 * This isn't really an error.  It's entirely possible for
12393		 * the abort and command completion to cross on the wire.
12394		 * This is more of an informative/diagnostic error.
12395		 */
12396#if 0
12397		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12398		       "%d:%d:%d:%d tag %d type %d\n",
12399		       io->io_hdr.nexus.initid.id,
12400		       io->io_hdr.nexus.targ_port,
12401		       io->io_hdr.nexus.targ_target.id,
12402		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12403		       io->taskio.tag_type);
12404#endif
12405	}
12406	return (0);
12407}
12408
12409static void
12410ctl_run_task(union ctl_io *io)
12411{
12412	struct ctl_softc *ctl_softc = control_softc;
12413	int retval = 1;
12414	const char *task_desc;
12415
12416	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12417
12418	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12419	    ("ctl_run_task: Unextected io_type %d\n",
12420	     io->io_hdr.io_type));
12421
12422	task_desc = ctl_scsi_task_string(&io->taskio);
12423	if (task_desc != NULL) {
12424#ifdef NEEDTOPORT
12425		csevent_log(CSC_CTL | CSC_SHELF_SW |
12426			    CTL_TASK_REPORT,
12427			    csevent_LogType_Trace,
12428			    csevent_Severity_Information,
12429			    csevent_AlertLevel_Green,
12430			    csevent_FRU_Firmware,
12431			    csevent_FRU_Unknown,
12432			    "CTL: received task: %s",task_desc);
12433#endif
12434	} else {
12435#ifdef NEEDTOPORT
12436		csevent_log(CSC_CTL | CSC_SHELF_SW |
12437			    CTL_TASK_REPORT,
12438			    csevent_LogType_Trace,
12439			    csevent_Severity_Information,
12440			    csevent_AlertLevel_Green,
12441			    csevent_FRU_Firmware,
12442			    csevent_FRU_Unknown,
12443			    "CTL: received unknown task "
12444			    "type: %d (%#x)",
12445			    io->taskio.task_action,
12446			    io->taskio.task_action);
12447#endif
12448	}
12449	switch (io->taskio.task_action) {
12450	case CTL_TASK_ABORT_TASK:
12451		retval = ctl_abort_task(io);
12452		break;
12453	case CTL_TASK_ABORT_TASK_SET:
12454	case CTL_TASK_CLEAR_TASK_SET:
12455		retval = ctl_abort_task_set(io);
12456		break;
12457	case CTL_TASK_CLEAR_ACA:
12458		break;
12459	case CTL_TASK_I_T_NEXUS_RESET:
12460		retval = ctl_i_t_nexus_reset(io);
12461		break;
12462	case CTL_TASK_LUN_RESET: {
12463		struct ctl_lun *lun;
12464		uint32_t targ_lun;
12465
12466		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12467		mtx_lock(&ctl_softc->ctl_lock);
12468		if ((targ_lun < CTL_MAX_LUNS)
12469		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12470			lun = ctl_softc->ctl_luns[targ_lun];
12471		else {
12472			mtx_unlock(&ctl_softc->ctl_lock);
12473			retval = 1;
12474			break;
12475		}
12476
12477		if (!(io->io_hdr.flags &
12478		    CTL_FLAG_FROM_OTHER_SC)) {
12479			union ctl_ha_msg msg_info;
12480
12481			io->io_hdr.flags |=
12482				CTL_FLAG_SENT_2OTHER_SC;
12483			msg_info.hdr.msg_type =
12484				CTL_MSG_MANAGE_TASKS;
12485			msg_info.hdr.nexus = io->io_hdr.nexus;
12486			msg_info.task.task_action =
12487				CTL_TASK_LUN_RESET;
12488			msg_info.hdr.original_sc = NULL;
12489			msg_info.hdr.serializing_sc = NULL;
12490			if (CTL_HA_STATUS_SUCCESS !=
12491			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12492			    (void *)&msg_info,
12493			    sizeof(msg_info), 0)) {
12494			}
12495		}
12496
12497		retval = ctl_lun_reset(lun, io,
12498				       CTL_UA_LUN_RESET);
12499		mtx_unlock(&ctl_softc->ctl_lock);
12500		break;
12501	}
12502	case CTL_TASK_TARGET_RESET:
12503		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12504		break;
12505	case CTL_TASK_BUS_RESET:
12506		retval = ctl_bus_reset(ctl_softc, io);
12507		break;
12508	case CTL_TASK_PORT_LOGIN:
12509		break;
12510	case CTL_TASK_PORT_LOGOUT:
12511		break;
12512	default:
12513		printf("ctl_run_task: got unknown task management event %d\n",
12514		       io->taskio.task_action);
12515		break;
12516	}
12517	if (retval == 0)
12518		io->io_hdr.status = CTL_SUCCESS;
12519	else
12520		io->io_hdr.status = CTL_ERROR;
12521	ctl_done(io);
12522}
12523
12524/*
12525 * For HA operation.  Handle commands that come in from the other
12526 * controller.
12527 */
12528static void
12529ctl_handle_isc(union ctl_io *io)
12530{
12531	int free_io;
12532	struct ctl_lun *lun;
12533	struct ctl_softc *ctl_softc;
12534	uint32_t targ_lun;
12535
12536	ctl_softc = control_softc;
12537
12538	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12539	lun = ctl_softc->ctl_luns[targ_lun];
12540
12541	switch (io->io_hdr.msg_type) {
12542	case CTL_MSG_SERIALIZE:
12543		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12544		break;
12545	case CTL_MSG_R2R: {
12546		const struct ctl_cmd_entry *entry;
12547
12548		/*
12549		 * This is only used in SER_ONLY mode.
12550		 */
12551		free_io = 0;
12552		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12553		mtx_lock(&lun->lun_lock);
12554		if (ctl_scsiio_lun_check(ctl_softc, lun,
12555		    entry, (struct ctl_scsiio *)io) != 0) {
12556			mtx_unlock(&lun->lun_lock);
12557			ctl_done(io);
12558			break;
12559		}
12560		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12561		mtx_unlock(&lun->lun_lock);
12562		ctl_enqueue_rtr(io);
12563		break;
12564	}
12565	case CTL_MSG_FINISH_IO:
12566		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12567			free_io = 0;
12568			ctl_done(io);
12569		} else {
12570			free_io = 1;
12571			mtx_lock(&lun->lun_lock);
12572			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12573				     ooa_links);
12574			ctl_check_blocked(lun);
12575			mtx_unlock(&lun->lun_lock);
12576		}
12577		break;
12578	case CTL_MSG_PERS_ACTION:
12579		ctl_hndl_per_res_out_on_other_sc(
12580			(union ctl_ha_msg *)&io->presio.pr_msg);
12581		free_io = 1;
12582		break;
12583	case CTL_MSG_BAD_JUJU:
12584		free_io = 0;
12585		ctl_done(io);
12586		break;
12587	case CTL_MSG_DATAMOVE:
12588		/* Only used in XFER mode */
12589		free_io = 0;
12590		ctl_datamove_remote(io);
12591		break;
12592	case CTL_MSG_DATAMOVE_DONE:
12593		/* Only used in XFER mode */
12594		free_io = 0;
12595		io->scsiio.be_move_done(io);
12596		break;
12597	default:
12598		free_io = 1;
12599		printf("%s: Invalid message type %d\n",
12600		       __func__, io->io_hdr.msg_type);
12601		break;
12602	}
12603	if (free_io)
12604		ctl_free_io(io);
12605
12606}
12607
12608
12609/*
12610 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12611 * there is no match.
12612 */
12613static ctl_lun_error_pattern
12614ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12615{
12616	const struct ctl_cmd_entry *entry;
12617	ctl_lun_error_pattern filtered_pattern, pattern;
12618
12619	pattern = desc->error_pattern;
12620
12621	/*
12622	 * XXX KDM we need more data passed into this function to match a
12623	 * custom pattern, and we actually need to implement custom pattern
12624	 * matching.
12625	 */
12626	if (pattern & CTL_LUN_PAT_CMD)
12627		return (CTL_LUN_PAT_CMD);
12628
12629	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12630		return (CTL_LUN_PAT_ANY);
12631
12632	entry = ctl_get_cmd_entry(ctsio, NULL);
12633
12634	filtered_pattern = entry->pattern & pattern;
12635
12636	/*
12637	 * If the user requested specific flags in the pattern (e.g.
12638	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12639	 * flags.
12640	 *
12641	 * If the user did not specify any flags, it doesn't matter whether
12642	 * or not the command supports the flags.
12643	 */
12644	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12645	     (pattern & ~CTL_LUN_PAT_MASK))
12646		return (CTL_LUN_PAT_NONE);
12647
12648	/*
12649	 * If the user asked for a range check, see if the requested LBA
12650	 * range overlaps with this command's LBA range.
12651	 */
12652	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12653		uint64_t lba1;
12654		uint64_t len1;
12655		ctl_action action;
12656		int retval;
12657
12658		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12659		if (retval != 0)
12660			return (CTL_LUN_PAT_NONE);
12661
12662		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12663					      desc->lba_range.len);
12664		/*
12665		 * A "pass" means that the LBA ranges don't overlap, so
12666		 * this doesn't match the user's range criteria.
12667		 */
12668		if (action == CTL_ACTION_PASS)
12669			return (CTL_LUN_PAT_NONE);
12670	}
12671
12672	return (filtered_pattern);
12673}
12674
12675static void
12676ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12677{
12678	struct ctl_error_desc *desc, *desc2;
12679
12680	mtx_assert(&lun->lun_lock, MA_OWNED);
12681
12682	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12683		ctl_lun_error_pattern pattern;
12684		/*
12685		 * Check to see whether this particular command matches
12686		 * the pattern in the descriptor.
12687		 */
12688		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12689		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12690			continue;
12691
12692		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12693		case CTL_LUN_INJ_ABORTED:
12694			ctl_set_aborted(&io->scsiio);
12695			break;
12696		case CTL_LUN_INJ_MEDIUM_ERR:
12697			ctl_set_medium_error(&io->scsiio);
12698			break;
12699		case CTL_LUN_INJ_UA:
12700			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12701			 * OCCURRED */
12702			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12703			break;
12704		case CTL_LUN_INJ_CUSTOM:
12705			/*
12706			 * We're assuming the user knows what he is doing.
12707			 * Just copy the sense information without doing
12708			 * checks.
12709			 */
12710			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12711			      ctl_min(sizeof(desc->custom_sense),
12712				      sizeof(io->scsiio.sense_data)));
12713			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12714			io->scsiio.sense_len = SSD_FULL_SIZE;
12715			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12716			break;
12717		case CTL_LUN_INJ_NONE:
12718		default:
12719			/*
12720			 * If this is an error injection type we don't know
12721			 * about, clear the continuous flag (if it is set)
12722			 * so it will get deleted below.
12723			 */
12724			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12725			break;
12726		}
12727		/*
12728		 * By default, each error injection action is a one-shot
12729		 */
12730		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12731			continue;
12732
12733		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12734
12735		free(desc, M_CTL);
12736	}
12737}
12738
12739#ifdef CTL_IO_DELAY
12740static void
12741ctl_datamove_timer_wakeup(void *arg)
12742{
12743	union ctl_io *io;
12744
12745	io = (union ctl_io *)arg;
12746
12747	ctl_datamove(io);
12748}
12749#endif /* CTL_IO_DELAY */
12750
12751void
12752ctl_datamove(union ctl_io *io)
12753{
12754	void (*fe_datamove)(union ctl_io *io);
12755
12756	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12757
12758	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12759
12760#ifdef CTL_TIME_IO
12761	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12762		char str[256];
12763		char path_str[64];
12764		struct sbuf sb;
12765
12766		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12767		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12768
12769		sbuf_cat(&sb, path_str);
12770		switch (io->io_hdr.io_type) {
12771		case CTL_IO_SCSI:
12772			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12773			sbuf_printf(&sb, "\n");
12774			sbuf_cat(&sb, path_str);
12775			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12776				    io->scsiio.tag_num, io->scsiio.tag_type);
12777			break;
12778		case CTL_IO_TASK:
12779			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12780				    "Tag Type: %d\n", io->taskio.task_action,
12781				    io->taskio.tag_num, io->taskio.tag_type);
12782			break;
12783		default:
12784			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12785			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12786			break;
12787		}
12788		sbuf_cat(&sb, path_str);
12789		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12790			    (intmax_t)time_uptime - io->io_hdr.start_time);
12791		sbuf_finish(&sb);
12792		printf("%s", sbuf_data(&sb));
12793	}
12794#endif /* CTL_TIME_IO */
12795
12796#ifdef CTL_IO_DELAY
12797	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12798		struct ctl_lun *lun;
12799
12800		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12801
12802		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12803	} else {
12804		struct ctl_lun *lun;
12805
12806		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12807		if ((lun != NULL)
12808		 && (lun->delay_info.datamove_delay > 0)) {
12809			struct callout *callout;
12810
12811			callout = (struct callout *)&io->io_hdr.timer_bytes;
12812			callout_init(callout, /*mpsafe*/ 1);
12813			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12814			callout_reset(callout,
12815				      lun->delay_info.datamove_delay * hz,
12816				      ctl_datamove_timer_wakeup, io);
12817			if (lun->delay_info.datamove_type ==
12818			    CTL_DELAY_TYPE_ONESHOT)
12819				lun->delay_info.datamove_delay = 0;
12820			return;
12821		}
12822	}
12823#endif
12824
12825	/*
12826	 * This command has been aborted.  Set the port status, so we fail
12827	 * the data move.
12828	 */
12829	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12830		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12831		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12832		       io->io_hdr.nexus.targ_port,
12833		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12834		       io->io_hdr.nexus.targ_lun);
12835		io->io_hdr.port_status = 31337;
12836		/*
12837		 * Note that the backend, in this case, will get the
12838		 * callback in its context.  In other cases it may get
12839		 * called in the frontend's interrupt thread context.
12840		 */
12841		io->scsiio.be_move_done(io);
12842		return;
12843	}
12844
12845	/*
12846	 * If we're in XFER mode and this I/O is from the other shelf
12847	 * controller, we need to send the DMA to the other side to
12848	 * actually transfer the data to/from the host.  In serialize only
12849	 * mode the transfer happens below CTL and ctl_datamove() is only
12850	 * called on the machine that originally received the I/O.
12851	 */
12852	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12853	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12854		union ctl_ha_msg msg;
12855		uint32_t sg_entries_sent;
12856		int do_sg_copy;
12857		int i;
12858
12859		memset(&msg, 0, sizeof(msg));
12860		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12861		msg.hdr.original_sc = io->io_hdr.original_sc;
12862		msg.hdr.serializing_sc = io;
12863		msg.hdr.nexus = io->io_hdr.nexus;
12864		msg.dt.flags = io->io_hdr.flags;
12865		/*
12866		 * We convert everything into a S/G list here.  We can't
12867		 * pass by reference, only by value between controllers.
12868		 * So we can't pass a pointer to the S/G list, only as many
12869		 * S/G entries as we can fit in here.  If it's possible for
12870		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12871		 * then we need to break this up into multiple transfers.
12872		 */
12873		if (io->scsiio.kern_sg_entries == 0) {
12874			msg.dt.kern_sg_entries = 1;
12875			/*
12876			 * If this is in cached memory, flush the cache
12877			 * before we send the DMA request to the other
12878			 * controller.  We want to do this in either the
12879			 * read or the write case.  The read case is
12880			 * straightforward.  In the write case, we want to
12881			 * make sure nothing is in the local cache that
12882			 * could overwrite the DMAed data.
12883			 */
12884			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12885				/*
12886				 * XXX KDM use bus_dmamap_sync() here.
12887				 */
12888			}
12889
12890			/*
12891			 * Convert to a physical address if this is a
12892			 * virtual address.
12893			 */
12894			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12895				msg.dt.sg_list[0].addr =
12896					io->scsiio.kern_data_ptr;
12897			} else {
12898				/*
12899				 * XXX KDM use busdma here!
12900				 */
12901#if 0
12902				msg.dt.sg_list[0].addr = (void *)
12903					vtophys(io->scsiio.kern_data_ptr);
12904#endif
12905			}
12906
12907			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12908			do_sg_copy = 0;
12909		} else {
12910			struct ctl_sg_entry *sgl;
12911
12912			do_sg_copy = 1;
12913			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12914			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12915			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12916				/*
12917				 * XXX KDM use bus_dmamap_sync() here.
12918				 */
12919			}
12920		}
12921
12922		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12923		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12924		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12925		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12926		msg.dt.sg_sequence = 0;
12927
12928		/*
12929		 * Loop until we've sent all of the S/G entries.  On the
12930		 * other end, we'll recompose these S/G entries into one
12931		 * contiguous list before passing it to the
12932		 */
12933		for (sg_entries_sent = 0; sg_entries_sent <
12934		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12935			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12936				sizeof(msg.dt.sg_list[0])),
12937				msg.dt.kern_sg_entries - sg_entries_sent);
12938
12939			if (do_sg_copy != 0) {
12940				struct ctl_sg_entry *sgl;
12941				int j;
12942
12943				sgl = (struct ctl_sg_entry *)
12944					io->scsiio.kern_data_ptr;
12945				/*
12946				 * If this is in cached memory, flush the cache
12947				 * before we send the DMA request to the other
12948				 * controller.  We want to do this in either
12949				 * the * read or the write case.  The read
12950				 * case is straightforward.  In the write
12951				 * case, we want to make sure nothing is
12952				 * in the local cache that could overwrite
12953				 * the DMAed data.
12954				 */
12955
12956				for (i = sg_entries_sent, j = 0;
12957				     i < msg.dt.cur_sg_entries; i++, j++) {
12958					if ((io->io_hdr.flags &
12959					     CTL_FLAG_NO_DATASYNC) == 0) {
12960						/*
12961						 * XXX KDM use bus_dmamap_sync()
12962						 */
12963					}
12964					if ((io->io_hdr.flags &
12965					     CTL_FLAG_BUS_ADDR) == 0) {
12966						/*
12967						 * XXX KDM use busdma.
12968						 */
12969#if 0
12970						msg.dt.sg_list[j].addr =(void *)
12971						       vtophys(sgl[i].addr);
12972#endif
12973					} else {
12974						msg.dt.sg_list[j].addr =
12975							sgl[i].addr;
12976					}
12977					msg.dt.sg_list[j].len = sgl[i].len;
12978				}
12979			}
12980
12981			sg_entries_sent += msg.dt.cur_sg_entries;
12982			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12983				msg.dt.sg_last = 1;
12984			else
12985				msg.dt.sg_last = 0;
12986
12987			/*
12988			 * XXX KDM drop and reacquire the lock here?
12989			 */
12990			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12991			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12992				/*
12993				 * XXX do something here.
12994				 */
12995			}
12996
12997			msg.dt.sent_sg_entries = sg_entries_sent;
12998		}
12999		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13000		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
13001			ctl_failover_io(io, /*have_lock*/ 0);
13002
13003	} else {
13004
13005		/*
13006		 * Lookup the fe_datamove() function for this particular
13007		 * front end.
13008		 */
13009		fe_datamove =
13010		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13011
13012		fe_datamove(io);
13013	}
13014}
13015
13016static void
13017ctl_send_datamove_done(union ctl_io *io, int have_lock)
13018{
13019	union ctl_ha_msg msg;
13020	int isc_status;
13021
13022	memset(&msg, 0, sizeof(msg));
13023
13024	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
13025	msg.hdr.original_sc = io;
13026	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13027	msg.hdr.nexus = io->io_hdr.nexus;
13028	msg.hdr.status = io->io_hdr.status;
13029	msg.scsi.tag_num = io->scsiio.tag_num;
13030	msg.scsi.tag_type = io->scsiio.tag_type;
13031	msg.scsi.scsi_status = io->scsiio.scsi_status;
13032	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13033	       sizeof(io->scsiio.sense_data));
13034	msg.scsi.sense_len = io->scsiio.sense_len;
13035	msg.scsi.sense_residual = io->scsiio.sense_residual;
13036	msg.scsi.fetd_status = io->io_hdr.port_status;
13037	msg.scsi.residual = io->scsiio.residual;
13038	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13039
13040	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13041		ctl_failover_io(io, /*have_lock*/ have_lock);
13042		return;
13043	}
13044
13045	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
13046	if (isc_status > CTL_HA_STATUS_SUCCESS) {
13047		/* XXX do something if this fails */
13048	}
13049
13050}
13051
13052/*
13053 * The DMA to the remote side is done, now we need to tell the other side
13054 * we're done so it can continue with its data movement.
13055 */
13056static void
13057ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
13058{
13059	union ctl_io *io;
13060
13061	io = rq->context;
13062
13063	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13064		printf("%s: ISC DMA write failed with error %d", __func__,
13065		       rq->ret);
13066		ctl_set_internal_failure(&io->scsiio,
13067					 /*sks_valid*/ 1,
13068					 /*retry_count*/ rq->ret);
13069	}
13070
13071	ctl_dt_req_free(rq);
13072
13073	/*
13074	 * In this case, we had to malloc the memory locally.  Free it.
13075	 */
13076	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13077		int i;
13078		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13079			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13080	}
13081	/*
13082	 * The data is in local and remote memory, so now we need to send
13083	 * status (good or back) back to the other side.
13084	 */
13085	ctl_send_datamove_done(io, /*have_lock*/ 0);
13086}
13087
13088/*
13089 * We've moved the data from the host/controller into local memory.  Now we
13090 * need to push it over to the remote controller's memory.
13091 */
13092static int
13093ctl_datamove_remote_dm_write_cb(union ctl_io *io)
13094{
13095	int retval;
13096
13097	retval = 0;
13098
13099	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
13100					  ctl_datamove_remote_write_cb);
13101
13102	return (retval);
13103}
13104
13105static void
13106ctl_datamove_remote_write(union ctl_io *io)
13107{
13108	int retval;
13109	void (*fe_datamove)(union ctl_io *io);
13110
13111	/*
13112	 * - Get the data from the host/HBA into local memory.
13113	 * - DMA memory from the local controller to the remote controller.
13114	 * - Send status back to the remote controller.
13115	 */
13116
13117	retval = ctl_datamove_remote_sgl_setup(io);
13118	if (retval != 0)
13119		return;
13120
13121	/* Switch the pointer over so the FETD knows what to do */
13122	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13123
13124	/*
13125	 * Use a custom move done callback, since we need to send completion
13126	 * back to the other controller, not to the backend on this side.
13127	 */
13128	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
13129
13130	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13131
13132	fe_datamove(io);
13133
13134	return;
13135
13136}
13137
13138static int
13139ctl_datamove_remote_dm_read_cb(union ctl_io *io)
13140{
13141#if 0
13142	char str[256];
13143	char path_str[64];
13144	struct sbuf sb;
13145#endif
13146
13147	/*
13148	 * In this case, we had to malloc the memory locally.  Free it.
13149	 */
13150	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13151		int i;
13152		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13153			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13154	}
13155
13156#if 0
13157	scsi_path_string(io, path_str, sizeof(path_str));
13158	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13159	sbuf_cat(&sb, path_str);
13160	scsi_command_string(&io->scsiio, NULL, &sb);
13161	sbuf_printf(&sb, "\n");
13162	sbuf_cat(&sb, path_str);
13163	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13164		    io->scsiio.tag_num, io->scsiio.tag_type);
13165	sbuf_cat(&sb, path_str);
13166	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
13167		    io->io_hdr.flags, io->io_hdr.status);
13168	sbuf_finish(&sb);
13169	printk("%s", sbuf_data(&sb));
13170#endif
13171
13172
13173	/*
13174	 * The read is done, now we need to send status (good or bad) back
13175	 * to the other side.
13176	 */
13177	ctl_send_datamove_done(io, /*have_lock*/ 0);
13178
13179	return (0);
13180}
13181
13182static void
13183ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
13184{
13185	union ctl_io *io;
13186	void (*fe_datamove)(union ctl_io *io);
13187
13188	io = rq->context;
13189
13190	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13191		printf("%s: ISC DMA read failed with error %d", __func__,
13192		       rq->ret);
13193		ctl_set_internal_failure(&io->scsiio,
13194					 /*sks_valid*/ 1,
13195					 /*retry_count*/ rq->ret);
13196	}
13197
13198	ctl_dt_req_free(rq);
13199
13200	/* Switch the pointer over so the FETD knows what to do */
13201	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13202
13203	/*
13204	 * Use a custom move done callback, since we need to send completion
13205	 * back to the other controller, not to the backend on this side.
13206	 */
13207	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13208
13209	/* XXX KDM add checks like the ones in ctl_datamove? */
13210
13211	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13212
13213	fe_datamove(io);
13214}
13215
13216static int
13217ctl_datamove_remote_sgl_setup(union ctl_io *io)
13218{
13219	struct ctl_sg_entry *local_sglist, *remote_sglist;
13220	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13221	struct ctl_softc *softc;
13222	int retval;
13223	int i;
13224
13225	retval = 0;
13226	softc = control_softc;
13227
13228	local_sglist = io->io_hdr.local_sglist;
13229	local_dma_sglist = io->io_hdr.local_dma_sglist;
13230	remote_sglist = io->io_hdr.remote_sglist;
13231	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13232
13233	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13234		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13235			local_sglist[i].len = remote_sglist[i].len;
13236
13237			/*
13238			 * XXX Detect the situation where the RS-level I/O
13239			 * redirector on the other side has already read the
13240			 * data off of the AOR RS on this side, and
13241			 * transferred it to remote (mirror) memory on the
13242			 * other side.  Since we already have the data in
13243			 * memory here, we just need to use it.
13244			 *
13245			 * XXX KDM this can probably be removed once we
13246			 * get the cache device code in and take the
13247			 * current AOR implementation out.
13248			 */
13249#ifdef NEEDTOPORT
13250			if ((remote_sglist[i].addr >=
13251			     (void *)vtophys(softc->mirr->addr))
13252			 && (remote_sglist[i].addr <
13253			     ((void *)vtophys(softc->mirr->addr) +
13254			     CacheMirrorOffset))) {
13255				local_sglist[i].addr = remote_sglist[i].addr -
13256					CacheMirrorOffset;
13257				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13258				     CTL_FLAG_DATA_IN)
13259					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13260			} else {
13261				local_sglist[i].addr = remote_sglist[i].addr +
13262					CacheMirrorOffset;
13263			}
13264#endif
13265#if 0
13266			printf("%s: local %p, remote %p, len %d\n",
13267			       __func__, local_sglist[i].addr,
13268			       remote_sglist[i].addr, local_sglist[i].len);
13269#endif
13270		}
13271	} else {
13272		uint32_t len_to_go;
13273
13274		/*
13275		 * In this case, we don't have automatically allocated
13276		 * memory for this I/O on this controller.  This typically
13277		 * happens with internal CTL I/O -- e.g. inquiry, mode
13278		 * sense, etc.  Anything coming from RAIDCore will have
13279		 * a mirror area available.
13280		 */
13281		len_to_go = io->scsiio.kern_data_len;
13282
13283		/*
13284		 * Clear the no datasync flag, we have to use malloced
13285		 * buffers.
13286		 */
13287		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13288
13289		/*
13290		 * The difficult thing here is that the size of the various
13291		 * S/G segments may be different than the size from the
13292		 * remote controller.  That'll make it harder when DMAing
13293		 * the data back to the other side.
13294		 */
13295		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13296		     sizeof(io->io_hdr.remote_sglist[0])) &&
13297		     (len_to_go > 0); i++) {
13298			local_sglist[i].len = ctl_min(len_to_go, 131072);
13299			CTL_SIZE_8B(local_dma_sglist[i].len,
13300				    local_sglist[i].len);
13301			local_sglist[i].addr =
13302				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13303
13304			local_dma_sglist[i].addr = local_sglist[i].addr;
13305
13306			if (local_sglist[i].addr == NULL) {
13307				int j;
13308
13309				printf("malloc failed for %zd bytes!",
13310				       local_dma_sglist[i].len);
13311				for (j = 0; j < i; j++) {
13312					free(local_sglist[j].addr, M_CTL);
13313				}
13314				ctl_set_internal_failure(&io->scsiio,
13315							 /*sks_valid*/ 1,
13316							 /*retry_count*/ 4857);
13317				retval = 1;
13318				goto bailout_error;
13319
13320			}
13321			/* XXX KDM do we need a sync here? */
13322
13323			len_to_go -= local_sglist[i].len;
13324		}
13325		/*
13326		 * Reset the number of S/G entries accordingly.  The
13327		 * original number of S/G entries is available in
13328		 * rem_sg_entries.
13329		 */
13330		io->scsiio.kern_sg_entries = i;
13331
13332#if 0
13333		printf("%s: kern_sg_entries = %d\n", __func__,
13334		       io->scsiio.kern_sg_entries);
13335		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13336			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13337			       local_sglist[i].addr, local_sglist[i].len,
13338			       local_dma_sglist[i].len);
13339#endif
13340	}
13341
13342
13343	return (retval);
13344
13345bailout_error:
13346
13347	ctl_send_datamove_done(io, /*have_lock*/ 0);
13348
13349	return (retval);
13350}
13351
13352static int
13353ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13354			 ctl_ha_dt_cb callback)
13355{
13356	struct ctl_ha_dt_req *rq;
13357	struct ctl_sg_entry *remote_sglist, *local_sglist;
13358	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13359	uint32_t local_used, remote_used, total_used;
13360	int retval;
13361	int i, j;
13362
13363	retval = 0;
13364
13365	rq = ctl_dt_req_alloc();
13366
13367	/*
13368	 * If we failed to allocate the request, and if the DMA didn't fail
13369	 * anyway, set busy status.  This is just a resource allocation
13370	 * failure.
13371	 */
13372	if ((rq == NULL)
13373	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13374		ctl_set_busy(&io->scsiio);
13375
13376	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13377
13378		if (rq != NULL)
13379			ctl_dt_req_free(rq);
13380
13381		/*
13382		 * The data move failed.  We need to return status back
13383		 * to the other controller.  No point in trying to DMA
13384		 * data to the remote controller.
13385		 */
13386
13387		ctl_send_datamove_done(io, /*have_lock*/ 0);
13388
13389		retval = 1;
13390
13391		goto bailout;
13392	}
13393
13394	local_sglist = io->io_hdr.local_sglist;
13395	local_dma_sglist = io->io_hdr.local_dma_sglist;
13396	remote_sglist = io->io_hdr.remote_sglist;
13397	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13398	local_used = 0;
13399	remote_used = 0;
13400	total_used = 0;
13401
13402	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13403		rq->ret = CTL_HA_STATUS_SUCCESS;
13404		rq->context = io;
13405		callback(rq);
13406		goto bailout;
13407	}
13408
13409	/*
13410	 * Pull/push the data over the wire from/to the other controller.
13411	 * This takes into account the possibility that the local and
13412	 * remote sglists may not be identical in terms of the size of
13413	 * the elements and the number of elements.
13414	 *
13415	 * One fundamental assumption here is that the length allocated for
13416	 * both the local and remote sglists is identical.  Otherwise, we've
13417	 * essentially got a coding error of some sort.
13418	 */
13419	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13420		int isc_ret;
13421		uint32_t cur_len, dma_length;
13422		uint8_t *tmp_ptr;
13423
13424		rq->id = CTL_HA_DATA_CTL;
13425		rq->command = command;
13426		rq->context = io;
13427
13428		/*
13429		 * Both pointers should be aligned.  But it is possible
13430		 * that the allocation length is not.  They should both
13431		 * also have enough slack left over at the end, though,
13432		 * to round up to the next 8 byte boundary.
13433		 */
13434		cur_len = ctl_min(local_sglist[i].len - local_used,
13435				  remote_sglist[j].len - remote_used);
13436
13437		/*
13438		 * In this case, we have a size issue and need to decrease
13439		 * the size, except in the case where we actually have less
13440		 * than 8 bytes left.  In that case, we need to increase
13441		 * the DMA length to get the last bit.
13442		 */
13443		if ((cur_len & 0x7) != 0) {
13444			if (cur_len > 0x7) {
13445				cur_len = cur_len - (cur_len & 0x7);
13446				dma_length = cur_len;
13447			} else {
13448				CTL_SIZE_8B(dma_length, cur_len);
13449			}
13450
13451		} else
13452			dma_length = cur_len;
13453
13454		/*
13455		 * If we had to allocate memory for this I/O, instead of using
13456		 * the non-cached mirror memory, we'll need to flush the cache
13457		 * before trying to DMA to the other controller.
13458		 *
13459		 * We could end up doing this multiple times for the same
13460		 * segment if we have a larger local segment than remote
13461		 * segment.  That shouldn't be an issue.
13462		 */
13463		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13464			/*
13465			 * XXX KDM use bus_dmamap_sync() here.
13466			 */
13467		}
13468
13469		rq->size = dma_length;
13470
13471		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13472		tmp_ptr += local_used;
13473
13474		/* Use physical addresses when talking to ISC hardware */
13475		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13476			/* XXX KDM use busdma */
13477#if 0
13478			rq->local = vtophys(tmp_ptr);
13479#endif
13480		} else
13481			rq->local = tmp_ptr;
13482
13483		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13484		tmp_ptr += remote_used;
13485		rq->remote = tmp_ptr;
13486
13487		rq->callback = NULL;
13488
13489		local_used += cur_len;
13490		if (local_used >= local_sglist[i].len) {
13491			i++;
13492			local_used = 0;
13493		}
13494
13495		remote_used += cur_len;
13496		if (remote_used >= remote_sglist[j].len) {
13497			j++;
13498			remote_used = 0;
13499		}
13500		total_used += cur_len;
13501
13502		if (total_used >= io->scsiio.kern_data_len)
13503			rq->callback = callback;
13504
13505		if ((rq->size & 0x7) != 0) {
13506			printf("%s: warning: size %d is not on 8b boundary\n",
13507			       __func__, rq->size);
13508		}
13509		if (((uintptr_t)rq->local & 0x7) != 0) {
13510			printf("%s: warning: local %p not on 8b boundary\n",
13511			       __func__, rq->local);
13512		}
13513		if (((uintptr_t)rq->remote & 0x7) != 0) {
13514			printf("%s: warning: remote %p not on 8b boundary\n",
13515			       __func__, rq->local);
13516		}
13517#if 0
13518		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13519		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13520		       rq->local, rq->remote, rq->size);
13521#endif
13522
13523		isc_ret = ctl_dt_single(rq);
13524		if (isc_ret == CTL_HA_STATUS_WAIT)
13525			continue;
13526
13527		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13528			rq->ret = CTL_HA_STATUS_SUCCESS;
13529		} else {
13530			rq->ret = isc_ret;
13531		}
13532		callback(rq);
13533		goto bailout;
13534	}
13535
13536bailout:
13537	return (retval);
13538
13539}
13540
13541static void
13542ctl_datamove_remote_read(union ctl_io *io)
13543{
13544	int retval;
13545	int i;
13546
13547	/*
13548	 * This will send an error to the other controller in the case of a
13549	 * failure.
13550	 */
13551	retval = ctl_datamove_remote_sgl_setup(io);
13552	if (retval != 0)
13553		return;
13554
13555	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13556					  ctl_datamove_remote_read_cb);
13557	if ((retval != 0)
13558	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13559		/*
13560		 * Make sure we free memory if there was an error..  The
13561		 * ctl_datamove_remote_xfer() function will send the
13562		 * datamove done message, or call the callback with an
13563		 * error if there is a problem.
13564		 */
13565		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13566			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13567	}
13568
13569	return;
13570}
13571
13572/*
13573 * Process a datamove request from the other controller.  This is used for
13574 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13575 * first.  Once that is complete, the data gets DMAed into the remote
13576 * controller's memory.  For reads, we DMA from the remote controller's
13577 * memory into our memory first, and then move it out to the FETD.
13578 */
13579static void
13580ctl_datamove_remote(union ctl_io *io)
13581{
13582	struct ctl_softc *softc;
13583
13584	softc = control_softc;
13585
13586	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13587
13588	/*
13589	 * Note that we look for an aborted I/O here, but don't do some of
13590	 * the other checks that ctl_datamove() normally does.
13591	 * We don't need to run the datamove delay code, since that should
13592	 * have been done if need be on the other controller.
13593	 */
13594	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13595		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13596		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13597		       io->io_hdr.nexus.targ_port,
13598		       io->io_hdr.nexus.targ_target.id,
13599		       io->io_hdr.nexus.targ_lun);
13600		io->io_hdr.port_status = 31338;
13601		ctl_send_datamove_done(io, /*have_lock*/ 0);
13602		return;
13603	}
13604
13605	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13606		ctl_datamove_remote_write(io);
13607	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13608		ctl_datamove_remote_read(io);
13609	} else {
13610		union ctl_ha_msg msg;
13611		struct scsi_sense_data *sense;
13612		uint8_t sks[3];
13613		int retry_count;
13614
13615		memset(&msg, 0, sizeof(msg));
13616
13617		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13618		msg.hdr.status = CTL_SCSI_ERROR;
13619		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13620
13621		retry_count = 4243;
13622
13623		sense = &msg.scsi.sense_data;
13624		sks[0] = SSD_SCS_VALID;
13625		sks[1] = (retry_count >> 8) & 0xff;
13626		sks[2] = retry_count & 0xff;
13627
13628		/* "Internal target failure" */
13629		scsi_set_sense_data(sense,
13630				    /*sense_format*/ SSD_TYPE_NONE,
13631				    /*current_error*/ 1,
13632				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13633				    /*asc*/ 0x44,
13634				    /*ascq*/ 0x00,
13635				    /*type*/ SSD_ELEM_SKS,
13636				    /*size*/ sizeof(sks),
13637				    /*data*/ sks,
13638				    SSD_ELEM_NONE);
13639
13640		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13641		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13642			ctl_failover_io(io, /*have_lock*/ 1);
13643			return;
13644		}
13645
13646		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13647		    CTL_HA_STATUS_SUCCESS) {
13648			/* XXX KDM what to do if this fails? */
13649		}
13650		return;
13651	}
13652
13653}
13654
13655static int
13656ctl_process_done(union ctl_io *io)
13657{
13658	struct ctl_lun *lun;
13659	struct ctl_softc *ctl_softc;
13660	void (*fe_done)(union ctl_io *io);
13661	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13662
13663	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13664
13665	fe_done =
13666	    control_softc->ctl_ports[targ_port]->fe_done;
13667
13668#ifdef CTL_TIME_IO
13669	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13670		char str[256];
13671		char path_str[64];
13672		struct sbuf sb;
13673
13674		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13675		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13676
13677		sbuf_cat(&sb, path_str);
13678		switch (io->io_hdr.io_type) {
13679		case CTL_IO_SCSI:
13680			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13681			sbuf_printf(&sb, "\n");
13682			sbuf_cat(&sb, path_str);
13683			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13684				    io->scsiio.tag_num, io->scsiio.tag_type);
13685			break;
13686		case CTL_IO_TASK:
13687			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13688				    "Tag Type: %d\n", io->taskio.task_action,
13689				    io->taskio.tag_num, io->taskio.tag_type);
13690			break;
13691		default:
13692			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13693			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13694			break;
13695		}
13696		sbuf_cat(&sb, path_str);
13697		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13698			    (intmax_t)time_uptime - io->io_hdr.start_time);
13699		sbuf_finish(&sb);
13700		printf("%s", sbuf_data(&sb));
13701	}
13702#endif /* CTL_TIME_IO */
13703
13704	switch (io->io_hdr.io_type) {
13705	case CTL_IO_SCSI:
13706		break;
13707	case CTL_IO_TASK:
13708		if (bootverbose || verbose > 0)
13709			ctl_io_error_print(io, NULL);
13710		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13711			ctl_free_io(io);
13712		else
13713			fe_done(io);
13714		return (CTL_RETVAL_COMPLETE);
13715		break;
13716	default:
13717		printf("ctl_process_done: invalid io type %d\n",
13718		       io->io_hdr.io_type);
13719		panic("ctl_process_done: invalid io type %d\n",
13720		      io->io_hdr.io_type);
13721		break; /* NOTREACHED */
13722	}
13723
13724	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13725	if (lun == NULL) {
13726		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13727				 io->io_hdr.nexus.targ_mapped_lun));
13728		fe_done(io);
13729		goto bailout;
13730	}
13731	ctl_softc = lun->ctl_softc;
13732
13733	mtx_lock(&lun->lun_lock);
13734
13735	/*
13736	 * Check to see if we have any errors to inject here.  We only
13737	 * inject errors for commands that don't already have errors set.
13738	 */
13739	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13740	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13741		ctl_inject_error(lun, io);
13742
13743	/*
13744	 * XXX KDM how do we treat commands that aren't completed
13745	 * successfully?
13746	 *
13747	 * XXX KDM should we also track I/O latency?
13748	 */
13749	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13750	    io->io_hdr.io_type == CTL_IO_SCSI) {
13751#ifdef CTL_TIME_IO
13752		struct bintime cur_bt;
13753#endif
13754		int type;
13755
13756		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13757		    CTL_FLAG_DATA_IN)
13758			type = CTL_STATS_READ;
13759		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13760		    CTL_FLAG_DATA_OUT)
13761			type = CTL_STATS_WRITE;
13762		else
13763			type = CTL_STATS_NO_IO;
13764
13765		lun->stats.ports[targ_port].bytes[type] +=
13766		    io->scsiio.kern_total_len;
13767		lun->stats.ports[targ_port].operations[type]++;
13768#ifdef CTL_TIME_IO
13769		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13770		   &io->io_hdr.dma_bt);
13771		lun->stats.ports[targ_port].num_dmas[type] +=
13772		    io->io_hdr.num_dmas;
13773		getbintime(&cur_bt);
13774		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13775		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13776#endif
13777	}
13778
13779	/*
13780	 * Remove this from the OOA queue.
13781	 */
13782	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13783
13784	/*
13785	 * Run through the blocked queue on this LUN and see if anything
13786	 * has become unblocked, now that this transaction is done.
13787	 */
13788	ctl_check_blocked(lun);
13789
13790	/*
13791	 * If the LUN has been invalidated, free it if there is nothing
13792	 * left on its OOA queue.
13793	 */
13794	if ((lun->flags & CTL_LUN_INVALID)
13795	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13796		mtx_unlock(&lun->lun_lock);
13797		mtx_lock(&ctl_softc->ctl_lock);
13798		ctl_free_lun(lun);
13799		mtx_unlock(&ctl_softc->ctl_lock);
13800	} else
13801		mtx_unlock(&lun->lun_lock);
13802
13803	/*
13804	 * If this command has been aborted, make sure we set the status
13805	 * properly.  The FETD is responsible for freeing the I/O and doing
13806	 * whatever it needs to do to clean up its state.
13807	 */
13808	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13809		ctl_set_task_aborted(&io->scsiio);
13810
13811	/*
13812	 * We print out status for every task management command.  For SCSI
13813	 * commands, we filter out any unit attention errors; they happen
13814	 * on every boot, and would clutter up the log.  Note:  task
13815	 * management commands aren't printed here, they are printed above,
13816	 * since they should never even make it down here.
13817	 */
13818	switch (io->io_hdr.io_type) {
13819	case CTL_IO_SCSI: {
13820		int error_code, sense_key, asc, ascq;
13821
13822		sense_key = 0;
13823
13824		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13825		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13826			/*
13827			 * Since this is just for printing, no need to
13828			 * show errors here.
13829			 */
13830			scsi_extract_sense_len(&io->scsiio.sense_data,
13831					       io->scsiio.sense_len,
13832					       &error_code,
13833					       &sense_key,
13834					       &asc,
13835					       &ascq,
13836					       /*show_errors*/ 0);
13837		}
13838
13839		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13840		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13841		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13842		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13843
13844			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13845				ctl_softc->skipped_prints++;
13846			} else {
13847				uint32_t skipped_prints;
13848
13849				skipped_prints = ctl_softc->skipped_prints;
13850
13851				ctl_softc->skipped_prints = 0;
13852				ctl_softc->last_print_jiffies = time_uptime;
13853
13854				if (skipped_prints > 0) {
13855#ifdef NEEDTOPORT
13856					csevent_log(CSC_CTL | CSC_SHELF_SW |
13857					    CTL_ERROR_REPORT,
13858					    csevent_LogType_Trace,
13859					    csevent_Severity_Information,
13860					    csevent_AlertLevel_Green,
13861					    csevent_FRU_Firmware,
13862					    csevent_FRU_Unknown,
13863					    "High CTL error volume, %d prints "
13864					    "skipped", skipped_prints);
13865#endif
13866				}
13867				if (bootverbose || verbose > 0)
13868					ctl_io_error_print(io, NULL);
13869			}
13870		}
13871		break;
13872	}
13873	case CTL_IO_TASK:
13874		if (bootverbose || verbose > 0)
13875			ctl_io_error_print(io, NULL);
13876		break;
13877	default:
13878		break;
13879	}
13880
13881	/*
13882	 * Tell the FETD or the other shelf controller we're done with this
13883	 * command.  Note that only SCSI commands get to this point.  Task
13884	 * management commands are completed above.
13885	 *
13886	 * We only send status to the other controller if we're in XFER
13887	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13888	 * received the I/O (from CTL's perspective), and so the status is
13889	 * generated there.
13890	 *
13891	 * XXX KDM if we hold the lock here, we could cause a deadlock
13892	 * if the frontend comes back in in this context to queue
13893	 * something.
13894	 */
13895	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13896	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13897		union ctl_ha_msg msg;
13898
13899		memset(&msg, 0, sizeof(msg));
13900		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13901		msg.hdr.original_sc = io->io_hdr.original_sc;
13902		msg.hdr.nexus = io->io_hdr.nexus;
13903		msg.hdr.status = io->io_hdr.status;
13904		msg.scsi.scsi_status = io->scsiio.scsi_status;
13905		msg.scsi.tag_num = io->scsiio.tag_num;
13906		msg.scsi.tag_type = io->scsiio.tag_type;
13907		msg.scsi.sense_len = io->scsiio.sense_len;
13908		msg.scsi.sense_residual = io->scsiio.sense_residual;
13909		msg.scsi.residual = io->scsiio.residual;
13910		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13911		       sizeof(io->scsiio.sense_data));
13912		/*
13913		 * We copy this whether or not this is an I/O-related
13914		 * command.  Otherwise, we'd have to go and check to see
13915		 * whether it's a read/write command, and it really isn't
13916		 * worth it.
13917		 */
13918		memcpy(&msg.scsi.lbalen,
13919		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13920		       sizeof(msg.scsi.lbalen));
13921
13922		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13923				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13924			/* XXX do something here */
13925		}
13926
13927		ctl_free_io(io);
13928	} else
13929		fe_done(io);
13930
13931bailout:
13932
13933	return (CTL_RETVAL_COMPLETE);
13934}
13935
13936#ifdef CTL_WITH_CA
13937/*
13938 * Front end should call this if it doesn't do autosense.  When the request
13939 * sense comes back in from the initiator, we'll dequeue this and send it.
13940 */
13941int
13942ctl_queue_sense(union ctl_io *io)
13943{
13944	struct ctl_lun *lun;
13945	struct ctl_softc *ctl_softc;
13946	uint32_t initidx, targ_lun;
13947
13948	ctl_softc = control_softc;
13949
13950	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13951
13952	/*
13953	 * LUN lookup will likely move to the ctl_work_thread() once we
13954	 * have our new queueing infrastructure (that doesn't put things on
13955	 * a per-LUN queue initially).  That is so that we can handle
13956	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13957	 * can't deal with that right now.
13958	 */
13959	mtx_lock(&ctl_softc->ctl_lock);
13960
13961	/*
13962	 * If we don't have a LUN for this, just toss the sense
13963	 * information.
13964	 */
13965	targ_lun = io->io_hdr.nexus.targ_lun;
13966	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13967	if ((targ_lun < CTL_MAX_LUNS)
13968	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13969		lun = ctl_softc->ctl_luns[targ_lun];
13970	else
13971		goto bailout;
13972
13973	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13974
13975	mtx_lock(&lun->lun_lock);
13976	/*
13977	 * Already have CA set for this LUN...toss the sense information.
13978	 */
13979	if (ctl_is_set(lun->have_ca, initidx)) {
13980		mtx_unlock(&lun->lun_lock);
13981		goto bailout;
13982	}
13983
13984	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13985	       ctl_min(sizeof(lun->pending_sense[initidx]),
13986	       sizeof(io->scsiio.sense_data)));
13987	ctl_set_mask(lun->have_ca, initidx);
13988	mtx_unlock(&lun->lun_lock);
13989
13990bailout:
13991	mtx_unlock(&ctl_softc->ctl_lock);
13992
13993	ctl_free_io(io);
13994
13995	return (CTL_RETVAL_COMPLETE);
13996}
13997#endif
13998
13999/*
14000 * Primary command inlet from frontend ports.  All SCSI and task I/O
14001 * requests must go through this function.
14002 */
14003int
14004ctl_queue(union ctl_io *io)
14005{
14006	struct ctl_softc *ctl_softc;
14007
14008	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
14009
14010	ctl_softc = control_softc;
14011
14012#ifdef CTL_TIME_IO
14013	io->io_hdr.start_time = time_uptime;
14014	getbintime(&io->io_hdr.start_bt);
14015#endif /* CTL_TIME_IO */
14016
14017	/* Map FE-specific LUN ID into global one. */
14018	io->io_hdr.nexus.targ_mapped_lun =
14019	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
14020
14021	switch (io->io_hdr.io_type) {
14022	case CTL_IO_SCSI:
14023	case CTL_IO_TASK:
14024		ctl_enqueue_incoming(io);
14025		break;
14026	default:
14027		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
14028		return (EINVAL);
14029	}
14030
14031	return (CTL_RETVAL_COMPLETE);
14032}
14033
14034#ifdef CTL_IO_DELAY
14035static void
14036ctl_done_timer_wakeup(void *arg)
14037{
14038	union ctl_io *io;
14039
14040	io = (union ctl_io *)arg;
14041	ctl_done(io);
14042}
14043#endif /* CTL_IO_DELAY */
14044
14045void
14046ctl_done(union ctl_io *io)
14047{
14048	struct ctl_softc *ctl_softc;
14049
14050	ctl_softc = control_softc;
14051
14052	/*
14053	 * Enable this to catch duplicate completion issues.
14054	 */
14055#if 0
14056	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
14057		printf("%s: type %d msg %d cdb %x iptl: "
14058		       "%d:%d:%d:%d tag 0x%04x "
14059		       "flag %#x status %x\n",
14060			__func__,
14061			io->io_hdr.io_type,
14062			io->io_hdr.msg_type,
14063			io->scsiio.cdb[0],
14064			io->io_hdr.nexus.initid.id,
14065			io->io_hdr.nexus.targ_port,
14066			io->io_hdr.nexus.targ_target.id,
14067			io->io_hdr.nexus.targ_lun,
14068			(io->io_hdr.io_type ==
14069			CTL_IO_TASK) ?
14070			io->taskio.tag_num :
14071			io->scsiio.tag_num,
14072		        io->io_hdr.flags,
14073			io->io_hdr.status);
14074	} else
14075		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
14076#endif
14077
14078	/*
14079	 * This is an internal copy of an I/O, and should not go through
14080	 * the normal done processing logic.
14081	 */
14082	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
14083		return;
14084
14085	/*
14086	 * We need to send a msg to the serializing shelf to finish the IO
14087	 * as well.  We don't send a finish message to the other shelf if
14088	 * this is a task management command.  Task management commands
14089	 * aren't serialized in the OOA queue, but rather just executed on
14090	 * both shelf controllers for commands that originated on that
14091	 * controller.
14092	 */
14093	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
14094	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
14095		union ctl_ha_msg msg_io;
14096
14097		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
14098		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
14099		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
14100		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
14101		}
14102		/* continue on to finish IO */
14103	}
14104#ifdef CTL_IO_DELAY
14105	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
14106		struct ctl_lun *lun;
14107
14108		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14109
14110		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
14111	} else {
14112		struct ctl_lun *lun;
14113
14114		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14115
14116		if ((lun != NULL)
14117		 && (lun->delay_info.done_delay > 0)) {
14118			struct callout *callout;
14119
14120			callout = (struct callout *)&io->io_hdr.timer_bytes;
14121			callout_init(callout, /*mpsafe*/ 1);
14122			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
14123			callout_reset(callout,
14124				      lun->delay_info.done_delay * hz,
14125				      ctl_done_timer_wakeup, io);
14126			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
14127				lun->delay_info.done_delay = 0;
14128			return;
14129		}
14130	}
14131#endif /* CTL_IO_DELAY */
14132
14133	ctl_enqueue_done(io);
14134}
14135
14136int
14137ctl_isc(struct ctl_scsiio *ctsio)
14138{
14139	struct ctl_lun *lun;
14140	int retval;
14141
14142	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14143
14144	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
14145
14146	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
14147
14148	retval = lun->backend->data_submit((union ctl_io *)ctsio);
14149
14150	return (retval);
14151}
14152
14153
14154static void
14155ctl_work_thread(void *arg)
14156{
14157	struct ctl_thread *thr = (struct ctl_thread *)arg;
14158	struct ctl_softc *softc = thr->ctl_softc;
14159	union ctl_io *io;
14160	int retval;
14161
14162	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
14163
14164	for (;;) {
14165		retval = 0;
14166
14167		/*
14168		 * We handle the queues in this order:
14169		 * - ISC
14170		 * - done queue (to free up resources, unblock other commands)
14171		 * - RtR queue
14172		 * - incoming queue
14173		 *
14174		 * If those queues are empty, we break out of the loop and
14175		 * go to sleep.
14176		 */
14177		mtx_lock(&thr->queue_lock);
14178		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
14179		if (io != NULL) {
14180			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
14181			mtx_unlock(&thr->queue_lock);
14182			ctl_handle_isc(io);
14183			continue;
14184		}
14185		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
14186		if (io != NULL) {
14187			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
14188			/* clear any blocked commands, call fe_done */
14189			mtx_unlock(&thr->queue_lock);
14190			retval = ctl_process_done(io);
14191			continue;
14192		}
14193		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
14194		if (io != NULL) {
14195			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
14196			mtx_unlock(&thr->queue_lock);
14197			if (io->io_hdr.io_type == CTL_IO_TASK)
14198				ctl_run_task(io);
14199			else
14200				ctl_scsiio_precheck(softc, &io->scsiio);
14201			continue;
14202		}
14203		if (!ctl_pause_rtr) {
14204			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
14205			if (io != NULL) {
14206				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
14207				mtx_unlock(&thr->queue_lock);
14208				retval = ctl_scsiio(&io->scsiio);
14209				if (retval != CTL_RETVAL_COMPLETE)
14210					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
14211				continue;
14212			}
14213		}
14214
14215		/* Sleep until we have something to do. */
14216		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
14217	}
14218}
14219
14220static void
14221ctl_lun_thread(void *arg)
14222{
14223	struct ctl_softc *softc = (struct ctl_softc *)arg;
14224	struct ctl_be_lun *be_lun;
14225	int retval;
14226
14227	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14228
14229	for (;;) {
14230		retval = 0;
14231		mtx_lock(&softc->ctl_lock);
14232		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14233		if (be_lun != NULL) {
14234			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14235			mtx_unlock(&softc->ctl_lock);
14236			ctl_create_lun(be_lun);
14237			continue;
14238		}
14239
14240		/* Sleep until we have something to do. */
14241		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14242		    PDROP | PRIBIO, "-", 0);
14243	}
14244}
14245
14246static void
14247ctl_enqueue_incoming(union ctl_io *io)
14248{
14249	struct ctl_softc *softc = control_softc;
14250	struct ctl_thread *thr;
14251	u_int idx;
14252
14253	idx = (io->io_hdr.nexus.targ_port * 127 +
14254	       io->io_hdr.nexus.initid.id) % worker_threads;
14255	thr = &softc->threads[idx];
14256	mtx_lock(&thr->queue_lock);
14257	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14258	mtx_unlock(&thr->queue_lock);
14259	wakeup(thr);
14260}
14261
14262static void
14263ctl_enqueue_rtr(union ctl_io *io)
14264{
14265	struct ctl_softc *softc = control_softc;
14266	struct ctl_thread *thr;
14267
14268	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14269	mtx_lock(&thr->queue_lock);
14270	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14271	mtx_unlock(&thr->queue_lock);
14272	wakeup(thr);
14273}
14274
14275static void
14276ctl_enqueue_done(union ctl_io *io)
14277{
14278	struct ctl_softc *softc = control_softc;
14279	struct ctl_thread *thr;
14280
14281	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14282	mtx_lock(&thr->queue_lock);
14283	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14284	mtx_unlock(&thr->queue_lock);
14285	wakeup(thr);
14286}
14287
14288static void
14289ctl_enqueue_isc(union ctl_io *io)
14290{
14291	struct ctl_softc *softc = control_softc;
14292	struct ctl_thread *thr;
14293
14294	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14295	mtx_lock(&thr->queue_lock);
14296	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14297	mtx_unlock(&thr->queue_lock);
14298	wakeup(thr);
14299}
14300
14301/* Initialization and failover */
14302
14303void
14304ctl_init_isc_msg(void)
14305{
14306	printf("CTL: Still calling this thing\n");
14307}
14308
14309/*
14310 * Init component
14311 * 	Initializes component into configuration defined by bootMode
14312 *	(see hasc-sv.c)
14313 *  	returns hasc_Status:
14314 * 		OK
14315 *		ERROR - fatal error
14316 */
14317static ctl_ha_comp_status
14318ctl_isc_init(struct ctl_ha_component *c)
14319{
14320	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14321
14322	c->status = ret;
14323	return ret;
14324}
14325
14326/* Start component
14327 * 	Starts component in state requested. If component starts successfully,
14328 *	it must set its own state to the requestrd state
14329 *	When requested state is HASC_STATE_HA, the component may refine it
14330 * 	by adding _SLAVE or _MASTER flags.
14331 *	Currently allowed state transitions are:
14332 *	UNKNOWN->HA		- initial startup
14333 *	UNKNOWN->SINGLE - initial startup when no parter detected
14334 *	HA->SINGLE		- failover
14335 * returns ctl_ha_comp_status:
14336 * 		OK	- component successfully started in requested state
14337 *		FAILED  - could not start the requested state, failover may
14338 * 			  be possible
14339 *		ERROR	- fatal error detected, no future startup possible
14340 */
14341static ctl_ha_comp_status
14342ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14343{
14344	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14345
14346	printf("%s: go\n", __func__);
14347
14348	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14349	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14350		ctl_is_single = 0;
14351		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14352		    != CTL_HA_STATUS_SUCCESS) {
14353			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14354			ret = CTL_HA_COMP_STATUS_ERROR;
14355		}
14356	} else if (CTL_HA_STATE_IS_HA(c->state)
14357		&& CTL_HA_STATE_IS_SINGLE(state)){
14358		// HA->SINGLE transition
14359	        ctl_failover();
14360		ctl_is_single = 1;
14361	} else {
14362		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14363		       c->state, state);
14364		ret = CTL_HA_COMP_STATUS_ERROR;
14365	}
14366	if (CTL_HA_STATE_IS_SINGLE(state))
14367		ctl_is_single = 1;
14368
14369	c->state = state;
14370	c->status = ret;
14371	return ret;
14372}
14373
14374/*
14375 * Quiesce component
14376 * The component must clear any error conditions (set status to OK) and
14377 * prepare itself to another Start call
14378 * returns ctl_ha_comp_status:
14379 * 	OK
14380 *	ERROR
14381 */
14382static ctl_ha_comp_status
14383ctl_isc_quiesce(struct ctl_ha_component *c)
14384{
14385	int ret = CTL_HA_COMP_STATUS_OK;
14386
14387	ctl_pause_rtr = 1;
14388	c->status = ret;
14389	return ret;
14390}
14391
14392struct ctl_ha_component ctl_ha_component_ctlisc =
14393{
14394	.name = "CTL ISC",
14395	.state = CTL_HA_STATE_UNKNOWN,
14396	.init = ctl_isc_init,
14397	.start = ctl_isc_start,
14398	.quiesce = ctl_isc_quiesce
14399};
14400
14401/*
14402 *  vim: ts=8
14403 */
14404