ctl.c revision 273311
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 273311 2014-10-20 07:33:41Z 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*/0,
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;
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	lun->ctl_softc = ctl_softc;
4615	TAILQ_INIT(&lun->ooa_queue);
4616	TAILQ_INIT(&lun->blocked_queue);
4617	STAILQ_INIT(&lun->error_list);
4618	ctl_tpc_lun_init(lun);
4619
4620	/*
4621	 * Initialize the mode page index.
4622	 */
4623	ctl_init_page_index(lun);
4624
4625	/*
4626	 * Set the poweron UA for all initiators on this LUN only.
4627	 */
4628	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4629		lun->pending_ua[i] = CTL_UA_POWERON;
4630
4631	/*
4632	 * Now, before we insert this lun on the lun list, set the lun
4633	 * inventory changed UA for all other luns.
4634	 */
4635	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4636		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4637			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4638		}
4639	}
4640
4641	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4642
4643	ctl_softc->ctl_luns[lun_number] = lun;
4644
4645	ctl_softc->num_luns++;
4646
4647	/* Setup statistics gathering */
4648	lun->stats.device_type = be_lun->lun_type;
4649	lun->stats.lun_number = lun_number;
4650	if (lun->stats.device_type == T_DIRECT)
4651		lun->stats.blocksize = be_lun->blocksize;
4652	else
4653		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4654	for (i = 0;i < CTL_MAX_PORTS;i++)
4655		lun->stats.ports[i].targ_port = i;
4656
4657	mtx_unlock(&ctl_softc->ctl_lock);
4658
4659	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4660
4661	/*
4662	 * Run through each registered FETD and bring it online if it isn't
4663	 * already.  Enable the target ID if it hasn't been enabled, and
4664	 * enable this particular LUN.
4665	 */
4666	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4667		int retval;
4668
4669		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4670		if (retval != 0) {
4671			printf("ctl_alloc_lun: FETD %s port %d returned error "
4672			       "%d for lun_enable on target %ju lun %d\n",
4673			       port->port_name, port->targ_port, retval,
4674			       (uintmax_t)target_id.id, lun_number);
4675		} else
4676			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4677	}
4678	return (0);
4679}
4680
4681/*
4682 * Delete a LUN.
4683 * Assumptions:
4684 * - LUN has already been marked invalid and any pending I/O has been taken
4685 *   care of.
4686 */
4687static int
4688ctl_free_lun(struct ctl_lun *lun)
4689{
4690	struct ctl_softc *softc;
4691#if 0
4692	struct ctl_port *port;
4693#endif
4694	struct ctl_lun *nlun;
4695	int i;
4696
4697	softc = lun->ctl_softc;
4698
4699	mtx_assert(&softc->ctl_lock, MA_OWNED);
4700
4701	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4702
4703	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4704
4705	softc->ctl_luns[lun->lun] = NULL;
4706
4707	if (!TAILQ_EMPTY(&lun->ooa_queue))
4708		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4709
4710	softc->num_luns--;
4711
4712	/*
4713	 * XXX KDM this scheme only works for a single target/multiple LUN
4714	 * setup.  It needs to be revamped for a multiple target scheme.
4715	 *
4716	 * XXX KDM this results in port->lun_disable() getting called twice,
4717	 * once when ctl_disable_lun() is called, and a second time here.
4718	 * We really need to re-think the LUN disable semantics.  There
4719	 * should probably be several steps/levels to LUN removal:
4720	 *  - disable
4721	 *  - invalidate
4722	 *  - free
4723 	 *
4724	 * Right now we only have a disable method when communicating to
4725	 * the front end ports, at least for individual LUNs.
4726	 */
4727#if 0
4728	STAILQ_FOREACH(port, &softc->port_list, links) {
4729		int retval;
4730
4731		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4732					 lun->lun);
4733		if (retval != 0) {
4734			printf("ctl_free_lun: FETD %s port %d returned error "
4735			       "%d for lun_disable on target %ju lun %jd\n",
4736			       port->port_name, port->targ_port, retval,
4737			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4738		}
4739
4740		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4741			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4742
4743			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4744			if (retval != 0) {
4745				printf("ctl_free_lun: FETD %s port %d "
4746				       "returned error %d for targ_disable on "
4747				       "target %ju\n", port->port_name,
4748				       port->targ_port, retval,
4749				       (uintmax_t)lun->target.id);
4750			} else
4751				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4752
4753			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4754				continue;
4755
4756#if 0
4757			port->port_offline(port->onoff_arg);
4758			port->status &= ~CTL_PORT_STATUS_ONLINE;
4759#endif
4760		}
4761	}
4762#endif
4763
4764	/*
4765	 * Tell the backend to free resources, if this LUN has a backend.
4766	 */
4767	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4768	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4769
4770	ctl_tpc_lun_shutdown(lun);
4771	mtx_destroy(&lun->lun_lock);
4772	free(lun->lun_devid, M_CTL);
4773	if (lun->flags & CTL_LUN_MALLOCED)
4774		free(lun, M_CTL);
4775
4776	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4777		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4778			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4779		}
4780	}
4781
4782	return (0);
4783}
4784
4785static void
4786ctl_create_lun(struct ctl_be_lun *be_lun)
4787{
4788	struct ctl_softc *ctl_softc;
4789
4790	ctl_softc = control_softc;
4791
4792	/*
4793	 * ctl_alloc_lun() should handle all potential failure cases.
4794	 */
4795	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4796}
4797
4798int
4799ctl_add_lun(struct ctl_be_lun *be_lun)
4800{
4801	struct ctl_softc *ctl_softc = control_softc;
4802
4803	mtx_lock(&ctl_softc->ctl_lock);
4804	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4805	mtx_unlock(&ctl_softc->ctl_lock);
4806	wakeup(&ctl_softc->pending_lun_queue);
4807
4808	return (0);
4809}
4810
4811int
4812ctl_enable_lun(struct ctl_be_lun *be_lun)
4813{
4814	struct ctl_softc *ctl_softc;
4815	struct ctl_port *port, *nport;
4816	struct ctl_lun *lun;
4817	int retval;
4818
4819	ctl_softc = control_softc;
4820
4821	lun = (struct ctl_lun *)be_lun->ctl_lun;
4822
4823	mtx_lock(&ctl_softc->ctl_lock);
4824	mtx_lock(&lun->lun_lock);
4825	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4826		/*
4827		 * eh?  Why did we get called if the LUN is already
4828		 * enabled?
4829		 */
4830		mtx_unlock(&lun->lun_lock);
4831		mtx_unlock(&ctl_softc->ctl_lock);
4832		return (0);
4833	}
4834	lun->flags &= ~CTL_LUN_DISABLED;
4835	mtx_unlock(&lun->lun_lock);
4836
4837	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4838		nport = STAILQ_NEXT(port, links);
4839
4840		/*
4841		 * Drop the lock while we call the FETD's enable routine.
4842		 * This can lead to a callback into CTL (at least in the
4843		 * case of the internal initiator frontend.
4844		 */
4845		mtx_unlock(&ctl_softc->ctl_lock);
4846		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4847		mtx_lock(&ctl_softc->ctl_lock);
4848		if (retval != 0) {
4849			printf("%s: FETD %s port %d returned error "
4850			       "%d for lun_enable on target %ju lun %jd\n",
4851			       __func__, port->port_name, port->targ_port, retval,
4852			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4853		}
4854#if 0
4855		 else {
4856            /* NOTE:  TODO:  why does lun enable affect port status? */
4857			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4858		}
4859#endif
4860	}
4861
4862	mtx_unlock(&ctl_softc->ctl_lock);
4863
4864	return (0);
4865}
4866
4867int
4868ctl_disable_lun(struct ctl_be_lun *be_lun)
4869{
4870	struct ctl_softc *ctl_softc;
4871	struct ctl_port *port;
4872	struct ctl_lun *lun;
4873	int retval;
4874
4875	ctl_softc = control_softc;
4876
4877	lun = (struct ctl_lun *)be_lun->ctl_lun;
4878
4879	mtx_lock(&ctl_softc->ctl_lock);
4880	mtx_lock(&lun->lun_lock);
4881	if (lun->flags & CTL_LUN_DISABLED) {
4882		mtx_unlock(&lun->lun_lock);
4883		mtx_unlock(&ctl_softc->ctl_lock);
4884		return (0);
4885	}
4886	lun->flags |= CTL_LUN_DISABLED;
4887	mtx_unlock(&lun->lun_lock);
4888
4889	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4890		mtx_unlock(&ctl_softc->ctl_lock);
4891		/*
4892		 * Drop the lock before we call the frontend's disable
4893		 * routine, to avoid lock order reversals.
4894		 *
4895		 * XXX KDM what happens if the frontend list changes while
4896		 * we're traversing it?  It's unlikely, but should be handled.
4897		 */
4898		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4899					 lun->lun);
4900		mtx_lock(&ctl_softc->ctl_lock);
4901		if (retval != 0) {
4902			printf("ctl_alloc_lun: FETD %s port %d returned error "
4903			       "%d for lun_disable on target %ju lun %jd\n",
4904			       port->port_name, port->targ_port, retval,
4905			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4906		}
4907	}
4908
4909	mtx_unlock(&ctl_softc->ctl_lock);
4910
4911	return (0);
4912}
4913
4914int
4915ctl_start_lun(struct ctl_be_lun *be_lun)
4916{
4917	struct ctl_softc *ctl_softc;
4918	struct ctl_lun *lun;
4919
4920	ctl_softc = control_softc;
4921
4922	lun = (struct ctl_lun *)be_lun->ctl_lun;
4923
4924	mtx_lock(&lun->lun_lock);
4925	lun->flags &= ~CTL_LUN_STOPPED;
4926	mtx_unlock(&lun->lun_lock);
4927
4928	return (0);
4929}
4930
4931int
4932ctl_stop_lun(struct ctl_be_lun *be_lun)
4933{
4934	struct ctl_softc *ctl_softc;
4935	struct ctl_lun *lun;
4936
4937	ctl_softc = control_softc;
4938
4939	lun = (struct ctl_lun *)be_lun->ctl_lun;
4940
4941	mtx_lock(&lun->lun_lock);
4942	lun->flags |= CTL_LUN_STOPPED;
4943	mtx_unlock(&lun->lun_lock);
4944
4945	return (0);
4946}
4947
4948int
4949ctl_lun_offline(struct ctl_be_lun *be_lun)
4950{
4951	struct ctl_softc *ctl_softc;
4952	struct ctl_lun *lun;
4953
4954	ctl_softc = control_softc;
4955
4956	lun = (struct ctl_lun *)be_lun->ctl_lun;
4957
4958	mtx_lock(&lun->lun_lock);
4959	lun->flags |= CTL_LUN_OFFLINE;
4960	mtx_unlock(&lun->lun_lock);
4961
4962	return (0);
4963}
4964
4965int
4966ctl_lun_online(struct ctl_be_lun *be_lun)
4967{
4968	struct ctl_softc *ctl_softc;
4969	struct ctl_lun *lun;
4970
4971	ctl_softc = control_softc;
4972
4973	lun = (struct ctl_lun *)be_lun->ctl_lun;
4974
4975	mtx_lock(&lun->lun_lock);
4976	lun->flags &= ~CTL_LUN_OFFLINE;
4977	mtx_unlock(&lun->lun_lock);
4978
4979	return (0);
4980}
4981
4982int
4983ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4984{
4985	struct ctl_softc *ctl_softc;
4986	struct ctl_lun *lun;
4987
4988	ctl_softc = control_softc;
4989
4990	lun = (struct ctl_lun *)be_lun->ctl_lun;
4991
4992	mtx_lock(&lun->lun_lock);
4993
4994	/*
4995	 * The LUN needs to be disabled before it can be marked invalid.
4996	 */
4997	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4998		mtx_unlock(&lun->lun_lock);
4999		return (-1);
5000	}
5001	/*
5002	 * Mark the LUN invalid.
5003	 */
5004	lun->flags |= CTL_LUN_INVALID;
5005
5006	/*
5007	 * If there is nothing in the OOA queue, go ahead and free the LUN.
5008	 * If we have something in the OOA queue, we'll free it when the
5009	 * last I/O completes.
5010	 */
5011	if (TAILQ_EMPTY(&lun->ooa_queue)) {
5012		mtx_unlock(&lun->lun_lock);
5013		mtx_lock(&ctl_softc->ctl_lock);
5014		ctl_free_lun(lun);
5015		mtx_unlock(&ctl_softc->ctl_lock);
5016	} else
5017		mtx_unlock(&lun->lun_lock);
5018
5019	return (0);
5020}
5021
5022int
5023ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5024{
5025	struct ctl_softc *ctl_softc;
5026	struct ctl_lun *lun;
5027
5028	ctl_softc = control_softc;
5029	lun = (struct ctl_lun *)be_lun->ctl_lun;
5030
5031	mtx_lock(&lun->lun_lock);
5032	lun->flags |= CTL_LUN_INOPERABLE;
5033	mtx_unlock(&lun->lun_lock);
5034
5035	return (0);
5036}
5037
5038int
5039ctl_lun_operable(struct ctl_be_lun *be_lun)
5040{
5041	struct ctl_softc *ctl_softc;
5042	struct ctl_lun *lun;
5043
5044	ctl_softc = control_softc;
5045	lun = (struct ctl_lun *)be_lun->ctl_lun;
5046
5047	mtx_lock(&lun->lun_lock);
5048	lun->flags &= ~CTL_LUN_INOPERABLE;
5049	mtx_unlock(&lun->lun_lock);
5050
5051	return (0);
5052}
5053
5054int
5055ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5056		   int lock)
5057{
5058	struct ctl_softc *softc;
5059	struct ctl_lun *lun;
5060	struct copan_aps_subpage *current_sp;
5061	struct ctl_page_index *page_index;
5062	int i;
5063
5064	softc = control_softc;
5065
5066	mtx_lock(&softc->ctl_lock);
5067
5068	lun = (struct ctl_lun *)be_lun->ctl_lun;
5069	mtx_lock(&lun->lun_lock);
5070
5071	page_index = NULL;
5072	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5073		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5074		     APS_PAGE_CODE)
5075			continue;
5076
5077		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5078			continue;
5079		page_index = &lun->mode_pages.index[i];
5080	}
5081
5082	if (page_index == NULL) {
5083		mtx_unlock(&lun->lun_lock);
5084		mtx_unlock(&softc->ctl_lock);
5085		printf("%s: APS subpage not found for lun %ju!\n", __func__,
5086		       (uintmax_t)lun->lun);
5087		return (1);
5088	}
5089#if 0
5090	if ((softc->aps_locked_lun != 0)
5091	 && (softc->aps_locked_lun != lun->lun)) {
5092		printf("%s: attempt to lock LUN %llu when %llu is already "
5093		       "locked\n");
5094		mtx_unlock(&lun->lun_lock);
5095		mtx_unlock(&softc->ctl_lock);
5096		return (1);
5097	}
5098#endif
5099
5100	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5101		(page_index->page_len * CTL_PAGE_CURRENT));
5102
5103	if (lock != 0) {
5104		current_sp->lock_active = APS_LOCK_ACTIVE;
5105		softc->aps_locked_lun = lun->lun;
5106	} else {
5107		current_sp->lock_active = 0;
5108		softc->aps_locked_lun = 0;
5109	}
5110
5111
5112	/*
5113	 * If we're in HA mode, try to send the lock message to the other
5114	 * side.
5115	 */
5116	if (ctl_is_single == 0) {
5117		int isc_retval;
5118		union ctl_ha_msg lock_msg;
5119
5120		lock_msg.hdr.nexus = *nexus;
5121		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5122		if (lock != 0)
5123			lock_msg.aps.lock_flag = 1;
5124		else
5125			lock_msg.aps.lock_flag = 0;
5126		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5127					 sizeof(lock_msg), 0);
5128		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5129			printf("%s: APS (lock=%d) error returned from "
5130			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5131			mtx_unlock(&lun->lun_lock);
5132			mtx_unlock(&softc->ctl_lock);
5133			return (1);
5134		}
5135	}
5136
5137	mtx_unlock(&lun->lun_lock);
5138	mtx_unlock(&softc->ctl_lock);
5139
5140	return (0);
5141}
5142
5143void
5144ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5145{
5146	struct ctl_lun *lun;
5147	struct ctl_softc *softc;
5148	int i;
5149
5150	softc = control_softc;
5151
5152	lun = (struct ctl_lun *)be_lun->ctl_lun;
5153
5154	mtx_lock(&lun->lun_lock);
5155
5156	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5157		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5158
5159	mtx_unlock(&lun->lun_lock);
5160}
5161
5162/*
5163 * Backend "memory move is complete" callback for requests that never
5164 * make it down to say RAIDCore's configuration code.
5165 */
5166int
5167ctl_config_move_done(union ctl_io *io)
5168{
5169	int retval;
5170
5171	retval = CTL_RETVAL_COMPLETE;
5172
5173
5174	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5175	/*
5176	 * XXX KDM this shouldn't happen, but what if it does?
5177	 */
5178	if (io->io_hdr.io_type != CTL_IO_SCSI)
5179		panic("I/O type isn't CTL_IO_SCSI!");
5180
5181	if ((io->io_hdr.port_status == 0)
5182	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5183	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5184		io->io_hdr.status = CTL_SUCCESS;
5185	else 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		/*
5189		 * For hardware error sense keys, the sense key
5190		 * specific value is defined to be a retry count,
5191		 * but we use it to pass back an internal FETD
5192		 * error code.  XXX KDM  Hopefully the FETD is only
5193		 * using 16 bits for an error code, since that's
5194		 * all the space we have in the sks field.
5195		 */
5196		ctl_set_internal_failure(&io->scsiio,
5197					 /*sks_valid*/ 1,
5198					 /*retry_count*/
5199					 io->io_hdr.port_status);
5200		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5201			free(io->scsiio.kern_data_ptr, M_CTL);
5202		ctl_done(io);
5203		goto bailout;
5204	}
5205
5206	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5207	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5208	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5209		/*
5210		 * XXX KDM just assuming a single pointer here, and not a
5211		 * S/G list.  If we start using S/G lists for config data,
5212		 * we'll need to know how to clean them up here as well.
5213		 */
5214		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5215			free(io->scsiio.kern_data_ptr, M_CTL);
5216		/* Hopefully the user has already set the status... */
5217		ctl_done(io);
5218	} else {
5219		/*
5220		 * XXX KDM now we need to continue data movement.  Some
5221		 * options:
5222		 * - call ctl_scsiio() again?  We don't do this for data
5223		 *   writes, because for those at least we know ahead of
5224		 *   time where the write will go and how long it is.  For
5225		 *   config writes, though, that information is largely
5226		 *   contained within the write itself, thus we need to
5227		 *   parse out the data again.
5228		 *
5229		 * - Call some other function once the data is in?
5230		 */
5231
5232		/*
5233		 * XXX KDM call ctl_scsiio() again for now, and check flag
5234		 * bits to see whether we're allocated or not.
5235		 */
5236		retval = ctl_scsiio(&io->scsiio);
5237	}
5238bailout:
5239	return (retval);
5240}
5241
5242/*
5243 * This gets called by a backend driver when it is done with a
5244 * data_submit method.
5245 */
5246void
5247ctl_data_submit_done(union ctl_io *io)
5248{
5249	/*
5250	 * If the IO_CONT flag is set, we need to call the supplied
5251	 * function to continue processing the I/O, instead of completing
5252	 * the I/O just yet.
5253	 *
5254	 * If there is an error, though, we don't want to keep processing.
5255	 * Instead, just send status back to the initiator.
5256	 */
5257	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5258	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5259	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5260	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5261		io->scsiio.io_cont(io);
5262		return;
5263	}
5264	ctl_done(io);
5265}
5266
5267/*
5268 * This gets called by a backend driver when it is done with a
5269 * configuration write.
5270 */
5271void
5272ctl_config_write_done(union ctl_io *io)
5273{
5274	uint8_t *buf;
5275
5276	/*
5277	 * If the IO_CONT flag is set, we need to call the supplied
5278	 * function to continue processing the I/O, instead of completing
5279	 * the I/O just yet.
5280	 *
5281	 * If there is an error, though, we don't want to keep processing.
5282	 * Instead, just send status back to the initiator.
5283	 */
5284	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5285	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5286	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5287	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5288		io->scsiio.io_cont(io);
5289		return;
5290	}
5291	/*
5292	 * Since a configuration write can be done for commands that actually
5293	 * have data allocated, like write buffer, and commands that have
5294	 * no data, like start/stop unit, we need to check here.
5295	 */
5296	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5297		buf = io->scsiio.kern_data_ptr;
5298	else
5299		buf = NULL;
5300	ctl_done(io);
5301	if (buf)
5302		free(buf, M_CTL);
5303}
5304
5305/*
5306 * SCSI release command.
5307 */
5308int
5309ctl_scsi_release(struct ctl_scsiio *ctsio)
5310{
5311	int length, longid, thirdparty_id, resv_id;
5312	struct ctl_softc *ctl_softc;
5313	struct ctl_lun *lun;
5314	uint32_t residx;
5315
5316	length = 0;
5317	resv_id = 0;
5318
5319	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5320
5321	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5322	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5323	ctl_softc = control_softc;
5324
5325	switch (ctsio->cdb[0]) {
5326	case RELEASE_10: {
5327		struct scsi_release_10 *cdb;
5328
5329		cdb = (struct scsi_release_10 *)ctsio->cdb;
5330
5331		if (cdb->byte2 & SR10_LONGID)
5332			longid = 1;
5333		else
5334			thirdparty_id = cdb->thirdparty_id;
5335
5336		resv_id = cdb->resv_id;
5337		length = scsi_2btoul(cdb->length);
5338		break;
5339	}
5340	}
5341
5342
5343	/*
5344	 * XXX KDM right now, we only support LUN reservation.  We don't
5345	 * support 3rd party reservations, or extent reservations, which
5346	 * might actually need the parameter list.  If we've gotten this
5347	 * far, we've got a LUN reservation.  Anything else got kicked out
5348	 * above.  So, according to SPC, ignore the length.
5349	 */
5350	length = 0;
5351
5352	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5353	 && (length > 0)) {
5354		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5355		ctsio->kern_data_len = length;
5356		ctsio->kern_total_len = length;
5357		ctsio->kern_data_resid = 0;
5358		ctsio->kern_rel_offset = 0;
5359		ctsio->kern_sg_entries = 0;
5360		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5361		ctsio->be_move_done = ctl_config_move_done;
5362		ctl_datamove((union ctl_io *)ctsio);
5363
5364		return (CTL_RETVAL_COMPLETE);
5365	}
5366
5367	if (length > 0)
5368		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5369
5370	mtx_lock(&lun->lun_lock);
5371
5372	/*
5373	 * According to SPC, it is not an error for an intiator to attempt
5374	 * to release a reservation on a LUN that isn't reserved, or that
5375	 * is reserved by another initiator.  The reservation can only be
5376	 * released, though, by the initiator who made it or by one of
5377	 * several reset type events.
5378	 */
5379	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5380			lun->flags &= ~CTL_LUN_RESERVED;
5381
5382	mtx_unlock(&lun->lun_lock);
5383
5384	ctsio->scsi_status = SCSI_STATUS_OK;
5385	ctsio->io_hdr.status = CTL_SUCCESS;
5386
5387	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5388		free(ctsio->kern_data_ptr, M_CTL);
5389		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5390	}
5391
5392	ctl_done((union ctl_io *)ctsio);
5393	return (CTL_RETVAL_COMPLETE);
5394}
5395
5396int
5397ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5398{
5399	int extent, thirdparty, longid;
5400	int resv_id, length;
5401	uint64_t thirdparty_id;
5402	struct ctl_softc *ctl_softc;
5403	struct ctl_lun *lun;
5404	uint32_t residx;
5405
5406	extent = 0;
5407	thirdparty = 0;
5408	longid = 0;
5409	resv_id = 0;
5410	length = 0;
5411	thirdparty_id = 0;
5412
5413	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5414
5415	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5416	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5417	ctl_softc = control_softc;
5418
5419	switch (ctsio->cdb[0]) {
5420	case RESERVE_10: {
5421		struct scsi_reserve_10 *cdb;
5422
5423		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5424
5425		if (cdb->byte2 & SR10_LONGID)
5426			longid = 1;
5427		else
5428			thirdparty_id = cdb->thirdparty_id;
5429
5430		resv_id = cdb->resv_id;
5431		length = scsi_2btoul(cdb->length);
5432		break;
5433	}
5434	}
5435
5436	/*
5437	 * XXX KDM right now, we only support LUN reservation.  We don't
5438	 * support 3rd party reservations, or extent reservations, which
5439	 * might actually need the parameter list.  If we've gotten this
5440	 * far, we've got a LUN reservation.  Anything else got kicked out
5441	 * above.  So, according to SPC, ignore the length.
5442	 */
5443	length = 0;
5444
5445	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5446	 && (length > 0)) {
5447		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5448		ctsio->kern_data_len = length;
5449		ctsio->kern_total_len = length;
5450		ctsio->kern_data_resid = 0;
5451		ctsio->kern_rel_offset = 0;
5452		ctsio->kern_sg_entries = 0;
5453		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5454		ctsio->be_move_done = ctl_config_move_done;
5455		ctl_datamove((union ctl_io *)ctsio);
5456
5457		return (CTL_RETVAL_COMPLETE);
5458	}
5459
5460	if (length > 0)
5461		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5462
5463	mtx_lock(&lun->lun_lock);
5464	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5465		ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5466		ctsio->io_hdr.status = CTL_SCSI_ERROR;
5467		goto bailout;
5468	}
5469
5470	lun->flags |= CTL_LUN_RESERVED;
5471	lun->res_idx = residx;
5472
5473	ctsio->scsi_status = SCSI_STATUS_OK;
5474	ctsio->io_hdr.status = CTL_SUCCESS;
5475
5476bailout:
5477	mtx_unlock(&lun->lun_lock);
5478
5479	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5480		free(ctsio->kern_data_ptr, M_CTL);
5481		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5482	}
5483
5484	ctl_done((union ctl_io *)ctsio);
5485	return (CTL_RETVAL_COMPLETE);
5486}
5487
5488int
5489ctl_start_stop(struct ctl_scsiio *ctsio)
5490{
5491	struct scsi_start_stop_unit *cdb;
5492	struct ctl_lun *lun;
5493	struct ctl_softc *ctl_softc;
5494	int retval;
5495
5496	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5497
5498	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5499	ctl_softc = control_softc;
5500	retval = 0;
5501
5502	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5503
5504	/*
5505	 * XXX KDM
5506	 * We don't support the immediate bit on a stop unit.  In order to
5507	 * do that, we would need to code up a way to know that a stop is
5508	 * pending, and hold off any new commands until it completes, one
5509	 * way or another.  Then we could accept or reject those commands
5510	 * depending on its status.  We would almost need to do the reverse
5511	 * of what we do below for an immediate start -- return the copy of
5512	 * the ctl_io to the FETD with status to send to the host (and to
5513	 * free the copy!) and then free the original I/O once the stop
5514	 * actually completes.  That way, the OOA queue mechanism can work
5515	 * to block commands that shouldn't proceed.  Another alternative
5516	 * would be to put the copy in the queue in place of the original,
5517	 * and return the original back to the caller.  That could be
5518	 * slightly safer..
5519	 */
5520	if ((cdb->byte2 & SSS_IMMED)
5521	 && ((cdb->how & SSS_START) == 0)) {
5522		ctl_set_invalid_field(ctsio,
5523				      /*sks_valid*/ 1,
5524				      /*command*/ 1,
5525				      /*field*/ 1,
5526				      /*bit_valid*/ 1,
5527				      /*bit*/ 0);
5528		ctl_done((union ctl_io *)ctsio);
5529		return (CTL_RETVAL_COMPLETE);
5530	}
5531
5532	if ((lun->flags & CTL_LUN_PR_RESERVED)
5533	 && ((cdb->how & SSS_START)==0)) {
5534		uint32_t residx;
5535
5536		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5537		if (!lun->per_res[residx].registered
5538		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5539
5540			ctl_set_reservation_conflict(ctsio);
5541			ctl_done((union ctl_io *)ctsio);
5542			return (CTL_RETVAL_COMPLETE);
5543		}
5544	}
5545
5546	/*
5547	 * If there is no backend on this device, we can't start or stop
5548	 * it.  In theory we shouldn't get any start/stop commands in the
5549	 * first place at this level if the LUN doesn't have a backend.
5550	 * That should get stopped by the command decode code.
5551	 */
5552	if (lun->backend == NULL) {
5553		ctl_set_invalid_opcode(ctsio);
5554		ctl_done((union ctl_io *)ctsio);
5555		return (CTL_RETVAL_COMPLETE);
5556	}
5557
5558	/*
5559	 * XXX KDM Copan-specific offline behavior.
5560	 * Figure out a reasonable way to port this?
5561	 */
5562#ifdef NEEDTOPORT
5563	mtx_lock(&lun->lun_lock);
5564
5565	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5566	 && (lun->flags & CTL_LUN_OFFLINE)) {
5567		/*
5568		 * If the LUN is offline, and the on/offline bit isn't set,
5569		 * reject the start or stop.  Otherwise, let it through.
5570		 */
5571		mtx_unlock(&lun->lun_lock);
5572		ctl_set_lun_not_ready(ctsio);
5573		ctl_done((union ctl_io *)ctsio);
5574	} else {
5575		mtx_unlock(&lun->lun_lock);
5576#endif /* NEEDTOPORT */
5577		/*
5578		 * This could be a start or a stop when we're online,
5579		 * or a stop/offline or start/online.  A start or stop when
5580		 * we're offline is covered in the case above.
5581		 */
5582		/*
5583		 * In the non-immediate case, we send the request to
5584		 * the backend and return status to the user when
5585		 * it is done.
5586		 *
5587		 * In the immediate case, we allocate a new ctl_io
5588		 * to hold a copy of the request, and send that to
5589		 * the backend.  We then set good status on the
5590		 * user's request and return it immediately.
5591		 */
5592		if (cdb->byte2 & SSS_IMMED) {
5593			union ctl_io *new_io;
5594
5595			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5596			if (new_io == NULL) {
5597				ctl_set_busy(ctsio);
5598				ctl_done((union ctl_io *)ctsio);
5599			} else {
5600				ctl_copy_io((union ctl_io *)ctsio,
5601					    new_io);
5602				retval = lun->backend->config_write(new_io);
5603				ctl_set_success(ctsio);
5604				ctl_done((union ctl_io *)ctsio);
5605			}
5606		} else {
5607			retval = lun->backend->config_write(
5608				(union ctl_io *)ctsio);
5609		}
5610#ifdef NEEDTOPORT
5611	}
5612#endif
5613	return (retval);
5614}
5615
5616/*
5617 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5618 * we don't really do anything with the LBA and length fields if the user
5619 * passes them in.  Instead we'll just flush out the cache for the entire
5620 * LUN.
5621 */
5622int
5623ctl_sync_cache(struct ctl_scsiio *ctsio)
5624{
5625	struct ctl_lun *lun;
5626	struct ctl_softc *ctl_softc;
5627	uint64_t starting_lba;
5628	uint32_t block_count;
5629	int retval;
5630
5631	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5632
5633	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5634	ctl_softc = control_softc;
5635	retval = 0;
5636
5637	switch (ctsio->cdb[0]) {
5638	case SYNCHRONIZE_CACHE: {
5639		struct scsi_sync_cache *cdb;
5640		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5641
5642		starting_lba = scsi_4btoul(cdb->begin_lba);
5643		block_count = scsi_2btoul(cdb->lb_count);
5644		break;
5645	}
5646	case SYNCHRONIZE_CACHE_16: {
5647		struct scsi_sync_cache_16 *cdb;
5648		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5649
5650		starting_lba = scsi_8btou64(cdb->begin_lba);
5651		block_count = scsi_4btoul(cdb->lb_count);
5652		break;
5653	}
5654	default:
5655		ctl_set_invalid_opcode(ctsio);
5656		ctl_done((union ctl_io *)ctsio);
5657		goto bailout;
5658		break; /* NOTREACHED */
5659	}
5660
5661	/*
5662	 * We check the LBA and length, but don't do anything with them.
5663	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5664	 * get flushed.  This check will just help satisfy anyone who wants
5665	 * to see an error for an out of range LBA.
5666	 */
5667	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5668		ctl_set_lba_out_of_range(ctsio);
5669		ctl_done((union ctl_io *)ctsio);
5670		goto bailout;
5671	}
5672
5673	/*
5674	 * If this LUN has no backend, we can't flush the cache anyway.
5675	 */
5676	if (lun->backend == NULL) {
5677		ctl_set_invalid_opcode(ctsio);
5678		ctl_done((union ctl_io *)ctsio);
5679		goto bailout;
5680	}
5681
5682	/*
5683	 * Check to see whether we're configured to send the SYNCHRONIZE
5684	 * CACHE command directly to the back end.
5685	 */
5686	mtx_lock(&lun->lun_lock);
5687	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5688	 && (++(lun->sync_count) >= lun->sync_interval)) {
5689		lun->sync_count = 0;
5690		mtx_unlock(&lun->lun_lock);
5691		retval = lun->backend->config_write((union ctl_io *)ctsio);
5692	} else {
5693		mtx_unlock(&lun->lun_lock);
5694		ctl_set_success(ctsio);
5695		ctl_done((union ctl_io *)ctsio);
5696	}
5697
5698bailout:
5699
5700	return (retval);
5701}
5702
5703int
5704ctl_format(struct ctl_scsiio *ctsio)
5705{
5706	struct scsi_format *cdb;
5707	struct ctl_lun *lun;
5708	struct ctl_softc *ctl_softc;
5709	int length, defect_list_len;
5710
5711	CTL_DEBUG_PRINT(("ctl_format\n"));
5712
5713	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5714	ctl_softc = control_softc;
5715
5716	cdb = (struct scsi_format *)ctsio->cdb;
5717
5718	length = 0;
5719	if (cdb->byte2 & SF_FMTDATA) {
5720		if (cdb->byte2 & SF_LONGLIST)
5721			length = sizeof(struct scsi_format_header_long);
5722		else
5723			length = sizeof(struct scsi_format_header_short);
5724	}
5725
5726	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5727	 && (length > 0)) {
5728		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5729		ctsio->kern_data_len = length;
5730		ctsio->kern_total_len = length;
5731		ctsio->kern_data_resid = 0;
5732		ctsio->kern_rel_offset = 0;
5733		ctsio->kern_sg_entries = 0;
5734		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5735		ctsio->be_move_done = ctl_config_move_done;
5736		ctl_datamove((union ctl_io *)ctsio);
5737
5738		return (CTL_RETVAL_COMPLETE);
5739	}
5740
5741	defect_list_len = 0;
5742
5743	if (cdb->byte2 & SF_FMTDATA) {
5744		if (cdb->byte2 & SF_LONGLIST) {
5745			struct scsi_format_header_long *header;
5746
5747			header = (struct scsi_format_header_long *)
5748				ctsio->kern_data_ptr;
5749
5750			defect_list_len = scsi_4btoul(header->defect_list_len);
5751			if (defect_list_len != 0) {
5752				ctl_set_invalid_field(ctsio,
5753						      /*sks_valid*/ 1,
5754						      /*command*/ 0,
5755						      /*field*/ 2,
5756						      /*bit_valid*/ 0,
5757						      /*bit*/ 0);
5758				goto bailout;
5759			}
5760		} else {
5761			struct scsi_format_header_short *header;
5762
5763			header = (struct scsi_format_header_short *)
5764				ctsio->kern_data_ptr;
5765
5766			defect_list_len = scsi_2btoul(header->defect_list_len);
5767			if (defect_list_len != 0) {
5768				ctl_set_invalid_field(ctsio,
5769						      /*sks_valid*/ 1,
5770						      /*command*/ 0,
5771						      /*field*/ 2,
5772						      /*bit_valid*/ 0,
5773						      /*bit*/ 0);
5774				goto bailout;
5775			}
5776		}
5777	}
5778
5779	/*
5780	 * The format command will clear out the "Medium format corrupted"
5781	 * status if set by the configuration code.  That status is really
5782	 * just a way to notify the host that we have lost the media, and
5783	 * get them to issue a command that will basically make them think
5784	 * they're blowing away the media.
5785	 */
5786	mtx_lock(&lun->lun_lock);
5787	lun->flags &= ~CTL_LUN_INOPERABLE;
5788	mtx_unlock(&lun->lun_lock);
5789
5790	ctsio->scsi_status = SCSI_STATUS_OK;
5791	ctsio->io_hdr.status = CTL_SUCCESS;
5792bailout:
5793
5794	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5795		free(ctsio->kern_data_ptr, M_CTL);
5796		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5797	}
5798
5799	ctl_done((union ctl_io *)ctsio);
5800	return (CTL_RETVAL_COMPLETE);
5801}
5802
5803int
5804ctl_read_buffer(struct ctl_scsiio *ctsio)
5805{
5806	struct scsi_read_buffer *cdb;
5807	struct ctl_lun *lun;
5808	int buffer_offset, len;
5809	static uint8_t descr[4];
5810	static uint8_t echo_descr[4] = { 0 };
5811
5812	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5813
5814	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5815	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5816
5817	if (lun->flags & CTL_LUN_PR_RESERVED) {
5818		uint32_t residx;
5819
5820		/*
5821		 * XXX KDM need a lock here.
5822		 */
5823		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5824		if ((lun->res_type == SPR_TYPE_EX_AC
5825		  && residx != lun->pr_res_idx)
5826		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5827		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5828		  && !lun->per_res[residx].registered)) {
5829			ctl_set_reservation_conflict(ctsio);
5830			ctl_done((union ctl_io *)ctsio);
5831			return (CTL_RETVAL_COMPLETE);
5832	        }
5833	}
5834
5835	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5836	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5837	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5838		ctl_set_invalid_field(ctsio,
5839				      /*sks_valid*/ 1,
5840				      /*command*/ 1,
5841				      /*field*/ 1,
5842				      /*bit_valid*/ 1,
5843				      /*bit*/ 4);
5844		ctl_done((union ctl_io *)ctsio);
5845		return (CTL_RETVAL_COMPLETE);
5846	}
5847
5848	len = scsi_3btoul(cdb->length);
5849	buffer_offset = scsi_3btoul(cdb->offset);
5850
5851	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5852		ctl_set_invalid_field(ctsio,
5853				      /*sks_valid*/ 1,
5854				      /*command*/ 1,
5855				      /*field*/ 6,
5856				      /*bit_valid*/ 0,
5857				      /*bit*/ 0);
5858		ctl_done((union ctl_io *)ctsio);
5859		return (CTL_RETVAL_COMPLETE);
5860	}
5861
5862	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5863		descr[0] = 0;
5864		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5865		ctsio->kern_data_ptr = descr;
5866		len = min(len, sizeof(descr));
5867	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5868		ctsio->kern_data_ptr = echo_descr;
5869		len = min(len, sizeof(echo_descr));
5870	} else
5871		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5872	ctsio->kern_data_len = len;
5873	ctsio->kern_total_len = len;
5874	ctsio->kern_data_resid = 0;
5875	ctsio->kern_rel_offset = 0;
5876	ctsio->kern_sg_entries = 0;
5877	ctsio->be_move_done = ctl_config_move_done;
5878	ctl_datamove((union ctl_io *)ctsio);
5879
5880	return (CTL_RETVAL_COMPLETE);
5881}
5882
5883int
5884ctl_write_buffer(struct ctl_scsiio *ctsio)
5885{
5886	struct scsi_write_buffer *cdb;
5887	struct ctl_lun *lun;
5888	int buffer_offset, len;
5889
5890	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5891
5892	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5893	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5894
5895	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5896		ctl_set_invalid_field(ctsio,
5897				      /*sks_valid*/ 1,
5898				      /*command*/ 1,
5899				      /*field*/ 1,
5900				      /*bit_valid*/ 1,
5901				      /*bit*/ 4);
5902		ctl_done((union ctl_io *)ctsio);
5903		return (CTL_RETVAL_COMPLETE);
5904	}
5905
5906	len = scsi_3btoul(cdb->length);
5907	buffer_offset = scsi_3btoul(cdb->offset);
5908
5909	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5910		ctl_set_invalid_field(ctsio,
5911				      /*sks_valid*/ 1,
5912				      /*command*/ 1,
5913				      /*field*/ 6,
5914				      /*bit_valid*/ 0,
5915				      /*bit*/ 0);
5916		ctl_done((union ctl_io *)ctsio);
5917		return (CTL_RETVAL_COMPLETE);
5918	}
5919
5920	/*
5921	 * If we've got a kernel request that hasn't been malloced yet,
5922	 * malloc it and tell the caller the data buffer is here.
5923	 */
5924	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5925		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5926		ctsio->kern_data_len = len;
5927		ctsio->kern_total_len = len;
5928		ctsio->kern_data_resid = 0;
5929		ctsio->kern_rel_offset = 0;
5930		ctsio->kern_sg_entries = 0;
5931		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5932		ctsio->be_move_done = ctl_config_move_done;
5933		ctl_datamove((union ctl_io *)ctsio);
5934
5935		return (CTL_RETVAL_COMPLETE);
5936	}
5937
5938	ctl_done((union ctl_io *)ctsio);
5939
5940	return (CTL_RETVAL_COMPLETE);
5941}
5942
5943int
5944ctl_write_same(struct ctl_scsiio *ctsio)
5945{
5946	struct ctl_lun *lun;
5947	struct ctl_lba_len_flags *lbalen;
5948	uint64_t lba;
5949	uint32_t num_blocks;
5950	int len, retval;
5951	uint8_t byte2;
5952
5953	retval = CTL_RETVAL_COMPLETE;
5954
5955	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5956
5957	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5958
5959	switch (ctsio->cdb[0]) {
5960	case WRITE_SAME_10: {
5961		struct scsi_write_same_10 *cdb;
5962
5963		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5964
5965		lba = scsi_4btoul(cdb->addr);
5966		num_blocks = scsi_2btoul(cdb->length);
5967		byte2 = cdb->byte2;
5968		break;
5969	}
5970	case WRITE_SAME_16: {
5971		struct scsi_write_same_16 *cdb;
5972
5973		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5974
5975		lba = scsi_8btou64(cdb->addr);
5976		num_blocks = scsi_4btoul(cdb->length);
5977		byte2 = cdb->byte2;
5978		break;
5979	}
5980	default:
5981		/*
5982		 * We got a command we don't support.  This shouldn't
5983		 * happen, commands should be filtered out above us.
5984		 */
5985		ctl_set_invalid_opcode(ctsio);
5986		ctl_done((union ctl_io *)ctsio);
5987
5988		return (CTL_RETVAL_COMPLETE);
5989		break; /* NOTREACHED */
5990	}
5991
5992	/* NDOB and ANCHOR flags can be used only together with UNMAP */
5993	if ((byte2 & SWS_UNMAP) == 0 &&
5994	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5995		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5996		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5997		ctl_done((union ctl_io *)ctsio);
5998		return (CTL_RETVAL_COMPLETE);
5999	}
6000
6001	/*
6002	 * The first check is to make sure we're in bounds, the second
6003	 * check is to catch wrap-around problems.  If the lba + num blocks
6004	 * is less than the lba, then we've wrapped around and the block
6005	 * range is invalid anyway.
6006	 */
6007	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6008	 || ((lba + num_blocks) < lba)) {
6009		ctl_set_lba_out_of_range(ctsio);
6010		ctl_done((union ctl_io *)ctsio);
6011		return (CTL_RETVAL_COMPLETE);
6012	}
6013
6014	/* Zero number of blocks means "to the last logical block" */
6015	if (num_blocks == 0) {
6016		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
6017			ctl_set_invalid_field(ctsio,
6018					      /*sks_valid*/ 0,
6019					      /*command*/ 1,
6020					      /*field*/ 0,
6021					      /*bit_valid*/ 0,
6022					      /*bit*/ 0);
6023			ctl_done((union ctl_io *)ctsio);
6024			return (CTL_RETVAL_COMPLETE);
6025		}
6026		num_blocks = (lun->be_lun->maxlba + 1) - lba;
6027	}
6028
6029	len = lun->be_lun->blocksize;
6030
6031	/*
6032	 * If we've got a kernel request that hasn't been malloced yet,
6033	 * malloc it and tell the caller the data buffer is here.
6034	 */
6035	if ((byte2 & SWS_NDOB) == 0 &&
6036	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6037		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6038		ctsio->kern_data_len = len;
6039		ctsio->kern_total_len = len;
6040		ctsio->kern_data_resid = 0;
6041		ctsio->kern_rel_offset = 0;
6042		ctsio->kern_sg_entries = 0;
6043		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6044		ctsio->be_move_done = ctl_config_move_done;
6045		ctl_datamove((union ctl_io *)ctsio);
6046
6047		return (CTL_RETVAL_COMPLETE);
6048	}
6049
6050	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6051	lbalen->lba = lba;
6052	lbalen->len = num_blocks;
6053	lbalen->flags = byte2;
6054	retval = lun->backend->config_write((union ctl_io *)ctsio);
6055
6056	return (retval);
6057}
6058
6059int
6060ctl_unmap(struct ctl_scsiio *ctsio)
6061{
6062	struct ctl_lun *lun;
6063	struct scsi_unmap *cdb;
6064	struct ctl_ptr_len_flags *ptrlen;
6065	struct scsi_unmap_header *hdr;
6066	struct scsi_unmap_desc *buf, *end, *endnz, *range;
6067	uint64_t lba;
6068	uint32_t num_blocks;
6069	int len, retval;
6070	uint8_t byte2;
6071
6072	retval = CTL_RETVAL_COMPLETE;
6073
6074	CTL_DEBUG_PRINT(("ctl_unmap\n"));
6075
6076	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6077	cdb = (struct scsi_unmap *)ctsio->cdb;
6078
6079	len = scsi_2btoul(cdb->length);
6080	byte2 = cdb->byte2;
6081
6082	/*
6083	 * If we've got a kernel request that hasn't been malloced yet,
6084	 * malloc it and tell the caller the data buffer is here.
6085	 */
6086	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6087		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6088		ctsio->kern_data_len = len;
6089		ctsio->kern_total_len = len;
6090		ctsio->kern_data_resid = 0;
6091		ctsio->kern_rel_offset = 0;
6092		ctsio->kern_sg_entries = 0;
6093		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6094		ctsio->be_move_done = ctl_config_move_done;
6095		ctl_datamove((union ctl_io *)ctsio);
6096
6097		return (CTL_RETVAL_COMPLETE);
6098	}
6099
6100	len = ctsio->kern_total_len - ctsio->kern_data_resid;
6101	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6102	if (len < sizeof (*hdr) ||
6103	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6104	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6105	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6106		ctl_set_invalid_field(ctsio,
6107				      /*sks_valid*/ 0,
6108				      /*command*/ 0,
6109				      /*field*/ 0,
6110				      /*bit_valid*/ 0,
6111				      /*bit*/ 0);
6112		ctl_done((union ctl_io *)ctsio);
6113		return (CTL_RETVAL_COMPLETE);
6114	}
6115	len = scsi_2btoul(hdr->desc_length);
6116	buf = (struct scsi_unmap_desc *)(hdr + 1);
6117	end = buf + len / sizeof(*buf);
6118
6119	endnz = buf;
6120	for (range = buf; range < end; range++) {
6121		lba = scsi_8btou64(range->lba);
6122		num_blocks = scsi_4btoul(range->length);
6123		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6124		 || ((lba + num_blocks) < lba)) {
6125			ctl_set_lba_out_of_range(ctsio);
6126			ctl_done((union ctl_io *)ctsio);
6127			return (CTL_RETVAL_COMPLETE);
6128		}
6129		if (num_blocks != 0)
6130			endnz = range + 1;
6131	}
6132
6133	/*
6134	 * Block backend can not handle zero last range.
6135	 * Filter it out and return if there is nothing left.
6136	 */
6137	len = (uint8_t *)endnz - (uint8_t *)buf;
6138	if (len == 0) {
6139		ctl_set_success(ctsio);
6140		ctl_done((union ctl_io *)ctsio);
6141		return (CTL_RETVAL_COMPLETE);
6142	}
6143
6144	mtx_lock(&lun->lun_lock);
6145	ptrlen = (struct ctl_ptr_len_flags *)
6146	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6147	ptrlen->ptr = (void *)buf;
6148	ptrlen->len = len;
6149	ptrlen->flags = byte2;
6150	ctl_check_blocked(lun);
6151	mtx_unlock(&lun->lun_lock);
6152
6153	retval = lun->backend->config_write((union ctl_io *)ctsio);
6154	return (retval);
6155}
6156
6157/*
6158 * Note that this function currently doesn't actually do anything inside
6159 * CTL to enforce things if the DQue bit is turned on.
6160 *
6161 * Also note that this function can't be used in the default case, because
6162 * the DQue bit isn't set in the changeable mask for the control mode page
6163 * anyway.  This is just here as an example for how to implement a page
6164 * handler, and a placeholder in case we want to allow the user to turn
6165 * tagged queueing on and off.
6166 *
6167 * The D_SENSE bit handling is functional, however, and will turn
6168 * descriptor sense on and off for a given LUN.
6169 */
6170int
6171ctl_control_page_handler(struct ctl_scsiio *ctsio,
6172			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6173{
6174	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6175	struct ctl_lun *lun;
6176	struct ctl_softc *softc;
6177	int set_ua;
6178	uint32_t initidx;
6179
6180	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6181	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6182	set_ua = 0;
6183
6184	user_cp = (struct scsi_control_page *)page_ptr;
6185	current_cp = (struct scsi_control_page *)
6186		(page_index->page_data + (page_index->page_len *
6187		CTL_PAGE_CURRENT));
6188	saved_cp = (struct scsi_control_page *)
6189		(page_index->page_data + (page_index->page_len *
6190		CTL_PAGE_SAVED));
6191
6192	softc = control_softc;
6193
6194	mtx_lock(&lun->lun_lock);
6195	if (((current_cp->rlec & SCP_DSENSE) == 0)
6196	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6197		/*
6198		 * Descriptor sense is currently turned off and the user
6199		 * wants to turn it on.
6200		 */
6201		current_cp->rlec |= SCP_DSENSE;
6202		saved_cp->rlec |= SCP_DSENSE;
6203		lun->flags |= CTL_LUN_SENSE_DESC;
6204		set_ua = 1;
6205	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6206		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6207		/*
6208		 * Descriptor sense is currently turned on, and the user
6209		 * wants to turn it off.
6210		 */
6211		current_cp->rlec &= ~SCP_DSENSE;
6212		saved_cp->rlec &= ~SCP_DSENSE;
6213		lun->flags &= ~CTL_LUN_SENSE_DESC;
6214		set_ua = 1;
6215	}
6216	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6217	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6218		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6219		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6220		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6221		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6222		set_ua = 1;
6223	}
6224	if (set_ua != 0) {
6225		int i;
6226		/*
6227		 * Let other initiators know that the mode
6228		 * parameters for this LUN have changed.
6229		 */
6230		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6231			if (i == initidx)
6232				continue;
6233
6234			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6235		}
6236	}
6237	mtx_unlock(&lun->lun_lock);
6238
6239	return (0);
6240}
6241
6242int
6243ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6244		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6245{
6246	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6247	struct ctl_lun *lun;
6248	int set_ua;
6249	uint32_t initidx;
6250
6251	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6252	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6253	set_ua = 0;
6254
6255	user_cp = (struct scsi_caching_page *)page_ptr;
6256	current_cp = (struct scsi_caching_page *)
6257		(page_index->page_data + (page_index->page_len *
6258		CTL_PAGE_CURRENT));
6259	saved_cp = (struct scsi_caching_page *)
6260		(page_index->page_data + (page_index->page_len *
6261		CTL_PAGE_SAVED));
6262
6263	mtx_lock(&lun->lun_lock);
6264	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6265	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6266		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6267		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6268		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6269		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6270		set_ua = 1;
6271	}
6272	if (set_ua != 0) {
6273		int i;
6274		/*
6275		 * Let other initiators know that the mode
6276		 * parameters for this LUN have changed.
6277		 */
6278		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6279			if (i == initidx)
6280				continue;
6281
6282			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6283		}
6284	}
6285	mtx_unlock(&lun->lun_lock);
6286
6287	return (0);
6288}
6289
6290int
6291ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6292		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6293{
6294	return (0);
6295}
6296
6297int
6298ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6299			   struct ctl_page_index *page_index, int pc)
6300{
6301	struct copan_power_subpage *page;
6302
6303	page = (struct copan_power_subpage *)page_index->page_data +
6304		(page_index->page_len * pc);
6305
6306	switch (pc) {
6307	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6308		/*
6309		 * We don't update the changable bits for this page.
6310		 */
6311		break;
6312	case SMS_PAGE_CTRL_CURRENT >> 6:
6313	case SMS_PAGE_CTRL_DEFAULT >> 6:
6314	case SMS_PAGE_CTRL_SAVED >> 6:
6315#ifdef NEEDTOPORT
6316		ctl_update_power_subpage(page);
6317#endif
6318		break;
6319	default:
6320#ifdef NEEDTOPORT
6321		EPRINT(0, "Invalid PC %d!!", pc);
6322#endif
6323		break;
6324	}
6325	return (0);
6326}
6327
6328
6329int
6330ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6331		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6332{
6333	struct copan_aps_subpage *user_sp;
6334	struct copan_aps_subpage *current_sp;
6335	union ctl_modepage_info *modepage_info;
6336	struct ctl_softc *softc;
6337	struct ctl_lun *lun;
6338	int retval;
6339
6340	retval = CTL_RETVAL_COMPLETE;
6341	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6342		     (page_index->page_len * CTL_PAGE_CURRENT));
6343	softc = control_softc;
6344	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6345
6346	user_sp = (struct copan_aps_subpage *)page_ptr;
6347
6348	modepage_info = (union ctl_modepage_info *)
6349		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6350
6351	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6352	modepage_info->header.subpage = page_index->subpage;
6353	modepage_info->aps.lock_active = user_sp->lock_active;
6354
6355	mtx_lock(&softc->ctl_lock);
6356
6357	/*
6358	 * If there is a request to lock the LUN and another LUN is locked
6359	 * this is an error. If the requested LUN is already locked ignore
6360	 * the request. If no LUN is locked attempt to lock it.
6361	 * if there is a request to unlock the LUN and the LUN is currently
6362	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6363	 * if another LUN is locked or no LUN is locked.
6364	 */
6365	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6366		if (softc->aps_locked_lun == lun->lun) {
6367			/*
6368			 * This LUN is already locked, so we're done.
6369			 */
6370			retval = CTL_RETVAL_COMPLETE;
6371		} else if (softc->aps_locked_lun == 0) {
6372			/*
6373			 * No one has the lock, pass the request to the
6374			 * backend.
6375			 */
6376			retval = lun->backend->config_write(
6377				(union ctl_io *)ctsio);
6378		} else {
6379			/*
6380			 * Someone else has the lock, throw out the request.
6381			 */
6382			ctl_set_already_locked(ctsio);
6383			free(ctsio->kern_data_ptr, M_CTL);
6384			ctl_done((union ctl_io *)ctsio);
6385
6386			/*
6387			 * Set the return value so that ctl_do_mode_select()
6388			 * won't try to complete the command.  We already
6389			 * completed it here.
6390			 */
6391			retval = CTL_RETVAL_ERROR;
6392		}
6393	} else if (softc->aps_locked_lun == lun->lun) {
6394		/*
6395		 * This LUN is locked, so pass the unlock request to the
6396		 * backend.
6397		 */
6398		retval = lun->backend->config_write((union ctl_io *)ctsio);
6399	}
6400	mtx_unlock(&softc->ctl_lock);
6401
6402	return (retval);
6403}
6404
6405int
6406ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6407				struct ctl_page_index *page_index,
6408				uint8_t *page_ptr)
6409{
6410	uint8_t *c;
6411	int i;
6412
6413	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6414	ctl_time_io_secs =
6415		(c[0] << 8) |
6416		(c[1] << 0) |
6417		0;
6418	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6419	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6420	printf("page data:");
6421	for (i=0; i<8; i++)
6422		printf(" %.2x",page_ptr[i]);
6423	printf("\n");
6424	return (0);
6425}
6426
6427int
6428ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6429			       struct ctl_page_index *page_index,
6430			       int pc)
6431{
6432	struct copan_debugconf_subpage *page;
6433
6434	page = (struct copan_debugconf_subpage *)page_index->page_data +
6435		(page_index->page_len * pc);
6436
6437	switch (pc) {
6438	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6439	case SMS_PAGE_CTRL_DEFAULT >> 6:
6440	case SMS_PAGE_CTRL_SAVED >> 6:
6441		/*
6442		 * We don't update the changable or default bits for this page.
6443		 */
6444		break;
6445	case SMS_PAGE_CTRL_CURRENT >> 6:
6446		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6447		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6448		break;
6449	default:
6450#ifdef NEEDTOPORT
6451		EPRINT(0, "Invalid PC %d!!", pc);
6452#endif /* NEEDTOPORT */
6453		break;
6454	}
6455	return (0);
6456}
6457
6458
6459static int
6460ctl_do_mode_select(union ctl_io *io)
6461{
6462	struct scsi_mode_page_header *page_header;
6463	struct ctl_page_index *page_index;
6464	struct ctl_scsiio *ctsio;
6465	int control_dev, page_len;
6466	int page_len_offset, page_len_size;
6467	union ctl_modepage_info *modepage_info;
6468	struct ctl_lun *lun;
6469	int *len_left, *len_used;
6470	int retval, i;
6471
6472	ctsio = &io->scsiio;
6473	page_index = NULL;
6474	page_len = 0;
6475	retval = CTL_RETVAL_COMPLETE;
6476
6477	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6478
6479	if (lun->be_lun->lun_type != T_DIRECT)
6480		control_dev = 1;
6481	else
6482		control_dev = 0;
6483
6484	modepage_info = (union ctl_modepage_info *)
6485		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6486	len_left = &modepage_info->header.len_left;
6487	len_used = &modepage_info->header.len_used;
6488
6489do_next_page:
6490
6491	page_header = (struct scsi_mode_page_header *)
6492		(ctsio->kern_data_ptr + *len_used);
6493
6494	if (*len_left == 0) {
6495		free(ctsio->kern_data_ptr, M_CTL);
6496		ctl_set_success(ctsio);
6497		ctl_done((union ctl_io *)ctsio);
6498		return (CTL_RETVAL_COMPLETE);
6499	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6500
6501		free(ctsio->kern_data_ptr, M_CTL);
6502		ctl_set_param_len_error(ctsio);
6503		ctl_done((union ctl_io *)ctsio);
6504		return (CTL_RETVAL_COMPLETE);
6505
6506	} else if ((page_header->page_code & SMPH_SPF)
6507		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6508
6509		free(ctsio->kern_data_ptr, M_CTL);
6510		ctl_set_param_len_error(ctsio);
6511		ctl_done((union ctl_io *)ctsio);
6512		return (CTL_RETVAL_COMPLETE);
6513	}
6514
6515
6516	/*
6517	 * XXX KDM should we do something with the block descriptor?
6518	 */
6519	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6520
6521		if ((control_dev != 0)
6522		 && (lun->mode_pages.index[i].page_flags &
6523		     CTL_PAGE_FLAG_DISK_ONLY))
6524			continue;
6525
6526		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6527		    (page_header->page_code & SMPH_PC_MASK))
6528			continue;
6529
6530		/*
6531		 * If neither page has a subpage code, then we've got a
6532		 * match.
6533		 */
6534		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6535		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6536			page_index = &lun->mode_pages.index[i];
6537			page_len = page_header->page_length;
6538			break;
6539		}
6540
6541		/*
6542		 * If both pages have subpages, then the subpage numbers
6543		 * have to match.
6544		 */
6545		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6546		  && (page_header->page_code & SMPH_SPF)) {
6547			struct scsi_mode_page_header_sp *sph;
6548
6549			sph = (struct scsi_mode_page_header_sp *)page_header;
6550
6551			if (lun->mode_pages.index[i].subpage ==
6552			    sph->subpage) {
6553				page_index = &lun->mode_pages.index[i];
6554				page_len = scsi_2btoul(sph->page_length);
6555				break;
6556			}
6557		}
6558	}
6559
6560	/*
6561	 * If we couldn't find the page, or if we don't have a mode select
6562	 * handler for it, send back an error to the user.
6563	 */
6564	if ((page_index == NULL)
6565	 || (page_index->select_handler == NULL)) {
6566		ctl_set_invalid_field(ctsio,
6567				      /*sks_valid*/ 1,
6568				      /*command*/ 0,
6569				      /*field*/ *len_used,
6570				      /*bit_valid*/ 0,
6571				      /*bit*/ 0);
6572		free(ctsio->kern_data_ptr, M_CTL);
6573		ctl_done((union ctl_io *)ctsio);
6574		return (CTL_RETVAL_COMPLETE);
6575	}
6576
6577	if (page_index->page_code & SMPH_SPF) {
6578		page_len_offset = 2;
6579		page_len_size = 2;
6580	} else {
6581		page_len_size = 1;
6582		page_len_offset = 1;
6583	}
6584
6585	/*
6586	 * If the length the initiator gives us isn't the one we specify in
6587	 * the mode page header, or if they didn't specify enough data in
6588	 * the CDB to avoid truncating this page, kick out the request.
6589	 */
6590	if ((page_len != (page_index->page_len - page_len_offset -
6591			  page_len_size))
6592	 || (*len_left < page_index->page_len)) {
6593
6594
6595		ctl_set_invalid_field(ctsio,
6596				      /*sks_valid*/ 1,
6597				      /*command*/ 0,
6598				      /*field*/ *len_used + page_len_offset,
6599				      /*bit_valid*/ 0,
6600				      /*bit*/ 0);
6601		free(ctsio->kern_data_ptr, M_CTL);
6602		ctl_done((union ctl_io *)ctsio);
6603		return (CTL_RETVAL_COMPLETE);
6604	}
6605
6606	/*
6607	 * Run through the mode page, checking to make sure that the bits
6608	 * the user changed are actually legal for him to change.
6609	 */
6610	for (i = 0; i < page_index->page_len; i++) {
6611		uint8_t *user_byte, *change_mask, *current_byte;
6612		int bad_bit;
6613		int j;
6614
6615		user_byte = (uint8_t *)page_header + i;
6616		change_mask = page_index->page_data +
6617			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6618		current_byte = page_index->page_data +
6619			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6620
6621		/*
6622		 * Check to see whether the user set any bits in this byte
6623		 * that he is not allowed to set.
6624		 */
6625		if ((*user_byte & ~(*change_mask)) ==
6626		    (*current_byte & ~(*change_mask)))
6627			continue;
6628
6629		/*
6630		 * Go through bit by bit to determine which one is illegal.
6631		 */
6632		bad_bit = 0;
6633		for (j = 7; j >= 0; j--) {
6634			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6635			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6636				bad_bit = i;
6637				break;
6638			}
6639		}
6640		ctl_set_invalid_field(ctsio,
6641				      /*sks_valid*/ 1,
6642				      /*command*/ 0,
6643				      /*field*/ *len_used + i,
6644				      /*bit_valid*/ 1,
6645				      /*bit*/ bad_bit);
6646		free(ctsio->kern_data_ptr, M_CTL);
6647		ctl_done((union ctl_io *)ctsio);
6648		return (CTL_RETVAL_COMPLETE);
6649	}
6650
6651	/*
6652	 * Decrement these before we call the page handler, since we may
6653	 * end up getting called back one way or another before the handler
6654	 * returns to this context.
6655	 */
6656	*len_left -= page_index->page_len;
6657	*len_used += page_index->page_len;
6658
6659	retval = page_index->select_handler(ctsio, page_index,
6660					    (uint8_t *)page_header);
6661
6662	/*
6663	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6664	 * wait until this queued command completes to finish processing
6665	 * the mode page.  If it returns anything other than
6666	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6667	 * already set the sense information, freed the data pointer, and
6668	 * completed the io for us.
6669	 */
6670	if (retval != CTL_RETVAL_COMPLETE)
6671		goto bailout_no_done;
6672
6673	/*
6674	 * If the initiator sent us more than one page, parse the next one.
6675	 */
6676	if (*len_left > 0)
6677		goto do_next_page;
6678
6679	ctl_set_success(ctsio);
6680	free(ctsio->kern_data_ptr, M_CTL);
6681	ctl_done((union ctl_io *)ctsio);
6682
6683bailout_no_done:
6684
6685	return (CTL_RETVAL_COMPLETE);
6686
6687}
6688
6689int
6690ctl_mode_select(struct ctl_scsiio *ctsio)
6691{
6692	int param_len, pf, sp;
6693	int header_size, bd_len;
6694	int len_left, len_used;
6695	struct ctl_page_index *page_index;
6696	struct ctl_lun *lun;
6697	int control_dev, page_len;
6698	union ctl_modepage_info *modepage_info;
6699	int retval;
6700
6701	pf = 0;
6702	sp = 0;
6703	page_len = 0;
6704	len_used = 0;
6705	len_left = 0;
6706	retval = 0;
6707	bd_len = 0;
6708	page_index = NULL;
6709
6710	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6711
6712	if (lun->be_lun->lun_type != T_DIRECT)
6713		control_dev = 1;
6714	else
6715		control_dev = 0;
6716
6717	switch (ctsio->cdb[0]) {
6718	case MODE_SELECT_6: {
6719		struct scsi_mode_select_6 *cdb;
6720
6721		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6722
6723		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6724		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6725
6726		param_len = cdb->length;
6727		header_size = sizeof(struct scsi_mode_header_6);
6728		break;
6729	}
6730	case MODE_SELECT_10: {
6731		struct scsi_mode_select_10 *cdb;
6732
6733		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6734
6735		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6736		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6737
6738		param_len = scsi_2btoul(cdb->length);
6739		header_size = sizeof(struct scsi_mode_header_10);
6740		break;
6741	}
6742	default:
6743		ctl_set_invalid_opcode(ctsio);
6744		ctl_done((union ctl_io *)ctsio);
6745		return (CTL_RETVAL_COMPLETE);
6746		break; /* NOTREACHED */
6747	}
6748
6749	/*
6750	 * From SPC-3:
6751	 * "A parameter list length of zero indicates that the Data-Out Buffer
6752	 * shall be empty. This condition shall not be considered as an error."
6753	 */
6754	if (param_len == 0) {
6755		ctl_set_success(ctsio);
6756		ctl_done((union ctl_io *)ctsio);
6757		return (CTL_RETVAL_COMPLETE);
6758	}
6759
6760	/*
6761	 * Since we'll hit this the first time through, prior to
6762	 * allocation, we don't need to free a data buffer here.
6763	 */
6764	if (param_len < header_size) {
6765		ctl_set_param_len_error(ctsio);
6766		ctl_done((union ctl_io *)ctsio);
6767		return (CTL_RETVAL_COMPLETE);
6768	}
6769
6770	/*
6771	 * Allocate the data buffer and grab the user's data.  In theory,
6772	 * we shouldn't have to sanity check the parameter list length here
6773	 * because the maximum size is 64K.  We should be able to malloc
6774	 * that much without too many problems.
6775	 */
6776	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6777		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6778		ctsio->kern_data_len = param_len;
6779		ctsio->kern_total_len = param_len;
6780		ctsio->kern_data_resid = 0;
6781		ctsio->kern_rel_offset = 0;
6782		ctsio->kern_sg_entries = 0;
6783		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6784		ctsio->be_move_done = ctl_config_move_done;
6785		ctl_datamove((union ctl_io *)ctsio);
6786
6787		return (CTL_RETVAL_COMPLETE);
6788	}
6789
6790	switch (ctsio->cdb[0]) {
6791	case MODE_SELECT_6: {
6792		struct scsi_mode_header_6 *mh6;
6793
6794		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6795		bd_len = mh6->blk_desc_len;
6796		break;
6797	}
6798	case MODE_SELECT_10: {
6799		struct scsi_mode_header_10 *mh10;
6800
6801		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6802		bd_len = scsi_2btoul(mh10->blk_desc_len);
6803		break;
6804	}
6805	default:
6806		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6807		break;
6808	}
6809
6810	if (param_len < (header_size + bd_len)) {
6811		free(ctsio->kern_data_ptr, M_CTL);
6812		ctl_set_param_len_error(ctsio);
6813		ctl_done((union ctl_io *)ctsio);
6814		return (CTL_RETVAL_COMPLETE);
6815	}
6816
6817	/*
6818	 * Set the IO_CONT flag, so that if this I/O gets passed to
6819	 * ctl_config_write_done(), it'll get passed back to
6820	 * ctl_do_mode_select() for further processing, or completion if
6821	 * we're all done.
6822	 */
6823	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6824	ctsio->io_cont = ctl_do_mode_select;
6825
6826	modepage_info = (union ctl_modepage_info *)
6827		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6828
6829	memset(modepage_info, 0, sizeof(*modepage_info));
6830
6831	len_left = param_len - header_size - bd_len;
6832	len_used = header_size + bd_len;
6833
6834	modepage_info->header.len_left = len_left;
6835	modepage_info->header.len_used = len_used;
6836
6837	return (ctl_do_mode_select((union ctl_io *)ctsio));
6838}
6839
6840int
6841ctl_mode_sense(struct ctl_scsiio *ctsio)
6842{
6843	struct ctl_lun *lun;
6844	int pc, page_code, dbd, llba, subpage;
6845	int alloc_len, page_len, header_len, total_len;
6846	struct scsi_mode_block_descr *block_desc;
6847	struct ctl_page_index *page_index;
6848	int control_dev;
6849
6850	dbd = 0;
6851	llba = 0;
6852	block_desc = NULL;
6853	page_index = NULL;
6854
6855	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6856
6857	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6858
6859	if (lun->be_lun->lun_type != T_DIRECT)
6860		control_dev = 1;
6861	else
6862		control_dev = 0;
6863
6864	if (lun->flags & CTL_LUN_PR_RESERVED) {
6865		uint32_t residx;
6866
6867		/*
6868		 * XXX KDM need a lock here.
6869		 */
6870		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6871		if ((lun->res_type == SPR_TYPE_EX_AC
6872		  && residx != lun->pr_res_idx)
6873		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6874		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6875		  && !lun->per_res[residx].registered)) {
6876			ctl_set_reservation_conflict(ctsio);
6877			ctl_done((union ctl_io *)ctsio);
6878			return (CTL_RETVAL_COMPLETE);
6879		}
6880	}
6881
6882	switch (ctsio->cdb[0]) {
6883	case MODE_SENSE_6: {
6884		struct scsi_mode_sense_6 *cdb;
6885
6886		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6887
6888		header_len = sizeof(struct scsi_mode_hdr_6);
6889		if (cdb->byte2 & SMS_DBD)
6890			dbd = 1;
6891		else
6892			header_len += sizeof(struct scsi_mode_block_descr);
6893
6894		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6895		page_code = cdb->page & SMS_PAGE_CODE;
6896		subpage = cdb->subpage;
6897		alloc_len = cdb->length;
6898		break;
6899	}
6900	case MODE_SENSE_10: {
6901		struct scsi_mode_sense_10 *cdb;
6902
6903		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6904
6905		header_len = sizeof(struct scsi_mode_hdr_10);
6906
6907		if (cdb->byte2 & SMS_DBD)
6908			dbd = 1;
6909		else
6910			header_len += sizeof(struct scsi_mode_block_descr);
6911		if (cdb->byte2 & SMS10_LLBAA)
6912			llba = 1;
6913		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6914		page_code = cdb->page & SMS_PAGE_CODE;
6915		subpage = cdb->subpage;
6916		alloc_len = scsi_2btoul(cdb->length);
6917		break;
6918	}
6919	default:
6920		ctl_set_invalid_opcode(ctsio);
6921		ctl_done((union ctl_io *)ctsio);
6922		return (CTL_RETVAL_COMPLETE);
6923		break; /* NOTREACHED */
6924	}
6925
6926	/*
6927	 * We have to make a first pass through to calculate the size of
6928	 * the pages that match the user's query.  Then we allocate enough
6929	 * memory to hold it, and actually copy the data into the buffer.
6930	 */
6931	switch (page_code) {
6932	case SMS_ALL_PAGES_PAGE: {
6933		int i;
6934
6935		page_len = 0;
6936
6937		/*
6938		 * At the moment, values other than 0 and 0xff here are
6939		 * reserved according to SPC-3.
6940		 */
6941		if ((subpage != SMS_SUBPAGE_PAGE_0)
6942		 && (subpage != SMS_SUBPAGE_ALL)) {
6943			ctl_set_invalid_field(ctsio,
6944					      /*sks_valid*/ 1,
6945					      /*command*/ 1,
6946					      /*field*/ 3,
6947					      /*bit_valid*/ 0,
6948					      /*bit*/ 0);
6949			ctl_done((union ctl_io *)ctsio);
6950			return (CTL_RETVAL_COMPLETE);
6951		}
6952
6953		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6954			if ((control_dev != 0)
6955			 && (lun->mode_pages.index[i].page_flags &
6956			     CTL_PAGE_FLAG_DISK_ONLY))
6957				continue;
6958
6959			/*
6960			 * We don't use this subpage if the user didn't
6961			 * request all subpages.
6962			 */
6963			if ((lun->mode_pages.index[i].subpage != 0)
6964			 && (subpage == SMS_SUBPAGE_PAGE_0))
6965				continue;
6966
6967#if 0
6968			printf("found page %#x len %d\n",
6969			       lun->mode_pages.index[i].page_code &
6970			       SMPH_PC_MASK,
6971			       lun->mode_pages.index[i].page_len);
6972#endif
6973			page_len += lun->mode_pages.index[i].page_len;
6974		}
6975		break;
6976	}
6977	default: {
6978		int i;
6979
6980		page_len = 0;
6981
6982		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6983			/* Look for the right page code */
6984			if ((lun->mode_pages.index[i].page_code &
6985			     SMPH_PC_MASK) != page_code)
6986				continue;
6987
6988			/* Look for the right subpage or the subpage wildcard*/
6989			if ((lun->mode_pages.index[i].subpage != subpage)
6990			 && (subpage != SMS_SUBPAGE_ALL))
6991				continue;
6992
6993			/* Make sure the page is supported for this dev type */
6994			if ((control_dev != 0)
6995			 && (lun->mode_pages.index[i].page_flags &
6996			     CTL_PAGE_FLAG_DISK_ONLY))
6997				continue;
6998
6999#if 0
7000			printf("found page %#x len %d\n",
7001			       lun->mode_pages.index[i].page_code &
7002			       SMPH_PC_MASK,
7003			       lun->mode_pages.index[i].page_len);
7004#endif
7005
7006			page_len += lun->mode_pages.index[i].page_len;
7007		}
7008
7009		if (page_len == 0) {
7010			ctl_set_invalid_field(ctsio,
7011					      /*sks_valid*/ 1,
7012					      /*command*/ 1,
7013					      /*field*/ 2,
7014					      /*bit_valid*/ 1,
7015					      /*bit*/ 5);
7016			ctl_done((union ctl_io *)ctsio);
7017			return (CTL_RETVAL_COMPLETE);
7018		}
7019		break;
7020	}
7021	}
7022
7023	total_len = header_len + page_len;
7024#if 0
7025	printf("header_len = %d, page_len = %d, total_len = %d\n",
7026	       header_len, page_len, total_len);
7027#endif
7028
7029	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7030	ctsio->kern_sg_entries = 0;
7031	ctsio->kern_data_resid = 0;
7032	ctsio->kern_rel_offset = 0;
7033	if (total_len < alloc_len) {
7034		ctsio->residual = alloc_len - total_len;
7035		ctsio->kern_data_len = total_len;
7036		ctsio->kern_total_len = total_len;
7037	} else {
7038		ctsio->residual = 0;
7039		ctsio->kern_data_len = alloc_len;
7040		ctsio->kern_total_len = alloc_len;
7041	}
7042
7043	switch (ctsio->cdb[0]) {
7044	case MODE_SENSE_6: {
7045		struct scsi_mode_hdr_6 *header;
7046
7047		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7048
7049		header->datalen = ctl_min(total_len - 1, 254);
7050		if (control_dev == 0)
7051			header->dev_specific = 0x10; /* DPOFUA */
7052		if (dbd)
7053			header->block_descr_len = 0;
7054		else
7055			header->block_descr_len =
7056				sizeof(struct scsi_mode_block_descr);
7057		block_desc = (struct scsi_mode_block_descr *)&header[1];
7058		break;
7059	}
7060	case MODE_SENSE_10: {
7061		struct scsi_mode_hdr_10 *header;
7062		int datalen;
7063
7064		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7065
7066		datalen = ctl_min(total_len - 2, 65533);
7067		scsi_ulto2b(datalen, header->datalen);
7068		if (control_dev == 0)
7069			header->dev_specific = 0x10; /* DPOFUA */
7070		if (dbd)
7071			scsi_ulto2b(0, header->block_descr_len);
7072		else
7073			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7074				    header->block_descr_len);
7075		block_desc = (struct scsi_mode_block_descr *)&header[1];
7076		break;
7077	}
7078	default:
7079		panic("invalid CDB type %#x", ctsio->cdb[0]);
7080		break; /* NOTREACHED */
7081	}
7082
7083	/*
7084	 * If we've got a disk, use its blocksize in the block
7085	 * descriptor.  Otherwise, just set it to 0.
7086	 */
7087	if (dbd == 0) {
7088		if (control_dev == 0)
7089			scsi_ulto3b(lun->be_lun->blocksize,
7090				    block_desc->block_len);
7091		else
7092			scsi_ulto3b(0, block_desc->block_len);
7093	}
7094
7095	switch (page_code) {
7096	case SMS_ALL_PAGES_PAGE: {
7097		int i, data_used;
7098
7099		data_used = header_len;
7100		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7101			struct ctl_page_index *page_index;
7102
7103			page_index = &lun->mode_pages.index[i];
7104
7105			if ((control_dev != 0)
7106			 && (page_index->page_flags &
7107			    CTL_PAGE_FLAG_DISK_ONLY))
7108				continue;
7109
7110			/*
7111			 * We don't use this subpage if the user didn't
7112			 * request all subpages.  We already checked (above)
7113			 * to make sure the user only specified a subpage
7114			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7115			 */
7116			if ((page_index->subpage != 0)
7117			 && (subpage == SMS_SUBPAGE_PAGE_0))
7118				continue;
7119
7120			/*
7121			 * Call the handler, if it exists, to update the
7122			 * page to the latest values.
7123			 */
7124			if (page_index->sense_handler != NULL)
7125				page_index->sense_handler(ctsio, page_index,pc);
7126
7127			memcpy(ctsio->kern_data_ptr + data_used,
7128			       page_index->page_data +
7129			       (page_index->page_len * pc),
7130			       page_index->page_len);
7131			data_used += page_index->page_len;
7132		}
7133		break;
7134	}
7135	default: {
7136		int i, data_used;
7137
7138		data_used = header_len;
7139
7140		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7141			struct ctl_page_index *page_index;
7142
7143			page_index = &lun->mode_pages.index[i];
7144
7145			/* Look for the right page code */
7146			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7147				continue;
7148
7149			/* Look for the right subpage or the subpage wildcard*/
7150			if ((page_index->subpage != subpage)
7151			 && (subpage != SMS_SUBPAGE_ALL))
7152				continue;
7153
7154			/* Make sure the page is supported for this dev type */
7155			if ((control_dev != 0)
7156			 && (page_index->page_flags &
7157			     CTL_PAGE_FLAG_DISK_ONLY))
7158				continue;
7159
7160			/*
7161			 * Call the handler, if it exists, to update the
7162			 * page to the latest values.
7163			 */
7164			if (page_index->sense_handler != NULL)
7165				page_index->sense_handler(ctsio, page_index,pc);
7166
7167			memcpy(ctsio->kern_data_ptr + data_used,
7168			       page_index->page_data +
7169			       (page_index->page_len * pc),
7170			       page_index->page_len);
7171			data_used += page_index->page_len;
7172		}
7173		break;
7174	}
7175	}
7176
7177	ctsio->scsi_status = SCSI_STATUS_OK;
7178
7179	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7180	ctsio->be_move_done = ctl_config_move_done;
7181	ctl_datamove((union ctl_io *)ctsio);
7182
7183	return (CTL_RETVAL_COMPLETE);
7184}
7185
7186int
7187ctl_read_capacity(struct ctl_scsiio *ctsio)
7188{
7189	struct scsi_read_capacity *cdb;
7190	struct scsi_read_capacity_data *data;
7191	struct ctl_lun *lun;
7192	uint32_t lba;
7193
7194	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7195
7196	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7197
7198	lba = scsi_4btoul(cdb->addr);
7199	if (((cdb->pmi & SRC_PMI) == 0)
7200	 && (lba != 0)) {
7201		ctl_set_invalid_field(/*ctsio*/ ctsio,
7202				      /*sks_valid*/ 1,
7203				      /*command*/ 1,
7204				      /*field*/ 2,
7205				      /*bit_valid*/ 0,
7206				      /*bit*/ 0);
7207		ctl_done((union ctl_io *)ctsio);
7208		return (CTL_RETVAL_COMPLETE);
7209	}
7210
7211	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7212
7213	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7214	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7215	ctsio->residual = 0;
7216	ctsio->kern_data_len = sizeof(*data);
7217	ctsio->kern_total_len = sizeof(*data);
7218	ctsio->kern_data_resid = 0;
7219	ctsio->kern_rel_offset = 0;
7220	ctsio->kern_sg_entries = 0;
7221
7222	/*
7223	 * If the maximum LBA is greater than 0xfffffffe, the user must
7224	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7225	 * serivce action set.
7226	 */
7227	if (lun->be_lun->maxlba > 0xfffffffe)
7228		scsi_ulto4b(0xffffffff, data->addr);
7229	else
7230		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7231
7232	/*
7233	 * XXX KDM this may not be 512 bytes...
7234	 */
7235	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7236
7237	ctsio->scsi_status = SCSI_STATUS_OK;
7238
7239	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7240	ctsio->be_move_done = ctl_config_move_done;
7241	ctl_datamove((union ctl_io *)ctsio);
7242
7243	return (CTL_RETVAL_COMPLETE);
7244}
7245
7246int
7247ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7248{
7249	struct scsi_read_capacity_16 *cdb;
7250	struct scsi_read_capacity_data_long *data;
7251	struct ctl_lun *lun;
7252	uint64_t lba;
7253	uint32_t alloc_len;
7254
7255	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7256
7257	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7258
7259	alloc_len = scsi_4btoul(cdb->alloc_len);
7260	lba = scsi_8btou64(cdb->addr);
7261
7262	if ((cdb->reladr & SRC16_PMI)
7263	 && (lba != 0)) {
7264		ctl_set_invalid_field(/*ctsio*/ ctsio,
7265				      /*sks_valid*/ 1,
7266				      /*command*/ 1,
7267				      /*field*/ 2,
7268				      /*bit_valid*/ 0,
7269				      /*bit*/ 0);
7270		ctl_done((union ctl_io *)ctsio);
7271		return (CTL_RETVAL_COMPLETE);
7272	}
7273
7274	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7275
7276	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7277	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7278
7279	if (sizeof(*data) < alloc_len) {
7280		ctsio->residual = alloc_len - sizeof(*data);
7281		ctsio->kern_data_len = sizeof(*data);
7282		ctsio->kern_total_len = sizeof(*data);
7283	} else {
7284		ctsio->residual = 0;
7285		ctsio->kern_data_len = alloc_len;
7286		ctsio->kern_total_len = alloc_len;
7287	}
7288	ctsio->kern_data_resid = 0;
7289	ctsio->kern_rel_offset = 0;
7290	ctsio->kern_sg_entries = 0;
7291
7292	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7293	/* XXX KDM this may not be 512 bytes... */
7294	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7295	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7296	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7297	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7298		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7299
7300	ctsio->scsi_status = SCSI_STATUS_OK;
7301
7302	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7303	ctsio->be_move_done = ctl_config_move_done;
7304	ctl_datamove((union ctl_io *)ctsio);
7305
7306	return (CTL_RETVAL_COMPLETE);
7307}
7308
7309int
7310ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7311{
7312	struct scsi_maintenance_in *cdb;
7313	int retval;
7314	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7315	int num_target_port_groups, num_target_ports, single;
7316	struct ctl_lun *lun;
7317	struct ctl_softc *softc;
7318	struct ctl_port *port;
7319	struct scsi_target_group_data *rtg_ptr;
7320	struct scsi_target_group_data_extended *rtg_ext_ptr;
7321	struct scsi_target_port_group_descriptor *tpg_desc;
7322
7323	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7324
7325	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7326	softc = control_softc;
7327	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7328
7329	retval = CTL_RETVAL_COMPLETE;
7330
7331	switch (cdb->byte2 & STG_PDF_MASK) {
7332	case STG_PDF_LENGTH:
7333		ext = 0;
7334		break;
7335	case STG_PDF_EXTENDED:
7336		ext = 1;
7337		break;
7338	default:
7339		ctl_set_invalid_field(/*ctsio*/ ctsio,
7340				      /*sks_valid*/ 1,
7341				      /*command*/ 1,
7342				      /*field*/ 2,
7343				      /*bit_valid*/ 1,
7344				      /*bit*/ 5);
7345		ctl_done((union ctl_io *)ctsio);
7346		return(retval);
7347	}
7348
7349	single = ctl_is_single;
7350	if (single)
7351		num_target_port_groups = 1;
7352	else
7353		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7354	num_target_ports = 0;
7355	mtx_lock(&softc->ctl_lock);
7356	STAILQ_FOREACH(port, &softc->port_list, links) {
7357		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7358			continue;
7359		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7360			continue;
7361		num_target_ports++;
7362	}
7363	mtx_unlock(&softc->ctl_lock);
7364
7365	if (ext)
7366		total_len = sizeof(struct scsi_target_group_data_extended);
7367	else
7368		total_len = sizeof(struct scsi_target_group_data);
7369	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7370		num_target_port_groups +
7371	    sizeof(struct scsi_target_port_descriptor) *
7372		num_target_ports * num_target_port_groups;
7373
7374	alloc_len = scsi_4btoul(cdb->length);
7375
7376	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7377
7378	ctsio->kern_sg_entries = 0;
7379
7380	if (total_len < alloc_len) {
7381		ctsio->residual = alloc_len - total_len;
7382		ctsio->kern_data_len = total_len;
7383		ctsio->kern_total_len = total_len;
7384	} else {
7385		ctsio->residual = 0;
7386		ctsio->kern_data_len = alloc_len;
7387		ctsio->kern_total_len = alloc_len;
7388	}
7389	ctsio->kern_data_resid = 0;
7390	ctsio->kern_rel_offset = 0;
7391
7392	if (ext) {
7393		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7394		    ctsio->kern_data_ptr;
7395		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7396		rtg_ext_ptr->format_type = 0x10;
7397		rtg_ext_ptr->implicit_transition_time = 0;
7398		tpg_desc = &rtg_ext_ptr->groups[0];
7399	} else {
7400		rtg_ptr = (struct scsi_target_group_data *)
7401		    ctsio->kern_data_ptr;
7402		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7403		tpg_desc = &rtg_ptr->groups[0];
7404	}
7405
7406	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7407	mtx_lock(&softc->ctl_lock);
7408	for (g = 0; g < num_target_port_groups; g++) {
7409		if (g == pg)
7410			tpg_desc->pref_state = TPG_PRIMARY |
7411			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7412		else
7413			tpg_desc->pref_state =
7414			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7415		tpg_desc->support = TPG_AO_SUP;
7416		if (!single)
7417			tpg_desc->support |= TPG_AN_SUP;
7418		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7419		tpg_desc->status = TPG_IMPLICIT;
7420		pc = 0;
7421		STAILQ_FOREACH(port, &softc->port_list, links) {
7422			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7423				continue;
7424			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7425			    CTL_MAX_LUNS)
7426				continue;
7427			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7428			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7429			    relative_target_port_identifier);
7430			pc++;
7431		}
7432		tpg_desc->target_port_count = pc;
7433		tpg_desc = (struct scsi_target_port_group_descriptor *)
7434		    &tpg_desc->descriptors[pc];
7435	}
7436	mtx_unlock(&softc->ctl_lock);
7437
7438	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7439	ctsio->be_move_done = ctl_config_move_done;
7440
7441	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7442			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7443			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7444			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7445			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7446
7447	ctl_datamove((union ctl_io *)ctsio);
7448	return(retval);
7449}
7450
7451int
7452ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7453{
7454	struct ctl_lun *lun;
7455	struct scsi_report_supported_opcodes *cdb;
7456	const struct ctl_cmd_entry *entry, *sentry;
7457	struct scsi_report_supported_opcodes_all *all;
7458	struct scsi_report_supported_opcodes_descr *descr;
7459	struct scsi_report_supported_opcodes_one *one;
7460	int retval;
7461	int alloc_len, total_len;
7462	int opcode, service_action, i, j, num;
7463
7464	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7465
7466	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7467	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7468
7469	retval = CTL_RETVAL_COMPLETE;
7470
7471	opcode = cdb->requested_opcode;
7472	service_action = scsi_2btoul(cdb->requested_service_action);
7473	switch (cdb->options & RSO_OPTIONS_MASK) {
7474	case RSO_OPTIONS_ALL:
7475		num = 0;
7476		for (i = 0; i < 256; i++) {
7477			entry = &ctl_cmd_table[i];
7478			if (entry->flags & CTL_CMD_FLAG_SA5) {
7479				for (j = 0; j < 32; j++) {
7480					sentry = &((const struct ctl_cmd_entry *)
7481					    entry->execute)[j];
7482					if (ctl_cmd_applicable(
7483					    lun->be_lun->lun_type, sentry))
7484						num++;
7485				}
7486			} else {
7487				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7488				    entry))
7489					num++;
7490			}
7491		}
7492		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7493		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7494		break;
7495	case RSO_OPTIONS_OC:
7496		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7497			ctl_set_invalid_field(/*ctsio*/ ctsio,
7498					      /*sks_valid*/ 1,
7499					      /*command*/ 1,
7500					      /*field*/ 2,
7501					      /*bit_valid*/ 1,
7502					      /*bit*/ 2);
7503			ctl_done((union ctl_io *)ctsio);
7504			return (CTL_RETVAL_COMPLETE);
7505		}
7506		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7507		break;
7508	case RSO_OPTIONS_OC_SA:
7509		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7510		    service_action >= 32) {
7511			ctl_set_invalid_field(/*ctsio*/ ctsio,
7512					      /*sks_valid*/ 1,
7513					      /*command*/ 1,
7514					      /*field*/ 2,
7515					      /*bit_valid*/ 1,
7516					      /*bit*/ 2);
7517			ctl_done((union ctl_io *)ctsio);
7518			return (CTL_RETVAL_COMPLETE);
7519		}
7520		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7521		break;
7522	default:
7523		ctl_set_invalid_field(/*ctsio*/ ctsio,
7524				      /*sks_valid*/ 1,
7525				      /*command*/ 1,
7526				      /*field*/ 2,
7527				      /*bit_valid*/ 1,
7528				      /*bit*/ 2);
7529		ctl_done((union ctl_io *)ctsio);
7530		return (CTL_RETVAL_COMPLETE);
7531	}
7532
7533	alloc_len = scsi_4btoul(cdb->length);
7534
7535	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7536
7537	ctsio->kern_sg_entries = 0;
7538
7539	if (total_len < alloc_len) {
7540		ctsio->residual = alloc_len - total_len;
7541		ctsio->kern_data_len = total_len;
7542		ctsio->kern_total_len = total_len;
7543	} else {
7544		ctsio->residual = 0;
7545		ctsio->kern_data_len = alloc_len;
7546		ctsio->kern_total_len = alloc_len;
7547	}
7548	ctsio->kern_data_resid = 0;
7549	ctsio->kern_rel_offset = 0;
7550
7551	switch (cdb->options & RSO_OPTIONS_MASK) {
7552	case RSO_OPTIONS_ALL:
7553		all = (struct scsi_report_supported_opcodes_all *)
7554		    ctsio->kern_data_ptr;
7555		num = 0;
7556		for (i = 0; i < 256; i++) {
7557			entry = &ctl_cmd_table[i];
7558			if (entry->flags & CTL_CMD_FLAG_SA5) {
7559				for (j = 0; j < 32; j++) {
7560					sentry = &((const struct ctl_cmd_entry *)
7561					    entry->execute)[j];
7562					if (!ctl_cmd_applicable(
7563					    lun->be_lun->lun_type, sentry))
7564						continue;
7565					descr = &all->descr[num++];
7566					descr->opcode = i;
7567					scsi_ulto2b(j, descr->service_action);
7568					descr->flags = RSO_SERVACTV;
7569					scsi_ulto2b(sentry->length,
7570					    descr->cdb_length);
7571				}
7572			} else {
7573				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7574				    entry))
7575					continue;
7576				descr = &all->descr[num++];
7577				descr->opcode = i;
7578				scsi_ulto2b(0, descr->service_action);
7579				descr->flags = 0;
7580				scsi_ulto2b(entry->length, descr->cdb_length);
7581			}
7582		}
7583		scsi_ulto4b(
7584		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7585		    all->length);
7586		break;
7587	case RSO_OPTIONS_OC:
7588		one = (struct scsi_report_supported_opcodes_one *)
7589		    ctsio->kern_data_ptr;
7590		entry = &ctl_cmd_table[opcode];
7591		goto fill_one;
7592	case RSO_OPTIONS_OC_SA:
7593		one = (struct scsi_report_supported_opcodes_one *)
7594		    ctsio->kern_data_ptr;
7595		entry = &ctl_cmd_table[opcode];
7596		entry = &((const struct ctl_cmd_entry *)
7597		    entry->execute)[service_action];
7598fill_one:
7599		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7600			one->support = 3;
7601			scsi_ulto2b(entry->length, one->cdb_length);
7602			one->cdb_usage[0] = opcode;
7603			memcpy(&one->cdb_usage[1], entry->usage,
7604			    entry->length - 1);
7605		} else
7606			one->support = 1;
7607		break;
7608	}
7609
7610	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7611	ctsio->be_move_done = ctl_config_move_done;
7612
7613	ctl_datamove((union ctl_io *)ctsio);
7614	return(retval);
7615}
7616
7617int
7618ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7619{
7620	struct ctl_lun *lun;
7621	struct scsi_report_supported_tmf *cdb;
7622	struct scsi_report_supported_tmf_data *data;
7623	int retval;
7624	int alloc_len, total_len;
7625
7626	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7627
7628	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7629	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7630
7631	retval = CTL_RETVAL_COMPLETE;
7632
7633	total_len = sizeof(struct scsi_report_supported_tmf_data);
7634	alloc_len = scsi_4btoul(cdb->length);
7635
7636	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7637
7638	ctsio->kern_sg_entries = 0;
7639
7640	if (total_len < alloc_len) {
7641		ctsio->residual = alloc_len - total_len;
7642		ctsio->kern_data_len = total_len;
7643		ctsio->kern_total_len = total_len;
7644	} else {
7645		ctsio->residual = 0;
7646		ctsio->kern_data_len = alloc_len;
7647		ctsio->kern_total_len = alloc_len;
7648	}
7649	ctsio->kern_data_resid = 0;
7650	ctsio->kern_rel_offset = 0;
7651
7652	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7653	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7654	data->byte2 |= RST_ITNRS;
7655
7656	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7657	ctsio->be_move_done = ctl_config_move_done;
7658
7659	ctl_datamove((union ctl_io *)ctsio);
7660	return (retval);
7661}
7662
7663int
7664ctl_report_timestamp(struct ctl_scsiio *ctsio)
7665{
7666	struct ctl_lun *lun;
7667	struct scsi_report_timestamp *cdb;
7668	struct scsi_report_timestamp_data *data;
7669	struct timeval tv;
7670	int64_t timestamp;
7671	int retval;
7672	int alloc_len, total_len;
7673
7674	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7675
7676	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7677	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7678
7679	retval = CTL_RETVAL_COMPLETE;
7680
7681	total_len = sizeof(struct scsi_report_timestamp_data);
7682	alloc_len = scsi_4btoul(cdb->length);
7683
7684	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7685
7686	ctsio->kern_sg_entries = 0;
7687
7688	if (total_len < alloc_len) {
7689		ctsio->residual = alloc_len - total_len;
7690		ctsio->kern_data_len = total_len;
7691		ctsio->kern_total_len = total_len;
7692	} else {
7693		ctsio->residual = 0;
7694		ctsio->kern_data_len = alloc_len;
7695		ctsio->kern_total_len = alloc_len;
7696	}
7697	ctsio->kern_data_resid = 0;
7698	ctsio->kern_rel_offset = 0;
7699
7700	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7701	scsi_ulto2b(sizeof(*data) - 2, data->length);
7702	data->origin = RTS_ORIG_OUTSIDE;
7703	getmicrotime(&tv);
7704	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7705	scsi_ulto4b(timestamp >> 16, data->timestamp);
7706	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7707
7708	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7709	ctsio->be_move_done = ctl_config_move_done;
7710
7711	ctl_datamove((union ctl_io *)ctsio);
7712	return (retval);
7713}
7714
7715int
7716ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7717{
7718	struct scsi_per_res_in *cdb;
7719	int alloc_len, total_len = 0;
7720	/* struct scsi_per_res_in_rsrv in_data; */
7721	struct ctl_lun *lun;
7722	struct ctl_softc *softc;
7723
7724	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7725
7726	softc = control_softc;
7727
7728	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7729
7730	alloc_len = scsi_2btoul(cdb->length);
7731
7732	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7733
7734retry:
7735	mtx_lock(&lun->lun_lock);
7736	switch (cdb->action) {
7737	case SPRI_RK: /* read keys */
7738		total_len = sizeof(struct scsi_per_res_in_keys) +
7739			lun->pr_key_count *
7740			sizeof(struct scsi_per_res_key);
7741		break;
7742	case SPRI_RR: /* read reservation */
7743		if (lun->flags & CTL_LUN_PR_RESERVED)
7744			total_len = sizeof(struct scsi_per_res_in_rsrv);
7745		else
7746			total_len = sizeof(struct scsi_per_res_in_header);
7747		break;
7748	case SPRI_RC: /* report capabilities */
7749		total_len = sizeof(struct scsi_per_res_cap);
7750		break;
7751	case SPRI_RS: /* read full status */
7752		total_len = sizeof(struct scsi_per_res_in_header) +
7753		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7754		    lun->pr_key_count;
7755		break;
7756	default:
7757		panic("Invalid PR type %x", cdb->action);
7758	}
7759	mtx_unlock(&lun->lun_lock);
7760
7761	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7762
7763	if (total_len < alloc_len) {
7764		ctsio->residual = alloc_len - total_len;
7765		ctsio->kern_data_len = total_len;
7766		ctsio->kern_total_len = total_len;
7767	} else {
7768		ctsio->residual = 0;
7769		ctsio->kern_data_len = alloc_len;
7770		ctsio->kern_total_len = alloc_len;
7771	}
7772
7773	ctsio->kern_data_resid = 0;
7774	ctsio->kern_rel_offset = 0;
7775	ctsio->kern_sg_entries = 0;
7776
7777	mtx_lock(&lun->lun_lock);
7778	switch (cdb->action) {
7779	case SPRI_RK: { // read keys
7780        struct scsi_per_res_in_keys *res_keys;
7781		int i, key_count;
7782
7783		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7784
7785		/*
7786		 * We had to drop the lock to allocate our buffer, which
7787		 * leaves time for someone to come in with another
7788		 * persistent reservation.  (That is unlikely, though,
7789		 * since this should be the only persistent reservation
7790		 * command active right now.)
7791		 */
7792		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7793		    (lun->pr_key_count *
7794		     sizeof(struct scsi_per_res_key)))){
7795			mtx_unlock(&lun->lun_lock);
7796			free(ctsio->kern_data_ptr, M_CTL);
7797			printf("%s: reservation length changed, retrying\n",
7798			       __func__);
7799			goto retry;
7800		}
7801
7802		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7803
7804		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7805			     lun->pr_key_count, res_keys->header.length);
7806
7807		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7808			if (!lun->per_res[i].registered)
7809				continue;
7810
7811			/*
7812			 * We used lun->pr_key_count to calculate the
7813			 * size to allocate.  If it turns out the number of
7814			 * initiators with the registered flag set is
7815			 * larger than that (i.e. they haven't been kept in
7816			 * sync), we've got a problem.
7817			 */
7818			if (key_count >= lun->pr_key_count) {
7819#ifdef NEEDTOPORT
7820				csevent_log(CSC_CTL | CSC_SHELF_SW |
7821					    CTL_PR_ERROR,
7822					    csevent_LogType_Fault,
7823					    csevent_AlertLevel_Yellow,
7824					    csevent_FRU_ShelfController,
7825					    csevent_FRU_Firmware,
7826				        csevent_FRU_Unknown,
7827					    "registered keys %d >= key "
7828					    "count %d", key_count,
7829					    lun->pr_key_count);
7830#endif
7831				key_count++;
7832				continue;
7833			}
7834			memcpy(res_keys->keys[key_count].key,
7835			       lun->per_res[i].res_key.key,
7836			       ctl_min(sizeof(res_keys->keys[key_count].key),
7837			       sizeof(lun->per_res[i].res_key)));
7838			key_count++;
7839		}
7840		break;
7841	}
7842	case SPRI_RR: { // read reservation
7843		struct scsi_per_res_in_rsrv *res;
7844		int tmp_len, header_only;
7845
7846		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7847
7848		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7849
7850		if (lun->flags & CTL_LUN_PR_RESERVED)
7851		{
7852			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7853			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7854				    res->header.length);
7855			header_only = 0;
7856		} else {
7857			tmp_len = sizeof(struct scsi_per_res_in_header);
7858			scsi_ulto4b(0, res->header.length);
7859			header_only = 1;
7860		}
7861
7862		/*
7863		 * We had to drop the lock to allocate our buffer, which
7864		 * leaves time for someone to come in with another
7865		 * persistent reservation.  (That is unlikely, though,
7866		 * since this should be the only persistent reservation
7867		 * command active right now.)
7868		 */
7869		if (tmp_len != total_len) {
7870			mtx_unlock(&lun->lun_lock);
7871			free(ctsio->kern_data_ptr, M_CTL);
7872			printf("%s: reservation status changed, retrying\n",
7873			       __func__);
7874			goto retry;
7875		}
7876
7877		/*
7878		 * No reservation held, so we're done.
7879		 */
7880		if (header_only != 0)
7881			break;
7882
7883		/*
7884		 * If the registration is an All Registrants type, the key
7885		 * is 0, since it doesn't really matter.
7886		 */
7887		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7888			memcpy(res->data.reservation,
7889			       &lun->per_res[lun->pr_res_idx].res_key,
7890			       sizeof(struct scsi_per_res_key));
7891		}
7892		res->data.scopetype = lun->res_type;
7893		break;
7894	}
7895	case SPRI_RC:     //report capabilities
7896	{
7897		struct scsi_per_res_cap *res_cap;
7898		uint16_t type_mask;
7899
7900		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7901		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7902		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7903		type_mask = SPRI_TM_WR_EX_AR |
7904			    SPRI_TM_EX_AC_RO |
7905			    SPRI_TM_WR_EX_RO |
7906			    SPRI_TM_EX_AC |
7907			    SPRI_TM_WR_EX |
7908			    SPRI_TM_EX_AC_AR;
7909		scsi_ulto2b(type_mask, res_cap->type_mask);
7910		break;
7911	}
7912	case SPRI_RS: { // read full status
7913		struct scsi_per_res_in_full *res_status;
7914		struct scsi_per_res_in_full_desc *res_desc;
7915		struct ctl_port *port;
7916		int i, len;
7917
7918		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7919
7920		/*
7921		 * We had to drop the lock to allocate our buffer, which
7922		 * leaves time for someone to come in with another
7923		 * persistent reservation.  (That is unlikely, though,
7924		 * since this should be the only persistent reservation
7925		 * command active right now.)
7926		 */
7927		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7928		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7929		     lun->pr_key_count)){
7930			mtx_unlock(&lun->lun_lock);
7931			free(ctsio->kern_data_ptr, M_CTL);
7932			printf("%s: reservation length changed, retrying\n",
7933			       __func__);
7934			goto retry;
7935		}
7936
7937		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7938
7939		res_desc = &res_status->desc[0];
7940		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7941			if (!lun->per_res[i].registered)
7942				continue;
7943
7944			memcpy(&res_desc->res_key, &lun->per_res[i].res_key.key,
7945			    sizeof(res_desc->res_key));
7946			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7947			    (lun->pr_res_idx == i ||
7948			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7949				res_desc->flags = SPRI_FULL_R_HOLDER;
7950				res_desc->scopetype = lun->res_type;
7951			}
7952			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7953			    res_desc->rel_trgt_port_id);
7954			len = 0;
7955			port = softc->ctl_ports[
7956			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7957			if (port != NULL)
7958				len = ctl_create_iid(port,
7959				    i % CTL_MAX_INIT_PER_PORT,
7960				    res_desc->transport_id);
7961			scsi_ulto4b(len, res_desc->additional_length);
7962			res_desc = (struct scsi_per_res_in_full_desc *)
7963			    &res_desc->transport_id[len];
7964		}
7965		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7966		    res_status->header.length);
7967		break;
7968	}
7969	default:
7970		/*
7971		 * This is a bug, because we just checked for this above,
7972		 * and should have returned an error.
7973		 */
7974		panic("Invalid PR type %x", cdb->action);
7975		break; /* NOTREACHED */
7976	}
7977	mtx_unlock(&lun->lun_lock);
7978
7979	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7980	ctsio->be_move_done = ctl_config_move_done;
7981
7982	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7983			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7984			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7985			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7986			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7987
7988	ctl_datamove((union ctl_io *)ctsio);
7989
7990	return (CTL_RETVAL_COMPLETE);
7991}
7992
7993/*
7994 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7995 * it should return.
7996 */
7997static int
7998ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7999		uint64_t sa_res_key, uint8_t type, uint32_t residx,
8000		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
8001		struct scsi_per_res_out_parms* param)
8002{
8003	union ctl_ha_msg persis_io;
8004	int retval, i;
8005	int isc_retval;
8006
8007	retval = 0;
8008
8009	mtx_lock(&lun->lun_lock);
8010	if (sa_res_key == 0) {
8011		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8012			/* validate scope and type */
8013			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8014			     SPR_LU_SCOPE) {
8015				mtx_unlock(&lun->lun_lock);
8016				ctl_set_invalid_field(/*ctsio*/ ctsio,
8017						      /*sks_valid*/ 1,
8018						      /*command*/ 1,
8019						      /*field*/ 2,
8020						      /*bit_valid*/ 1,
8021						      /*bit*/ 4);
8022				ctl_done((union ctl_io *)ctsio);
8023				return (1);
8024			}
8025
8026		        if (type>8 || type==2 || type==4 || type==0) {
8027				mtx_unlock(&lun->lun_lock);
8028				ctl_set_invalid_field(/*ctsio*/ ctsio,
8029       	           				      /*sks_valid*/ 1,
8030						      /*command*/ 1,
8031						      /*field*/ 2,
8032						      /*bit_valid*/ 1,
8033						      /*bit*/ 0);
8034				ctl_done((union ctl_io *)ctsio);
8035				return (1);
8036		        }
8037
8038			/* temporarily unregister this nexus */
8039			lun->per_res[residx].registered = 0;
8040
8041			/*
8042			 * Unregister everybody else and build UA for
8043			 * them
8044			 */
8045			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8046				if (lun->per_res[i].registered == 0)
8047					continue;
8048
8049				if (!persis_offset
8050				 && i <CTL_MAX_INITIATORS)
8051					lun->pending_ua[i] |=
8052						CTL_UA_REG_PREEMPT;
8053				else if (persis_offset
8054				      && i >= persis_offset)
8055					lun->pending_ua[i-persis_offset] |=
8056						CTL_UA_REG_PREEMPT;
8057				lun->per_res[i].registered = 0;
8058				memset(&lun->per_res[i].res_key, 0,
8059				       sizeof(struct scsi_per_res_key));
8060			}
8061			lun->per_res[residx].registered = 1;
8062			lun->pr_key_count = 1;
8063			lun->res_type = type;
8064			if (lun->res_type != SPR_TYPE_WR_EX_AR
8065			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8066				lun->pr_res_idx = residx;
8067
8068			/* send msg to other side */
8069			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8070			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8071			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8072			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8073			persis_io.pr.pr_info.res_type = type;
8074			memcpy(persis_io.pr.pr_info.sa_res_key,
8075			       param->serv_act_res_key,
8076			       sizeof(param->serv_act_res_key));
8077			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8078			     &persis_io, sizeof(persis_io), 0)) >
8079			     CTL_HA_STATUS_SUCCESS) {
8080				printf("CTL:Persis Out error returned "
8081				       "from ctl_ha_msg_send %d\n",
8082				       isc_retval);
8083			}
8084		} else {
8085			/* not all registrants */
8086			mtx_unlock(&lun->lun_lock);
8087			free(ctsio->kern_data_ptr, M_CTL);
8088			ctl_set_invalid_field(ctsio,
8089					      /*sks_valid*/ 1,
8090					      /*command*/ 0,
8091					      /*field*/ 8,
8092					      /*bit_valid*/ 0,
8093					      /*bit*/ 0);
8094			ctl_done((union ctl_io *)ctsio);
8095			return (1);
8096		}
8097	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8098		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8099		int found = 0;
8100
8101		if (res_key == sa_res_key) {
8102			/* special case */
8103			/*
8104			 * The spec implies this is not good but doesn't
8105			 * say what to do. There are two choices either
8106			 * generate a res conflict or check condition
8107			 * with illegal field in parameter data. Since
8108			 * that is what is done when the sa_res_key is
8109			 * zero I'll take that approach since this has
8110			 * to do with the sa_res_key.
8111			 */
8112			mtx_unlock(&lun->lun_lock);
8113			free(ctsio->kern_data_ptr, M_CTL);
8114			ctl_set_invalid_field(ctsio,
8115					      /*sks_valid*/ 1,
8116					      /*command*/ 0,
8117					      /*field*/ 8,
8118					      /*bit_valid*/ 0,
8119					      /*bit*/ 0);
8120			ctl_done((union ctl_io *)ctsio);
8121			return (1);
8122		}
8123
8124		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8125			if (lun->per_res[i].registered
8126			 && memcmp(param->serv_act_res_key,
8127			    lun->per_res[i].res_key.key,
8128			    sizeof(struct scsi_per_res_key)) != 0)
8129				continue;
8130
8131			found = 1;
8132			lun->per_res[i].registered = 0;
8133			memset(&lun->per_res[i].res_key, 0,
8134			       sizeof(struct scsi_per_res_key));
8135			lun->pr_key_count--;
8136
8137			if (!persis_offset && i < CTL_MAX_INITIATORS)
8138				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8139			else if (persis_offset && i >= persis_offset)
8140				lun->pending_ua[i-persis_offset] |=
8141					CTL_UA_REG_PREEMPT;
8142		}
8143		if (!found) {
8144			mtx_unlock(&lun->lun_lock);
8145			free(ctsio->kern_data_ptr, M_CTL);
8146			ctl_set_reservation_conflict(ctsio);
8147			ctl_done((union ctl_io *)ctsio);
8148			return (CTL_RETVAL_COMPLETE);
8149		}
8150		/* send msg to other side */
8151		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8152		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8153		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8154		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8155		persis_io.pr.pr_info.res_type = type;
8156		memcpy(persis_io.pr.pr_info.sa_res_key,
8157		       param->serv_act_res_key,
8158		       sizeof(param->serv_act_res_key));
8159		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8160		     &persis_io, sizeof(persis_io), 0)) >
8161		     CTL_HA_STATUS_SUCCESS) {
8162			printf("CTL:Persis Out error returned from "
8163			       "ctl_ha_msg_send %d\n", isc_retval);
8164		}
8165	} else {
8166		/* Reserved but not all registrants */
8167		/* sa_res_key is res holder */
8168		if (memcmp(param->serv_act_res_key,
8169                   lun->per_res[lun->pr_res_idx].res_key.key,
8170                   sizeof(struct scsi_per_res_key)) == 0) {
8171			/* validate scope and type */
8172			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8173			     SPR_LU_SCOPE) {
8174				mtx_unlock(&lun->lun_lock);
8175				ctl_set_invalid_field(/*ctsio*/ ctsio,
8176						      /*sks_valid*/ 1,
8177						      /*command*/ 1,
8178						      /*field*/ 2,
8179						      /*bit_valid*/ 1,
8180						      /*bit*/ 4);
8181				ctl_done((union ctl_io *)ctsio);
8182				return (1);
8183			}
8184
8185			if (type>8 || type==2 || type==4 || type==0) {
8186				mtx_unlock(&lun->lun_lock);
8187				ctl_set_invalid_field(/*ctsio*/ ctsio,
8188						      /*sks_valid*/ 1,
8189						      /*command*/ 1,
8190						      /*field*/ 2,
8191						      /*bit_valid*/ 1,
8192						      /*bit*/ 0);
8193				ctl_done((union ctl_io *)ctsio);
8194				return (1);
8195			}
8196
8197			/*
8198			 * Do the following:
8199			 * if sa_res_key != res_key remove all
8200			 * registrants w/sa_res_key and generate UA
8201			 * for these registrants(Registrations
8202			 * Preempted) if it wasn't an exclusive
8203			 * reservation generate UA(Reservations
8204			 * Preempted) for all other registered nexuses
8205			 * if the type has changed. Establish the new
8206			 * reservation and holder. If res_key and
8207			 * sa_res_key are the same do the above
8208			 * except don't unregister the res holder.
8209			 */
8210
8211			/*
8212			 * Temporarily unregister so it won't get
8213			 * removed or UA generated
8214			 */
8215			lun->per_res[residx].registered = 0;
8216			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8217				if (lun->per_res[i].registered == 0)
8218					continue;
8219
8220				if (memcmp(param->serv_act_res_key,
8221				    lun->per_res[i].res_key.key,
8222				    sizeof(struct scsi_per_res_key)) == 0) {
8223					lun->per_res[i].registered = 0;
8224					memset(&lun->per_res[i].res_key,
8225					       0,
8226					       sizeof(struct scsi_per_res_key));
8227					lun->pr_key_count--;
8228
8229					if (!persis_offset
8230					 && i < CTL_MAX_INITIATORS)
8231						lun->pending_ua[i] |=
8232							CTL_UA_REG_PREEMPT;
8233					else if (persis_offset
8234					      && i >= persis_offset)
8235						lun->pending_ua[i-persis_offset] |=
8236						  CTL_UA_REG_PREEMPT;
8237				} else if (type != lun->res_type
8238					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8239					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8240						if (!persis_offset
8241						 && i < CTL_MAX_INITIATORS)
8242							lun->pending_ua[i] |=
8243							CTL_UA_RES_RELEASE;
8244						else if (persis_offset
8245						      && i >= persis_offset)
8246							lun->pending_ua[
8247							i-persis_offset] |=
8248							CTL_UA_RES_RELEASE;
8249				}
8250			}
8251			lun->per_res[residx].registered = 1;
8252			lun->res_type = type;
8253			if (lun->res_type != SPR_TYPE_WR_EX_AR
8254			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8255				lun->pr_res_idx = residx;
8256			else
8257				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8258
8259			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8260			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8261			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8262			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8263			persis_io.pr.pr_info.res_type = type;
8264			memcpy(persis_io.pr.pr_info.sa_res_key,
8265			       param->serv_act_res_key,
8266			       sizeof(param->serv_act_res_key));
8267			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8268			     &persis_io, sizeof(persis_io), 0)) >
8269			     CTL_HA_STATUS_SUCCESS) {
8270				printf("CTL:Persis Out error returned "
8271				       "from ctl_ha_msg_send %d\n",
8272				       isc_retval);
8273			}
8274		} else {
8275			/*
8276			 * sa_res_key is not the res holder just
8277			 * remove registrants
8278			 */
8279			int found=0;
8280
8281			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8282				if (memcmp(param->serv_act_res_key,
8283				    lun->per_res[i].res_key.key,
8284				    sizeof(struct scsi_per_res_key)) != 0)
8285					continue;
8286
8287				found = 1;
8288				lun->per_res[i].registered = 0;
8289				memset(&lun->per_res[i].res_key, 0,
8290				       sizeof(struct scsi_per_res_key));
8291				lun->pr_key_count--;
8292
8293				if (!persis_offset
8294				 && i < CTL_MAX_INITIATORS)
8295					lun->pending_ua[i] |=
8296						CTL_UA_REG_PREEMPT;
8297				else if (persis_offset
8298				      && i >= persis_offset)
8299					lun->pending_ua[i-persis_offset] |=
8300						CTL_UA_REG_PREEMPT;
8301			}
8302
8303			if (!found) {
8304				mtx_unlock(&lun->lun_lock);
8305				free(ctsio->kern_data_ptr, M_CTL);
8306				ctl_set_reservation_conflict(ctsio);
8307				ctl_done((union ctl_io *)ctsio);
8308		        	return (1);
8309			}
8310			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8311			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8312			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8313			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8314			persis_io.pr.pr_info.res_type = type;
8315			memcpy(persis_io.pr.pr_info.sa_res_key,
8316			       param->serv_act_res_key,
8317			       sizeof(param->serv_act_res_key));
8318			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8319			     &persis_io, sizeof(persis_io), 0)) >
8320			     CTL_HA_STATUS_SUCCESS) {
8321				printf("CTL:Persis Out error returned "
8322				       "from ctl_ha_msg_send %d\n",
8323				isc_retval);
8324			}
8325		}
8326	}
8327
8328	lun->PRGeneration++;
8329	mtx_unlock(&lun->lun_lock);
8330
8331	return (retval);
8332}
8333
8334static void
8335ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8336{
8337	int i;
8338
8339	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8340	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8341	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8342		   msg->pr.pr_info.sa_res_key,
8343		   sizeof(struct scsi_per_res_key)) != 0) {
8344		uint64_t sa_res_key;
8345		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8346
8347		if (sa_res_key == 0) {
8348			/* temporarily unregister this nexus */
8349			lun->per_res[msg->pr.pr_info.residx].registered = 0;
8350
8351			/*
8352			 * Unregister everybody else and build UA for
8353			 * them
8354			 */
8355			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8356				if (lun->per_res[i].registered == 0)
8357					continue;
8358
8359				if (!persis_offset
8360				 && i < CTL_MAX_INITIATORS)
8361					lun->pending_ua[i] |=
8362						CTL_UA_REG_PREEMPT;
8363				else if (persis_offset && i >= persis_offset)
8364					lun->pending_ua[i - persis_offset] |=
8365						CTL_UA_REG_PREEMPT;
8366				lun->per_res[i].registered = 0;
8367				memset(&lun->per_res[i].res_key, 0,
8368				       sizeof(struct scsi_per_res_key));
8369			}
8370
8371			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8372			lun->pr_key_count = 1;
8373			lun->res_type = msg->pr.pr_info.res_type;
8374			if (lun->res_type != SPR_TYPE_WR_EX_AR
8375			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8376				lun->pr_res_idx = msg->pr.pr_info.residx;
8377		} else {
8378		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8379				if (memcmp(msg->pr.pr_info.sa_res_key,
8380		                   lun->per_res[i].res_key.key,
8381		                   sizeof(struct scsi_per_res_key)) != 0)
8382					continue;
8383
8384				lun->per_res[i].registered = 0;
8385				memset(&lun->per_res[i].res_key, 0,
8386				       sizeof(struct scsi_per_res_key));
8387				lun->pr_key_count--;
8388
8389				if (!persis_offset
8390				 && i < persis_offset)
8391					lun->pending_ua[i] |=
8392						CTL_UA_REG_PREEMPT;
8393				else if (persis_offset
8394				      && i >= persis_offset)
8395					lun->pending_ua[i - persis_offset] |=
8396						CTL_UA_REG_PREEMPT;
8397			}
8398		}
8399	} else {
8400		/*
8401		 * Temporarily unregister so it won't get removed
8402		 * or UA generated
8403		 */
8404		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8405		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8406			if (lun->per_res[i].registered == 0)
8407				continue;
8408
8409			if (memcmp(msg->pr.pr_info.sa_res_key,
8410	                   lun->per_res[i].res_key.key,
8411	                   sizeof(struct scsi_per_res_key)) == 0) {
8412				lun->per_res[i].registered = 0;
8413				memset(&lun->per_res[i].res_key, 0,
8414				       sizeof(struct scsi_per_res_key));
8415				lun->pr_key_count--;
8416				if (!persis_offset
8417				 && i < CTL_MAX_INITIATORS)
8418					lun->pending_ua[i] |=
8419						CTL_UA_REG_PREEMPT;
8420				else if (persis_offset
8421				      && i >= persis_offset)
8422					lun->pending_ua[i - persis_offset] |=
8423						CTL_UA_REG_PREEMPT;
8424			} else if (msg->pr.pr_info.res_type != lun->res_type
8425				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8426				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8427					if (!persis_offset
8428					 && i < persis_offset)
8429						lun->pending_ua[i] |=
8430							CTL_UA_RES_RELEASE;
8431					else if (persis_offset
8432					      && i >= persis_offset)
8433					lun->pending_ua[i - persis_offset] |=
8434						CTL_UA_RES_RELEASE;
8435			}
8436		}
8437		lun->per_res[msg->pr.pr_info.residx].registered = 1;
8438		lun->res_type = msg->pr.pr_info.res_type;
8439		if (lun->res_type != SPR_TYPE_WR_EX_AR
8440		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8441			lun->pr_res_idx = msg->pr.pr_info.residx;
8442		else
8443			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8444	}
8445	lun->PRGeneration++;
8446
8447}
8448
8449
8450int
8451ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8452{
8453	int retval;
8454	int isc_retval;
8455	u_int32_t param_len;
8456	struct scsi_per_res_out *cdb;
8457	struct ctl_lun *lun;
8458	struct scsi_per_res_out_parms* param;
8459	struct ctl_softc *softc;
8460	uint32_t residx;
8461	uint64_t res_key, sa_res_key;
8462	uint8_t type;
8463	union ctl_ha_msg persis_io;
8464	int    i;
8465
8466	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8467
8468	retval = CTL_RETVAL_COMPLETE;
8469
8470	softc = control_softc;
8471
8472	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8473	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8474
8475	/*
8476	 * We only support whole-LUN scope.  The scope & type are ignored for
8477	 * register, register and ignore existing key and clear.
8478	 * We sometimes ignore scope and type on preempts too!!
8479	 * Verify reservation type here as well.
8480	 */
8481	type = cdb->scope_type & SPR_TYPE_MASK;
8482	if ((cdb->action == SPRO_RESERVE)
8483	 || (cdb->action == SPRO_RELEASE)) {
8484		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8485			ctl_set_invalid_field(/*ctsio*/ ctsio,
8486					      /*sks_valid*/ 1,
8487					      /*command*/ 1,
8488					      /*field*/ 2,
8489					      /*bit_valid*/ 1,
8490					      /*bit*/ 4);
8491			ctl_done((union ctl_io *)ctsio);
8492			return (CTL_RETVAL_COMPLETE);
8493		}
8494
8495		if (type>8 || type==2 || type==4 || type==0) {
8496			ctl_set_invalid_field(/*ctsio*/ ctsio,
8497					      /*sks_valid*/ 1,
8498					      /*command*/ 1,
8499					      /*field*/ 2,
8500					      /*bit_valid*/ 1,
8501					      /*bit*/ 0);
8502			ctl_done((union ctl_io *)ctsio);
8503			return (CTL_RETVAL_COMPLETE);
8504		}
8505	}
8506
8507	param_len = scsi_4btoul(cdb->length);
8508
8509	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8510		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8511		ctsio->kern_data_len = param_len;
8512		ctsio->kern_total_len = param_len;
8513		ctsio->kern_data_resid = 0;
8514		ctsio->kern_rel_offset = 0;
8515		ctsio->kern_sg_entries = 0;
8516		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8517		ctsio->be_move_done = ctl_config_move_done;
8518		ctl_datamove((union ctl_io *)ctsio);
8519
8520		return (CTL_RETVAL_COMPLETE);
8521	}
8522
8523	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8524
8525	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8526	res_key = scsi_8btou64(param->res_key.key);
8527	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8528
8529	/*
8530	 * Validate the reservation key here except for SPRO_REG_IGNO
8531	 * This must be done for all other service actions
8532	 */
8533	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8534		mtx_lock(&lun->lun_lock);
8535		if (lun->per_res[residx].registered) {
8536		    if (memcmp(param->res_key.key,
8537			       lun->per_res[residx].res_key.key,
8538			       ctl_min(sizeof(param->res_key),
8539			       sizeof(lun->per_res[residx].res_key))) != 0) {
8540				/*
8541				 * The current key passed in doesn't match
8542				 * the one the initiator previously
8543				 * registered.
8544				 */
8545				mtx_unlock(&lun->lun_lock);
8546				free(ctsio->kern_data_ptr, M_CTL);
8547				ctl_set_reservation_conflict(ctsio);
8548				ctl_done((union ctl_io *)ctsio);
8549				return (CTL_RETVAL_COMPLETE);
8550			}
8551		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8552			/*
8553			 * We are not registered
8554			 */
8555			mtx_unlock(&lun->lun_lock);
8556			free(ctsio->kern_data_ptr, M_CTL);
8557			ctl_set_reservation_conflict(ctsio);
8558			ctl_done((union ctl_io *)ctsio);
8559			return (CTL_RETVAL_COMPLETE);
8560		} else if (res_key != 0) {
8561			/*
8562			 * We are not registered and trying to register but
8563			 * the register key isn't zero.
8564			 */
8565			mtx_unlock(&lun->lun_lock);
8566			free(ctsio->kern_data_ptr, M_CTL);
8567			ctl_set_reservation_conflict(ctsio);
8568			ctl_done((union ctl_io *)ctsio);
8569			return (CTL_RETVAL_COMPLETE);
8570		}
8571		mtx_unlock(&lun->lun_lock);
8572	}
8573
8574	switch (cdb->action & SPRO_ACTION_MASK) {
8575	case SPRO_REGISTER:
8576	case SPRO_REG_IGNO: {
8577
8578#if 0
8579		printf("Registration received\n");
8580#endif
8581
8582		/*
8583		 * We don't support any of these options, as we report in
8584		 * the read capabilities request (see
8585		 * ctl_persistent_reserve_in(), above).
8586		 */
8587		if ((param->flags & SPR_SPEC_I_PT)
8588		 || (param->flags & SPR_ALL_TG_PT)
8589		 || (param->flags & SPR_APTPL)) {
8590			int bit_ptr;
8591
8592			if (param->flags & SPR_APTPL)
8593				bit_ptr = 0;
8594			else if (param->flags & SPR_ALL_TG_PT)
8595				bit_ptr = 2;
8596			else /* SPR_SPEC_I_PT */
8597				bit_ptr = 3;
8598
8599			free(ctsio->kern_data_ptr, M_CTL);
8600			ctl_set_invalid_field(ctsio,
8601					      /*sks_valid*/ 1,
8602					      /*command*/ 0,
8603					      /*field*/ 20,
8604					      /*bit_valid*/ 1,
8605					      /*bit*/ bit_ptr);
8606			ctl_done((union ctl_io *)ctsio);
8607			return (CTL_RETVAL_COMPLETE);
8608		}
8609
8610		mtx_lock(&lun->lun_lock);
8611
8612		/*
8613		 * The initiator wants to clear the
8614		 * key/unregister.
8615		 */
8616		if (sa_res_key == 0) {
8617			if ((res_key == 0
8618			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8619			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8620			  && !lun->per_res[residx].registered)) {
8621				mtx_unlock(&lun->lun_lock);
8622				goto done;
8623			}
8624
8625			lun->per_res[residx].registered = 0;
8626			memset(&lun->per_res[residx].res_key,
8627			       0, sizeof(lun->per_res[residx].res_key));
8628			lun->pr_key_count--;
8629
8630			if (residx == lun->pr_res_idx) {
8631				lun->flags &= ~CTL_LUN_PR_RESERVED;
8632				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8633
8634				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8635				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8636				 && lun->pr_key_count) {
8637					/*
8638					 * If the reservation is a registrants
8639					 * only type we need to generate a UA
8640					 * for other registered inits.  The
8641					 * sense code should be RESERVATIONS
8642					 * RELEASED
8643					 */
8644
8645					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8646						if (lun->per_res[
8647						    i+persis_offset].registered
8648						    == 0)
8649							continue;
8650						lun->pending_ua[i] |=
8651							CTL_UA_RES_RELEASE;
8652					}
8653				}
8654				lun->res_type = 0;
8655			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8656				if (lun->pr_key_count==0) {
8657					lun->flags &= ~CTL_LUN_PR_RESERVED;
8658					lun->res_type = 0;
8659					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8660				}
8661			}
8662			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8663			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8664			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8665			persis_io.pr.pr_info.residx = residx;
8666			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8667			     &persis_io, sizeof(persis_io), 0 )) >
8668			     CTL_HA_STATUS_SUCCESS) {
8669				printf("CTL:Persis Out error returned from "
8670				       "ctl_ha_msg_send %d\n", isc_retval);
8671			}
8672		} else /* sa_res_key != 0 */ {
8673
8674			/*
8675			 * If we aren't registered currently then increment
8676			 * the key count and set the registered flag.
8677			 */
8678			if (!lun->per_res[residx].registered) {
8679				lun->pr_key_count++;
8680				lun->per_res[residx].registered = 1;
8681			}
8682
8683			memcpy(&lun->per_res[residx].res_key,
8684			       param->serv_act_res_key,
8685			       ctl_min(sizeof(param->serv_act_res_key),
8686			       sizeof(lun->per_res[residx].res_key)));
8687
8688			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8689			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8690			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8691			persis_io.pr.pr_info.residx = residx;
8692			memcpy(persis_io.pr.pr_info.sa_res_key,
8693			       param->serv_act_res_key,
8694			       sizeof(param->serv_act_res_key));
8695			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8696			     &persis_io, sizeof(persis_io), 0)) >
8697			     CTL_HA_STATUS_SUCCESS) {
8698				printf("CTL:Persis Out error returned from "
8699				       "ctl_ha_msg_send %d\n", isc_retval);
8700			}
8701		}
8702		lun->PRGeneration++;
8703		mtx_unlock(&lun->lun_lock);
8704
8705		break;
8706	}
8707	case SPRO_RESERVE:
8708#if 0
8709                printf("Reserve executed type %d\n", type);
8710#endif
8711		mtx_lock(&lun->lun_lock);
8712		if (lun->flags & CTL_LUN_PR_RESERVED) {
8713			/*
8714			 * if this isn't the reservation holder and it's
8715			 * not a "all registrants" type or if the type is
8716			 * different then we have a conflict
8717			 */
8718			if ((lun->pr_res_idx != residx
8719			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8720			 || lun->res_type != type) {
8721				mtx_unlock(&lun->lun_lock);
8722				free(ctsio->kern_data_ptr, M_CTL);
8723				ctl_set_reservation_conflict(ctsio);
8724				ctl_done((union ctl_io *)ctsio);
8725				return (CTL_RETVAL_COMPLETE);
8726			}
8727			mtx_unlock(&lun->lun_lock);
8728		} else /* create a reservation */ {
8729			/*
8730			 * If it's not an "all registrants" type record
8731			 * reservation holder
8732			 */
8733			if (type != SPR_TYPE_WR_EX_AR
8734			 && type != SPR_TYPE_EX_AC_AR)
8735				lun->pr_res_idx = residx; /* Res holder */
8736			else
8737				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8738
8739			lun->flags |= CTL_LUN_PR_RESERVED;
8740			lun->res_type = type;
8741
8742			mtx_unlock(&lun->lun_lock);
8743
8744			/* send msg to other side */
8745			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8746			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8747			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8748			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8749			persis_io.pr.pr_info.res_type = type;
8750			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8751			     &persis_io, sizeof(persis_io), 0)) >
8752			     CTL_HA_STATUS_SUCCESS) {
8753				printf("CTL:Persis Out error returned from "
8754				       "ctl_ha_msg_send %d\n", isc_retval);
8755			}
8756		}
8757		break;
8758
8759	case SPRO_RELEASE:
8760		mtx_lock(&lun->lun_lock);
8761		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8762			/* No reservation exists return good status */
8763			mtx_unlock(&lun->lun_lock);
8764			goto done;
8765		}
8766		/*
8767		 * Is this nexus a reservation holder?
8768		 */
8769		if (lun->pr_res_idx != residx
8770		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8771			/*
8772			 * not a res holder return good status but
8773			 * do nothing
8774			 */
8775			mtx_unlock(&lun->lun_lock);
8776			goto done;
8777		}
8778
8779		if (lun->res_type != type) {
8780			mtx_unlock(&lun->lun_lock);
8781			free(ctsio->kern_data_ptr, M_CTL);
8782			ctl_set_illegal_pr_release(ctsio);
8783			ctl_done((union ctl_io *)ctsio);
8784			return (CTL_RETVAL_COMPLETE);
8785		}
8786
8787		/* okay to release */
8788		lun->flags &= ~CTL_LUN_PR_RESERVED;
8789		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8790		lun->res_type = 0;
8791
8792		/*
8793		 * if this isn't an exclusive access
8794		 * res generate UA for all other
8795		 * registrants.
8796		 */
8797		if (type != SPR_TYPE_EX_AC
8798		 && type != SPR_TYPE_WR_EX) {
8799			/*
8800			 * temporarily unregister so we don't generate UA
8801			 */
8802			lun->per_res[residx].registered = 0;
8803
8804			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8805				if (lun->per_res[i+persis_offset].registered
8806				    == 0)
8807					continue;
8808				lun->pending_ua[i] |=
8809					CTL_UA_RES_RELEASE;
8810			}
8811
8812			lun->per_res[residx].registered = 1;
8813		}
8814		mtx_unlock(&lun->lun_lock);
8815		/* Send msg to other side */
8816		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8817		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8818		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8819		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8820		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8821			printf("CTL:Persis Out error returned from "
8822			       "ctl_ha_msg_send %d\n", isc_retval);
8823		}
8824		break;
8825
8826	case SPRO_CLEAR:
8827		/* send msg to other side */
8828
8829		mtx_lock(&lun->lun_lock);
8830		lun->flags &= ~CTL_LUN_PR_RESERVED;
8831		lun->res_type = 0;
8832		lun->pr_key_count = 0;
8833		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8834
8835
8836		memset(&lun->per_res[residx].res_key,
8837		       0, sizeof(lun->per_res[residx].res_key));
8838		lun->per_res[residx].registered = 0;
8839
8840		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8841			if (lun->per_res[i].registered) {
8842				if (!persis_offset && i < CTL_MAX_INITIATORS)
8843					lun->pending_ua[i] |=
8844						CTL_UA_RES_PREEMPT;
8845				else if (persis_offset && i >= persis_offset)
8846					lun->pending_ua[i-persis_offset] |=
8847					    CTL_UA_RES_PREEMPT;
8848
8849				memset(&lun->per_res[i].res_key,
8850				       0, sizeof(struct scsi_per_res_key));
8851				lun->per_res[i].registered = 0;
8852			}
8853		lun->PRGeneration++;
8854		mtx_unlock(&lun->lun_lock);
8855		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8856		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8857		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8858		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8859		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8860			printf("CTL:Persis Out error returned from "
8861			       "ctl_ha_msg_send %d\n", isc_retval);
8862		}
8863		break;
8864
8865	case SPRO_PREEMPT: {
8866		int nretval;
8867
8868		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8869					  residx, ctsio, cdb, param);
8870		if (nretval != 0)
8871			return (CTL_RETVAL_COMPLETE);
8872		break;
8873	}
8874	default:
8875		panic("Invalid PR type %x", cdb->action);
8876	}
8877
8878done:
8879	free(ctsio->kern_data_ptr, M_CTL);
8880	ctl_set_success(ctsio);
8881	ctl_done((union ctl_io *)ctsio);
8882
8883	return (retval);
8884}
8885
8886/*
8887 * This routine is for handling a message from the other SC pertaining to
8888 * persistent reserve out. All the error checking will have been done
8889 * so only perorming the action need be done here to keep the two
8890 * in sync.
8891 */
8892static void
8893ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8894{
8895	struct ctl_lun *lun;
8896	struct ctl_softc *softc;
8897	int i;
8898	uint32_t targ_lun;
8899
8900	softc = control_softc;
8901
8902	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8903	lun = softc->ctl_luns[targ_lun];
8904	mtx_lock(&lun->lun_lock);
8905	switch(msg->pr.pr_info.action) {
8906	case CTL_PR_REG_KEY:
8907		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8908			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8909			lun->pr_key_count++;
8910		}
8911		lun->PRGeneration++;
8912		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8913		       msg->pr.pr_info.sa_res_key,
8914		       sizeof(struct scsi_per_res_key));
8915		break;
8916
8917	case CTL_PR_UNREG_KEY:
8918		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8919		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8920		       0, sizeof(struct scsi_per_res_key));
8921		lun->pr_key_count--;
8922
8923		/* XXX Need to see if the reservation has been released */
8924		/* if so do we need to generate UA? */
8925		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8926			lun->flags &= ~CTL_LUN_PR_RESERVED;
8927			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8928
8929			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8930			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8931			 && lun->pr_key_count) {
8932				/*
8933				 * If the reservation is a registrants
8934				 * only type we need to generate a UA
8935				 * for other registered inits.  The
8936				 * sense code should be RESERVATIONS
8937				 * RELEASED
8938				 */
8939
8940				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8941					if (lun->per_res[i+
8942					    persis_offset].registered == 0)
8943						continue;
8944
8945					lun->pending_ua[i] |=
8946						CTL_UA_RES_RELEASE;
8947				}
8948			}
8949			lun->res_type = 0;
8950		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8951			if (lun->pr_key_count==0) {
8952				lun->flags &= ~CTL_LUN_PR_RESERVED;
8953				lun->res_type = 0;
8954				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8955			}
8956		}
8957		lun->PRGeneration++;
8958		break;
8959
8960	case CTL_PR_RESERVE:
8961		lun->flags |= CTL_LUN_PR_RESERVED;
8962		lun->res_type = msg->pr.pr_info.res_type;
8963		lun->pr_res_idx = msg->pr.pr_info.residx;
8964
8965		break;
8966
8967	case CTL_PR_RELEASE:
8968		/*
8969		 * if this isn't an exclusive access res generate UA for all
8970		 * other registrants.
8971		 */
8972		if (lun->res_type != SPR_TYPE_EX_AC
8973		 && lun->res_type != SPR_TYPE_WR_EX) {
8974			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8975				if (lun->per_res[i+persis_offset].registered)
8976					lun->pending_ua[i] |=
8977						CTL_UA_RES_RELEASE;
8978		}
8979
8980		lun->flags &= ~CTL_LUN_PR_RESERVED;
8981		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8982		lun->res_type = 0;
8983		break;
8984
8985	case CTL_PR_PREEMPT:
8986		ctl_pro_preempt_other(lun, msg);
8987		break;
8988	case CTL_PR_CLEAR:
8989		lun->flags &= ~CTL_LUN_PR_RESERVED;
8990		lun->res_type = 0;
8991		lun->pr_key_count = 0;
8992		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8993
8994		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8995			if (lun->per_res[i].registered == 0)
8996				continue;
8997			if (!persis_offset
8998			 && i < CTL_MAX_INITIATORS)
8999				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
9000			else if (persis_offset
9001			      && i >= persis_offset)
9002				lun->pending_ua[i-persis_offset] |=
9003					CTL_UA_RES_PREEMPT;
9004			memset(&lun->per_res[i].res_key, 0,
9005			       sizeof(struct scsi_per_res_key));
9006			lun->per_res[i].registered = 0;
9007		}
9008		lun->PRGeneration++;
9009		break;
9010	}
9011
9012	mtx_unlock(&lun->lun_lock);
9013}
9014
9015int
9016ctl_read_write(struct ctl_scsiio *ctsio)
9017{
9018	struct ctl_lun *lun;
9019	struct ctl_lba_len_flags *lbalen;
9020	uint64_t lba;
9021	uint32_t num_blocks;
9022	int flags, retval;
9023	int isread;
9024
9025	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9026
9027	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
9028
9029	flags = 0;
9030	retval = CTL_RETVAL_COMPLETE;
9031
9032	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
9033	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
9034	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
9035		uint32_t residx;
9036
9037		/*
9038		 * XXX KDM need a lock here.
9039		 */
9040		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9041		if ((lun->res_type == SPR_TYPE_EX_AC
9042		  && residx != lun->pr_res_idx)
9043		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
9044		   || lun->res_type == SPR_TYPE_EX_AC_AR)
9045		  && !lun->per_res[residx].registered)) {
9046			ctl_set_reservation_conflict(ctsio);
9047			ctl_done((union ctl_io *)ctsio);
9048			return (CTL_RETVAL_COMPLETE);
9049	        }
9050	}
9051
9052	switch (ctsio->cdb[0]) {
9053	case READ_6:
9054	case WRITE_6: {
9055		struct scsi_rw_6 *cdb;
9056
9057		cdb = (struct scsi_rw_6 *)ctsio->cdb;
9058
9059		lba = scsi_3btoul(cdb->addr);
9060		/* only 5 bits are valid in the most significant address byte */
9061		lba &= 0x1fffff;
9062		num_blocks = cdb->length;
9063		/*
9064		 * This is correct according to SBC-2.
9065		 */
9066		if (num_blocks == 0)
9067			num_blocks = 256;
9068		break;
9069	}
9070	case READ_10:
9071	case WRITE_10: {
9072		struct scsi_rw_10 *cdb;
9073
9074		cdb = (struct scsi_rw_10 *)ctsio->cdb;
9075		if (cdb->byte2 & SRW10_FUA)
9076			flags |= CTL_LLF_FUA;
9077		if (cdb->byte2 & SRW10_DPO)
9078			flags |= CTL_LLF_DPO;
9079		lba = scsi_4btoul(cdb->addr);
9080		num_blocks = scsi_2btoul(cdb->length);
9081		break;
9082	}
9083	case WRITE_VERIFY_10: {
9084		struct scsi_write_verify_10 *cdb;
9085
9086		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9087		flags |= CTL_LLF_FUA;
9088		if (cdb->byte2 & SWV_DPO)
9089			flags |= CTL_LLF_DPO;
9090		lba = scsi_4btoul(cdb->addr);
9091		num_blocks = scsi_2btoul(cdb->length);
9092		break;
9093	}
9094	case READ_12:
9095	case WRITE_12: {
9096		struct scsi_rw_12 *cdb;
9097
9098		cdb = (struct scsi_rw_12 *)ctsio->cdb;
9099		if (cdb->byte2 & SRW12_FUA)
9100			flags |= CTL_LLF_FUA;
9101		if (cdb->byte2 & SRW12_DPO)
9102			flags |= CTL_LLF_DPO;
9103		lba = scsi_4btoul(cdb->addr);
9104		num_blocks = scsi_4btoul(cdb->length);
9105		break;
9106	}
9107	case WRITE_VERIFY_12: {
9108		struct scsi_write_verify_12 *cdb;
9109
9110		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9111		flags |= CTL_LLF_FUA;
9112		if (cdb->byte2 & SWV_DPO)
9113			flags |= CTL_LLF_DPO;
9114		lba = scsi_4btoul(cdb->addr);
9115		num_blocks = scsi_4btoul(cdb->length);
9116		break;
9117	}
9118	case READ_16:
9119	case WRITE_16: {
9120		struct scsi_rw_16 *cdb;
9121
9122		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9123		if (cdb->byte2 & SRW12_FUA)
9124			flags |= CTL_LLF_FUA;
9125		if (cdb->byte2 & SRW12_DPO)
9126			flags |= CTL_LLF_DPO;
9127		lba = scsi_8btou64(cdb->addr);
9128		num_blocks = scsi_4btoul(cdb->length);
9129		break;
9130	}
9131	case WRITE_ATOMIC_16: {
9132		struct scsi_rw_16 *cdb;
9133
9134		if (lun->be_lun->atomicblock == 0) {
9135			ctl_set_invalid_opcode(ctsio);
9136			ctl_done((union ctl_io *)ctsio);
9137			return (CTL_RETVAL_COMPLETE);
9138		}
9139
9140		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9141		if (cdb->byte2 & SRW12_FUA)
9142			flags |= CTL_LLF_FUA;
9143		if (cdb->byte2 & SRW12_DPO)
9144			flags |= CTL_LLF_DPO;
9145		lba = scsi_8btou64(cdb->addr);
9146		num_blocks = scsi_4btoul(cdb->length);
9147		if (num_blocks > lun->be_lun->atomicblock) {
9148			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
9149			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
9150			    /*bit*/ 0);
9151			ctl_done((union ctl_io *)ctsio);
9152			return (CTL_RETVAL_COMPLETE);
9153		}
9154		break;
9155	}
9156	case WRITE_VERIFY_16: {
9157		struct scsi_write_verify_16 *cdb;
9158
9159		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9160		flags |= CTL_LLF_FUA;
9161		if (cdb->byte2 & SWV_DPO)
9162			flags |= CTL_LLF_DPO;
9163		lba = scsi_8btou64(cdb->addr);
9164		num_blocks = scsi_4btoul(cdb->length);
9165		break;
9166	}
9167	default:
9168		/*
9169		 * We got a command we don't support.  This shouldn't
9170		 * happen, commands should be filtered out above us.
9171		 */
9172		ctl_set_invalid_opcode(ctsio);
9173		ctl_done((union ctl_io *)ctsio);
9174
9175		return (CTL_RETVAL_COMPLETE);
9176		break; /* NOTREACHED */
9177	}
9178
9179	/*
9180	 * The first check is to make sure we're in bounds, the second
9181	 * check is to catch wrap-around problems.  If the lba + num blocks
9182	 * is less than the lba, then we've wrapped around and the block
9183	 * range is invalid anyway.
9184	 */
9185	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9186	 || ((lba + num_blocks) < lba)) {
9187		ctl_set_lba_out_of_range(ctsio);
9188		ctl_done((union ctl_io *)ctsio);
9189		return (CTL_RETVAL_COMPLETE);
9190	}
9191
9192	/*
9193	 * According to SBC-3, a transfer length of 0 is not an error.
9194	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9195	 * translates to 256 blocks for those commands.
9196	 */
9197	if (num_blocks == 0) {
9198		ctl_set_success(ctsio);
9199		ctl_done((union ctl_io *)ctsio);
9200		return (CTL_RETVAL_COMPLETE);
9201	}
9202
9203	/* Set FUA and/or DPO if caches are disabled. */
9204	if (isread) {
9205		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9206		    SCP_RCD) != 0)
9207			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9208	} else {
9209		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9210		    SCP_WCE) == 0)
9211			flags |= CTL_LLF_FUA;
9212	}
9213
9214	lbalen = (struct ctl_lba_len_flags *)
9215	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9216	lbalen->lba = lba;
9217	lbalen->len = num_blocks;
9218	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9219
9220	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9221	ctsio->kern_rel_offset = 0;
9222
9223	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9224
9225	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9226
9227	return (retval);
9228}
9229
9230static int
9231ctl_cnw_cont(union ctl_io *io)
9232{
9233	struct ctl_scsiio *ctsio;
9234	struct ctl_lun *lun;
9235	struct ctl_lba_len_flags *lbalen;
9236	int retval;
9237
9238	ctsio = &io->scsiio;
9239	ctsio->io_hdr.status = CTL_STATUS_NONE;
9240	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9241	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9242	lbalen = (struct ctl_lba_len_flags *)
9243	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9244	lbalen->flags &= ~CTL_LLF_COMPARE;
9245	lbalen->flags |= CTL_LLF_WRITE;
9246
9247	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9248	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9249	return (retval);
9250}
9251
9252int
9253ctl_cnw(struct ctl_scsiio *ctsio)
9254{
9255	struct ctl_lun *lun;
9256	struct ctl_lba_len_flags *lbalen;
9257	uint64_t lba;
9258	uint32_t num_blocks;
9259	int flags, retval;
9260
9261	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9262
9263	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9264
9265	flags = 0;
9266	retval = CTL_RETVAL_COMPLETE;
9267
9268	switch (ctsio->cdb[0]) {
9269	case COMPARE_AND_WRITE: {
9270		struct scsi_compare_and_write *cdb;
9271
9272		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9273		if (cdb->byte2 & SRW10_FUA)
9274			flags |= CTL_LLF_FUA;
9275		if (cdb->byte2 & SRW10_DPO)
9276			flags |= CTL_LLF_DPO;
9277		lba = scsi_8btou64(cdb->addr);
9278		num_blocks = cdb->length;
9279		break;
9280	}
9281	default:
9282		/*
9283		 * We got a command we don't support.  This shouldn't
9284		 * happen, commands should be filtered out above us.
9285		 */
9286		ctl_set_invalid_opcode(ctsio);
9287		ctl_done((union ctl_io *)ctsio);
9288
9289		return (CTL_RETVAL_COMPLETE);
9290		break; /* NOTREACHED */
9291	}
9292
9293	/*
9294	 * The first check is to make sure we're in bounds, the second
9295	 * check is to catch wrap-around problems.  If the lba + num blocks
9296	 * is less than the lba, then we've wrapped around and the block
9297	 * range is invalid anyway.
9298	 */
9299	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9300	 || ((lba + num_blocks) < lba)) {
9301		ctl_set_lba_out_of_range(ctsio);
9302		ctl_done((union ctl_io *)ctsio);
9303		return (CTL_RETVAL_COMPLETE);
9304	}
9305
9306	/*
9307	 * According to SBC-3, a transfer length of 0 is not an error.
9308	 */
9309	if (num_blocks == 0) {
9310		ctl_set_success(ctsio);
9311		ctl_done((union ctl_io *)ctsio);
9312		return (CTL_RETVAL_COMPLETE);
9313	}
9314
9315	/* Set FUA if write cache is disabled. */
9316	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9317	    SCP_WCE) == 0)
9318		flags |= CTL_LLF_FUA;
9319
9320	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9321	ctsio->kern_rel_offset = 0;
9322
9323	/*
9324	 * Set the IO_CONT flag, so that if this I/O gets passed to
9325	 * ctl_data_submit_done(), it'll get passed back to
9326	 * ctl_ctl_cnw_cont() for further processing.
9327	 */
9328	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9329	ctsio->io_cont = ctl_cnw_cont;
9330
9331	lbalen = (struct ctl_lba_len_flags *)
9332	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9333	lbalen->lba = lba;
9334	lbalen->len = num_blocks;
9335	lbalen->flags = CTL_LLF_COMPARE | flags;
9336
9337	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9338	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9339	return (retval);
9340}
9341
9342int
9343ctl_verify(struct ctl_scsiio *ctsio)
9344{
9345	struct ctl_lun *lun;
9346	struct ctl_lba_len_flags *lbalen;
9347	uint64_t lba;
9348	uint32_t num_blocks;
9349	int bytchk, flags;
9350	int retval;
9351
9352	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9353
9354	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9355
9356	bytchk = 0;
9357	flags = CTL_LLF_FUA;
9358	retval = CTL_RETVAL_COMPLETE;
9359
9360	switch (ctsio->cdb[0]) {
9361	case VERIFY_10: {
9362		struct scsi_verify_10 *cdb;
9363
9364		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9365		if (cdb->byte2 & SVFY_BYTCHK)
9366			bytchk = 1;
9367		if (cdb->byte2 & SVFY_DPO)
9368			flags |= CTL_LLF_DPO;
9369		lba = scsi_4btoul(cdb->addr);
9370		num_blocks = scsi_2btoul(cdb->length);
9371		break;
9372	}
9373	case VERIFY_12: {
9374		struct scsi_verify_12 *cdb;
9375
9376		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9377		if (cdb->byte2 & SVFY_BYTCHK)
9378			bytchk = 1;
9379		if (cdb->byte2 & SVFY_DPO)
9380			flags |= CTL_LLF_DPO;
9381		lba = scsi_4btoul(cdb->addr);
9382		num_blocks = scsi_4btoul(cdb->length);
9383		break;
9384	}
9385	case VERIFY_16: {
9386		struct scsi_rw_16 *cdb;
9387
9388		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9389		if (cdb->byte2 & SVFY_BYTCHK)
9390			bytchk = 1;
9391		if (cdb->byte2 & SVFY_DPO)
9392			flags |= CTL_LLF_DPO;
9393		lba = scsi_8btou64(cdb->addr);
9394		num_blocks = scsi_4btoul(cdb->length);
9395		break;
9396	}
9397	default:
9398		/*
9399		 * We got a command we don't support.  This shouldn't
9400		 * happen, commands should be filtered out above us.
9401		 */
9402		ctl_set_invalid_opcode(ctsio);
9403		ctl_done((union ctl_io *)ctsio);
9404		return (CTL_RETVAL_COMPLETE);
9405	}
9406
9407	/*
9408	 * The first check is to make sure we're in bounds, the second
9409	 * check is to catch wrap-around problems.  If the lba + num blocks
9410	 * is less than the lba, then we've wrapped around and the block
9411	 * range is invalid anyway.
9412	 */
9413	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9414	 || ((lba + num_blocks) < lba)) {
9415		ctl_set_lba_out_of_range(ctsio);
9416		ctl_done((union ctl_io *)ctsio);
9417		return (CTL_RETVAL_COMPLETE);
9418	}
9419
9420	/*
9421	 * According to SBC-3, a transfer length of 0 is not an error.
9422	 */
9423	if (num_blocks == 0) {
9424		ctl_set_success(ctsio);
9425		ctl_done((union ctl_io *)ctsio);
9426		return (CTL_RETVAL_COMPLETE);
9427	}
9428
9429	lbalen = (struct ctl_lba_len_flags *)
9430	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9431	lbalen->lba = lba;
9432	lbalen->len = num_blocks;
9433	if (bytchk) {
9434		lbalen->flags = CTL_LLF_COMPARE | flags;
9435		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9436	} else {
9437		lbalen->flags = CTL_LLF_VERIFY | flags;
9438		ctsio->kern_total_len = 0;
9439	}
9440	ctsio->kern_rel_offset = 0;
9441
9442	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9443	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9444	return (retval);
9445}
9446
9447int
9448ctl_report_luns(struct ctl_scsiio *ctsio)
9449{
9450	struct scsi_report_luns *cdb;
9451	struct scsi_report_luns_data *lun_data;
9452	struct ctl_lun *lun, *request_lun;
9453	int num_luns, retval;
9454	uint32_t alloc_len, lun_datalen;
9455	int num_filled, well_known;
9456	uint32_t initidx, targ_lun_id, lun_id;
9457
9458	retval = CTL_RETVAL_COMPLETE;
9459	well_known = 0;
9460
9461	cdb = (struct scsi_report_luns *)ctsio->cdb;
9462
9463	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9464
9465	mtx_lock(&control_softc->ctl_lock);
9466	num_luns = control_softc->num_luns;
9467	mtx_unlock(&control_softc->ctl_lock);
9468
9469	switch (cdb->select_report) {
9470	case RPL_REPORT_DEFAULT:
9471	case RPL_REPORT_ALL:
9472		break;
9473	case RPL_REPORT_WELLKNOWN:
9474		well_known = 1;
9475		num_luns = 0;
9476		break;
9477	default:
9478		ctl_set_invalid_field(ctsio,
9479				      /*sks_valid*/ 1,
9480				      /*command*/ 1,
9481				      /*field*/ 2,
9482				      /*bit_valid*/ 0,
9483				      /*bit*/ 0);
9484		ctl_done((union ctl_io *)ctsio);
9485		return (retval);
9486		break; /* NOTREACHED */
9487	}
9488
9489	alloc_len = scsi_4btoul(cdb->length);
9490	/*
9491	 * The initiator has to allocate at least 16 bytes for this request,
9492	 * so he can at least get the header and the first LUN.  Otherwise
9493	 * we reject the request (per SPC-3 rev 14, section 6.21).
9494	 */
9495	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9496	    sizeof(struct scsi_report_luns_lundata))) {
9497		ctl_set_invalid_field(ctsio,
9498				      /*sks_valid*/ 1,
9499				      /*command*/ 1,
9500				      /*field*/ 6,
9501				      /*bit_valid*/ 0,
9502				      /*bit*/ 0);
9503		ctl_done((union ctl_io *)ctsio);
9504		return (retval);
9505	}
9506
9507	request_lun = (struct ctl_lun *)
9508		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9509
9510	lun_datalen = sizeof(*lun_data) +
9511		(num_luns * sizeof(struct scsi_report_luns_lundata));
9512
9513	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9514	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9515	ctsio->kern_sg_entries = 0;
9516
9517	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9518
9519	mtx_lock(&control_softc->ctl_lock);
9520	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9521		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9522		if (lun_id >= CTL_MAX_LUNS)
9523			continue;
9524		lun = control_softc->ctl_luns[lun_id];
9525		if (lun == NULL)
9526			continue;
9527
9528		if (targ_lun_id <= 0xff) {
9529			/*
9530			 * Peripheral addressing method, bus number 0.
9531			 */
9532			lun_data->luns[num_filled].lundata[0] =
9533				RPL_LUNDATA_ATYP_PERIPH;
9534			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9535			num_filled++;
9536		} else if (targ_lun_id <= 0x3fff) {
9537			/*
9538			 * Flat addressing method.
9539			 */
9540			lun_data->luns[num_filled].lundata[0] =
9541				RPL_LUNDATA_ATYP_FLAT |
9542				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9543#ifdef OLDCTLHEADERS
9544				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9545				(targ_lun_id & SRLD_BUS_LUN_MASK);
9546#endif
9547			lun_data->luns[num_filled].lundata[1] =
9548#ifdef OLDCTLHEADERS
9549				targ_lun_id >> SRLD_BUS_LUN_BITS;
9550#endif
9551				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9552			num_filled++;
9553		} else {
9554			printf("ctl_report_luns: bogus LUN number %jd, "
9555			       "skipping\n", (intmax_t)targ_lun_id);
9556		}
9557		/*
9558		 * According to SPC-3, rev 14 section 6.21:
9559		 *
9560		 * "The execution of a REPORT LUNS command to any valid and
9561		 * installed logical unit shall clear the REPORTED LUNS DATA
9562		 * HAS CHANGED unit attention condition for all logical
9563		 * units of that target with respect to the requesting
9564		 * initiator. A valid and installed logical unit is one
9565		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9566		 * INQUIRY data (see 6.4.2)."
9567		 *
9568		 * If request_lun is NULL, the LUN this report luns command
9569		 * was issued to is either disabled or doesn't exist. In that
9570		 * case, we shouldn't clear any pending lun change unit
9571		 * attention.
9572		 */
9573		if (request_lun != NULL) {
9574			mtx_lock(&lun->lun_lock);
9575			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9576			mtx_unlock(&lun->lun_lock);
9577		}
9578	}
9579	mtx_unlock(&control_softc->ctl_lock);
9580
9581	/*
9582	 * It's quite possible that we've returned fewer LUNs than we allocated
9583	 * space for.  Trim it.
9584	 */
9585	lun_datalen = sizeof(*lun_data) +
9586		(num_filled * sizeof(struct scsi_report_luns_lundata));
9587
9588	if (lun_datalen < alloc_len) {
9589		ctsio->residual = alloc_len - lun_datalen;
9590		ctsio->kern_data_len = lun_datalen;
9591		ctsio->kern_total_len = lun_datalen;
9592	} else {
9593		ctsio->residual = 0;
9594		ctsio->kern_data_len = alloc_len;
9595		ctsio->kern_total_len = alloc_len;
9596	}
9597	ctsio->kern_data_resid = 0;
9598	ctsio->kern_rel_offset = 0;
9599	ctsio->kern_sg_entries = 0;
9600
9601	/*
9602	 * We set this to the actual data length, regardless of how much
9603	 * space we actually have to return results.  If the user looks at
9604	 * this value, he'll know whether or not he allocated enough space
9605	 * and reissue the command if necessary.  We don't support well
9606	 * known logical units, so if the user asks for that, return none.
9607	 */
9608	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9609
9610	/*
9611	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9612	 * this request.
9613	 */
9614	ctsio->scsi_status = SCSI_STATUS_OK;
9615
9616	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9617	ctsio->be_move_done = ctl_config_move_done;
9618	ctl_datamove((union ctl_io *)ctsio);
9619
9620	return (retval);
9621}
9622
9623int
9624ctl_request_sense(struct ctl_scsiio *ctsio)
9625{
9626	struct scsi_request_sense *cdb;
9627	struct scsi_sense_data *sense_ptr;
9628	struct ctl_lun *lun;
9629	uint32_t initidx;
9630	int have_error;
9631	scsi_sense_data_type sense_format;
9632
9633	cdb = (struct scsi_request_sense *)ctsio->cdb;
9634
9635	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9636
9637	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9638
9639	/*
9640	 * Determine which sense format the user wants.
9641	 */
9642	if (cdb->byte2 & SRS_DESC)
9643		sense_format = SSD_TYPE_DESC;
9644	else
9645		sense_format = SSD_TYPE_FIXED;
9646
9647	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9648	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9649	ctsio->kern_sg_entries = 0;
9650
9651	/*
9652	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9653	 * larger than the largest allowed value for the length field in the
9654	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9655	 */
9656	ctsio->residual = 0;
9657	ctsio->kern_data_len = cdb->length;
9658	ctsio->kern_total_len = cdb->length;
9659
9660	ctsio->kern_data_resid = 0;
9661	ctsio->kern_rel_offset = 0;
9662	ctsio->kern_sg_entries = 0;
9663
9664	/*
9665	 * If we don't have a LUN, we don't have any pending sense.
9666	 */
9667	if (lun == NULL)
9668		goto no_sense;
9669
9670	have_error = 0;
9671	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9672	/*
9673	 * Check for pending sense, and then for pending unit attentions.
9674	 * Pending sense gets returned first, then pending unit attentions.
9675	 */
9676	mtx_lock(&lun->lun_lock);
9677#ifdef CTL_WITH_CA
9678	if (ctl_is_set(lun->have_ca, initidx)) {
9679		scsi_sense_data_type stored_format;
9680
9681		/*
9682		 * Check to see which sense format was used for the stored
9683		 * sense data.
9684		 */
9685		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9686
9687		/*
9688		 * If the user requested a different sense format than the
9689		 * one we stored, then we need to convert it to the other
9690		 * format.  If we're going from descriptor to fixed format
9691		 * sense data, we may lose things in translation, depending
9692		 * on what options were used.
9693		 *
9694		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9695		 * for some reason we'll just copy it out as-is.
9696		 */
9697		if ((stored_format == SSD_TYPE_FIXED)
9698		 && (sense_format == SSD_TYPE_DESC))
9699			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9700			    &lun->pending_sense[initidx],
9701			    (struct scsi_sense_data_desc *)sense_ptr);
9702		else if ((stored_format == SSD_TYPE_DESC)
9703		      && (sense_format == SSD_TYPE_FIXED))
9704			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9705			    &lun->pending_sense[initidx],
9706			    (struct scsi_sense_data_fixed *)sense_ptr);
9707		else
9708			memcpy(sense_ptr, &lun->pending_sense[initidx],
9709			       ctl_min(sizeof(*sense_ptr),
9710			       sizeof(lun->pending_sense[initidx])));
9711
9712		ctl_clear_mask(lun->have_ca, initidx);
9713		have_error = 1;
9714	} else
9715#endif
9716	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9717		ctl_ua_type ua_type;
9718
9719		ua_type = ctl_build_ua(&lun->pending_ua[initidx],
9720				       sense_ptr, sense_format);
9721		if (ua_type != CTL_UA_NONE)
9722			have_error = 1;
9723	}
9724	mtx_unlock(&lun->lun_lock);
9725
9726	/*
9727	 * We already have a pending error, return it.
9728	 */
9729	if (have_error != 0) {
9730		/*
9731		 * We report the SCSI status as OK, since the status of the
9732		 * request sense command itself is OK.
9733		 */
9734		ctsio->scsi_status = SCSI_STATUS_OK;
9735
9736		/*
9737		 * We report 0 for the sense length, because we aren't doing
9738		 * autosense in this case.  We're reporting sense as
9739		 * parameter data.
9740		 */
9741		ctsio->sense_len = 0;
9742		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9743		ctsio->be_move_done = ctl_config_move_done;
9744		ctl_datamove((union ctl_io *)ctsio);
9745
9746		return (CTL_RETVAL_COMPLETE);
9747	}
9748
9749no_sense:
9750
9751	/*
9752	 * No sense information to report, so we report that everything is
9753	 * okay.
9754	 */
9755	ctl_set_sense_data(sense_ptr,
9756			   lun,
9757			   sense_format,
9758			   /*current_error*/ 1,
9759			   /*sense_key*/ SSD_KEY_NO_SENSE,
9760			   /*asc*/ 0x00,
9761			   /*ascq*/ 0x00,
9762			   SSD_ELEM_NONE);
9763
9764	ctsio->scsi_status = SCSI_STATUS_OK;
9765
9766	/*
9767	 * We report 0 for the sense length, because we aren't doing
9768	 * autosense in this case.  We're reporting sense as parameter data.
9769	 */
9770	ctsio->sense_len = 0;
9771	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9772	ctsio->be_move_done = ctl_config_move_done;
9773	ctl_datamove((union ctl_io *)ctsio);
9774
9775	return (CTL_RETVAL_COMPLETE);
9776}
9777
9778int
9779ctl_tur(struct ctl_scsiio *ctsio)
9780{
9781	struct ctl_lun *lun;
9782
9783	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9784
9785	CTL_DEBUG_PRINT(("ctl_tur\n"));
9786
9787	if (lun == NULL)
9788		return (EINVAL);
9789
9790	ctsio->scsi_status = SCSI_STATUS_OK;
9791	ctsio->io_hdr.status = CTL_SUCCESS;
9792
9793	ctl_done((union ctl_io *)ctsio);
9794
9795	return (CTL_RETVAL_COMPLETE);
9796}
9797
9798#ifdef notyet
9799static int
9800ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9801{
9802
9803}
9804#endif
9805
9806static int
9807ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9808{
9809	struct scsi_vpd_supported_pages *pages;
9810	int sup_page_size;
9811	struct ctl_lun *lun;
9812
9813	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9814
9815	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9816	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9817	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9818	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9819	ctsio->kern_sg_entries = 0;
9820
9821	if (sup_page_size < alloc_len) {
9822		ctsio->residual = alloc_len - sup_page_size;
9823		ctsio->kern_data_len = sup_page_size;
9824		ctsio->kern_total_len = sup_page_size;
9825	} else {
9826		ctsio->residual = 0;
9827		ctsio->kern_data_len = alloc_len;
9828		ctsio->kern_total_len = alloc_len;
9829	}
9830	ctsio->kern_data_resid = 0;
9831	ctsio->kern_rel_offset = 0;
9832	ctsio->kern_sg_entries = 0;
9833
9834	/*
9835	 * The control device is always connected.  The disk device, on the
9836	 * other hand, may not be online all the time.  Need to change this
9837	 * to figure out whether the disk device is actually online or not.
9838	 */
9839	if (lun != NULL)
9840		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9841				lun->be_lun->lun_type;
9842	else
9843		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9844
9845	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9846	/* Supported VPD pages */
9847	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9848	/* Serial Number */
9849	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9850	/* Device Identification */
9851	pages->page_list[2] = SVPD_DEVICE_ID;
9852	/* Extended INQUIRY Data */
9853	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9854	/* Mode Page Policy */
9855	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9856	/* SCSI Ports */
9857	pages->page_list[5] = SVPD_SCSI_PORTS;
9858	/* Third-party Copy */
9859	pages->page_list[6] = SVPD_SCSI_TPC;
9860	/* Block limits */
9861	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9862	/* Block Device Characteristics */
9863	pages->page_list[8] = SVPD_BDC;
9864	/* Logical Block Provisioning */
9865	pages->page_list[9] = SVPD_LBP;
9866
9867	ctsio->scsi_status = SCSI_STATUS_OK;
9868
9869	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9870	ctsio->be_move_done = ctl_config_move_done;
9871	ctl_datamove((union ctl_io *)ctsio);
9872
9873	return (CTL_RETVAL_COMPLETE);
9874}
9875
9876static int
9877ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9878{
9879	struct scsi_vpd_unit_serial_number *sn_ptr;
9880	struct ctl_lun *lun;
9881
9882	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9883
9884	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9885	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9886	ctsio->kern_sg_entries = 0;
9887
9888	if (sizeof(*sn_ptr) < alloc_len) {
9889		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9890		ctsio->kern_data_len = sizeof(*sn_ptr);
9891		ctsio->kern_total_len = sizeof(*sn_ptr);
9892	} else {
9893		ctsio->residual = 0;
9894		ctsio->kern_data_len = alloc_len;
9895		ctsio->kern_total_len = alloc_len;
9896	}
9897	ctsio->kern_data_resid = 0;
9898	ctsio->kern_rel_offset = 0;
9899	ctsio->kern_sg_entries = 0;
9900
9901	/*
9902	 * The control device is always connected.  The disk device, on the
9903	 * other hand, may not be online all the time.  Need to change this
9904	 * to figure out whether the disk device is actually online or not.
9905	 */
9906	if (lun != NULL)
9907		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9908				  lun->be_lun->lun_type;
9909	else
9910		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9911
9912	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9913	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9914	/*
9915	 * If we don't have a LUN, we just leave the serial number as
9916	 * all spaces.
9917	 */
9918	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9919	if (lun != NULL) {
9920		strncpy((char *)sn_ptr->serial_num,
9921			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9922	}
9923	ctsio->scsi_status = SCSI_STATUS_OK;
9924
9925	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9926	ctsio->be_move_done = ctl_config_move_done;
9927	ctl_datamove((union ctl_io *)ctsio);
9928
9929	return (CTL_RETVAL_COMPLETE);
9930}
9931
9932
9933static int
9934ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9935{
9936	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9937	struct ctl_lun *lun;
9938	int data_len;
9939
9940	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9941
9942	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9943	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9944	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9945	ctsio->kern_sg_entries = 0;
9946
9947	if (data_len < alloc_len) {
9948		ctsio->residual = alloc_len - data_len;
9949		ctsio->kern_data_len = data_len;
9950		ctsio->kern_total_len = data_len;
9951	} else {
9952		ctsio->residual = 0;
9953		ctsio->kern_data_len = alloc_len;
9954		ctsio->kern_total_len = alloc_len;
9955	}
9956	ctsio->kern_data_resid = 0;
9957	ctsio->kern_rel_offset = 0;
9958	ctsio->kern_sg_entries = 0;
9959
9960	/*
9961	 * The control device is always connected.  The disk device, on the
9962	 * other hand, may not be online all the time.
9963	 */
9964	if (lun != NULL)
9965		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9966				     lun->be_lun->lun_type;
9967	else
9968		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9969	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9970	eid_ptr->page_length = data_len - 4;
9971	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9972	eid_ptr->flags3 = SVPD_EID_V_SUP;
9973
9974	ctsio->scsi_status = SCSI_STATUS_OK;
9975	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9976	ctsio->be_move_done = ctl_config_move_done;
9977	ctl_datamove((union ctl_io *)ctsio);
9978
9979	return (CTL_RETVAL_COMPLETE);
9980}
9981
9982static int
9983ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9984{
9985	struct scsi_vpd_mode_page_policy *mpp_ptr;
9986	struct ctl_lun *lun;
9987	int data_len;
9988
9989	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9990
9991	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9992	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9993
9994	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9995	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9996	ctsio->kern_sg_entries = 0;
9997
9998	if (data_len < alloc_len) {
9999		ctsio->residual = alloc_len - data_len;
10000		ctsio->kern_data_len = data_len;
10001		ctsio->kern_total_len = data_len;
10002	} else {
10003		ctsio->residual = 0;
10004		ctsio->kern_data_len = alloc_len;
10005		ctsio->kern_total_len = alloc_len;
10006	}
10007	ctsio->kern_data_resid = 0;
10008	ctsio->kern_rel_offset = 0;
10009	ctsio->kern_sg_entries = 0;
10010
10011	/*
10012	 * The control device is always connected.  The disk device, on the
10013	 * other hand, may not be online all the time.
10014	 */
10015	if (lun != NULL)
10016		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10017				     lun->be_lun->lun_type;
10018	else
10019		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10020	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
10021	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
10022	mpp_ptr->descr[0].page_code = 0x3f;
10023	mpp_ptr->descr[0].subpage_code = 0xff;
10024	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
10025
10026	ctsio->scsi_status = SCSI_STATUS_OK;
10027	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10028	ctsio->be_move_done = ctl_config_move_done;
10029	ctl_datamove((union ctl_io *)ctsio);
10030
10031	return (CTL_RETVAL_COMPLETE);
10032}
10033
10034static int
10035ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
10036{
10037	struct scsi_vpd_device_id *devid_ptr;
10038	struct scsi_vpd_id_descriptor *desc;
10039	struct ctl_softc *ctl_softc;
10040	struct ctl_lun *lun;
10041	struct ctl_port *port;
10042	int data_len;
10043	uint8_t proto;
10044
10045	ctl_softc = control_softc;
10046
10047	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
10048	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10049
10050	data_len = sizeof(struct scsi_vpd_device_id) +
10051	    sizeof(struct scsi_vpd_id_descriptor) +
10052		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
10053	    sizeof(struct scsi_vpd_id_descriptor) +
10054		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
10055	if (lun && lun->lun_devid)
10056		data_len += lun->lun_devid->len;
10057	if (port->port_devid)
10058		data_len += port->port_devid->len;
10059	if (port->target_devid)
10060		data_len += port->target_devid->len;
10061
10062	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10063	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
10064	ctsio->kern_sg_entries = 0;
10065
10066	if (data_len < alloc_len) {
10067		ctsio->residual = alloc_len - data_len;
10068		ctsio->kern_data_len = data_len;
10069		ctsio->kern_total_len = data_len;
10070	} else {
10071		ctsio->residual = 0;
10072		ctsio->kern_data_len = alloc_len;
10073		ctsio->kern_total_len = alloc_len;
10074	}
10075	ctsio->kern_data_resid = 0;
10076	ctsio->kern_rel_offset = 0;
10077	ctsio->kern_sg_entries = 0;
10078
10079	/*
10080	 * The control device is always connected.  The disk device, on the
10081	 * other hand, may not be online all the time.
10082	 */
10083	if (lun != NULL)
10084		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10085				     lun->be_lun->lun_type;
10086	else
10087		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10088	devid_ptr->page_code = SVPD_DEVICE_ID;
10089	scsi_ulto2b(data_len - 4, devid_ptr->length);
10090
10091	if (port->port_type == CTL_PORT_FC)
10092		proto = SCSI_PROTO_FC << 4;
10093	else if (port->port_type == CTL_PORT_ISCSI)
10094		proto = SCSI_PROTO_ISCSI << 4;
10095	else
10096		proto = SCSI_PROTO_SPI << 4;
10097	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
10098
10099	/*
10100	 * We're using a LUN association here.  i.e., this device ID is a
10101	 * per-LUN identifier.
10102	 */
10103	if (lun && lun->lun_devid) {
10104		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
10105		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10106		    lun->lun_devid->len);
10107	}
10108
10109	/*
10110	 * This is for the WWPN which is a port association.
10111	 */
10112	if (port->port_devid) {
10113		memcpy(desc, port->port_devid->data, port->port_devid->len);
10114		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10115		    port->port_devid->len);
10116	}
10117
10118	/*
10119	 * This is for the Relative Target Port(type 4h) identifier
10120	 */
10121	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10122	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10123	    SVPD_ID_TYPE_RELTARG;
10124	desc->length = 4;
10125	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
10126	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10127	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
10128
10129	/*
10130	 * This is for the Target Port Group(type 5h) identifier
10131	 */
10132	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10133	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10134	    SVPD_ID_TYPE_TPORTGRP;
10135	desc->length = 4;
10136	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
10137	    &desc->identifier[2]);
10138	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10139	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10140
10141	/*
10142	 * This is for the Target identifier
10143	 */
10144	if (port->target_devid) {
10145		memcpy(desc, port->target_devid->data, port->target_devid->len);
10146	}
10147
10148	ctsio->scsi_status = SCSI_STATUS_OK;
10149	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10150	ctsio->be_move_done = ctl_config_move_done;
10151	ctl_datamove((union ctl_io *)ctsio);
10152
10153	return (CTL_RETVAL_COMPLETE);
10154}
10155
10156static int
10157ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10158{
10159	struct ctl_softc *softc = control_softc;
10160	struct scsi_vpd_scsi_ports *sp;
10161	struct scsi_vpd_port_designation *pd;
10162	struct scsi_vpd_port_designation_cont *pdc;
10163	struct ctl_lun *lun;
10164	struct ctl_port *port;
10165	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10166	int num_target_port_groups, single;
10167
10168	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10169
10170	single = ctl_is_single;
10171	if (single)
10172		num_target_port_groups = 1;
10173	else
10174		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10175	num_target_ports = 0;
10176	iid_len = 0;
10177	id_len = 0;
10178	mtx_lock(&softc->ctl_lock);
10179	STAILQ_FOREACH(port, &softc->port_list, links) {
10180		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10181			continue;
10182		if (lun != NULL &&
10183		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10184		    CTL_MAX_LUNS)
10185			continue;
10186		num_target_ports++;
10187		if (port->init_devid)
10188			iid_len += port->init_devid->len;
10189		if (port->port_devid)
10190			id_len += port->port_devid->len;
10191	}
10192	mtx_unlock(&softc->ctl_lock);
10193
10194	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10195	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10196	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10197	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10198	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10199	ctsio->kern_sg_entries = 0;
10200
10201	if (data_len < alloc_len) {
10202		ctsio->residual = alloc_len - data_len;
10203		ctsio->kern_data_len = data_len;
10204		ctsio->kern_total_len = data_len;
10205	} else {
10206		ctsio->residual = 0;
10207		ctsio->kern_data_len = alloc_len;
10208		ctsio->kern_total_len = alloc_len;
10209	}
10210	ctsio->kern_data_resid = 0;
10211	ctsio->kern_rel_offset = 0;
10212	ctsio->kern_sg_entries = 0;
10213
10214	/*
10215	 * The control device is always connected.  The disk device, on the
10216	 * other hand, may not be online all the time.  Need to change this
10217	 * to figure out whether the disk device is actually online or not.
10218	 */
10219	if (lun != NULL)
10220		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10221				  lun->be_lun->lun_type;
10222	else
10223		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10224
10225	sp->page_code = SVPD_SCSI_PORTS;
10226	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10227	    sp->page_length);
10228	pd = &sp->design[0];
10229
10230	mtx_lock(&softc->ctl_lock);
10231	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10232		pg = 0;
10233	else
10234		pg = 1;
10235	for (g = 0; g < num_target_port_groups; g++) {
10236		STAILQ_FOREACH(port, &softc->port_list, links) {
10237			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10238				continue;
10239			if (lun != NULL &&
10240			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10241			    CTL_MAX_LUNS)
10242				continue;
10243			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10244			scsi_ulto2b(p, pd->relative_port_id);
10245			if (port->init_devid && g == pg) {
10246				iid_len = port->init_devid->len;
10247				memcpy(pd->initiator_transportid,
10248				    port->init_devid->data, port->init_devid->len);
10249			} else
10250				iid_len = 0;
10251			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10252			pdc = (struct scsi_vpd_port_designation_cont *)
10253			    (&pd->initiator_transportid[iid_len]);
10254			if (port->port_devid && g == pg) {
10255				id_len = port->port_devid->len;
10256				memcpy(pdc->target_port_descriptors,
10257				    port->port_devid->data, port->port_devid->len);
10258			} else
10259				id_len = 0;
10260			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10261			pd = (struct scsi_vpd_port_designation *)
10262			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10263		}
10264	}
10265	mtx_unlock(&softc->ctl_lock);
10266
10267	ctsio->scsi_status = SCSI_STATUS_OK;
10268	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10269	ctsio->be_move_done = ctl_config_move_done;
10270	ctl_datamove((union ctl_io *)ctsio);
10271
10272	return (CTL_RETVAL_COMPLETE);
10273}
10274
10275static int
10276ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10277{
10278	struct scsi_vpd_block_limits *bl_ptr;
10279	struct ctl_lun *lun;
10280	int bs;
10281
10282	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10283
10284	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10285	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10286	ctsio->kern_sg_entries = 0;
10287
10288	if (sizeof(*bl_ptr) < alloc_len) {
10289		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10290		ctsio->kern_data_len = sizeof(*bl_ptr);
10291		ctsio->kern_total_len = sizeof(*bl_ptr);
10292	} else {
10293		ctsio->residual = 0;
10294		ctsio->kern_data_len = alloc_len;
10295		ctsio->kern_total_len = alloc_len;
10296	}
10297	ctsio->kern_data_resid = 0;
10298	ctsio->kern_rel_offset = 0;
10299	ctsio->kern_sg_entries = 0;
10300
10301	/*
10302	 * The control device is always connected.  The disk device, on the
10303	 * other hand, may not be online all the time.  Need to change this
10304	 * to figure out whether the disk device is actually online or not.
10305	 */
10306	if (lun != NULL)
10307		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10308				  lun->be_lun->lun_type;
10309	else
10310		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10311
10312	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10313	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
10314	bl_ptr->max_cmp_write_len = 0xff;
10315	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10316	if (lun != NULL) {
10317		bs = lun->be_lun->blocksize;
10318		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10319		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10320			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10321			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10322			if (lun->be_lun->pblockexp != 0) {
10323				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10324				    bl_ptr->opt_unmap_grain);
10325				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10326				    bl_ptr->unmap_grain_align);
10327			}
10328		}
10329		scsi_ulto4b(lun->be_lun->atomicblock,
10330		    bl_ptr->max_atomic_transfer_length);
10331		scsi_ulto4b(0, bl_ptr->atomic_alignment);
10332		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10333	}
10334	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10335
10336	ctsio->scsi_status = SCSI_STATUS_OK;
10337	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10338	ctsio->be_move_done = ctl_config_move_done;
10339	ctl_datamove((union ctl_io *)ctsio);
10340
10341	return (CTL_RETVAL_COMPLETE);
10342}
10343
10344static int
10345ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10346{
10347	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10348	struct ctl_lun *lun;
10349
10350	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10351
10352	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10353	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10354	ctsio->kern_sg_entries = 0;
10355
10356	if (sizeof(*bdc_ptr) < alloc_len) {
10357		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10358		ctsio->kern_data_len = sizeof(*bdc_ptr);
10359		ctsio->kern_total_len = sizeof(*bdc_ptr);
10360	} else {
10361		ctsio->residual = 0;
10362		ctsio->kern_data_len = alloc_len;
10363		ctsio->kern_total_len = alloc_len;
10364	}
10365	ctsio->kern_data_resid = 0;
10366	ctsio->kern_rel_offset = 0;
10367	ctsio->kern_sg_entries = 0;
10368
10369	/*
10370	 * The control device is always connected.  The disk device, on the
10371	 * other hand, may not be online all the time.  Need to change this
10372	 * to figure out whether the disk device is actually online or not.
10373	 */
10374	if (lun != NULL)
10375		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10376				  lun->be_lun->lun_type;
10377	else
10378		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10379	bdc_ptr->page_code = SVPD_BDC;
10380	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10381	scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10382	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10383
10384	ctsio->scsi_status = SCSI_STATUS_OK;
10385	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10386	ctsio->be_move_done = ctl_config_move_done;
10387	ctl_datamove((union ctl_io *)ctsio);
10388
10389	return (CTL_RETVAL_COMPLETE);
10390}
10391
10392static int
10393ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10394{
10395	struct scsi_vpd_logical_block_prov *lbp_ptr;
10396	struct ctl_lun *lun;
10397
10398	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10399
10400	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10401	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10402	ctsio->kern_sg_entries = 0;
10403
10404	if (sizeof(*lbp_ptr) < alloc_len) {
10405		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10406		ctsio->kern_data_len = sizeof(*lbp_ptr);
10407		ctsio->kern_total_len = sizeof(*lbp_ptr);
10408	} else {
10409		ctsio->residual = 0;
10410		ctsio->kern_data_len = alloc_len;
10411		ctsio->kern_total_len = alloc_len;
10412	}
10413	ctsio->kern_data_resid = 0;
10414	ctsio->kern_rel_offset = 0;
10415	ctsio->kern_sg_entries = 0;
10416
10417	/*
10418	 * The control device is always connected.  The disk device, on the
10419	 * other hand, may not be online all the time.  Need to change this
10420	 * to figure out whether the disk device is actually online or not.
10421	 */
10422	if (lun != NULL)
10423		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10424				  lun->be_lun->lun_type;
10425	else
10426		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10427
10428	lbp_ptr->page_code = SVPD_LBP;
10429	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10430	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10431		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10432		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10433		lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10434	}
10435
10436	ctsio->scsi_status = SCSI_STATUS_OK;
10437	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10438	ctsio->be_move_done = ctl_config_move_done;
10439	ctl_datamove((union ctl_io *)ctsio);
10440
10441	return (CTL_RETVAL_COMPLETE);
10442}
10443
10444static int
10445ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10446{
10447	struct scsi_inquiry *cdb;
10448	struct ctl_lun *lun;
10449	int alloc_len, retval;
10450
10451	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10452	cdb = (struct scsi_inquiry *)ctsio->cdb;
10453
10454	retval = CTL_RETVAL_COMPLETE;
10455
10456	alloc_len = scsi_2btoul(cdb->length);
10457
10458	switch (cdb->page_code) {
10459	case SVPD_SUPPORTED_PAGES:
10460		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10461		break;
10462	case SVPD_UNIT_SERIAL_NUMBER:
10463		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10464		break;
10465	case SVPD_DEVICE_ID:
10466		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10467		break;
10468	case SVPD_EXTENDED_INQUIRY_DATA:
10469		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10470		break;
10471	case SVPD_MODE_PAGE_POLICY:
10472		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10473		break;
10474	case SVPD_SCSI_PORTS:
10475		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10476		break;
10477	case SVPD_SCSI_TPC:
10478		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10479		break;
10480	case SVPD_BLOCK_LIMITS:
10481		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10482		break;
10483	case SVPD_BDC:
10484		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10485		break;
10486	case SVPD_LBP:
10487		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10488		break;
10489	default:
10490		ctl_set_invalid_field(ctsio,
10491				      /*sks_valid*/ 1,
10492				      /*command*/ 1,
10493				      /*field*/ 2,
10494				      /*bit_valid*/ 0,
10495				      /*bit*/ 0);
10496		ctl_done((union ctl_io *)ctsio);
10497		retval = CTL_RETVAL_COMPLETE;
10498		break;
10499	}
10500
10501	return (retval);
10502}
10503
10504static int
10505ctl_inquiry_std(struct ctl_scsiio *ctsio)
10506{
10507	struct scsi_inquiry_data *inq_ptr;
10508	struct scsi_inquiry *cdb;
10509	struct ctl_softc *ctl_softc;
10510	struct ctl_lun *lun;
10511	char *val;
10512	uint32_t alloc_len, data_len;
10513	ctl_port_type port_type;
10514
10515	ctl_softc = control_softc;
10516
10517	/*
10518	 * Figure out whether we're talking to a Fibre Channel port or not.
10519	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10520	 * SCSI front ends.
10521	 */
10522	port_type = ctl_softc->ctl_ports[
10523	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10524	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10525		port_type = CTL_PORT_SCSI;
10526
10527	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10528	cdb = (struct scsi_inquiry *)ctsio->cdb;
10529	alloc_len = scsi_2btoul(cdb->length);
10530
10531	/*
10532	 * We malloc the full inquiry data size here and fill it
10533	 * in.  If the user only asks for less, we'll give him
10534	 * that much.
10535	 */
10536	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10537	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10538	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10539	ctsio->kern_sg_entries = 0;
10540	ctsio->kern_data_resid = 0;
10541	ctsio->kern_rel_offset = 0;
10542
10543	if (data_len < alloc_len) {
10544		ctsio->residual = alloc_len - data_len;
10545		ctsio->kern_data_len = data_len;
10546		ctsio->kern_total_len = data_len;
10547	} else {
10548		ctsio->residual = 0;
10549		ctsio->kern_data_len = alloc_len;
10550		ctsio->kern_total_len = alloc_len;
10551	}
10552
10553	/*
10554	 * If we have a LUN configured, report it as connected.  Otherwise,
10555	 * report that it is offline or no device is supported, depending
10556	 * on the value of inquiry_pq_no_lun.
10557	 *
10558	 * According to the spec (SPC-4 r34), the peripheral qualifier
10559	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10560	 *
10561	 * "A peripheral device having the specified peripheral device type
10562	 * is not connected to this logical unit. However, the device
10563	 * server is capable of supporting the specified peripheral device
10564	 * type on this logical unit."
10565	 *
10566	 * According to the same spec, the peripheral qualifier
10567	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10568	 *
10569	 * "The device server is not capable of supporting a peripheral
10570	 * device on this logical unit. For this peripheral qualifier the
10571	 * peripheral device type shall be set to 1Fh. All other peripheral
10572	 * device type values are reserved for this peripheral qualifier."
10573	 *
10574	 * Given the text, it would seem that we probably want to report that
10575	 * the LUN is offline here.  There is no LUN connected, but we can
10576	 * support a LUN at the given LUN number.
10577	 *
10578	 * In the real world, though, it sounds like things are a little
10579	 * different:
10580	 *
10581	 * - Linux, when presented with a LUN with the offline peripheral
10582	 *   qualifier, will create an sg driver instance for it.  So when
10583	 *   you attach it to CTL, you wind up with a ton of sg driver
10584	 *   instances.  (One for every LUN that Linux bothered to probe.)
10585	 *   Linux does this despite the fact that it issues a REPORT LUNs
10586	 *   to LUN 0 to get the inventory of supported LUNs.
10587	 *
10588	 * - There is other anecdotal evidence (from Emulex folks) about
10589	 *   arrays that use the offline peripheral qualifier for LUNs that
10590	 *   are on the "passive" path in an active/passive array.
10591	 *
10592	 * So the solution is provide a hopefully reasonable default
10593	 * (return bad/no LUN) and allow the user to change the behavior
10594	 * with a tunable/sysctl variable.
10595	 */
10596	if (lun != NULL)
10597		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10598				  lun->be_lun->lun_type;
10599	else if (ctl_softc->inquiry_pq_no_lun == 0)
10600		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10601	else
10602		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10603
10604	/* RMB in byte 2 is 0 */
10605	inq_ptr->version = SCSI_REV_SPC4;
10606
10607	/*
10608	 * According to SAM-3, even if a device only supports a single
10609	 * level of LUN addressing, it should still set the HISUP bit:
10610	 *
10611	 * 4.9.1 Logical unit numbers overview
10612	 *
10613	 * All logical unit number formats described in this standard are
10614	 * hierarchical in structure even when only a single level in that
10615	 * hierarchy is used. The HISUP bit shall be set to one in the
10616	 * standard INQUIRY data (see SPC-2) when any logical unit number
10617	 * format described in this standard is used.  Non-hierarchical
10618	 * formats are outside the scope of this standard.
10619	 *
10620	 * Therefore we set the HiSup bit here.
10621	 *
10622	 * The reponse format is 2, per SPC-3.
10623	 */
10624	inq_ptr->response_format = SID_HiSup | 2;
10625
10626	inq_ptr->additional_length = data_len -
10627	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10628	CTL_DEBUG_PRINT(("additional_length = %d\n",
10629			 inq_ptr->additional_length));
10630
10631	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10632	/* 16 bit addressing */
10633	if (port_type == CTL_PORT_SCSI)
10634		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10635	/* XXX set the SID_MultiP bit here if we're actually going to
10636	   respond on multiple ports */
10637	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10638
10639	/* 16 bit data bus, synchronous transfers */
10640	if (port_type == CTL_PORT_SCSI)
10641		inq_ptr->flags = SID_WBus16 | SID_Sync;
10642	/*
10643	 * XXX KDM do we want to support tagged queueing on the control
10644	 * device at all?
10645	 */
10646	if ((lun == NULL)
10647	 || (lun->be_lun->lun_type != T_PROCESSOR))
10648		inq_ptr->flags |= SID_CmdQue;
10649	/*
10650	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10651	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10652	 * name and 4 bytes for the revision.
10653	 */
10654	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10655	    "vendor")) == NULL) {
10656		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10657	} else {
10658		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10659		strncpy(inq_ptr->vendor, val,
10660		    min(sizeof(inq_ptr->vendor), strlen(val)));
10661	}
10662	if (lun == NULL) {
10663		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10664		    sizeof(inq_ptr->product));
10665	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10666		switch (lun->be_lun->lun_type) {
10667		case T_DIRECT:
10668			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10669			    sizeof(inq_ptr->product));
10670			break;
10671		case T_PROCESSOR:
10672			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10673			    sizeof(inq_ptr->product));
10674			break;
10675		default:
10676			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10677			    sizeof(inq_ptr->product));
10678			break;
10679		}
10680	} else {
10681		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10682		strncpy(inq_ptr->product, val,
10683		    min(sizeof(inq_ptr->product), strlen(val)));
10684	}
10685
10686	/*
10687	 * XXX make this a macro somewhere so it automatically gets
10688	 * incremented when we make changes.
10689	 */
10690	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10691	    "revision")) == NULL) {
10692		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10693	} else {
10694		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10695		strncpy(inq_ptr->revision, val,
10696		    min(sizeof(inq_ptr->revision), strlen(val)));
10697	}
10698
10699	/*
10700	 * For parallel SCSI, we support double transition and single
10701	 * transition clocking.  We also support QAS (Quick Arbitration
10702	 * and Selection) and Information Unit transfers on both the
10703	 * control and array devices.
10704	 */
10705	if (port_type == CTL_PORT_SCSI)
10706		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10707				    SID_SPI_IUS;
10708
10709	/* SAM-5 (no version claimed) */
10710	scsi_ulto2b(0x00A0, inq_ptr->version1);
10711	/* SPC-4 (no version claimed) */
10712	scsi_ulto2b(0x0460, inq_ptr->version2);
10713	if (port_type == CTL_PORT_FC) {
10714		/* FCP-2 ANSI INCITS.350:2003 */
10715		scsi_ulto2b(0x0917, inq_ptr->version3);
10716	} else if (port_type == CTL_PORT_SCSI) {
10717		/* SPI-4 ANSI INCITS.362:200x */
10718		scsi_ulto2b(0x0B56, inq_ptr->version3);
10719	} else if (port_type == CTL_PORT_ISCSI) {
10720		/* iSCSI (no version claimed) */
10721		scsi_ulto2b(0x0960, inq_ptr->version3);
10722	} else if (port_type == CTL_PORT_SAS) {
10723		/* SAS (no version claimed) */
10724		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10725	}
10726
10727	if (lun == NULL) {
10728		/* SBC-4 (no version claimed) */
10729		scsi_ulto2b(0x0600, inq_ptr->version4);
10730	} else {
10731		switch (lun->be_lun->lun_type) {
10732		case T_DIRECT:
10733			/* SBC-4 (no version claimed) */
10734			scsi_ulto2b(0x0600, inq_ptr->version4);
10735			break;
10736		case T_PROCESSOR:
10737		default:
10738			break;
10739		}
10740	}
10741
10742	ctsio->scsi_status = SCSI_STATUS_OK;
10743	if (ctsio->kern_data_len > 0) {
10744		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10745		ctsio->be_move_done = ctl_config_move_done;
10746		ctl_datamove((union ctl_io *)ctsio);
10747	} else {
10748		ctsio->io_hdr.status = CTL_SUCCESS;
10749		ctl_done((union ctl_io *)ctsio);
10750	}
10751
10752	return (CTL_RETVAL_COMPLETE);
10753}
10754
10755int
10756ctl_inquiry(struct ctl_scsiio *ctsio)
10757{
10758	struct scsi_inquiry *cdb;
10759	int retval;
10760
10761	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10762
10763	cdb = (struct scsi_inquiry *)ctsio->cdb;
10764	if (cdb->byte2 & SI_EVPD)
10765		retval = ctl_inquiry_evpd(ctsio);
10766	else if (cdb->page_code == 0)
10767		retval = ctl_inquiry_std(ctsio);
10768	else {
10769		ctl_set_invalid_field(ctsio,
10770				      /*sks_valid*/ 1,
10771				      /*command*/ 1,
10772				      /*field*/ 2,
10773				      /*bit_valid*/ 0,
10774				      /*bit*/ 0);
10775		ctl_done((union ctl_io *)ctsio);
10776		return (CTL_RETVAL_COMPLETE);
10777	}
10778
10779	return (retval);
10780}
10781
10782/*
10783 * For known CDB types, parse the LBA and length.
10784 */
10785static int
10786ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10787{
10788	if (io->io_hdr.io_type != CTL_IO_SCSI)
10789		return (1);
10790
10791	switch (io->scsiio.cdb[0]) {
10792	case COMPARE_AND_WRITE: {
10793		struct scsi_compare_and_write *cdb;
10794
10795		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10796
10797		*lba = scsi_8btou64(cdb->addr);
10798		*len = cdb->length;
10799		break;
10800	}
10801	case READ_6:
10802	case WRITE_6: {
10803		struct scsi_rw_6 *cdb;
10804
10805		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10806
10807		*lba = scsi_3btoul(cdb->addr);
10808		/* only 5 bits are valid in the most significant address byte */
10809		*lba &= 0x1fffff;
10810		*len = cdb->length;
10811		break;
10812	}
10813	case READ_10:
10814	case WRITE_10: {
10815		struct scsi_rw_10 *cdb;
10816
10817		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10818
10819		*lba = scsi_4btoul(cdb->addr);
10820		*len = scsi_2btoul(cdb->length);
10821		break;
10822	}
10823	case WRITE_VERIFY_10: {
10824		struct scsi_write_verify_10 *cdb;
10825
10826		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10827
10828		*lba = scsi_4btoul(cdb->addr);
10829		*len = scsi_2btoul(cdb->length);
10830		break;
10831	}
10832	case READ_12:
10833	case WRITE_12: {
10834		struct scsi_rw_12 *cdb;
10835
10836		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10837
10838		*lba = scsi_4btoul(cdb->addr);
10839		*len = scsi_4btoul(cdb->length);
10840		break;
10841	}
10842	case WRITE_VERIFY_12: {
10843		struct scsi_write_verify_12 *cdb;
10844
10845		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10846
10847		*lba = scsi_4btoul(cdb->addr);
10848		*len = scsi_4btoul(cdb->length);
10849		break;
10850	}
10851	case READ_16:
10852	case WRITE_16:
10853	case WRITE_ATOMIC_16: {
10854		struct scsi_rw_16 *cdb;
10855
10856		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10857
10858		*lba = scsi_8btou64(cdb->addr);
10859		*len = scsi_4btoul(cdb->length);
10860		break;
10861	}
10862	case WRITE_VERIFY_16: {
10863		struct scsi_write_verify_16 *cdb;
10864
10865		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10866
10867		*lba = scsi_8btou64(cdb->addr);
10868		*len = scsi_4btoul(cdb->length);
10869		break;
10870	}
10871	case WRITE_SAME_10: {
10872		struct scsi_write_same_10 *cdb;
10873
10874		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10875
10876		*lba = scsi_4btoul(cdb->addr);
10877		*len = scsi_2btoul(cdb->length);
10878		break;
10879	}
10880	case WRITE_SAME_16: {
10881		struct scsi_write_same_16 *cdb;
10882
10883		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10884
10885		*lba = scsi_8btou64(cdb->addr);
10886		*len = scsi_4btoul(cdb->length);
10887		break;
10888	}
10889	case VERIFY_10: {
10890		struct scsi_verify_10 *cdb;
10891
10892		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10893
10894		*lba = scsi_4btoul(cdb->addr);
10895		*len = scsi_2btoul(cdb->length);
10896		break;
10897	}
10898	case VERIFY_12: {
10899		struct scsi_verify_12 *cdb;
10900
10901		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10902
10903		*lba = scsi_4btoul(cdb->addr);
10904		*len = scsi_4btoul(cdb->length);
10905		break;
10906	}
10907	case VERIFY_16: {
10908		struct scsi_verify_16 *cdb;
10909
10910		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10911
10912		*lba = scsi_8btou64(cdb->addr);
10913		*len = scsi_4btoul(cdb->length);
10914		break;
10915	}
10916	case UNMAP: {
10917		*lba = 0;
10918		*len = UINT64_MAX;
10919		break;
10920	}
10921	default:
10922		return (1);
10923		break; /* NOTREACHED */
10924	}
10925
10926	return (0);
10927}
10928
10929static ctl_action
10930ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10931{
10932	uint64_t endlba1, endlba2;
10933
10934	endlba1 = lba1 + len1 - 1;
10935	endlba2 = lba2 + len2 - 1;
10936
10937	if ((endlba1 < lba2)
10938	 || (endlba2 < lba1))
10939		return (CTL_ACTION_PASS);
10940	else
10941		return (CTL_ACTION_BLOCK);
10942}
10943
10944static int
10945ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10946{
10947	struct ctl_ptr_len_flags *ptrlen;
10948	struct scsi_unmap_desc *buf, *end, *range;
10949	uint64_t lba;
10950	uint32_t len;
10951
10952	/* If not UNMAP -- go other way. */
10953	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10954	    io->scsiio.cdb[0] != UNMAP)
10955		return (CTL_ACTION_ERROR);
10956
10957	/* If UNMAP without data -- block and wait for data. */
10958	ptrlen = (struct ctl_ptr_len_flags *)
10959	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10960	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10961	    ptrlen->ptr == NULL)
10962		return (CTL_ACTION_BLOCK);
10963
10964	/* UNMAP with data -- check for collision. */
10965	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10966	end = buf + ptrlen->len / sizeof(*buf);
10967	for (range = buf; range < end; range++) {
10968		lba = scsi_8btou64(range->lba);
10969		len = scsi_4btoul(range->length);
10970		if ((lba < lba2 + len2) && (lba + len > lba2))
10971			return (CTL_ACTION_BLOCK);
10972	}
10973	return (CTL_ACTION_PASS);
10974}
10975
10976static ctl_action
10977ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10978{
10979	uint64_t lba1, lba2;
10980	uint64_t len1, len2;
10981	int retval;
10982
10983	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10984		return (CTL_ACTION_ERROR);
10985
10986	retval = ctl_extent_check_unmap(io2, lba1, len1);
10987	if (retval != CTL_ACTION_ERROR)
10988		return (retval);
10989
10990	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10991		return (CTL_ACTION_ERROR);
10992
10993	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10994}
10995
10996static ctl_action
10997ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10998    union ctl_io *ooa_io)
10999{
11000	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
11001	ctl_serialize_action *serialize_row;
11002
11003	/*
11004	 * The initiator attempted multiple untagged commands at the same
11005	 * time.  Can't do that.
11006	 */
11007	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11008	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11009	 && ((pending_io->io_hdr.nexus.targ_port ==
11010	      ooa_io->io_hdr.nexus.targ_port)
11011	  && (pending_io->io_hdr.nexus.initid.id ==
11012	      ooa_io->io_hdr.nexus.initid.id))
11013	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
11014		return (CTL_ACTION_OVERLAP);
11015
11016	/*
11017	 * The initiator attempted to send multiple tagged commands with
11018	 * the same ID.  (It's fine if different initiators have the same
11019	 * tag ID.)
11020	 *
11021	 * Even if all of those conditions are true, we don't kill the I/O
11022	 * if the command ahead of us has been aborted.  We won't end up
11023	 * sending it to the FETD, and it's perfectly legal to resend a
11024	 * command with the same tag number as long as the previous
11025	 * instance of this tag number has been aborted somehow.
11026	 */
11027	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11028	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11029	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11030	 && ((pending_io->io_hdr.nexus.targ_port ==
11031	      ooa_io->io_hdr.nexus.targ_port)
11032	  && (pending_io->io_hdr.nexus.initid.id ==
11033	      ooa_io->io_hdr.nexus.initid.id))
11034	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
11035		return (CTL_ACTION_OVERLAP_TAG);
11036
11037	/*
11038	 * If we get a head of queue tag, SAM-3 says that we should
11039	 * immediately execute it.
11040	 *
11041	 * What happens if this command would normally block for some other
11042	 * reason?  e.g. a request sense with a head of queue tag
11043	 * immediately after a write.  Normally that would block, but this
11044	 * will result in its getting executed immediately...
11045	 *
11046	 * We currently return "pass" instead of "skip", so we'll end up
11047	 * going through the rest of the queue to check for overlapped tags.
11048	 *
11049	 * XXX KDM check for other types of blockage first??
11050	 */
11051	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11052		return (CTL_ACTION_PASS);
11053
11054	/*
11055	 * Ordered tags have to block until all items ahead of them
11056	 * have completed.  If we get called with an ordered tag, we always
11057	 * block, if something else is ahead of us in the queue.
11058	 */
11059	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11060		return (CTL_ACTION_BLOCK);
11061
11062	/*
11063	 * Simple tags get blocked until all head of queue and ordered tags
11064	 * ahead of them have completed.  I'm lumping untagged commands in
11065	 * with simple tags here.  XXX KDM is that the right thing to do?
11066	 */
11067	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11068	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11069	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11070	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11071		return (CTL_ACTION_BLOCK);
11072
11073	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11074	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11075
11076	serialize_row = ctl_serialize_table[ooa_entry->seridx];
11077
11078	switch (serialize_row[pending_entry->seridx]) {
11079	case CTL_SER_BLOCK:
11080		return (CTL_ACTION_BLOCK);
11081	case CTL_SER_EXTENT:
11082		return (ctl_extent_check(pending_io, ooa_io));
11083	case CTL_SER_EXTENTOPT:
11084		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11085		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11086			return (ctl_extent_check(pending_io, ooa_io));
11087		/* FALLTHROUGH */
11088	case CTL_SER_PASS:
11089		return (CTL_ACTION_PASS);
11090	case CTL_SER_BLOCKOPT:
11091		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11092		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11093			return (CTL_ACTION_BLOCK);
11094		return (CTL_ACTION_PASS);
11095	case CTL_SER_SKIP:
11096		return (CTL_ACTION_SKIP);
11097	default:
11098		panic("invalid serialization value %d",
11099		      serialize_row[pending_entry->seridx]);
11100	}
11101
11102	return (CTL_ACTION_ERROR);
11103}
11104
11105/*
11106 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11107 * Assumptions:
11108 * - pending_io is generally either incoming, or on the blocked queue
11109 * - starting I/O is the I/O we want to start the check with.
11110 */
11111static ctl_action
11112ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11113	      union ctl_io *starting_io)
11114{
11115	union ctl_io *ooa_io;
11116	ctl_action action;
11117
11118	mtx_assert(&lun->lun_lock, MA_OWNED);
11119
11120	/*
11121	 * Run back along the OOA queue, starting with the current
11122	 * blocked I/O and going through every I/O before it on the
11123	 * queue.  If starting_io is NULL, we'll just end up returning
11124	 * CTL_ACTION_PASS.
11125	 */
11126	for (ooa_io = starting_io; ooa_io != NULL;
11127	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11128	     ooa_links)){
11129
11130		/*
11131		 * This routine just checks to see whether
11132		 * cur_blocked is blocked by ooa_io, which is ahead
11133		 * of it in the queue.  It doesn't queue/dequeue
11134		 * cur_blocked.
11135		 */
11136		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11137		switch (action) {
11138		case CTL_ACTION_BLOCK:
11139		case CTL_ACTION_OVERLAP:
11140		case CTL_ACTION_OVERLAP_TAG:
11141		case CTL_ACTION_SKIP:
11142		case CTL_ACTION_ERROR:
11143			return (action);
11144			break; /* NOTREACHED */
11145		case CTL_ACTION_PASS:
11146			break;
11147		default:
11148			panic("invalid action %d", action);
11149			break;  /* NOTREACHED */
11150		}
11151	}
11152
11153	return (CTL_ACTION_PASS);
11154}
11155
11156/*
11157 * Assumptions:
11158 * - An I/O has just completed, and has been removed from the per-LUN OOA
11159 *   queue, so some items on the blocked queue may now be unblocked.
11160 */
11161static int
11162ctl_check_blocked(struct ctl_lun *lun)
11163{
11164	union ctl_io *cur_blocked, *next_blocked;
11165
11166	mtx_assert(&lun->lun_lock, MA_OWNED);
11167
11168	/*
11169	 * Run forward from the head of the blocked queue, checking each
11170	 * entry against the I/Os prior to it on the OOA queue to see if
11171	 * there is still any blockage.
11172	 *
11173	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11174	 * with our removing a variable on it while it is traversing the
11175	 * list.
11176	 */
11177	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11178	     cur_blocked != NULL; cur_blocked = next_blocked) {
11179		union ctl_io *prev_ooa;
11180		ctl_action action;
11181
11182		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11183							  blocked_links);
11184
11185		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11186						      ctl_ooaq, ooa_links);
11187
11188		/*
11189		 * If cur_blocked happens to be the first item in the OOA
11190		 * queue now, prev_ooa will be NULL, and the action
11191		 * returned will just be CTL_ACTION_PASS.
11192		 */
11193		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11194
11195		switch (action) {
11196		case CTL_ACTION_BLOCK:
11197			/* Nothing to do here, still blocked */
11198			break;
11199		case CTL_ACTION_OVERLAP:
11200		case CTL_ACTION_OVERLAP_TAG:
11201			/*
11202			 * This shouldn't happen!  In theory we've already
11203			 * checked this command for overlap...
11204			 */
11205			break;
11206		case CTL_ACTION_PASS:
11207		case CTL_ACTION_SKIP: {
11208			struct ctl_softc *softc;
11209			const struct ctl_cmd_entry *entry;
11210			uint32_t initidx;
11211			int isc_retval;
11212
11213			/*
11214			 * The skip case shouldn't happen, this transaction
11215			 * should have never made it onto the blocked queue.
11216			 */
11217			/*
11218			 * This I/O is no longer blocked, we can remove it
11219			 * from the blocked queue.  Since this is a TAILQ
11220			 * (doubly linked list), we can do O(1) removals
11221			 * from any place on the list.
11222			 */
11223			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11224				     blocked_links);
11225			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11226
11227			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11228				/*
11229				 * Need to send IO back to original side to
11230				 * run
11231				 */
11232				union ctl_ha_msg msg_info;
11233
11234				msg_info.hdr.original_sc =
11235					cur_blocked->io_hdr.original_sc;
11236				msg_info.hdr.serializing_sc = cur_blocked;
11237				msg_info.hdr.msg_type = CTL_MSG_R2R;
11238				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11239				     &msg_info, sizeof(msg_info), 0)) >
11240				     CTL_HA_STATUS_SUCCESS) {
11241					printf("CTL:Check Blocked error from "
11242					       "ctl_ha_msg_send %d\n",
11243					       isc_retval);
11244				}
11245				break;
11246			}
11247			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11248			softc = control_softc;
11249
11250			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11251
11252			/*
11253			 * Check this I/O for LUN state changes that may
11254			 * have happened while this command was blocked.
11255			 * The LUN state may have been changed by a command
11256			 * ahead of us in the queue, so we need to re-check
11257			 * for any states that can be caused by SCSI
11258			 * commands.
11259			 */
11260			if (ctl_scsiio_lun_check(softc, lun, entry,
11261						 &cur_blocked->scsiio) == 0) {
11262				cur_blocked->io_hdr.flags |=
11263				                      CTL_FLAG_IS_WAS_ON_RTR;
11264				ctl_enqueue_rtr(cur_blocked);
11265			} else
11266				ctl_done(cur_blocked);
11267			break;
11268		}
11269		default:
11270			/*
11271			 * This probably shouldn't happen -- we shouldn't
11272			 * get CTL_ACTION_ERROR, or anything else.
11273			 */
11274			break;
11275		}
11276	}
11277
11278	return (CTL_RETVAL_COMPLETE);
11279}
11280
11281/*
11282 * This routine (with one exception) checks LUN flags that can be set by
11283 * commands ahead of us in the OOA queue.  These flags have to be checked
11284 * when a command initially comes in, and when we pull a command off the
11285 * blocked queue and are preparing to execute it.  The reason we have to
11286 * check these flags for commands on the blocked queue is that the LUN
11287 * state may have been changed by a command ahead of us while we're on the
11288 * blocked queue.
11289 *
11290 * Ordering is somewhat important with these checks, so please pay
11291 * careful attention to the placement of any new checks.
11292 */
11293static int
11294ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11295    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11296{
11297	int retval;
11298	uint32_t residx;
11299
11300	retval = 0;
11301
11302	mtx_assert(&lun->lun_lock, MA_OWNED);
11303
11304	/*
11305	 * If this shelf is a secondary shelf controller, we have to reject
11306	 * any media access commands.
11307	 */
11308#if 0
11309	/* No longer needed for HA */
11310	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11311	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11312		ctl_set_lun_standby(ctsio);
11313		retval = 1;
11314		goto bailout;
11315	}
11316#endif
11317
11318	/*
11319	 * Check for a reservation conflict.  If this command isn't allowed
11320	 * even on reserved LUNs, and if this initiator isn't the one who
11321	 * reserved us, reject the command with a reservation conflict.
11322	 */
11323	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11324	if ((lun->flags & CTL_LUN_RESERVED)
11325	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11326		if (lun->res_idx != residx) {
11327			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11328			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11329			retval = 1;
11330			goto bailout;
11331		}
11332	}
11333
11334	if ((lun->flags & CTL_LUN_PR_RESERVED)
11335	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11336		/*
11337		 * if we aren't registered or it's a res holder type
11338		 * reservation and this isn't the res holder then set a
11339		 * conflict.
11340		 * NOTE: Commands which might be allowed on write exclusive
11341		 * type reservations are checked in the particular command
11342		 * for a conflict. Read and SSU are the only ones.
11343		 */
11344		if (!lun->per_res[residx].registered
11345		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11346			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11347			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11348			retval = 1;
11349			goto bailout;
11350		}
11351
11352	}
11353
11354	if ((lun->flags & CTL_LUN_OFFLINE)
11355	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11356		ctl_set_lun_not_ready(ctsio);
11357		retval = 1;
11358		goto bailout;
11359	}
11360
11361	/*
11362	 * If the LUN is stopped, see if this particular command is allowed
11363	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11364	 */
11365	if ((lun->flags & CTL_LUN_STOPPED)
11366	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11367		/* "Logical unit not ready, initializing cmd. required" */
11368		ctl_set_lun_stopped(ctsio);
11369		retval = 1;
11370		goto bailout;
11371	}
11372
11373	if ((lun->flags & CTL_LUN_INOPERABLE)
11374	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11375		/* "Medium format corrupted" */
11376		ctl_set_medium_format_corrupted(ctsio);
11377		retval = 1;
11378		goto bailout;
11379	}
11380
11381bailout:
11382	return (retval);
11383
11384}
11385
11386static void
11387ctl_failover_io(union ctl_io *io, int have_lock)
11388{
11389	ctl_set_busy(&io->scsiio);
11390	ctl_done(io);
11391}
11392
11393static void
11394ctl_failover(void)
11395{
11396	struct ctl_lun *lun;
11397	struct ctl_softc *ctl_softc;
11398	union ctl_io *next_io, *pending_io;
11399	union ctl_io *io;
11400	int lun_idx;
11401	int i;
11402
11403	ctl_softc = control_softc;
11404
11405	mtx_lock(&ctl_softc->ctl_lock);
11406	/*
11407	 * Remove any cmds from the other SC from the rtr queue.  These
11408	 * will obviously only be for LUNs for which we're the primary.
11409	 * We can't send status or get/send data for these commands.
11410	 * Since they haven't been executed yet, we can just remove them.
11411	 * We'll either abort them or delete them below, depending on
11412	 * which HA mode we're in.
11413	 */
11414#ifdef notyet
11415	mtx_lock(&ctl_softc->queue_lock);
11416	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11417	     io != NULL; io = next_io) {
11418		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11419		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11420			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11421				      ctl_io_hdr, links);
11422	}
11423	mtx_unlock(&ctl_softc->queue_lock);
11424#endif
11425
11426	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11427		lun = ctl_softc->ctl_luns[lun_idx];
11428		if (lun==NULL)
11429			continue;
11430
11431		/*
11432		 * Processor LUNs are primary on both sides.
11433		 * XXX will this always be true?
11434		 */
11435		if (lun->be_lun->lun_type == T_PROCESSOR)
11436			continue;
11437
11438		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11439		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11440			printf("FAILOVER: primary lun %d\n", lun_idx);
11441		        /*
11442			 * Remove all commands from the other SC. First from the
11443			 * blocked queue then from the ooa queue. Once we have
11444			 * removed them. Call ctl_check_blocked to see if there
11445			 * is anything that can run.
11446			 */
11447			for (io = (union ctl_io *)TAILQ_FIRST(
11448			     &lun->blocked_queue); io != NULL; io = next_io) {
11449
11450		        	next_io = (union ctl_io *)TAILQ_NEXT(
11451				    &io->io_hdr, blocked_links);
11452
11453				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11454					TAILQ_REMOVE(&lun->blocked_queue,
11455						     &io->io_hdr,blocked_links);
11456					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11457					TAILQ_REMOVE(&lun->ooa_queue,
11458						     &io->io_hdr, ooa_links);
11459
11460					ctl_free_io(io);
11461				}
11462			}
11463
11464			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11465	     		     io != NULL; io = next_io) {
11466
11467		        	next_io = (union ctl_io *)TAILQ_NEXT(
11468				    &io->io_hdr, ooa_links);
11469
11470				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11471
11472					TAILQ_REMOVE(&lun->ooa_queue,
11473						&io->io_hdr,
11474					     	ooa_links);
11475
11476					ctl_free_io(io);
11477				}
11478			}
11479			ctl_check_blocked(lun);
11480		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11481			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11482
11483			printf("FAILOVER: primary lun %d\n", lun_idx);
11484			/*
11485			 * Abort all commands from the other SC.  We can't
11486			 * send status back for them now.  These should get
11487			 * cleaned up when they are completed or come out
11488			 * for a datamove operation.
11489			 */
11490			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11491	     		     io != NULL; io = next_io) {
11492		        	next_io = (union ctl_io *)TAILQ_NEXT(
11493					&io->io_hdr, ooa_links);
11494
11495				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11496					io->io_hdr.flags |= CTL_FLAG_ABORT;
11497			}
11498		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11499			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11500
11501			printf("FAILOVER: secondary lun %d\n", lun_idx);
11502
11503			lun->flags |= CTL_LUN_PRIMARY_SC;
11504
11505			/*
11506			 * We send all I/O that was sent to this controller
11507			 * and redirected to the other side back with
11508			 * busy status, and have the initiator retry it.
11509			 * Figuring out how much data has been transferred,
11510			 * etc. and picking up where we left off would be
11511			 * very tricky.
11512			 *
11513			 * XXX KDM need to remove I/O from the blocked
11514			 * queue as well!
11515			 */
11516			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11517			     &lun->ooa_queue); pending_io != NULL;
11518			     pending_io = next_io) {
11519
11520				next_io =  (union ctl_io *)TAILQ_NEXT(
11521					&pending_io->io_hdr, ooa_links);
11522
11523				pending_io->io_hdr.flags &=
11524					~CTL_FLAG_SENT_2OTHER_SC;
11525
11526				if (pending_io->io_hdr.flags &
11527				    CTL_FLAG_IO_ACTIVE) {
11528					pending_io->io_hdr.flags |=
11529						CTL_FLAG_FAILOVER;
11530				} else {
11531					ctl_set_busy(&pending_io->scsiio);
11532					ctl_done(pending_io);
11533				}
11534			}
11535
11536			/*
11537			 * Build Unit Attention
11538			 */
11539			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11540				lun->pending_ua[i] |=
11541				                     CTL_UA_ASYM_ACC_CHANGE;
11542			}
11543		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11544			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11545			printf("FAILOVER: secondary lun %d\n", lun_idx);
11546			/*
11547			 * if the first io on the OOA is not on the RtR queue
11548			 * add it.
11549			 */
11550			lun->flags |= CTL_LUN_PRIMARY_SC;
11551
11552			pending_io = (union ctl_io *)TAILQ_FIRST(
11553			    &lun->ooa_queue);
11554			if (pending_io==NULL) {
11555				printf("Nothing on OOA queue\n");
11556				continue;
11557			}
11558
11559			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11560			if ((pending_io->io_hdr.flags &
11561			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11562				pending_io->io_hdr.flags |=
11563				    CTL_FLAG_IS_WAS_ON_RTR;
11564				ctl_enqueue_rtr(pending_io);
11565			}
11566#if 0
11567			else
11568			{
11569				printf("Tag 0x%04x is running\n",
11570				      pending_io->scsiio.tag_num);
11571			}
11572#endif
11573
11574			next_io = (union ctl_io *)TAILQ_NEXT(
11575			    &pending_io->io_hdr, ooa_links);
11576			for (pending_io=next_io; pending_io != NULL;
11577			     pending_io = next_io) {
11578				pending_io->io_hdr.flags &=
11579				    ~CTL_FLAG_SENT_2OTHER_SC;
11580				next_io = (union ctl_io *)TAILQ_NEXT(
11581					&pending_io->io_hdr, ooa_links);
11582				if (pending_io->io_hdr.flags &
11583				    CTL_FLAG_IS_WAS_ON_RTR) {
11584#if 0
11585				        printf("Tag 0x%04x is running\n",
11586				      		pending_io->scsiio.tag_num);
11587#endif
11588					continue;
11589				}
11590
11591				switch (ctl_check_ooa(lun, pending_io,
11592			            (union ctl_io *)TAILQ_PREV(
11593				    &pending_io->io_hdr, ctl_ooaq,
11594				    ooa_links))) {
11595
11596				case CTL_ACTION_BLOCK:
11597					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11598							  &pending_io->io_hdr,
11599							  blocked_links);
11600					pending_io->io_hdr.flags |=
11601					    CTL_FLAG_BLOCKED;
11602					break;
11603				case CTL_ACTION_PASS:
11604				case CTL_ACTION_SKIP:
11605					pending_io->io_hdr.flags |=
11606					    CTL_FLAG_IS_WAS_ON_RTR;
11607					ctl_enqueue_rtr(pending_io);
11608					break;
11609				case CTL_ACTION_OVERLAP:
11610					ctl_set_overlapped_cmd(
11611					    (struct ctl_scsiio *)pending_io);
11612					ctl_done(pending_io);
11613					break;
11614				case CTL_ACTION_OVERLAP_TAG:
11615					ctl_set_overlapped_tag(
11616					    (struct ctl_scsiio *)pending_io,
11617					    pending_io->scsiio.tag_num & 0xff);
11618					ctl_done(pending_io);
11619					break;
11620				case CTL_ACTION_ERROR:
11621				default:
11622					ctl_set_internal_failure(
11623						(struct ctl_scsiio *)pending_io,
11624						0,  // sks_valid
11625						0); //retry count
11626					ctl_done(pending_io);
11627					break;
11628				}
11629			}
11630
11631			/*
11632			 * Build Unit Attention
11633			 */
11634			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11635				lun->pending_ua[i] |=
11636				                     CTL_UA_ASYM_ACC_CHANGE;
11637			}
11638		} else {
11639			panic("Unhandled HA mode failover, LUN flags = %#x, "
11640			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11641		}
11642	}
11643	ctl_pause_rtr = 0;
11644	mtx_unlock(&ctl_softc->ctl_lock);
11645}
11646
11647static int
11648ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11649{
11650	struct ctl_lun *lun;
11651	const struct ctl_cmd_entry *entry;
11652	uint32_t initidx, targ_lun;
11653	int retval;
11654
11655	retval = 0;
11656
11657	lun = NULL;
11658
11659	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11660	if ((targ_lun < CTL_MAX_LUNS)
11661	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11662		lun = ctl_softc->ctl_luns[targ_lun];
11663		/*
11664		 * If the LUN is invalid, pretend that it doesn't exist.
11665		 * It will go away as soon as all pending I/O has been
11666		 * completed.
11667		 */
11668		if (lun->flags & CTL_LUN_DISABLED) {
11669			lun = NULL;
11670		} else {
11671			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11672			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11673				lun->be_lun;
11674			if (lun->be_lun->lun_type == T_PROCESSOR) {
11675				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11676			}
11677
11678			/*
11679			 * Every I/O goes into the OOA queue for a
11680			 * particular LUN, and stays there until completion.
11681			 */
11682			mtx_lock(&lun->lun_lock);
11683			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11684			    ooa_links);
11685		}
11686	} else {
11687		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11688		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11689	}
11690
11691	/* Get command entry and return error if it is unsuppotyed. */
11692	entry = ctl_validate_command(ctsio);
11693	if (entry == NULL) {
11694		if (lun)
11695			mtx_unlock(&lun->lun_lock);
11696		return (retval);
11697	}
11698
11699	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11700	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11701
11702	/*
11703	 * Check to see whether we can send this command to LUNs that don't
11704	 * exist.  This should pretty much only be the case for inquiry
11705	 * and request sense.  Further checks, below, really require having
11706	 * a LUN, so we can't really check the command anymore.  Just put
11707	 * it on the rtr queue.
11708	 */
11709	if (lun == NULL) {
11710		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11711			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11712			ctl_enqueue_rtr((union ctl_io *)ctsio);
11713			return (retval);
11714		}
11715
11716		ctl_set_unsupported_lun(ctsio);
11717		ctl_done((union ctl_io *)ctsio);
11718		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11719		return (retval);
11720	} else {
11721		/*
11722		 * Make sure we support this particular command on this LUN.
11723		 * e.g., we don't support writes to the control LUN.
11724		 */
11725		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11726			mtx_unlock(&lun->lun_lock);
11727			ctl_set_invalid_opcode(ctsio);
11728			ctl_done((union ctl_io *)ctsio);
11729			return (retval);
11730		}
11731	}
11732
11733	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11734
11735#ifdef CTL_WITH_CA
11736	/*
11737	 * If we've got a request sense, it'll clear the contingent
11738	 * allegiance condition.  Otherwise, if we have a CA condition for
11739	 * this initiator, clear it, because it sent down a command other
11740	 * than request sense.
11741	 */
11742	if ((ctsio->cdb[0] != REQUEST_SENSE)
11743	 && (ctl_is_set(lun->have_ca, initidx)))
11744		ctl_clear_mask(lun->have_ca, initidx);
11745#endif
11746
11747	/*
11748	 * If the command has this flag set, it handles its own unit
11749	 * attention reporting, we shouldn't do anything.  Otherwise we
11750	 * check for any pending unit attentions, and send them back to the
11751	 * initiator.  We only do this when a command initially comes in,
11752	 * not when we pull it off the blocked queue.
11753	 *
11754	 * According to SAM-3, section 5.3.2, the order that things get
11755	 * presented back to the host is basically unit attentions caused
11756	 * by some sort of reset event, busy status, reservation conflicts
11757	 * or task set full, and finally any other status.
11758	 *
11759	 * One issue here is that some of the unit attentions we report
11760	 * don't fall into the "reset" category (e.g. "reported luns data
11761	 * has changed").  So reporting it here, before the reservation
11762	 * check, may be technically wrong.  I guess the only thing to do
11763	 * would be to check for and report the reset events here, and then
11764	 * check for the other unit attention types after we check for a
11765	 * reservation conflict.
11766	 *
11767	 * XXX KDM need to fix this
11768	 */
11769	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11770		ctl_ua_type ua_type;
11771
11772		if (lun->pending_ua[initidx] != CTL_UA_NONE) {
11773			scsi_sense_data_type sense_format;
11774
11775			if (lun != NULL)
11776				sense_format = (lun->flags &
11777				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11778				    SSD_TYPE_FIXED;
11779			else
11780				sense_format = SSD_TYPE_FIXED;
11781
11782			ua_type = ctl_build_ua(&lun->pending_ua[initidx],
11783			    &ctsio->sense_data, sense_format);
11784			if (ua_type != CTL_UA_NONE) {
11785				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11786				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11787						       CTL_AUTOSENSE;
11788				ctsio->sense_len = SSD_FULL_SIZE;
11789				mtx_unlock(&lun->lun_lock);
11790				ctl_done((union ctl_io *)ctsio);
11791				return (retval);
11792			}
11793		}
11794	}
11795
11796
11797	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11798		mtx_unlock(&lun->lun_lock);
11799		ctl_done((union ctl_io *)ctsio);
11800		return (retval);
11801	}
11802
11803	/*
11804	 * XXX CHD this is where we want to send IO to other side if
11805	 * this LUN is secondary on this SC. We will need to make a copy
11806	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11807	 * the copy we send as FROM_OTHER.
11808	 * We also need to stuff the address of the original IO so we can
11809	 * find it easily. Something similar will need be done on the other
11810	 * side so when we are done we can find the copy.
11811	 */
11812	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11813		union ctl_ha_msg msg_info;
11814		int isc_retval;
11815
11816		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11817
11818		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11819		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11820#if 0
11821		printf("1. ctsio %p\n", ctsio);
11822#endif
11823		msg_info.hdr.serializing_sc = NULL;
11824		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11825		msg_info.scsi.tag_num = ctsio->tag_num;
11826		msg_info.scsi.tag_type = ctsio->tag_type;
11827		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11828
11829		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11830
11831		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11832		    (void *)&msg_info, sizeof(msg_info), 0)) >
11833		    CTL_HA_STATUS_SUCCESS) {
11834			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11835			       isc_retval);
11836			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11837		} else {
11838#if 0
11839			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11840#endif
11841		}
11842
11843		/*
11844		 * XXX KDM this I/O is off the incoming queue, but hasn't
11845		 * been inserted on any other queue.  We may need to come
11846		 * up with a holding queue while we wait for serialization
11847		 * so that we have an idea of what we're waiting for from
11848		 * the other side.
11849		 */
11850		mtx_unlock(&lun->lun_lock);
11851		return (retval);
11852	}
11853
11854	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11855			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11856			      ctl_ooaq, ooa_links))) {
11857	case CTL_ACTION_BLOCK:
11858		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11859		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11860				  blocked_links);
11861		mtx_unlock(&lun->lun_lock);
11862		return (retval);
11863	case CTL_ACTION_PASS:
11864	case CTL_ACTION_SKIP:
11865		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11866		mtx_unlock(&lun->lun_lock);
11867		ctl_enqueue_rtr((union ctl_io *)ctsio);
11868		break;
11869	case CTL_ACTION_OVERLAP:
11870		mtx_unlock(&lun->lun_lock);
11871		ctl_set_overlapped_cmd(ctsio);
11872		ctl_done((union ctl_io *)ctsio);
11873		break;
11874	case CTL_ACTION_OVERLAP_TAG:
11875		mtx_unlock(&lun->lun_lock);
11876		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11877		ctl_done((union ctl_io *)ctsio);
11878		break;
11879	case CTL_ACTION_ERROR:
11880	default:
11881		mtx_unlock(&lun->lun_lock);
11882		ctl_set_internal_failure(ctsio,
11883					 /*sks_valid*/ 0,
11884					 /*retry_count*/ 0);
11885		ctl_done((union ctl_io *)ctsio);
11886		break;
11887	}
11888	return (retval);
11889}
11890
11891const struct ctl_cmd_entry *
11892ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11893{
11894	const struct ctl_cmd_entry *entry;
11895	int service_action;
11896
11897	entry = &ctl_cmd_table[ctsio->cdb[0]];
11898	if (sa)
11899		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11900	if (entry->flags & CTL_CMD_FLAG_SA5) {
11901		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11902		entry = &((const struct ctl_cmd_entry *)
11903		    entry->execute)[service_action];
11904	}
11905	return (entry);
11906}
11907
11908const struct ctl_cmd_entry *
11909ctl_validate_command(struct ctl_scsiio *ctsio)
11910{
11911	const struct ctl_cmd_entry *entry;
11912	int i, sa;
11913	uint8_t diff;
11914
11915	entry = ctl_get_cmd_entry(ctsio, &sa);
11916	if (entry->execute == NULL) {
11917		if (sa)
11918			ctl_set_invalid_field(ctsio,
11919					      /*sks_valid*/ 1,
11920					      /*command*/ 1,
11921					      /*field*/ 1,
11922					      /*bit_valid*/ 1,
11923					      /*bit*/ 4);
11924		else
11925			ctl_set_invalid_opcode(ctsio);
11926		ctl_done((union ctl_io *)ctsio);
11927		return (NULL);
11928	}
11929	KASSERT(entry->length > 0,
11930	    ("Not defined length for command 0x%02x/0x%02x",
11931	     ctsio->cdb[0], ctsio->cdb[1]));
11932	for (i = 1; i < entry->length; i++) {
11933		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11934		if (diff == 0)
11935			continue;
11936		ctl_set_invalid_field(ctsio,
11937				      /*sks_valid*/ 1,
11938				      /*command*/ 1,
11939				      /*field*/ i,
11940				      /*bit_valid*/ 1,
11941				      /*bit*/ fls(diff) - 1);
11942		ctl_done((union ctl_io *)ctsio);
11943		return (NULL);
11944	}
11945	return (entry);
11946}
11947
11948static int
11949ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11950{
11951
11952	switch (lun_type) {
11953	case T_PROCESSOR:
11954		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11955		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11956			return (0);
11957		break;
11958	case T_DIRECT:
11959		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11960		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11961			return (0);
11962		break;
11963	default:
11964		return (0);
11965	}
11966	return (1);
11967}
11968
11969static int
11970ctl_scsiio(struct ctl_scsiio *ctsio)
11971{
11972	int retval;
11973	const struct ctl_cmd_entry *entry;
11974
11975	retval = CTL_RETVAL_COMPLETE;
11976
11977	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11978
11979	entry = ctl_get_cmd_entry(ctsio, NULL);
11980
11981	/*
11982	 * If this I/O has been aborted, just send it straight to
11983	 * ctl_done() without executing it.
11984	 */
11985	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11986		ctl_done((union ctl_io *)ctsio);
11987		goto bailout;
11988	}
11989
11990	/*
11991	 * All the checks should have been handled by ctl_scsiio_precheck().
11992	 * We should be clear now to just execute the I/O.
11993	 */
11994	retval = entry->execute(ctsio);
11995
11996bailout:
11997	return (retval);
11998}
11999
12000/*
12001 * Since we only implement one target right now, a bus reset simply resets
12002 * our single target.
12003 */
12004static int
12005ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
12006{
12007	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
12008}
12009
12010static int
12011ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
12012		 ctl_ua_type ua_type)
12013{
12014	struct ctl_lun *lun;
12015	int retval;
12016
12017	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12018		union ctl_ha_msg msg_info;
12019
12020		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
12021		msg_info.hdr.nexus = io->io_hdr.nexus;
12022		if (ua_type==CTL_UA_TARG_RESET)
12023			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
12024		else
12025			msg_info.task.task_action = CTL_TASK_BUS_RESET;
12026		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12027		msg_info.hdr.original_sc = NULL;
12028		msg_info.hdr.serializing_sc = NULL;
12029		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12030		    (void *)&msg_info, sizeof(msg_info), 0)) {
12031		}
12032	}
12033	retval = 0;
12034
12035	mtx_lock(&ctl_softc->ctl_lock);
12036	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
12037		retval += ctl_lun_reset(lun, io, ua_type);
12038	mtx_unlock(&ctl_softc->ctl_lock);
12039
12040	return (retval);
12041}
12042
12043/*
12044 * The LUN should always be set.  The I/O is optional, and is used to
12045 * distinguish between I/Os sent by this initiator, and by other
12046 * initiators.  We set unit attention for initiators other than this one.
12047 * SAM-3 is vague on this point.  It does say that a unit attention should
12048 * be established for other initiators when a LUN is reset (see section
12049 * 5.7.3), but it doesn't specifically say that the unit attention should
12050 * be established for this particular initiator when a LUN is reset.  Here
12051 * is the relevant text, from SAM-3 rev 8:
12052 *
12053 * 5.7.2 When a SCSI initiator port aborts its own tasks
12054 *
12055 * When a SCSI initiator port causes its own task(s) to be aborted, no
12056 * notification that the task(s) have been aborted shall be returned to
12057 * the SCSI initiator port other than the completion response for the
12058 * command or task management function action that caused the task(s) to
12059 * be aborted and notification(s) associated with related effects of the
12060 * action (e.g., a reset unit attention condition).
12061 *
12062 * XXX KDM for now, we're setting unit attention for all initiators.
12063 */
12064static int
12065ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
12066{
12067	union ctl_io *xio;
12068#if 0
12069	uint32_t initindex;
12070#endif
12071	int i;
12072
12073	mtx_lock(&lun->lun_lock);
12074	/*
12075	 * Run through the OOA queue and abort each I/O.
12076	 */
12077#if 0
12078	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12079#endif
12080	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12081	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12082		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
12083	}
12084
12085	/*
12086	 * This version sets unit attention for every
12087	 */
12088#if 0
12089	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12090	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12091		if (initindex == i)
12092			continue;
12093		lun->pending_ua[i] |= ua_type;
12094	}
12095#endif
12096
12097	/*
12098	 * A reset (any kind, really) clears reservations established with
12099	 * RESERVE/RELEASE.  It does not clear reservations established
12100	 * with PERSISTENT RESERVE OUT, but we don't support that at the
12101	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
12102	 * reservations made with the RESERVE/RELEASE commands, because
12103	 * those commands are obsolete in SPC-3.
12104	 */
12105	lun->flags &= ~CTL_LUN_RESERVED;
12106
12107	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12108#ifdef CTL_WITH_CA
12109		ctl_clear_mask(lun->have_ca, i);
12110#endif
12111		lun->pending_ua[i] |= ua_type;
12112	}
12113	mtx_unlock(&lun->lun_lock);
12114
12115	return (0);
12116}
12117
12118static void
12119ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
12120    int other_sc)
12121{
12122	union ctl_io *xio;
12123
12124	mtx_assert(&lun->lun_lock, MA_OWNED);
12125
12126	/*
12127	 * Run through the OOA queue and attempt to find the given I/O.
12128	 * The target port, initiator ID, tag type and tag number have to
12129	 * match the values that we got from the initiator.  If we have an
12130	 * untagged command to abort, simply abort the first untagged command
12131	 * we come to.  We only allow one untagged command at a time of course.
12132	 */
12133	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12134	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12135
12136		if ((targ_port == UINT32_MAX ||
12137		     targ_port == xio->io_hdr.nexus.targ_port) &&
12138		    (init_id == UINT32_MAX ||
12139		     init_id == xio->io_hdr.nexus.initid.id)) {
12140			if (targ_port != xio->io_hdr.nexus.targ_port ||
12141			    init_id != xio->io_hdr.nexus.initid.id)
12142				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12143			xio->io_hdr.flags |= CTL_FLAG_ABORT;
12144			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12145				union ctl_ha_msg msg_info;
12146
12147				msg_info.hdr.nexus = xio->io_hdr.nexus;
12148				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12149				msg_info.task.tag_num = xio->scsiio.tag_num;
12150				msg_info.task.tag_type = xio->scsiio.tag_type;
12151				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12152				msg_info.hdr.original_sc = NULL;
12153				msg_info.hdr.serializing_sc = NULL;
12154				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12155				    (void *)&msg_info, sizeof(msg_info), 0);
12156			}
12157		}
12158	}
12159}
12160
12161static int
12162ctl_abort_task_set(union ctl_io *io)
12163{
12164	struct ctl_softc *softc = control_softc;
12165	struct ctl_lun *lun;
12166	uint32_t targ_lun;
12167
12168	/*
12169	 * Look up the LUN.
12170	 */
12171	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12172	mtx_lock(&softc->ctl_lock);
12173	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12174		lun = softc->ctl_luns[targ_lun];
12175	else {
12176		mtx_unlock(&softc->ctl_lock);
12177		return (1);
12178	}
12179
12180	mtx_lock(&lun->lun_lock);
12181	mtx_unlock(&softc->ctl_lock);
12182	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12183		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12184		    io->io_hdr.nexus.initid.id,
12185		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12186	} else { /* CTL_TASK_CLEAR_TASK_SET */
12187		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12188		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12189	}
12190	mtx_unlock(&lun->lun_lock);
12191	return (0);
12192}
12193
12194static int
12195ctl_i_t_nexus_reset(union ctl_io *io)
12196{
12197	struct ctl_softc *softc = control_softc;
12198	struct ctl_lun *lun;
12199	uint32_t initindex, residx;
12200
12201	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12202	residx = ctl_get_resindex(&io->io_hdr.nexus);
12203	mtx_lock(&softc->ctl_lock);
12204	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12205		mtx_lock(&lun->lun_lock);
12206		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12207		    io->io_hdr.nexus.initid.id,
12208		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12209#ifdef CTL_WITH_CA
12210		ctl_clear_mask(lun->have_ca, initindex);
12211#endif
12212		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
12213			lun->flags &= ~CTL_LUN_RESERVED;
12214		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12215		mtx_unlock(&lun->lun_lock);
12216	}
12217	mtx_unlock(&softc->ctl_lock);
12218	return (0);
12219}
12220
12221static int
12222ctl_abort_task(union ctl_io *io)
12223{
12224	union ctl_io *xio;
12225	struct ctl_lun *lun;
12226	struct ctl_softc *ctl_softc;
12227#if 0
12228	struct sbuf sb;
12229	char printbuf[128];
12230#endif
12231	int found;
12232	uint32_t targ_lun;
12233
12234	ctl_softc = control_softc;
12235	found = 0;
12236
12237	/*
12238	 * Look up the LUN.
12239	 */
12240	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12241	mtx_lock(&ctl_softc->ctl_lock);
12242	if ((targ_lun < CTL_MAX_LUNS)
12243	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12244		lun = ctl_softc->ctl_luns[targ_lun];
12245	else {
12246		mtx_unlock(&ctl_softc->ctl_lock);
12247		return (1);
12248	}
12249
12250#if 0
12251	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12252	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12253#endif
12254
12255	mtx_lock(&lun->lun_lock);
12256	mtx_unlock(&ctl_softc->ctl_lock);
12257	/*
12258	 * Run through the OOA queue and attempt to find the given I/O.
12259	 * The target port, initiator ID, tag type and tag number have to
12260	 * match the values that we got from the initiator.  If we have an
12261	 * untagged command to abort, simply abort the first untagged command
12262	 * we come to.  We only allow one untagged command at a time of course.
12263	 */
12264#if 0
12265	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12266#endif
12267	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12268	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12269#if 0
12270		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12271
12272		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12273			    lun->lun, xio->scsiio.tag_num,
12274			    xio->scsiio.tag_type,
12275			    (xio->io_hdr.blocked_links.tqe_prev
12276			    == NULL) ? "" : " BLOCKED",
12277			    (xio->io_hdr.flags &
12278			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12279			    (xio->io_hdr.flags &
12280			    CTL_FLAG_ABORT) ? " ABORT" : "",
12281			    (xio->io_hdr.flags &
12282			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12283		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12284		sbuf_finish(&sb);
12285		printf("%s\n", sbuf_data(&sb));
12286#endif
12287
12288		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12289		 && (xio->io_hdr.nexus.initid.id ==
12290		     io->io_hdr.nexus.initid.id)) {
12291			/*
12292			 * If the abort says that the task is untagged, the
12293			 * task in the queue must be untagged.  Otherwise,
12294			 * we just check to see whether the tag numbers
12295			 * match.  This is because the QLogic firmware
12296			 * doesn't pass back the tag type in an abort
12297			 * request.
12298			 */
12299#if 0
12300			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12301			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12302			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12303#endif
12304			/*
12305			 * XXX KDM we've got problems with FC, because it
12306			 * doesn't send down a tag type with aborts.  So we
12307			 * can only really go by the tag number...
12308			 * This may cause problems with parallel SCSI.
12309			 * Need to figure that out!!
12310			 */
12311			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12312				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12313				found = 1;
12314				if ((io->io_hdr.flags &
12315				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12316				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12317					union ctl_ha_msg msg_info;
12318
12319					io->io_hdr.flags |=
12320					                CTL_FLAG_SENT_2OTHER_SC;
12321					msg_info.hdr.nexus = io->io_hdr.nexus;
12322					msg_info.task.task_action =
12323						CTL_TASK_ABORT_TASK;
12324					msg_info.task.tag_num =
12325						io->taskio.tag_num;
12326					msg_info.task.tag_type =
12327						io->taskio.tag_type;
12328					msg_info.hdr.msg_type =
12329						CTL_MSG_MANAGE_TASKS;
12330					msg_info.hdr.original_sc = NULL;
12331					msg_info.hdr.serializing_sc = NULL;
12332#if 0
12333					printf("Sent Abort to other side\n");
12334#endif
12335					if (CTL_HA_STATUS_SUCCESS !=
12336					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12337		    				(void *)&msg_info,
12338						sizeof(msg_info), 0)) {
12339					}
12340				}
12341#if 0
12342				printf("ctl_abort_task: found I/O to abort\n");
12343#endif
12344				break;
12345			}
12346		}
12347	}
12348	mtx_unlock(&lun->lun_lock);
12349
12350	if (found == 0) {
12351		/*
12352		 * This isn't really an error.  It's entirely possible for
12353		 * the abort and command completion to cross on the wire.
12354		 * This is more of an informative/diagnostic error.
12355		 */
12356#if 0
12357		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12358		       "%d:%d:%d:%d tag %d type %d\n",
12359		       io->io_hdr.nexus.initid.id,
12360		       io->io_hdr.nexus.targ_port,
12361		       io->io_hdr.nexus.targ_target.id,
12362		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12363		       io->taskio.tag_type);
12364#endif
12365	}
12366	return (0);
12367}
12368
12369static void
12370ctl_run_task(union ctl_io *io)
12371{
12372	struct ctl_softc *ctl_softc = control_softc;
12373	int retval = 1;
12374	const char *task_desc;
12375
12376	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12377
12378	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12379	    ("ctl_run_task: Unextected io_type %d\n",
12380	     io->io_hdr.io_type));
12381
12382	task_desc = ctl_scsi_task_string(&io->taskio);
12383	if (task_desc != NULL) {
12384#ifdef NEEDTOPORT
12385		csevent_log(CSC_CTL | CSC_SHELF_SW |
12386			    CTL_TASK_REPORT,
12387			    csevent_LogType_Trace,
12388			    csevent_Severity_Information,
12389			    csevent_AlertLevel_Green,
12390			    csevent_FRU_Firmware,
12391			    csevent_FRU_Unknown,
12392			    "CTL: received task: %s",task_desc);
12393#endif
12394	} else {
12395#ifdef NEEDTOPORT
12396		csevent_log(CSC_CTL | CSC_SHELF_SW |
12397			    CTL_TASK_REPORT,
12398			    csevent_LogType_Trace,
12399			    csevent_Severity_Information,
12400			    csevent_AlertLevel_Green,
12401			    csevent_FRU_Firmware,
12402			    csevent_FRU_Unknown,
12403			    "CTL: received unknown task "
12404			    "type: %d (%#x)",
12405			    io->taskio.task_action,
12406			    io->taskio.task_action);
12407#endif
12408	}
12409	switch (io->taskio.task_action) {
12410	case CTL_TASK_ABORT_TASK:
12411		retval = ctl_abort_task(io);
12412		break;
12413	case CTL_TASK_ABORT_TASK_SET:
12414	case CTL_TASK_CLEAR_TASK_SET:
12415		retval = ctl_abort_task_set(io);
12416		break;
12417	case CTL_TASK_CLEAR_ACA:
12418		break;
12419	case CTL_TASK_I_T_NEXUS_RESET:
12420		retval = ctl_i_t_nexus_reset(io);
12421		break;
12422	case CTL_TASK_LUN_RESET: {
12423		struct ctl_lun *lun;
12424		uint32_t targ_lun;
12425
12426		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12427		mtx_lock(&ctl_softc->ctl_lock);
12428		if ((targ_lun < CTL_MAX_LUNS)
12429		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12430			lun = ctl_softc->ctl_luns[targ_lun];
12431		else {
12432			mtx_unlock(&ctl_softc->ctl_lock);
12433			retval = 1;
12434			break;
12435		}
12436
12437		if (!(io->io_hdr.flags &
12438		    CTL_FLAG_FROM_OTHER_SC)) {
12439			union ctl_ha_msg msg_info;
12440
12441			io->io_hdr.flags |=
12442				CTL_FLAG_SENT_2OTHER_SC;
12443			msg_info.hdr.msg_type =
12444				CTL_MSG_MANAGE_TASKS;
12445			msg_info.hdr.nexus = io->io_hdr.nexus;
12446			msg_info.task.task_action =
12447				CTL_TASK_LUN_RESET;
12448			msg_info.hdr.original_sc = NULL;
12449			msg_info.hdr.serializing_sc = NULL;
12450			if (CTL_HA_STATUS_SUCCESS !=
12451			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12452			    (void *)&msg_info,
12453			    sizeof(msg_info), 0)) {
12454			}
12455		}
12456
12457		retval = ctl_lun_reset(lun, io,
12458				       CTL_UA_LUN_RESET);
12459		mtx_unlock(&ctl_softc->ctl_lock);
12460		break;
12461	}
12462	case CTL_TASK_TARGET_RESET:
12463		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12464		break;
12465	case CTL_TASK_BUS_RESET:
12466		retval = ctl_bus_reset(ctl_softc, io);
12467		break;
12468	case CTL_TASK_PORT_LOGIN:
12469		break;
12470	case CTL_TASK_PORT_LOGOUT:
12471		break;
12472	default:
12473		printf("ctl_run_task: got unknown task management event %d\n",
12474		       io->taskio.task_action);
12475		break;
12476	}
12477	if (retval == 0)
12478		io->io_hdr.status = CTL_SUCCESS;
12479	else
12480		io->io_hdr.status = CTL_ERROR;
12481	ctl_done(io);
12482}
12483
12484/*
12485 * For HA operation.  Handle commands that come in from the other
12486 * controller.
12487 */
12488static void
12489ctl_handle_isc(union ctl_io *io)
12490{
12491	int free_io;
12492	struct ctl_lun *lun;
12493	struct ctl_softc *ctl_softc;
12494	uint32_t targ_lun;
12495
12496	ctl_softc = control_softc;
12497
12498	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12499	lun = ctl_softc->ctl_luns[targ_lun];
12500
12501	switch (io->io_hdr.msg_type) {
12502	case CTL_MSG_SERIALIZE:
12503		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12504		break;
12505	case CTL_MSG_R2R: {
12506		const struct ctl_cmd_entry *entry;
12507
12508		/*
12509		 * This is only used in SER_ONLY mode.
12510		 */
12511		free_io = 0;
12512		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12513		mtx_lock(&lun->lun_lock);
12514		if (ctl_scsiio_lun_check(ctl_softc, lun,
12515		    entry, (struct ctl_scsiio *)io) != 0) {
12516			mtx_unlock(&lun->lun_lock);
12517			ctl_done(io);
12518			break;
12519		}
12520		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12521		mtx_unlock(&lun->lun_lock);
12522		ctl_enqueue_rtr(io);
12523		break;
12524	}
12525	case CTL_MSG_FINISH_IO:
12526		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12527			free_io = 0;
12528			ctl_done(io);
12529		} else {
12530			free_io = 1;
12531			mtx_lock(&lun->lun_lock);
12532			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12533				     ooa_links);
12534			ctl_check_blocked(lun);
12535			mtx_unlock(&lun->lun_lock);
12536		}
12537		break;
12538	case CTL_MSG_PERS_ACTION:
12539		ctl_hndl_per_res_out_on_other_sc(
12540			(union ctl_ha_msg *)&io->presio.pr_msg);
12541		free_io = 1;
12542		break;
12543	case CTL_MSG_BAD_JUJU:
12544		free_io = 0;
12545		ctl_done(io);
12546		break;
12547	case CTL_MSG_DATAMOVE:
12548		/* Only used in XFER mode */
12549		free_io = 0;
12550		ctl_datamove_remote(io);
12551		break;
12552	case CTL_MSG_DATAMOVE_DONE:
12553		/* Only used in XFER mode */
12554		free_io = 0;
12555		io->scsiio.be_move_done(io);
12556		break;
12557	default:
12558		free_io = 1;
12559		printf("%s: Invalid message type %d\n",
12560		       __func__, io->io_hdr.msg_type);
12561		break;
12562	}
12563	if (free_io)
12564		ctl_free_io(io);
12565
12566}
12567
12568
12569/*
12570 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12571 * there is no match.
12572 */
12573static ctl_lun_error_pattern
12574ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12575{
12576	const struct ctl_cmd_entry *entry;
12577	ctl_lun_error_pattern filtered_pattern, pattern;
12578
12579	pattern = desc->error_pattern;
12580
12581	/*
12582	 * XXX KDM we need more data passed into this function to match a
12583	 * custom pattern, and we actually need to implement custom pattern
12584	 * matching.
12585	 */
12586	if (pattern & CTL_LUN_PAT_CMD)
12587		return (CTL_LUN_PAT_CMD);
12588
12589	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12590		return (CTL_LUN_PAT_ANY);
12591
12592	entry = ctl_get_cmd_entry(ctsio, NULL);
12593
12594	filtered_pattern = entry->pattern & pattern;
12595
12596	/*
12597	 * If the user requested specific flags in the pattern (e.g.
12598	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12599	 * flags.
12600	 *
12601	 * If the user did not specify any flags, it doesn't matter whether
12602	 * or not the command supports the flags.
12603	 */
12604	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12605	     (pattern & ~CTL_LUN_PAT_MASK))
12606		return (CTL_LUN_PAT_NONE);
12607
12608	/*
12609	 * If the user asked for a range check, see if the requested LBA
12610	 * range overlaps with this command's LBA range.
12611	 */
12612	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12613		uint64_t lba1;
12614		uint64_t len1;
12615		ctl_action action;
12616		int retval;
12617
12618		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12619		if (retval != 0)
12620			return (CTL_LUN_PAT_NONE);
12621
12622		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12623					      desc->lba_range.len);
12624		/*
12625		 * A "pass" means that the LBA ranges don't overlap, so
12626		 * this doesn't match the user's range criteria.
12627		 */
12628		if (action == CTL_ACTION_PASS)
12629			return (CTL_LUN_PAT_NONE);
12630	}
12631
12632	return (filtered_pattern);
12633}
12634
12635static void
12636ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12637{
12638	struct ctl_error_desc *desc, *desc2;
12639
12640	mtx_assert(&lun->lun_lock, MA_OWNED);
12641
12642	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12643		ctl_lun_error_pattern pattern;
12644		/*
12645		 * Check to see whether this particular command matches
12646		 * the pattern in the descriptor.
12647		 */
12648		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12649		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12650			continue;
12651
12652		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12653		case CTL_LUN_INJ_ABORTED:
12654			ctl_set_aborted(&io->scsiio);
12655			break;
12656		case CTL_LUN_INJ_MEDIUM_ERR:
12657			ctl_set_medium_error(&io->scsiio);
12658			break;
12659		case CTL_LUN_INJ_UA:
12660			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12661			 * OCCURRED */
12662			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12663			break;
12664		case CTL_LUN_INJ_CUSTOM:
12665			/*
12666			 * We're assuming the user knows what he is doing.
12667			 * Just copy the sense information without doing
12668			 * checks.
12669			 */
12670			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12671			      ctl_min(sizeof(desc->custom_sense),
12672				      sizeof(io->scsiio.sense_data)));
12673			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12674			io->scsiio.sense_len = SSD_FULL_SIZE;
12675			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12676			break;
12677		case CTL_LUN_INJ_NONE:
12678		default:
12679			/*
12680			 * If this is an error injection type we don't know
12681			 * about, clear the continuous flag (if it is set)
12682			 * so it will get deleted below.
12683			 */
12684			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12685			break;
12686		}
12687		/*
12688		 * By default, each error injection action is a one-shot
12689		 */
12690		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12691			continue;
12692
12693		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12694
12695		free(desc, M_CTL);
12696	}
12697}
12698
12699#ifdef CTL_IO_DELAY
12700static void
12701ctl_datamove_timer_wakeup(void *arg)
12702{
12703	union ctl_io *io;
12704
12705	io = (union ctl_io *)arg;
12706
12707	ctl_datamove(io);
12708}
12709#endif /* CTL_IO_DELAY */
12710
12711void
12712ctl_datamove(union ctl_io *io)
12713{
12714	void (*fe_datamove)(union ctl_io *io);
12715
12716	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12717
12718	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12719
12720#ifdef CTL_TIME_IO
12721	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12722		char str[256];
12723		char path_str[64];
12724		struct sbuf sb;
12725
12726		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12727		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12728
12729		sbuf_cat(&sb, path_str);
12730		switch (io->io_hdr.io_type) {
12731		case CTL_IO_SCSI:
12732			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12733			sbuf_printf(&sb, "\n");
12734			sbuf_cat(&sb, path_str);
12735			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12736				    io->scsiio.tag_num, io->scsiio.tag_type);
12737			break;
12738		case CTL_IO_TASK:
12739			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12740				    "Tag Type: %d\n", io->taskio.task_action,
12741				    io->taskio.tag_num, io->taskio.tag_type);
12742			break;
12743		default:
12744			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12745			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12746			break;
12747		}
12748		sbuf_cat(&sb, path_str);
12749		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12750			    (intmax_t)time_uptime - io->io_hdr.start_time);
12751		sbuf_finish(&sb);
12752		printf("%s", sbuf_data(&sb));
12753	}
12754#endif /* CTL_TIME_IO */
12755
12756#ifdef CTL_IO_DELAY
12757	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12758		struct ctl_lun *lun;
12759
12760		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12761
12762		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12763	} else {
12764		struct ctl_lun *lun;
12765
12766		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12767		if ((lun != NULL)
12768		 && (lun->delay_info.datamove_delay > 0)) {
12769			struct callout *callout;
12770
12771			callout = (struct callout *)&io->io_hdr.timer_bytes;
12772			callout_init(callout, /*mpsafe*/ 1);
12773			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12774			callout_reset(callout,
12775				      lun->delay_info.datamove_delay * hz,
12776				      ctl_datamove_timer_wakeup, io);
12777			if (lun->delay_info.datamove_type ==
12778			    CTL_DELAY_TYPE_ONESHOT)
12779				lun->delay_info.datamove_delay = 0;
12780			return;
12781		}
12782	}
12783#endif
12784
12785	/*
12786	 * This command has been aborted.  Set the port status, so we fail
12787	 * the data move.
12788	 */
12789	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12790		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12791		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12792		       io->io_hdr.nexus.targ_port,
12793		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12794		       io->io_hdr.nexus.targ_lun);
12795		io->io_hdr.port_status = 31337;
12796		/*
12797		 * Note that the backend, in this case, will get the
12798		 * callback in its context.  In other cases it may get
12799		 * called in the frontend's interrupt thread context.
12800		 */
12801		io->scsiio.be_move_done(io);
12802		return;
12803	}
12804
12805	/*
12806	 * If we're in XFER mode and this I/O is from the other shelf
12807	 * controller, we need to send the DMA to the other side to
12808	 * actually transfer the data to/from the host.  In serialize only
12809	 * mode the transfer happens below CTL and ctl_datamove() is only
12810	 * called on the machine that originally received the I/O.
12811	 */
12812	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12813	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12814		union ctl_ha_msg msg;
12815		uint32_t sg_entries_sent;
12816		int do_sg_copy;
12817		int i;
12818
12819		memset(&msg, 0, sizeof(msg));
12820		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12821		msg.hdr.original_sc = io->io_hdr.original_sc;
12822		msg.hdr.serializing_sc = io;
12823		msg.hdr.nexus = io->io_hdr.nexus;
12824		msg.dt.flags = io->io_hdr.flags;
12825		/*
12826		 * We convert everything into a S/G list here.  We can't
12827		 * pass by reference, only by value between controllers.
12828		 * So we can't pass a pointer to the S/G list, only as many
12829		 * S/G entries as we can fit in here.  If it's possible for
12830		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12831		 * then we need to break this up into multiple transfers.
12832		 */
12833		if (io->scsiio.kern_sg_entries == 0) {
12834			msg.dt.kern_sg_entries = 1;
12835			/*
12836			 * If this is in cached memory, flush the cache
12837			 * before we send the DMA request to the other
12838			 * controller.  We want to do this in either the
12839			 * read or the write case.  The read case is
12840			 * straightforward.  In the write case, we want to
12841			 * make sure nothing is in the local cache that
12842			 * could overwrite the DMAed data.
12843			 */
12844			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12845				/*
12846				 * XXX KDM use bus_dmamap_sync() here.
12847				 */
12848			}
12849
12850			/*
12851			 * Convert to a physical address if this is a
12852			 * virtual address.
12853			 */
12854			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12855				msg.dt.sg_list[0].addr =
12856					io->scsiio.kern_data_ptr;
12857			} else {
12858				/*
12859				 * XXX KDM use busdma here!
12860				 */
12861#if 0
12862				msg.dt.sg_list[0].addr = (void *)
12863					vtophys(io->scsiio.kern_data_ptr);
12864#endif
12865			}
12866
12867			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12868			do_sg_copy = 0;
12869		} else {
12870			struct ctl_sg_entry *sgl;
12871
12872			do_sg_copy = 1;
12873			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12874			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12875			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12876				/*
12877				 * XXX KDM use bus_dmamap_sync() here.
12878				 */
12879			}
12880		}
12881
12882		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12883		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12884		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12885		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12886		msg.dt.sg_sequence = 0;
12887
12888		/*
12889		 * Loop until we've sent all of the S/G entries.  On the
12890		 * other end, we'll recompose these S/G entries into one
12891		 * contiguous list before passing it to the
12892		 */
12893		for (sg_entries_sent = 0; sg_entries_sent <
12894		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12895			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12896				sizeof(msg.dt.sg_list[0])),
12897				msg.dt.kern_sg_entries - sg_entries_sent);
12898
12899			if (do_sg_copy != 0) {
12900				struct ctl_sg_entry *sgl;
12901				int j;
12902
12903				sgl = (struct ctl_sg_entry *)
12904					io->scsiio.kern_data_ptr;
12905				/*
12906				 * If this is in cached memory, flush the cache
12907				 * before we send the DMA request to the other
12908				 * controller.  We want to do this in either
12909				 * the * read or the write case.  The read
12910				 * case is straightforward.  In the write
12911				 * case, we want to make sure nothing is
12912				 * in the local cache that could overwrite
12913				 * the DMAed data.
12914				 */
12915
12916				for (i = sg_entries_sent, j = 0;
12917				     i < msg.dt.cur_sg_entries; i++, j++) {
12918					if ((io->io_hdr.flags &
12919					     CTL_FLAG_NO_DATASYNC) == 0) {
12920						/*
12921						 * XXX KDM use bus_dmamap_sync()
12922						 */
12923					}
12924					if ((io->io_hdr.flags &
12925					     CTL_FLAG_BUS_ADDR) == 0) {
12926						/*
12927						 * XXX KDM use busdma.
12928						 */
12929#if 0
12930						msg.dt.sg_list[j].addr =(void *)
12931						       vtophys(sgl[i].addr);
12932#endif
12933					} else {
12934						msg.dt.sg_list[j].addr =
12935							sgl[i].addr;
12936					}
12937					msg.dt.sg_list[j].len = sgl[i].len;
12938				}
12939			}
12940
12941			sg_entries_sent += msg.dt.cur_sg_entries;
12942			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12943				msg.dt.sg_last = 1;
12944			else
12945				msg.dt.sg_last = 0;
12946
12947			/*
12948			 * XXX KDM drop and reacquire the lock here?
12949			 */
12950			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12951			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12952				/*
12953				 * XXX do something here.
12954				 */
12955			}
12956
12957			msg.dt.sent_sg_entries = sg_entries_sent;
12958		}
12959		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12960		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12961			ctl_failover_io(io, /*have_lock*/ 0);
12962
12963	} else {
12964
12965		/*
12966		 * Lookup the fe_datamove() function for this particular
12967		 * front end.
12968		 */
12969		fe_datamove =
12970		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12971
12972		fe_datamove(io);
12973	}
12974}
12975
12976static void
12977ctl_send_datamove_done(union ctl_io *io, int have_lock)
12978{
12979	union ctl_ha_msg msg;
12980	int isc_status;
12981
12982	memset(&msg, 0, sizeof(msg));
12983
12984	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12985	msg.hdr.original_sc = io;
12986	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12987	msg.hdr.nexus = io->io_hdr.nexus;
12988	msg.hdr.status = io->io_hdr.status;
12989	msg.scsi.tag_num = io->scsiio.tag_num;
12990	msg.scsi.tag_type = io->scsiio.tag_type;
12991	msg.scsi.scsi_status = io->scsiio.scsi_status;
12992	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12993	       sizeof(io->scsiio.sense_data));
12994	msg.scsi.sense_len = io->scsiio.sense_len;
12995	msg.scsi.sense_residual = io->scsiio.sense_residual;
12996	msg.scsi.fetd_status = io->io_hdr.port_status;
12997	msg.scsi.residual = io->scsiio.residual;
12998	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12999
13000	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13001		ctl_failover_io(io, /*have_lock*/ have_lock);
13002		return;
13003	}
13004
13005	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
13006	if (isc_status > CTL_HA_STATUS_SUCCESS) {
13007		/* XXX do something if this fails */
13008	}
13009
13010}
13011
13012/*
13013 * The DMA to the remote side is done, now we need to tell the other side
13014 * we're done so it can continue with its data movement.
13015 */
13016static void
13017ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
13018{
13019	union ctl_io *io;
13020
13021	io = rq->context;
13022
13023	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13024		printf("%s: ISC DMA write failed with error %d", __func__,
13025		       rq->ret);
13026		ctl_set_internal_failure(&io->scsiio,
13027					 /*sks_valid*/ 1,
13028					 /*retry_count*/ rq->ret);
13029	}
13030
13031	ctl_dt_req_free(rq);
13032
13033	/*
13034	 * In this case, we had to malloc the memory locally.  Free it.
13035	 */
13036	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13037		int i;
13038		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13039			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13040	}
13041	/*
13042	 * The data is in local and remote memory, so now we need to send
13043	 * status (good or back) back to the other side.
13044	 */
13045	ctl_send_datamove_done(io, /*have_lock*/ 0);
13046}
13047
13048/*
13049 * We've moved the data from the host/controller into local memory.  Now we
13050 * need to push it over to the remote controller's memory.
13051 */
13052static int
13053ctl_datamove_remote_dm_write_cb(union ctl_io *io)
13054{
13055	int retval;
13056
13057	retval = 0;
13058
13059	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
13060					  ctl_datamove_remote_write_cb);
13061
13062	return (retval);
13063}
13064
13065static void
13066ctl_datamove_remote_write(union ctl_io *io)
13067{
13068	int retval;
13069	void (*fe_datamove)(union ctl_io *io);
13070
13071	/*
13072	 * - Get the data from the host/HBA into local memory.
13073	 * - DMA memory from the local controller to the remote controller.
13074	 * - Send status back to the remote controller.
13075	 */
13076
13077	retval = ctl_datamove_remote_sgl_setup(io);
13078	if (retval != 0)
13079		return;
13080
13081	/* Switch the pointer over so the FETD knows what to do */
13082	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13083
13084	/*
13085	 * Use a custom move done callback, since we need to send completion
13086	 * back to the other controller, not to the backend on this side.
13087	 */
13088	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
13089
13090	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13091
13092	fe_datamove(io);
13093
13094	return;
13095
13096}
13097
13098static int
13099ctl_datamove_remote_dm_read_cb(union ctl_io *io)
13100{
13101#if 0
13102	char str[256];
13103	char path_str[64];
13104	struct sbuf sb;
13105#endif
13106
13107	/*
13108	 * In this case, we had to malloc the memory locally.  Free it.
13109	 */
13110	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13111		int i;
13112		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13113			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13114	}
13115
13116#if 0
13117	scsi_path_string(io, path_str, sizeof(path_str));
13118	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13119	sbuf_cat(&sb, path_str);
13120	scsi_command_string(&io->scsiio, NULL, &sb);
13121	sbuf_printf(&sb, "\n");
13122	sbuf_cat(&sb, path_str);
13123	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13124		    io->scsiio.tag_num, io->scsiio.tag_type);
13125	sbuf_cat(&sb, path_str);
13126	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
13127		    io->io_hdr.flags, io->io_hdr.status);
13128	sbuf_finish(&sb);
13129	printk("%s", sbuf_data(&sb));
13130#endif
13131
13132
13133	/*
13134	 * The read is done, now we need to send status (good or bad) back
13135	 * to the other side.
13136	 */
13137	ctl_send_datamove_done(io, /*have_lock*/ 0);
13138
13139	return (0);
13140}
13141
13142static void
13143ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
13144{
13145	union ctl_io *io;
13146	void (*fe_datamove)(union ctl_io *io);
13147
13148	io = rq->context;
13149
13150	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13151		printf("%s: ISC DMA read failed with error %d", __func__,
13152		       rq->ret);
13153		ctl_set_internal_failure(&io->scsiio,
13154					 /*sks_valid*/ 1,
13155					 /*retry_count*/ rq->ret);
13156	}
13157
13158	ctl_dt_req_free(rq);
13159
13160	/* Switch the pointer over so the FETD knows what to do */
13161	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13162
13163	/*
13164	 * Use a custom move done callback, since we need to send completion
13165	 * back to the other controller, not to the backend on this side.
13166	 */
13167	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13168
13169	/* XXX KDM add checks like the ones in ctl_datamove? */
13170
13171	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13172
13173	fe_datamove(io);
13174}
13175
13176static int
13177ctl_datamove_remote_sgl_setup(union ctl_io *io)
13178{
13179	struct ctl_sg_entry *local_sglist, *remote_sglist;
13180	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13181	struct ctl_softc *softc;
13182	int retval;
13183	int i;
13184
13185	retval = 0;
13186	softc = control_softc;
13187
13188	local_sglist = io->io_hdr.local_sglist;
13189	local_dma_sglist = io->io_hdr.local_dma_sglist;
13190	remote_sglist = io->io_hdr.remote_sglist;
13191	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13192
13193	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13194		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13195			local_sglist[i].len = remote_sglist[i].len;
13196
13197			/*
13198			 * XXX Detect the situation where the RS-level I/O
13199			 * redirector on the other side has already read the
13200			 * data off of the AOR RS on this side, and
13201			 * transferred it to remote (mirror) memory on the
13202			 * other side.  Since we already have the data in
13203			 * memory here, we just need to use it.
13204			 *
13205			 * XXX KDM this can probably be removed once we
13206			 * get the cache device code in and take the
13207			 * current AOR implementation out.
13208			 */
13209#ifdef NEEDTOPORT
13210			if ((remote_sglist[i].addr >=
13211			     (void *)vtophys(softc->mirr->addr))
13212			 && (remote_sglist[i].addr <
13213			     ((void *)vtophys(softc->mirr->addr) +
13214			     CacheMirrorOffset))) {
13215				local_sglist[i].addr = remote_sglist[i].addr -
13216					CacheMirrorOffset;
13217				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13218				     CTL_FLAG_DATA_IN)
13219					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13220			} else {
13221				local_sglist[i].addr = remote_sglist[i].addr +
13222					CacheMirrorOffset;
13223			}
13224#endif
13225#if 0
13226			printf("%s: local %p, remote %p, len %d\n",
13227			       __func__, local_sglist[i].addr,
13228			       remote_sglist[i].addr, local_sglist[i].len);
13229#endif
13230		}
13231	} else {
13232		uint32_t len_to_go;
13233
13234		/*
13235		 * In this case, we don't have automatically allocated
13236		 * memory for this I/O on this controller.  This typically
13237		 * happens with internal CTL I/O -- e.g. inquiry, mode
13238		 * sense, etc.  Anything coming from RAIDCore will have
13239		 * a mirror area available.
13240		 */
13241		len_to_go = io->scsiio.kern_data_len;
13242
13243		/*
13244		 * Clear the no datasync flag, we have to use malloced
13245		 * buffers.
13246		 */
13247		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13248
13249		/*
13250		 * The difficult thing here is that the size of the various
13251		 * S/G segments may be different than the size from the
13252		 * remote controller.  That'll make it harder when DMAing
13253		 * the data back to the other side.
13254		 */
13255		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13256		     sizeof(io->io_hdr.remote_sglist[0])) &&
13257		     (len_to_go > 0); i++) {
13258			local_sglist[i].len = ctl_min(len_to_go, 131072);
13259			CTL_SIZE_8B(local_dma_sglist[i].len,
13260				    local_sglist[i].len);
13261			local_sglist[i].addr =
13262				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13263
13264			local_dma_sglist[i].addr = local_sglist[i].addr;
13265
13266			if (local_sglist[i].addr == NULL) {
13267				int j;
13268
13269				printf("malloc failed for %zd bytes!",
13270				       local_dma_sglist[i].len);
13271				for (j = 0; j < i; j++) {
13272					free(local_sglist[j].addr, M_CTL);
13273				}
13274				ctl_set_internal_failure(&io->scsiio,
13275							 /*sks_valid*/ 1,
13276							 /*retry_count*/ 4857);
13277				retval = 1;
13278				goto bailout_error;
13279
13280			}
13281			/* XXX KDM do we need a sync here? */
13282
13283			len_to_go -= local_sglist[i].len;
13284		}
13285		/*
13286		 * Reset the number of S/G entries accordingly.  The
13287		 * original number of S/G entries is available in
13288		 * rem_sg_entries.
13289		 */
13290		io->scsiio.kern_sg_entries = i;
13291
13292#if 0
13293		printf("%s: kern_sg_entries = %d\n", __func__,
13294		       io->scsiio.kern_sg_entries);
13295		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13296			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13297			       local_sglist[i].addr, local_sglist[i].len,
13298			       local_dma_sglist[i].len);
13299#endif
13300	}
13301
13302
13303	return (retval);
13304
13305bailout_error:
13306
13307	ctl_send_datamove_done(io, /*have_lock*/ 0);
13308
13309	return (retval);
13310}
13311
13312static int
13313ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13314			 ctl_ha_dt_cb callback)
13315{
13316	struct ctl_ha_dt_req *rq;
13317	struct ctl_sg_entry *remote_sglist, *local_sglist;
13318	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13319	uint32_t local_used, remote_used, total_used;
13320	int retval;
13321	int i, j;
13322
13323	retval = 0;
13324
13325	rq = ctl_dt_req_alloc();
13326
13327	/*
13328	 * If we failed to allocate the request, and if the DMA didn't fail
13329	 * anyway, set busy status.  This is just a resource allocation
13330	 * failure.
13331	 */
13332	if ((rq == NULL)
13333	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13334		ctl_set_busy(&io->scsiio);
13335
13336	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13337
13338		if (rq != NULL)
13339			ctl_dt_req_free(rq);
13340
13341		/*
13342		 * The data move failed.  We need to return status back
13343		 * to the other controller.  No point in trying to DMA
13344		 * data to the remote controller.
13345		 */
13346
13347		ctl_send_datamove_done(io, /*have_lock*/ 0);
13348
13349		retval = 1;
13350
13351		goto bailout;
13352	}
13353
13354	local_sglist = io->io_hdr.local_sglist;
13355	local_dma_sglist = io->io_hdr.local_dma_sglist;
13356	remote_sglist = io->io_hdr.remote_sglist;
13357	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13358	local_used = 0;
13359	remote_used = 0;
13360	total_used = 0;
13361
13362	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13363		rq->ret = CTL_HA_STATUS_SUCCESS;
13364		rq->context = io;
13365		callback(rq);
13366		goto bailout;
13367	}
13368
13369	/*
13370	 * Pull/push the data over the wire from/to the other controller.
13371	 * This takes into account the possibility that the local and
13372	 * remote sglists may not be identical in terms of the size of
13373	 * the elements and the number of elements.
13374	 *
13375	 * One fundamental assumption here is that the length allocated for
13376	 * both the local and remote sglists is identical.  Otherwise, we've
13377	 * essentially got a coding error of some sort.
13378	 */
13379	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13380		int isc_ret;
13381		uint32_t cur_len, dma_length;
13382		uint8_t *tmp_ptr;
13383
13384		rq->id = CTL_HA_DATA_CTL;
13385		rq->command = command;
13386		rq->context = io;
13387
13388		/*
13389		 * Both pointers should be aligned.  But it is possible
13390		 * that the allocation length is not.  They should both
13391		 * also have enough slack left over at the end, though,
13392		 * to round up to the next 8 byte boundary.
13393		 */
13394		cur_len = ctl_min(local_sglist[i].len - local_used,
13395				  remote_sglist[j].len - remote_used);
13396
13397		/*
13398		 * In this case, we have a size issue and need to decrease
13399		 * the size, except in the case where we actually have less
13400		 * than 8 bytes left.  In that case, we need to increase
13401		 * the DMA length to get the last bit.
13402		 */
13403		if ((cur_len & 0x7) != 0) {
13404			if (cur_len > 0x7) {
13405				cur_len = cur_len - (cur_len & 0x7);
13406				dma_length = cur_len;
13407			} else {
13408				CTL_SIZE_8B(dma_length, cur_len);
13409			}
13410
13411		} else
13412			dma_length = cur_len;
13413
13414		/*
13415		 * If we had to allocate memory for this I/O, instead of using
13416		 * the non-cached mirror memory, we'll need to flush the cache
13417		 * before trying to DMA to the other controller.
13418		 *
13419		 * We could end up doing this multiple times for the same
13420		 * segment if we have a larger local segment than remote
13421		 * segment.  That shouldn't be an issue.
13422		 */
13423		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13424			/*
13425			 * XXX KDM use bus_dmamap_sync() here.
13426			 */
13427		}
13428
13429		rq->size = dma_length;
13430
13431		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13432		tmp_ptr += local_used;
13433
13434		/* Use physical addresses when talking to ISC hardware */
13435		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13436			/* XXX KDM use busdma */
13437#if 0
13438			rq->local = vtophys(tmp_ptr);
13439#endif
13440		} else
13441			rq->local = tmp_ptr;
13442
13443		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13444		tmp_ptr += remote_used;
13445		rq->remote = tmp_ptr;
13446
13447		rq->callback = NULL;
13448
13449		local_used += cur_len;
13450		if (local_used >= local_sglist[i].len) {
13451			i++;
13452			local_used = 0;
13453		}
13454
13455		remote_used += cur_len;
13456		if (remote_used >= remote_sglist[j].len) {
13457			j++;
13458			remote_used = 0;
13459		}
13460		total_used += cur_len;
13461
13462		if (total_used >= io->scsiio.kern_data_len)
13463			rq->callback = callback;
13464
13465		if ((rq->size & 0x7) != 0) {
13466			printf("%s: warning: size %d is not on 8b boundary\n",
13467			       __func__, rq->size);
13468		}
13469		if (((uintptr_t)rq->local & 0x7) != 0) {
13470			printf("%s: warning: local %p not on 8b boundary\n",
13471			       __func__, rq->local);
13472		}
13473		if (((uintptr_t)rq->remote & 0x7) != 0) {
13474			printf("%s: warning: remote %p not on 8b boundary\n",
13475			       __func__, rq->local);
13476		}
13477#if 0
13478		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13479		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13480		       rq->local, rq->remote, rq->size);
13481#endif
13482
13483		isc_ret = ctl_dt_single(rq);
13484		if (isc_ret == CTL_HA_STATUS_WAIT)
13485			continue;
13486
13487		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13488			rq->ret = CTL_HA_STATUS_SUCCESS;
13489		} else {
13490			rq->ret = isc_ret;
13491		}
13492		callback(rq);
13493		goto bailout;
13494	}
13495
13496bailout:
13497	return (retval);
13498
13499}
13500
13501static void
13502ctl_datamove_remote_read(union ctl_io *io)
13503{
13504	int retval;
13505	int i;
13506
13507	/*
13508	 * This will send an error to the other controller in the case of a
13509	 * failure.
13510	 */
13511	retval = ctl_datamove_remote_sgl_setup(io);
13512	if (retval != 0)
13513		return;
13514
13515	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13516					  ctl_datamove_remote_read_cb);
13517	if ((retval != 0)
13518	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13519		/*
13520		 * Make sure we free memory if there was an error..  The
13521		 * ctl_datamove_remote_xfer() function will send the
13522		 * datamove done message, or call the callback with an
13523		 * error if there is a problem.
13524		 */
13525		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13526			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13527	}
13528
13529	return;
13530}
13531
13532/*
13533 * Process a datamove request from the other controller.  This is used for
13534 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13535 * first.  Once that is complete, the data gets DMAed into the remote
13536 * controller's memory.  For reads, we DMA from the remote controller's
13537 * memory into our memory first, and then move it out to the FETD.
13538 */
13539static void
13540ctl_datamove_remote(union ctl_io *io)
13541{
13542	struct ctl_softc *softc;
13543
13544	softc = control_softc;
13545
13546	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13547
13548	/*
13549	 * Note that we look for an aborted I/O here, but don't do some of
13550	 * the other checks that ctl_datamove() normally does.
13551	 * We don't need to run the datamove delay code, since that should
13552	 * have been done if need be on the other controller.
13553	 */
13554	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13555		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13556		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13557		       io->io_hdr.nexus.targ_port,
13558		       io->io_hdr.nexus.targ_target.id,
13559		       io->io_hdr.nexus.targ_lun);
13560		io->io_hdr.port_status = 31338;
13561		ctl_send_datamove_done(io, /*have_lock*/ 0);
13562		return;
13563	}
13564
13565	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13566		ctl_datamove_remote_write(io);
13567	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13568		ctl_datamove_remote_read(io);
13569	} else {
13570		union ctl_ha_msg msg;
13571		struct scsi_sense_data *sense;
13572		uint8_t sks[3];
13573		int retry_count;
13574
13575		memset(&msg, 0, sizeof(msg));
13576
13577		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13578		msg.hdr.status = CTL_SCSI_ERROR;
13579		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13580
13581		retry_count = 4243;
13582
13583		sense = &msg.scsi.sense_data;
13584		sks[0] = SSD_SCS_VALID;
13585		sks[1] = (retry_count >> 8) & 0xff;
13586		sks[2] = retry_count & 0xff;
13587
13588		/* "Internal target failure" */
13589		scsi_set_sense_data(sense,
13590				    /*sense_format*/ SSD_TYPE_NONE,
13591				    /*current_error*/ 1,
13592				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13593				    /*asc*/ 0x44,
13594				    /*ascq*/ 0x00,
13595				    /*type*/ SSD_ELEM_SKS,
13596				    /*size*/ sizeof(sks),
13597				    /*data*/ sks,
13598				    SSD_ELEM_NONE);
13599
13600		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13601		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13602			ctl_failover_io(io, /*have_lock*/ 1);
13603			return;
13604		}
13605
13606		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13607		    CTL_HA_STATUS_SUCCESS) {
13608			/* XXX KDM what to do if this fails? */
13609		}
13610		return;
13611	}
13612
13613}
13614
13615static int
13616ctl_process_done(union ctl_io *io)
13617{
13618	struct ctl_lun *lun;
13619	struct ctl_softc *ctl_softc;
13620	void (*fe_done)(union ctl_io *io);
13621	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13622
13623	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13624
13625	fe_done =
13626	    control_softc->ctl_ports[targ_port]->fe_done;
13627
13628#ifdef CTL_TIME_IO
13629	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13630		char str[256];
13631		char path_str[64];
13632		struct sbuf sb;
13633
13634		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13635		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13636
13637		sbuf_cat(&sb, path_str);
13638		switch (io->io_hdr.io_type) {
13639		case CTL_IO_SCSI:
13640			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13641			sbuf_printf(&sb, "\n");
13642			sbuf_cat(&sb, path_str);
13643			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13644				    io->scsiio.tag_num, io->scsiio.tag_type);
13645			break;
13646		case CTL_IO_TASK:
13647			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13648				    "Tag Type: %d\n", io->taskio.task_action,
13649				    io->taskio.tag_num, io->taskio.tag_type);
13650			break;
13651		default:
13652			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13653			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13654			break;
13655		}
13656		sbuf_cat(&sb, path_str);
13657		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13658			    (intmax_t)time_uptime - io->io_hdr.start_time);
13659		sbuf_finish(&sb);
13660		printf("%s", sbuf_data(&sb));
13661	}
13662#endif /* CTL_TIME_IO */
13663
13664	switch (io->io_hdr.io_type) {
13665	case CTL_IO_SCSI:
13666		break;
13667	case CTL_IO_TASK:
13668		if (bootverbose || verbose > 0)
13669			ctl_io_error_print(io, NULL);
13670		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13671			ctl_free_io(io);
13672		else
13673			fe_done(io);
13674		return (CTL_RETVAL_COMPLETE);
13675		break;
13676	default:
13677		printf("ctl_process_done: invalid io type %d\n",
13678		       io->io_hdr.io_type);
13679		panic("ctl_process_done: invalid io type %d\n",
13680		      io->io_hdr.io_type);
13681		break; /* NOTREACHED */
13682	}
13683
13684	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13685	if (lun == NULL) {
13686		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13687				 io->io_hdr.nexus.targ_mapped_lun));
13688		fe_done(io);
13689		goto bailout;
13690	}
13691	ctl_softc = lun->ctl_softc;
13692
13693	mtx_lock(&lun->lun_lock);
13694
13695	/*
13696	 * Check to see if we have any errors to inject here.  We only
13697	 * inject errors for commands that don't already have errors set.
13698	 */
13699	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13700	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13701		ctl_inject_error(lun, io);
13702
13703	/*
13704	 * XXX KDM how do we treat commands that aren't completed
13705	 * successfully?
13706	 *
13707	 * XXX KDM should we also track I/O latency?
13708	 */
13709	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13710	    io->io_hdr.io_type == CTL_IO_SCSI) {
13711#ifdef CTL_TIME_IO
13712		struct bintime cur_bt;
13713#endif
13714		int type;
13715
13716		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13717		    CTL_FLAG_DATA_IN)
13718			type = CTL_STATS_READ;
13719		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13720		    CTL_FLAG_DATA_OUT)
13721			type = CTL_STATS_WRITE;
13722		else
13723			type = CTL_STATS_NO_IO;
13724
13725		lun->stats.ports[targ_port].bytes[type] +=
13726		    io->scsiio.kern_total_len;
13727		lun->stats.ports[targ_port].operations[type]++;
13728#ifdef CTL_TIME_IO
13729		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13730		   &io->io_hdr.dma_bt);
13731		lun->stats.ports[targ_port].num_dmas[type] +=
13732		    io->io_hdr.num_dmas;
13733		getbintime(&cur_bt);
13734		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13735		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13736#endif
13737	}
13738
13739	/*
13740	 * Remove this from the OOA queue.
13741	 */
13742	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13743
13744	/*
13745	 * Run through the blocked queue on this LUN and see if anything
13746	 * has become unblocked, now that this transaction is done.
13747	 */
13748	ctl_check_blocked(lun);
13749
13750	/*
13751	 * If the LUN has been invalidated, free it if there is nothing
13752	 * left on its OOA queue.
13753	 */
13754	if ((lun->flags & CTL_LUN_INVALID)
13755	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13756		mtx_unlock(&lun->lun_lock);
13757		mtx_lock(&ctl_softc->ctl_lock);
13758		ctl_free_lun(lun);
13759		mtx_unlock(&ctl_softc->ctl_lock);
13760	} else
13761		mtx_unlock(&lun->lun_lock);
13762
13763	/*
13764	 * If this command has been aborted, make sure we set the status
13765	 * properly.  The FETD is responsible for freeing the I/O and doing
13766	 * whatever it needs to do to clean up its state.
13767	 */
13768	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13769		ctl_set_task_aborted(&io->scsiio);
13770
13771	/*
13772	 * We print out status for every task management command.  For SCSI
13773	 * commands, we filter out any unit attention errors; they happen
13774	 * on every boot, and would clutter up the log.  Note:  task
13775	 * management commands aren't printed here, they are printed above,
13776	 * since they should never even make it down here.
13777	 */
13778	switch (io->io_hdr.io_type) {
13779	case CTL_IO_SCSI: {
13780		int error_code, sense_key, asc, ascq;
13781
13782		sense_key = 0;
13783
13784		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13785		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13786			/*
13787			 * Since this is just for printing, no need to
13788			 * show errors here.
13789			 */
13790			scsi_extract_sense_len(&io->scsiio.sense_data,
13791					       io->scsiio.sense_len,
13792					       &error_code,
13793					       &sense_key,
13794					       &asc,
13795					       &ascq,
13796					       /*show_errors*/ 0);
13797		}
13798
13799		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13800		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13801		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13802		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13803
13804			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13805				ctl_softc->skipped_prints++;
13806			} else {
13807				uint32_t skipped_prints;
13808
13809				skipped_prints = ctl_softc->skipped_prints;
13810
13811				ctl_softc->skipped_prints = 0;
13812				ctl_softc->last_print_jiffies = time_uptime;
13813
13814				if (skipped_prints > 0) {
13815#ifdef NEEDTOPORT
13816					csevent_log(CSC_CTL | CSC_SHELF_SW |
13817					    CTL_ERROR_REPORT,
13818					    csevent_LogType_Trace,
13819					    csevent_Severity_Information,
13820					    csevent_AlertLevel_Green,
13821					    csevent_FRU_Firmware,
13822					    csevent_FRU_Unknown,
13823					    "High CTL error volume, %d prints "
13824					    "skipped", skipped_prints);
13825#endif
13826				}
13827				if (bootverbose || verbose > 0)
13828					ctl_io_error_print(io, NULL);
13829			}
13830		}
13831		break;
13832	}
13833	case CTL_IO_TASK:
13834		if (bootverbose || verbose > 0)
13835			ctl_io_error_print(io, NULL);
13836		break;
13837	default:
13838		break;
13839	}
13840
13841	/*
13842	 * Tell the FETD or the other shelf controller we're done with this
13843	 * command.  Note that only SCSI commands get to this point.  Task
13844	 * management commands are completed above.
13845	 *
13846	 * We only send status to the other controller if we're in XFER
13847	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13848	 * received the I/O (from CTL's perspective), and so the status is
13849	 * generated there.
13850	 *
13851	 * XXX KDM if we hold the lock here, we could cause a deadlock
13852	 * if the frontend comes back in in this context to queue
13853	 * something.
13854	 */
13855	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13856	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13857		union ctl_ha_msg msg;
13858
13859		memset(&msg, 0, sizeof(msg));
13860		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13861		msg.hdr.original_sc = io->io_hdr.original_sc;
13862		msg.hdr.nexus = io->io_hdr.nexus;
13863		msg.hdr.status = io->io_hdr.status;
13864		msg.scsi.scsi_status = io->scsiio.scsi_status;
13865		msg.scsi.tag_num = io->scsiio.tag_num;
13866		msg.scsi.tag_type = io->scsiio.tag_type;
13867		msg.scsi.sense_len = io->scsiio.sense_len;
13868		msg.scsi.sense_residual = io->scsiio.sense_residual;
13869		msg.scsi.residual = io->scsiio.residual;
13870		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13871		       sizeof(io->scsiio.sense_data));
13872		/*
13873		 * We copy this whether or not this is an I/O-related
13874		 * command.  Otherwise, we'd have to go and check to see
13875		 * whether it's a read/write command, and it really isn't
13876		 * worth it.
13877		 */
13878		memcpy(&msg.scsi.lbalen,
13879		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13880		       sizeof(msg.scsi.lbalen));
13881
13882		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13883				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13884			/* XXX do something here */
13885		}
13886
13887		ctl_free_io(io);
13888	} else
13889		fe_done(io);
13890
13891bailout:
13892
13893	return (CTL_RETVAL_COMPLETE);
13894}
13895
13896#ifdef CTL_WITH_CA
13897/*
13898 * Front end should call this if it doesn't do autosense.  When the request
13899 * sense comes back in from the initiator, we'll dequeue this and send it.
13900 */
13901int
13902ctl_queue_sense(union ctl_io *io)
13903{
13904	struct ctl_lun *lun;
13905	struct ctl_softc *ctl_softc;
13906	uint32_t initidx, targ_lun;
13907
13908	ctl_softc = control_softc;
13909
13910	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13911
13912	/*
13913	 * LUN lookup will likely move to the ctl_work_thread() once we
13914	 * have our new queueing infrastructure (that doesn't put things on
13915	 * a per-LUN queue initially).  That is so that we can handle
13916	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13917	 * can't deal with that right now.
13918	 */
13919	mtx_lock(&ctl_softc->ctl_lock);
13920
13921	/*
13922	 * If we don't have a LUN for this, just toss the sense
13923	 * information.
13924	 */
13925	targ_lun = io->io_hdr.nexus.targ_lun;
13926	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13927	if ((targ_lun < CTL_MAX_LUNS)
13928	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13929		lun = ctl_softc->ctl_luns[targ_lun];
13930	else
13931		goto bailout;
13932
13933	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13934
13935	mtx_lock(&lun->lun_lock);
13936	/*
13937	 * Already have CA set for this LUN...toss the sense information.
13938	 */
13939	if (ctl_is_set(lun->have_ca, initidx)) {
13940		mtx_unlock(&lun->lun_lock);
13941		goto bailout;
13942	}
13943
13944	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13945	       ctl_min(sizeof(lun->pending_sense[initidx]),
13946	       sizeof(io->scsiio.sense_data)));
13947	ctl_set_mask(lun->have_ca, initidx);
13948	mtx_unlock(&lun->lun_lock);
13949
13950bailout:
13951	mtx_unlock(&ctl_softc->ctl_lock);
13952
13953	ctl_free_io(io);
13954
13955	return (CTL_RETVAL_COMPLETE);
13956}
13957#endif
13958
13959/*
13960 * Primary command inlet from frontend ports.  All SCSI and task I/O
13961 * requests must go through this function.
13962 */
13963int
13964ctl_queue(union ctl_io *io)
13965{
13966	struct ctl_softc *ctl_softc;
13967
13968	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13969
13970	ctl_softc = control_softc;
13971
13972#ifdef CTL_TIME_IO
13973	io->io_hdr.start_time = time_uptime;
13974	getbintime(&io->io_hdr.start_bt);
13975#endif /* CTL_TIME_IO */
13976
13977	/* Map FE-specific LUN ID into global one. */
13978	io->io_hdr.nexus.targ_mapped_lun =
13979	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13980
13981	switch (io->io_hdr.io_type) {
13982	case CTL_IO_SCSI:
13983	case CTL_IO_TASK:
13984		ctl_enqueue_incoming(io);
13985		break;
13986	default:
13987		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13988		return (EINVAL);
13989	}
13990
13991	return (CTL_RETVAL_COMPLETE);
13992}
13993
13994#ifdef CTL_IO_DELAY
13995static void
13996ctl_done_timer_wakeup(void *arg)
13997{
13998	union ctl_io *io;
13999
14000	io = (union ctl_io *)arg;
14001	ctl_done(io);
14002}
14003#endif /* CTL_IO_DELAY */
14004
14005void
14006ctl_done(union ctl_io *io)
14007{
14008	struct ctl_softc *ctl_softc;
14009
14010	ctl_softc = control_softc;
14011
14012	/*
14013	 * Enable this to catch duplicate completion issues.
14014	 */
14015#if 0
14016	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
14017		printf("%s: type %d msg %d cdb %x iptl: "
14018		       "%d:%d:%d:%d tag 0x%04x "
14019		       "flag %#x status %x\n",
14020			__func__,
14021			io->io_hdr.io_type,
14022			io->io_hdr.msg_type,
14023			io->scsiio.cdb[0],
14024			io->io_hdr.nexus.initid.id,
14025			io->io_hdr.nexus.targ_port,
14026			io->io_hdr.nexus.targ_target.id,
14027			io->io_hdr.nexus.targ_lun,
14028			(io->io_hdr.io_type ==
14029			CTL_IO_TASK) ?
14030			io->taskio.tag_num :
14031			io->scsiio.tag_num,
14032		        io->io_hdr.flags,
14033			io->io_hdr.status);
14034	} else
14035		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
14036#endif
14037
14038	/*
14039	 * This is an internal copy of an I/O, and should not go through
14040	 * the normal done processing logic.
14041	 */
14042	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
14043		return;
14044
14045	/*
14046	 * We need to send a msg to the serializing shelf to finish the IO
14047	 * as well.  We don't send a finish message to the other shelf if
14048	 * this is a task management command.  Task management commands
14049	 * aren't serialized in the OOA queue, but rather just executed on
14050	 * both shelf controllers for commands that originated on that
14051	 * controller.
14052	 */
14053	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
14054	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
14055		union ctl_ha_msg msg_io;
14056
14057		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
14058		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
14059		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
14060		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
14061		}
14062		/* continue on to finish IO */
14063	}
14064#ifdef CTL_IO_DELAY
14065	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
14066		struct ctl_lun *lun;
14067
14068		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14069
14070		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
14071	} else {
14072		struct ctl_lun *lun;
14073
14074		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14075
14076		if ((lun != NULL)
14077		 && (lun->delay_info.done_delay > 0)) {
14078			struct callout *callout;
14079
14080			callout = (struct callout *)&io->io_hdr.timer_bytes;
14081			callout_init(callout, /*mpsafe*/ 1);
14082			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
14083			callout_reset(callout,
14084				      lun->delay_info.done_delay * hz,
14085				      ctl_done_timer_wakeup, io);
14086			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
14087				lun->delay_info.done_delay = 0;
14088			return;
14089		}
14090	}
14091#endif /* CTL_IO_DELAY */
14092
14093	ctl_enqueue_done(io);
14094}
14095
14096int
14097ctl_isc(struct ctl_scsiio *ctsio)
14098{
14099	struct ctl_lun *lun;
14100	int retval;
14101
14102	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14103
14104	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
14105
14106	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
14107
14108	retval = lun->backend->data_submit((union ctl_io *)ctsio);
14109
14110	return (retval);
14111}
14112
14113
14114static void
14115ctl_work_thread(void *arg)
14116{
14117	struct ctl_thread *thr = (struct ctl_thread *)arg;
14118	struct ctl_softc *softc = thr->ctl_softc;
14119	union ctl_io *io;
14120	int retval;
14121
14122	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
14123
14124	for (;;) {
14125		retval = 0;
14126
14127		/*
14128		 * We handle the queues in this order:
14129		 * - ISC
14130		 * - done queue (to free up resources, unblock other commands)
14131		 * - RtR queue
14132		 * - incoming queue
14133		 *
14134		 * If those queues are empty, we break out of the loop and
14135		 * go to sleep.
14136		 */
14137		mtx_lock(&thr->queue_lock);
14138		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
14139		if (io != NULL) {
14140			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
14141			mtx_unlock(&thr->queue_lock);
14142			ctl_handle_isc(io);
14143			continue;
14144		}
14145		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
14146		if (io != NULL) {
14147			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
14148			/* clear any blocked commands, call fe_done */
14149			mtx_unlock(&thr->queue_lock);
14150			retval = ctl_process_done(io);
14151			continue;
14152		}
14153		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
14154		if (io != NULL) {
14155			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
14156			mtx_unlock(&thr->queue_lock);
14157			if (io->io_hdr.io_type == CTL_IO_TASK)
14158				ctl_run_task(io);
14159			else
14160				ctl_scsiio_precheck(softc, &io->scsiio);
14161			continue;
14162		}
14163		if (!ctl_pause_rtr) {
14164			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
14165			if (io != NULL) {
14166				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
14167				mtx_unlock(&thr->queue_lock);
14168				retval = ctl_scsiio(&io->scsiio);
14169				if (retval != CTL_RETVAL_COMPLETE)
14170					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
14171				continue;
14172			}
14173		}
14174
14175		/* Sleep until we have something to do. */
14176		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
14177	}
14178}
14179
14180static void
14181ctl_lun_thread(void *arg)
14182{
14183	struct ctl_softc *softc = (struct ctl_softc *)arg;
14184	struct ctl_be_lun *be_lun;
14185	int retval;
14186
14187	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14188
14189	for (;;) {
14190		retval = 0;
14191		mtx_lock(&softc->ctl_lock);
14192		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14193		if (be_lun != NULL) {
14194			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14195			mtx_unlock(&softc->ctl_lock);
14196			ctl_create_lun(be_lun);
14197			continue;
14198		}
14199
14200		/* Sleep until we have something to do. */
14201		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14202		    PDROP | PRIBIO, "-", 0);
14203	}
14204}
14205
14206static void
14207ctl_enqueue_incoming(union ctl_io *io)
14208{
14209	struct ctl_softc *softc = control_softc;
14210	struct ctl_thread *thr;
14211	u_int idx;
14212
14213	idx = (io->io_hdr.nexus.targ_port * 127 +
14214	       io->io_hdr.nexus.initid.id) % worker_threads;
14215	thr = &softc->threads[idx];
14216	mtx_lock(&thr->queue_lock);
14217	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14218	mtx_unlock(&thr->queue_lock);
14219	wakeup(thr);
14220}
14221
14222static void
14223ctl_enqueue_rtr(union ctl_io *io)
14224{
14225	struct ctl_softc *softc = control_softc;
14226	struct ctl_thread *thr;
14227
14228	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14229	mtx_lock(&thr->queue_lock);
14230	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14231	mtx_unlock(&thr->queue_lock);
14232	wakeup(thr);
14233}
14234
14235static void
14236ctl_enqueue_done(union ctl_io *io)
14237{
14238	struct ctl_softc *softc = control_softc;
14239	struct ctl_thread *thr;
14240
14241	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14242	mtx_lock(&thr->queue_lock);
14243	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14244	mtx_unlock(&thr->queue_lock);
14245	wakeup(thr);
14246}
14247
14248static void
14249ctl_enqueue_isc(union ctl_io *io)
14250{
14251	struct ctl_softc *softc = control_softc;
14252	struct ctl_thread *thr;
14253
14254	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14255	mtx_lock(&thr->queue_lock);
14256	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14257	mtx_unlock(&thr->queue_lock);
14258	wakeup(thr);
14259}
14260
14261/* Initialization and failover */
14262
14263void
14264ctl_init_isc_msg(void)
14265{
14266	printf("CTL: Still calling this thing\n");
14267}
14268
14269/*
14270 * Init component
14271 * 	Initializes component into configuration defined by bootMode
14272 *	(see hasc-sv.c)
14273 *  	returns hasc_Status:
14274 * 		OK
14275 *		ERROR - fatal error
14276 */
14277static ctl_ha_comp_status
14278ctl_isc_init(struct ctl_ha_component *c)
14279{
14280	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14281
14282	c->status = ret;
14283	return ret;
14284}
14285
14286/* Start component
14287 * 	Starts component in state requested. If component starts successfully,
14288 *	it must set its own state to the requestrd state
14289 *	When requested state is HASC_STATE_HA, the component may refine it
14290 * 	by adding _SLAVE or _MASTER flags.
14291 *	Currently allowed state transitions are:
14292 *	UNKNOWN->HA		- initial startup
14293 *	UNKNOWN->SINGLE - initial startup when no parter detected
14294 *	HA->SINGLE		- failover
14295 * returns ctl_ha_comp_status:
14296 * 		OK	- component successfully started in requested state
14297 *		FAILED  - could not start the requested state, failover may
14298 * 			  be possible
14299 *		ERROR	- fatal error detected, no future startup possible
14300 */
14301static ctl_ha_comp_status
14302ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14303{
14304	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14305
14306	printf("%s: go\n", __func__);
14307
14308	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14309	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14310		ctl_is_single = 0;
14311		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14312		    != CTL_HA_STATUS_SUCCESS) {
14313			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14314			ret = CTL_HA_COMP_STATUS_ERROR;
14315		}
14316	} else if (CTL_HA_STATE_IS_HA(c->state)
14317		&& CTL_HA_STATE_IS_SINGLE(state)){
14318		// HA->SINGLE transition
14319	        ctl_failover();
14320		ctl_is_single = 1;
14321	} else {
14322		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14323		       c->state, state);
14324		ret = CTL_HA_COMP_STATUS_ERROR;
14325	}
14326	if (CTL_HA_STATE_IS_SINGLE(state))
14327		ctl_is_single = 1;
14328
14329	c->state = state;
14330	c->status = ret;
14331	return ret;
14332}
14333
14334/*
14335 * Quiesce component
14336 * The component must clear any error conditions (set status to OK) and
14337 * prepare itself to another Start call
14338 * returns ctl_ha_comp_status:
14339 * 	OK
14340 *	ERROR
14341 */
14342static ctl_ha_comp_status
14343ctl_isc_quiesce(struct ctl_ha_component *c)
14344{
14345	int ret = CTL_HA_COMP_STATUS_OK;
14346
14347	ctl_pause_rtr = 1;
14348	c->status = ret;
14349	return ret;
14350}
14351
14352struct ctl_ha_component ctl_ha_component_ctlisc =
14353{
14354	.name = "CTL ISC",
14355	.state = CTL_HA_STATE_UNKNOWN,
14356	.init = ctl_isc_init,
14357	.start = ctl_isc_start,
14358	.quiesce = ctl_isc_quiesce
14359};
14360
14361/*
14362 *  vim: ts=8
14363 */
14364