ctl.c revision 273315
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 273315 2014-10-20 07:41:37Z mav $");
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/kernel.h>
50#include <sys/types.h>
51#include <sys/kthread.h>
52#include <sys/bio.h>
53#include <sys/fcntl.h>
54#include <sys/lock.h>
55#include <sys/module.h>
56#include <sys/mutex.h>
57#include <sys/condvar.h>
58#include <sys/malloc.h>
59#include <sys/conf.h>
60#include <sys/ioccom.h>
61#include <sys/queue.h>
62#include <sys/sbuf.h>
63#include <sys/smp.h>
64#include <sys/endian.h>
65#include <sys/sysctl.h>
66
67#include <cam/cam.h>
68#include <cam/scsi/scsi_all.h>
69#include <cam/scsi/scsi_da.h>
70#include <cam/ctl/ctl_io.h>
71#include <cam/ctl/ctl.h>
72#include <cam/ctl/ctl_frontend.h>
73#include <cam/ctl/ctl_frontend_internal.h>
74#include <cam/ctl/ctl_util.h>
75#include <cam/ctl/ctl_backend.h>
76#include <cam/ctl/ctl_ioctl.h>
77#include <cam/ctl/ctl_ha.h>
78#include <cam/ctl/ctl_private.h>
79#include <cam/ctl/ctl_debug.h>
80#include <cam/ctl/ctl_scsi_all.h>
81#include <cam/ctl/ctl_error.h>
82
83struct ctl_softc *control_softc = NULL;
84
85/*
86 * Size and alignment macros needed for Copan-specific HA hardware.  These
87 * can go away when the HA code is re-written, and uses busdma for any
88 * hardware.
89 */
90#define	CTL_ALIGN_8B(target, source, type)				\
91	if (((uint32_t)source & 0x7) != 0)				\
92		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
93	else								\
94		target = (type)source;
95
96#define	CTL_SIZE_8B(target, size)					\
97	if ((size & 0x7) != 0)						\
98		target = size + (0x8 - (size & 0x7));			\
99	else								\
100		target = size;
101
102#define CTL_ALIGN_8B_MARGIN	16
103
104/*
105 * Template mode pages.
106 */
107
108/*
109 * Note that these are default values only.  The actual values will be
110 * filled in when the user does a mode sense.
111 */
112static struct copan_power_subpage power_page_default = {
113	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
114	/*subpage*/ PWR_SUBPAGE_CODE,
115	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
116			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
117	/*page_version*/ PWR_VERSION,
118	/* total_luns */ 26,
119	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
120	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
121		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
122		      0, 0, 0, 0, 0, 0}
123};
124
125static struct copan_power_subpage power_page_changeable = {
126	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
127	/*subpage*/ PWR_SUBPAGE_CODE,
128	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
129			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
130	/*page_version*/ 0,
131	/* total_luns */ 0,
132	/* max_active_luns*/ 0,
133	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
134		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
135		      0, 0, 0, 0, 0, 0}
136};
137
138static struct copan_aps_subpage aps_page_default = {
139	APS_PAGE_CODE | SMPH_SPF, //page_code
140	APS_SUBPAGE_CODE, //subpage
141	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
142	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
143	APS_VERSION, //page_version
144	0, //lock_active
145	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
146	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
147	0, 0, 0, 0, 0} //reserved
148};
149
150static struct copan_aps_subpage aps_page_changeable = {
151	APS_PAGE_CODE | SMPH_SPF, //page_code
152	APS_SUBPAGE_CODE, //subpage
153	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
154	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
155	0, //page_version
156	0, //lock_active
157	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
158	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
159	0, 0, 0, 0, 0} //reserved
160};
161
162static struct copan_debugconf_subpage debugconf_page_default = {
163	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
164	DBGCNF_SUBPAGE_CODE,		/* subpage */
165	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
166	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
167	DBGCNF_VERSION,			/* page_version */
168	{CTL_TIME_IO_DEFAULT_SECS>>8,
169	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
170};
171
172static struct copan_debugconf_subpage debugconf_page_changeable = {
173	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
174	DBGCNF_SUBPAGE_CODE,		/* subpage */
175	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
176	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
177	0,				/* page_version */
178	{0xff,0xff},			/* ctl_time_io_secs */
179};
180
181static struct scsi_format_page format_page_default = {
182	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
183	/*page_length*/sizeof(struct scsi_format_page) - 2,
184	/*tracks_per_zone*/ {0, 0},
185	/*alt_sectors_per_zone*/ {0, 0},
186	/*alt_tracks_per_zone*/ {0, 0},
187	/*alt_tracks_per_lun*/ {0, 0},
188	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
189			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
190	/*bytes_per_sector*/ {0, 0},
191	/*interleave*/ {0, 0},
192	/*track_skew*/ {0, 0},
193	/*cylinder_skew*/ {0, 0},
194	/*flags*/ SFP_HSEC,
195	/*reserved*/ {0, 0, 0}
196};
197
198static struct scsi_format_page format_page_changeable = {
199	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
200	/*page_length*/sizeof(struct scsi_format_page) - 2,
201	/*tracks_per_zone*/ {0, 0},
202	/*alt_sectors_per_zone*/ {0, 0},
203	/*alt_tracks_per_zone*/ {0, 0},
204	/*alt_tracks_per_lun*/ {0, 0},
205	/*sectors_per_track*/ {0, 0},
206	/*bytes_per_sector*/ {0, 0},
207	/*interleave*/ {0, 0},
208	/*track_skew*/ {0, 0},
209	/*cylinder_skew*/ {0, 0},
210	/*flags*/ 0,
211	/*reserved*/ {0, 0, 0}
212};
213
214static struct scsi_rigid_disk_page rigid_disk_page_default = {
215	/*page_code*/SMS_RIGID_DISK_PAGE,
216	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
217	/*cylinders*/ {0, 0, 0},
218	/*heads*/ CTL_DEFAULT_HEADS,
219	/*start_write_precomp*/ {0, 0, 0},
220	/*start_reduced_current*/ {0, 0, 0},
221	/*step_rate*/ {0, 0},
222	/*landing_zone_cylinder*/ {0, 0, 0},
223	/*rpl*/ SRDP_RPL_DISABLED,
224	/*rotational_offset*/ 0,
225	/*reserved1*/ 0,
226	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
227			   CTL_DEFAULT_ROTATION_RATE & 0xff},
228	/*reserved2*/ {0, 0}
229};
230
231static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
232	/*page_code*/SMS_RIGID_DISK_PAGE,
233	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
234	/*cylinders*/ {0, 0, 0},
235	/*heads*/ 0,
236	/*start_write_precomp*/ {0, 0, 0},
237	/*start_reduced_current*/ {0, 0, 0},
238	/*step_rate*/ {0, 0},
239	/*landing_zone_cylinder*/ {0, 0, 0},
240	/*rpl*/ 0,
241	/*rotational_offset*/ 0,
242	/*reserved1*/ 0,
243	/*rotation_rate*/ {0, 0},
244	/*reserved2*/ {0, 0}
245};
246
247static struct scsi_caching_page caching_page_default = {
248	/*page_code*/SMS_CACHING_PAGE,
249	/*page_length*/sizeof(struct scsi_caching_page) - 2,
250	/*flags1*/ SCP_DISC | SCP_WCE,
251	/*ret_priority*/ 0,
252	/*disable_pf_transfer_len*/ {0xff, 0xff},
253	/*min_prefetch*/ {0, 0},
254	/*max_prefetch*/ {0xff, 0xff},
255	/*max_pf_ceiling*/ {0xff, 0xff},
256	/*flags2*/ 0,
257	/*cache_segments*/ 0,
258	/*cache_seg_size*/ {0, 0},
259	/*reserved*/ 0,
260	/*non_cache_seg_size*/ {0, 0, 0}
261};
262
263static struct scsi_caching_page caching_page_changeable = {
264	/*page_code*/SMS_CACHING_PAGE,
265	/*page_length*/sizeof(struct scsi_caching_page) - 2,
266	/*flags1*/ SCP_WCE | SCP_RCD,
267	/*ret_priority*/ 0,
268	/*disable_pf_transfer_len*/ {0, 0},
269	/*min_prefetch*/ {0, 0},
270	/*max_prefetch*/ {0, 0},
271	/*max_pf_ceiling*/ {0, 0},
272	/*flags2*/ 0,
273	/*cache_segments*/ 0,
274	/*cache_seg_size*/ {0, 0},
275	/*reserved*/ 0,
276	/*non_cache_seg_size*/ {0, 0, 0}
277};
278
279static struct scsi_control_page control_page_default = {
280	/*page_code*/SMS_CONTROL_MODE_PAGE,
281	/*page_length*/sizeof(struct scsi_control_page) - 2,
282	/*rlec*/0,
283	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
284	/*eca_and_aen*/0,
285	/*flags4*/SCP_TAS,
286	/*aen_holdoff_period*/{0, 0},
287	/*busy_timeout_period*/{0, 0},
288	/*extended_selftest_completion_time*/{0, 0}
289};
290
291static struct scsi_control_page control_page_changeable = {
292	/*page_code*/SMS_CONTROL_MODE_PAGE,
293	/*page_length*/sizeof(struct scsi_control_page) - 2,
294	/*rlec*/SCP_DSENSE,
295	/*queue_flags*/SCP_QUEUE_ALG_MASK,
296	/*eca_and_aen*/SCP_SWP,
297	/*flags4*/0,
298	/*aen_holdoff_period*/{0, 0},
299	/*busy_timeout_period*/{0, 0},
300	/*extended_selftest_completion_time*/{0, 0}
301};
302
303
304/*
305 * XXX KDM move these into the softc.
306 */
307static int rcv_sync_msg;
308static int persis_offset;
309static uint8_t ctl_pause_rtr;
310static int     ctl_is_single = 1;
311static int     index_to_aps_page;
312
313SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
314static int worker_threads = -1;
315TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
316SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
317    &worker_threads, 1, "Number of worker threads");
318static int verbose = 0;
319TUNABLE_INT("kern.cam.ctl.verbose", &verbose);
320SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
321    &verbose, 0, "Show SCSI errors returned to initiator");
322
323/*
324 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
325 * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
326 * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
327 * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
328 */
329#define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
330
331static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
332				  int param);
333static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
334static int ctl_init(void);
335void ctl_shutdown(void);
336static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
337static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
338static void ctl_ioctl_online(void *arg);
339static void ctl_ioctl_offline(void *arg);
340static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
341static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
342static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
343static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
344static int ctl_ioctl_submit_wait(union ctl_io *io);
345static void ctl_ioctl_datamove(union ctl_io *io);
346static void ctl_ioctl_done(union ctl_io *io);
347static void ctl_ioctl_hard_startstop_callback(void *arg,
348					      struct cfi_metatask *metatask);
349static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
350static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
351			      struct ctl_ooa *ooa_hdr,
352			      struct ctl_ooa_entry *kern_entries);
353static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
354		     struct thread *td);
355static uint32_t ctl_map_lun(int port_num, uint32_t lun);
356static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
357#ifdef unused
358static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
359				   uint32_t targ_target, uint32_t targ_lun,
360				   int can_wait);
361static void ctl_kfree_io(union ctl_io *io);
362#endif /* unused */
363static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
364			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
365static int ctl_free_lun(struct ctl_lun *lun);
366static void ctl_create_lun(struct ctl_be_lun *be_lun);
367/**
368static void ctl_failover_change_pages(struct ctl_softc *softc,
369				      struct ctl_scsiio *ctsio, int master);
370**/
371
372static int ctl_do_mode_select(union ctl_io *io);
373static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
374			   uint64_t res_key, uint64_t sa_res_key,
375			   uint8_t type, uint32_t residx,
376			   struct ctl_scsiio *ctsio,
377			   struct scsi_per_res_out *cdb,
378			   struct scsi_per_res_out_parms* param);
379static void ctl_pro_preempt_other(struct ctl_lun *lun,
380				  union ctl_ha_msg *msg);
381static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
382static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
383static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
384static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
385static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
386static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
387static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
388					 int alloc_len);
389static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
390					 int alloc_len);
391static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
392static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
393static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
394static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
395static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
396static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
397static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
398    union ctl_io *pending_io, union ctl_io *ooa_io);
399static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
400				union ctl_io *starting_io);
401static int ctl_check_blocked(struct ctl_lun *lun);
402static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
403				struct ctl_lun *lun,
404				const struct ctl_cmd_entry *entry,
405				struct ctl_scsiio *ctsio);
406//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
407static void ctl_failover(void);
408static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
409			       struct ctl_scsiio *ctsio);
410static int ctl_scsiio(struct ctl_scsiio *ctsio);
411
412static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
413static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
414			    ctl_ua_type ua_type);
415static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
416			 ctl_ua_type ua_type);
417static int ctl_abort_task(union ctl_io *io);
418static int ctl_abort_task_set(union ctl_io *io);
419static int ctl_i_t_nexus_reset(union ctl_io *io);
420static void ctl_run_task(union ctl_io *io);
421#ifdef CTL_IO_DELAY
422static void ctl_datamove_timer_wakeup(void *arg);
423static void ctl_done_timer_wakeup(void *arg);
424#endif /* CTL_IO_DELAY */
425
426static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
427static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
428static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
429static void ctl_datamove_remote_write(union ctl_io *io);
430static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
431static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
432static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
433static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
434				    ctl_ha_dt_cb callback);
435static void ctl_datamove_remote_read(union ctl_io *io);
436static void ctl_datamove_remote(union ctl_io *io);
437static int ctl_process_done(union ctl_io *io);
438static void ctl_lun_thread(void *arg);
439static void ctl_work_thread(void *arg);
440static void ctl_enqueue_incoming(union ctl_io *io);
441static void ctl_enqueue_rtr(union ctl_io *io);
442static void ctl_enqueue_done(union ctl_io *io);
443static void ctl_enqueue_isc(union ctl_io *io);
444static const struct ctl_cmd_entry *
445    ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
446static const struct ctl_cmd_entry *
447    ctl_validate_command(struct ctl_scsiio *ctsio);
448static int ctl_cmd_applicable(uint8_t lun_type,
449    const struct ctl_cmd_entry *entry);
450
451/*
452 * Load the serialization table.  This isn't very pretty, but is probably
453 * the easiest way to do it.
454 */
455#include "ctl_ser_table.c"
456
457/*
458 * We only need to define open, close and ioctl routines for this driver.
459 */
460static struct cdevsw ctl_cdevsw = {
461	.d_version =	D_VERSION,
462	.d_flags =	0,
463	.d_open =	ctl_open,
464	.d_close =	ctl_close,
465	.d_ioctl =	ctl_ioctl,
466	.d_name =	"ctl",
467};
468
469
470MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
471MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
472
473static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
474
475static moduledata_t ctl_moduledata = {
476	"ctl",
477	ctl_module_event_handler,
478	NULL
479};
480
481DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
482MODULE_VERSION(ctl, 1);
483
484static struct ctl_frontend ioctl_frontend =
485{
486	.name = "ioctl",
487};
488
489static void
490ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
491			    union ctl_ha_msg *msg_info)
492{
493	struct ctl_scsiio *ctsio;
494
495	if (msg_info->hdr.original_sc == NULL) {
496		printf("%s: original_sc == NULL!\n", __func__);
497		/* XXX KDM now what? */
498		return;
499	}
500
501	ctsio = &msg_info->hdr.original_sc->scsiio;
502	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
503	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
504	ctsio->io_hdr.status = msg_info->hdr.status;
505	ctsio->scsi_status = msg_info->scsi.scsi_status;
506	ctsio->sense_len = msg_info->scsi.sense_len;
507	ctsio->sense_residual = msg_info->scsi.sense_residual;
508	ctsio->residual = msg_info->scsi.residual;
509	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
510	       sizeof(ctsio->sense_data));
511	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
512	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
513	ctl_enqueue_isc((union ctl_io *)ctsio);
514}
515
516static void
517ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
518				union ctl_ha_msg *msg_info)
519{
520	struct ctl_scsiio *ctsio;
521
522	if (msg_info->hdr.serializing_sc == NULL) {
523		printf("%s: serializing_sc == NULL!\n", __func__);
524		/* XXX KDM now what? */
525		return;
526	}
527
528	ctsio = &msg_info->hdr.serializing_sc->scsiio;
529#if 0
530	/*
531	 * Attempt to catch the situation where an I/O has
532	 * been freed, and we're using it again.
533	 */
534	if (ctsio->io_hdr.io_type == 0xff) {
535		union ctl_io *tmp_io;
536		tmp_io = (union ctl_io *)ctsio;
537		printf("%s: %p use after free!\n", __func__,
538		       ctsio);
539		printf("%s: type %d msg %d cdb %x iptl: "
540		       "%d:%d:%d:%d tag 0x%04x "
541		       "flag %#x status %x\n",
542			__func__,
543			tmp_io->io_hdr.io_type,
544			tmp_io->io_hdr.msg_type,
545			tmp_io->scsiio.cdb[0],
546			tmp_io->io_hdr.nexus.initid.id,
547			tmp_io->io_hdr.nexus.targ_port,
548			tmp_io->io_hdr.nexus.targ_target.id,
549			tmp_io->io_hdr.nexus.targ_lun,
550			(tmp_io->io_hdr.io_type ==
551			CTL_IO_TASK) ?
552			tmp_io->taskio.tag_num :
553			tmp_io->scsiio.tag_num,
554		        tmp_io->io_hdr.flags,
555			tmp_io->io_hdr.status);
556	}
557#endif
558	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
559	ctl_enqueue_isc((union ctl_io *)ctsio);
560}
561
562/*
563 * ISC (Inter Shelf Communication) event handler.  Events from the HA
564 * subsystem come in here.
565 */
566static void
567ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
568{
569	struct ctl_softc *ctl_softc;
570	union ctl_io *io;
571	struct ctl_prio *presio;
572	ctl_ha_status isc_status;
573
574	ctl_softc = control_softc;
575	io = NULL;
576
577
578#if 0
579	printf("CTL: Isc Msg event %d\n", event);
580#endif
581	if (event == CTL_HA_EVT_MSG_RECV) {
582		union ctl_ha_msg msg_info;
583
584		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
585					     sizeof(msg_info), /*wait*/ 0);
586#if 0
587		printf("CTL: msg_type %d\n", msg_info.msg_type);
588#endif
589		if (isc_status != 0) {
590			printf("Error receiving message, status = %d\n",
591			       isc_status);
592			return;
593		}
594
595		switch (msg_info.hdr.msg_type) {
596		case CTL_MSG_SERIALIZE:
597#if 0
598			printf("Serialize\n");
599#endif
600			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
601			if (io == NULL) {
602				printf("ctl_isc_event_handler: can't allocate "
603				       "ctl_io!\n");
604				/* Bad Juju */
605				/* Need to set busy and send msg back */
606				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
607				msg_info.hdr.status = CTL_SCSI_ERROR;
608				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
609				msg_info.scsi.sense_len = 0;
610			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
611				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
612				}
613				goto bailout;
614			}
615			ctl_zero_io(io);
616			// populate ctsio from msg_info
617			io->io_hdr.io_type = CTL_IO_SCSI;
618			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
619			io->io_hdr.original_sc = msg_info.hdr.original_sc;
620#if 0
621			printf("pOrig %x\n", (int)msg_info.original_sc);
622#endif
623			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
624					    CTL_FLAG_IO_ACTIVE;
625			/*
626			 * If we're in serialization-only mode, we don't
627			 * want to go through full done processing.  Thus
628			 * the COPY flag.
629			 *
630			 * XXX KDM add another flag that is more specific.
631			 */
632			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
633				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
634			io->io_hdr.nexus = msg_info.hdr.nexus;
635#if 0
636			printf("targ %d, port %d, iid %d, lun %d\n",
637			       io->io_hdr.nexus.targ_target.id,
638			       io->io_hdr.nexus.targ_port,
639			       io->io_hdr.nexus.initid.id,
640			       io->io_hdr.nexus.targ_lun);
641#endif
642			io->scsiio.tag_num = msg_info.scsi.tag_num;
643			io->scsiio.tag_type = msg_info.scsi.tag_type;
644			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
645			       CTL_MAX_CDBLEN);
646			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
647				const struct ctl_cmd_entry *entry;
648
649				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
650				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
651				io->io_hdr.flags |=
652					entry->flags & CTL_FLAG_DATA_MASK;
653			}
654			ctl_enqueue_isc(io);
655			break;
656
657		/* Performed on the Originating SC, XFER mode only */
658		case CTL_MSG_DATAMOVE: {
659			struct ctl_sg_entry *sgl;
660			int i, j;
661
662			io = msg_info.hdr.original_sc;
663			if (io == NULL) {
664				printf("%s: original_sc == NULL!\n", __func__);
665				/* XXX KDM do something here */
666				break;
667			}
668			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
669			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
670			/*
671			 * Keep track of this, we need to send it back over
672			 * when the datamove is complete.
673			 */
674			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
675
676			if (msg_info.dt.sg_sequence == 0) {
677				/*
678				 * XXX KDM we use the preallocated S/G list
679				 * here, but we'll need to change this to
680				 * dynamic allocation if we need larger S/G
681				 * lists.
682				 */
683				if (msg_info.dt.kern_sg_entries >
684				    sizeof(io->io_hdr.remote_sglist) /
685				    sizeof(io->io_hdr.remote_sglist[0])) {
686					printf("%s: number of S/G entries "
687					    "needed %u > allocated num %zd\n",
688					    __func__,
689					    msg_info.dt.kern_sg_entries,
690					    sizeof(io->io_hdr.remote_sglist)/
691					    sizeof(io->io_hdr.remote_sglist[0]));
692
693					/*
694					 * XXX KDM send a message back to
695					 * the other side to shut down the
696					 * DMA.  The error will come back
697					 * through via the normal channel.
698					 */
699					break;
700				}
701				sgl = io->io_hdr.remote_sglist;
702				memset(sgl, 0,
703				       sizeof(io->io_hdr.remote_sglist));
704
705				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
706
707				io->scsiio.kern_sg_entries =
708					msg_info.dt.kern_sg_entries;
709				io->scsiio.rem_sg_entries =
710					msg_info.dt.kern_sg_entries;
711				io->scsiio.kern_data_len =
712					msg_info.dt.kern_data_len;
713				io->scsiio.kern_total_len =
714					msg_info.dt.kern_total_len;
715				io->scsiio.kern_data_resid =
716					msg_info.dt.kern_data_resid;
717				io->scsiio.kern_rel_offset =
718					msg_info.dt.kern_rel_offset;
719				/*
720				 * Clear out per-DMA flags.
721				 */
722				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
723				/*
724				 * Add per-DMA flags that are set for this
725				 * particular DMA request.
726				 */
727				io->io_hdr.flags |= msg_info.dt.flags &
728						    CTL_FLAG_RDMA_MASK;
729			} else
730				sgl = (struct ctl_sg_entry *)
731					io->scsiio.kern_data_ptr;
732
733			for (i = msg_info.dt.sent_sg_entries, j = 0;
734			     i < (msg_info.dt.sent_sg_entries +
735			     msg_info.dt.cur_sg_entries); i++, j++) {
736				sgl[i].addr = msg_info.dt.sg_list[j].addr;
737				sgl[i].len = msg_info.dt.sg_list[j].len;
738
739#if 0
740				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
741				       __func__,
742				       msg_info.dt.sg_list[j].addr,
743				       msg_info.dt.sg_list[j].len,
744				       sgl[i].addr, sgl[i].len, j, i);
745#endif
746			}
747#if 0
748			memcpy(&sgl[msg_info.dt.sent_sg_entries],
749			       msg_info.dt.sg_list,
750			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
751#endif
752
753			/*
754			 * If this is the last piece of the I/O, we've got
755			 * the full S/G list.  Queue processing in the thread.
756			 * Otherwise wait for the next piece.
757			 */
758			if (msg_info.dt.sg_last != 0)
759				ctl_enqueue_isc(io);
760			break;
761		}
762		/* Performed on the Serializing (primary) SC, XFER mode only */
763		case CTL_MSG_DATAMOVE_DONE: {
764			if (msg_info.hdr.serializing_sc == NULL) {
765				printf("%s: serializing_sc == NULL!\n",
766				       __func__);
767				/* XXX KDM now what? */
768				break;
769			}
770			/*
771			 * We grab the sense information here in case
772			 * there was a failure, so we can return status
773			 * back to the initiator.
774			 */
775			io = msg_info.hdr.serializing_sc;
776			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
777			io->io_hdr.status = msg_info.hdr.status;
778			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
779			io->scsiio.sense_len = msg_info.scsi.sense_len;
780			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
781			io->io_hdr.port_status = msg_info.scsi.fetd_status;
782			io->scsiio.residual = msg_info.scsi.residual;
783			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
784			       sizeof(io->scsiio.sense_data));
785			ctl_enqueue_isc(io);
786			break;
787		}
788
789		/* Preformed on Originating SC, SER_ONLY mode */
790		case CTL_MSG_R2R:
791			io = msg_info.hdr.original_sc;
792			if (io == NULL) {
793				printf("%s: Major Bummer\n", __func__);
794				return;
795			} else {
796#if 0
797				printf("pOrig %x\n",(int) ctsio);
798#endif
799			}
800			io->io_hdr.msg_type = CTL_MSG_R2R;
801			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
802			ctl_enqueue_isc(io);
803			break;
804
805		/*
806		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
807		 * mode.
808		 * Performed on the Originating (i.e. secondary) SC in XFER
809		 * mode
810		 */
811		case CTL_MSG_FINISH_IO:
812			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
813				ctl_isc_handler_finish_xfer(ctl_softc,
814							    &msg_info);
815			else
816				ctl_isc_handler_finish_ser_only(ctl_softc,
817								&msg_info);
818			break;
819
820		/* Preformed on Originating SC */
821		case CTL_MSG_BAD_JUJU:
822			io = msg_info.hdr.original_sc;
823			if (io == NULL) {
824				printf("%s: Bad JUJU!, original_sc is NULL!\n",
825				       __func__);
826				break;
827			}
828			ctl_copy_sense_data(&msg_info, io);
829			/*
830			 * IO should have already been cleaned up on other
831			 * SC so clear this flag so we won't send a message
832			 * back to finish the IO there.
833			 */
834			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
835			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
836
837			/* io = msg_info.hdr.serializing_sc; */
838			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
839			ctl_enqueue_isc(io);
840			break;
841
842		/* Handle resets sent from the other side */
843		case CTL_MSG_MANAGE_TASKS: {
844			struct ctl_taskio *taskio;
845			taskio = (struct ctl_taskio *)ctl_alloc_io(
846				(void *)ctl_softc->othersc_pool);
847			if (taskio == NULL) {
848				printf("ctl_isc_event_handler: can't allocate "
849				       "ctl_io!\n");
850				/* Bad Juju */
851				/* should I just call the proper reset func
852				   here??? */
853				goto bailout;
854			}
855			ctl_zero_io((union ctl_io *)taskio);
856			taskio->io_hdr.io_type = CTL_IO_TASK;
857			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
858			taskio->io_hdr.nexus = msg_info.hdr.nexus;
859			taskio->task_action = msg_info.task.task_action;
860			taskio->tag_num = msg_info.task.tag_num;
861			taskio->tag_type = msg_info.task.tag_type;
862#ifdef CTL_TIME_IO
863			taskio->io_hdr.start_time = time_uptime;
864			getbintime(&taskio->io_hdr.start_bt);
865#if 0
866			cs_prof_gettime(&taskio->io_hdr.start_ticks);
867#endif
868#endif /* CTL_TIME_IO */
869			ctl_run_task((union ctl_io *)taskio);
870			break;
871		}
872		/* Persistent Reserve action which needs attention */
873		case CTL_MSG_PERS_ACTION:
874			presio = (struct ctl_prio *)ctl_alloc_io(
875				(void *)ctl_softc->othersc_pool);
876			if (presio == NULL) {
877				printf("ctl_isc_event_handler: can't allocate "
878				       "ctl_io!\n");
879				/* Bad Juju */
880				/* Need to set busy and send msg back */
881				goto bailout;
882			}
883			ctl_zero_io((union ctl_io *)presio);
884			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
885			presio->pr_msg = msg_info.pr;
886			ctl_enqueue_isc((union ctl_io *)presio);
887			break;
888		case CTL_MSG_SYNC_FE:
889			rcv_sync_msg = 1;
890			break;
891		case CTL_MSG_APS_LOCK: {
892			// It's quicker to execute this then to
893			// queue it.
894			struct ctl_lun *lun;
895			struct ctl_page_index *page_index;
896			struct copan_aps_subpage *current_sp;
897			uint32_t targ_lun;
898
899			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
900			lun = ctl_softc->ctl_luns[targ_lun];
901			mtx_lock(&lun->lun_lock);
902			page_index = &lun->mode_pages.index[index_to_aps_page];
903			current_sp = (struct copan_aps_subpage *)
904				     (page_index->page_data +
905				     (page_index->page_len * CTL_PAGE_CURRENT));
906
907			current_sp->lock_active = msg_info.aps.lock_flag;
908			mtx_unlock(&lun->lun_lock);
909		        break;
910		}
911		default:
912		        printf("How did I get here?\n");
913		}
914	} else if (event == CTL_HA_EVT_MSG_SENT) {
915		if (param != CTL_HA_STATUS_SUCCESS) {
916			printf("Bad status from ctl_ha_msg_send status %d\n",
917			       param);
918		}
919		return;
920	} else if (event == CTL_HA_EVT_DISCONNECT) {
921		printf("CTL: Got a disconnect from Isc\n");
922		return;
923	} else {
924		printf("ctl_isc_event_handler: Unknown event %d\n", event);
925		return;
926	}
927
928bailout:
929	return;
930}
931
932static void
933ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
934{
935	struct scsi_sense_data *sense;
936
937	sense = &dest->scsiio.sense_data;
938	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
939	dest->scsiio.scsi_status = src->scsi.scsi_status;
940	dest->scsiio.sense_len = src->scsi.sense_len;
941	dest->io_hdr.status = src->hdr.status;
942}
943
944static int
945ctl_init(void)
946{
947	struct ctl_softc *softc;
948	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
949	struct ctl_port *port;
950        uint8_t sc_id =0;
951	int i, error, retval;
952	//int isc_retval;
953
954	retval = 0;
955	ctl_pause_rtr = 0;
956        rcv_sync_msg = 0;
957
958	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
959			       M_WAITOK | M_ZERO);
960	softc = control_softc;
961
962	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
963			      "cam/ctl");
964
965	softc->dev->si_drv1 = softc;
966
967	/*
968	 * By default, return a "bad LUN" peripheral qualifier for unknown
969	 * LUNs.  The user can override this default using the tunable or
970	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
971	 */
972	softc->inquiry_pq_no_lun = 1;
973	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
974			  &softc->inquiry_pq_no_lun);
975	sysctl_ctx_init(&softc->sysctl_ctx);
976	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
977		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
978		CTLFLAG_RD, 0, "CAM Target Layer");
979
980	if (softc->sysctl_tree == NULL) {
981		printf("%s: unable to allocate sysctl tree\n", __func__);
982		destroy_dev(softc->dev);
983		free(control_softc, M_DEVBUF);
984		control_softc = NULL;
985		return (ENOMEM);
986	}
987
988	SYSCTL_ADD_INT(&softc->sysctl_ctx,
989		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
990		       "inquiry_pq_no_lun", CTLFLAG_RW,
991		       &softc->inquiry_pq_no_lun, 0,
992		       "Report no lun possible for invalid LUNs");
993
994	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
995	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
996	softc->open_count = 0;
997
998	/*
999	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1000	 * the drive.
1001	 */
1002	softc->flags = CTL_FLAG_REAL_SYNC;
1003
1004	/*
1005	 * In Copan's HA scheme, the "master" and "slave" roles are
1006	 * figured out through the slot the controller is in.  Although it
1007	 * is an active/active system, someone has to be in charge.
1008 	 */
1009#ifdef NEEDTOPORT
1010        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1011#endif
1012
1013        if (sc_id == 0) {
1014		softc->flags |= CTL_FLAG_MASTER_SHELF;
1015		persis_offset = 0;
1016	} else
1017		persis_offset = CTL_MAX_INITIATORS;
1018
1019	/*
1020	 * XXX KDM need to figure out where we want to get our target ID
1021	 * and WWID.  Is it different on each port?
1022	 */
1023	softc->target.id = 0;
1024	softc->target.wwid[0] = 0x12345678;
1025	softc->target.wwid[1] = 0x87654321;
1026	STAILQ_INIT(&softc->lun_list);
1027	STAILQ_INIT(&softc->pending_lun_queue);
1028	STAILQ_INIT(&softc->fe_list);
1029	STAILQ_INIT(&softc->port_list);
1030	STAILQ_INIT(&softc->be_list);
1031	STAILQ_INIT(&softc->io_pools);
1032	ctl_tpc_init(softc);
1033
1034	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1035			    &internal_pool)!= 0){
1036		printf("ctl: can't allocate %d entry internal pool, "
1037		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1038		return (ENOMEM);
1039	}
1040
1041	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1042			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1043		printf("ctl: can't allocate %d entry emergency pool, "
1044		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1045		ctl_pool_free(internal_pool);
1046		return (ENOMEM);
1047	}
1048
1049	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1050	                    &other_pool) != 0)
1051	{
1052		printf("ctl: can't allocate %d entry other SC pool, "
1053		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1054		ctl_pool_free(internal_pool);
1055		ctl_pool_free(emergency_pool);
1056		return (ENOMEM);
1057	}
1058
1059	softc->internal_pool = internal_pool;
1060	softc->emergency_pool = emergency_pool;
1061	softc->othersc_pool = other_pool;
1062
1063	if (worker_threads <= 0)
1064		worker_threads = max(1, mp_ncpus / 4);
1065	if (worker_threads > CTL_MAX_THREADS)
1066		worker_threads = CTL_MAX_THREADS;
1067
1068	for (i = 0; i < worker_threads; i++) {
1069		struct ctl_thread *thr = &softc->threads[i];
1070
1071		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1072		thr->ctl_softc = softc;
1073		STAILQ_INIT(&thr->incoming_queue);
1074		STAILQ_INIT(&thr->rtr_queue);
1075		STAILQ_INIT(&thr->done_queue);
1076		STAILQ_INIT(&thr->isc_queue);
1077
1078		error = kproc_kthread_add(ctl_work_thread, thr,
1079		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1080		if (error != 0) {
1081			printf("error creating CTL work thread!\n");
1082			ctl_pool_free(internal_pool);
1083			ctl_pool_free(emergency_pool);
1084			ctl_pool_free(other_pool);
1085			return (error);
1086		}
1087	}
1088	error = kproc_kthread_add(ctl_lun_thread, softc,
1089	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1090	if (error != 0) {
1091		printf("error creating CTL lun thread!\n");
1092		ctl_pool_free(internal_pool);
1093		ctl_pool_free(emergency_pool);
1094		ctl_pool_free(other_pool);
1095		return (error);
1096	}
1097	if (bootverbose)
1098		printf("ctl: CAM Target Layer loaded\n");
1099
1100	/*
1101	 * Initialize the ioctl front end.
1102	 */
1103	ctl_frontend_register(&ioctl_frontend);
1104	port = &softc->ioctl_info.port;
1105	port->frontend = &ioctl_frontend;
1106	sprintf(softc->ioctl_info.port_name, "ioctl");
1107	port->port_type = CTL_PORT_IOCTL;
1108	port->num_requested_ctl_io = 100;
1109	port->port_name = softc->ioctl_info.port_name;
1110	port->port_online = ctl_ioctl_online;
1111	port->port_offline = ctl_ioctl_offline;
1112	port->onoff_arg = &softc->ioctl_info;
1113	port->lun_enable = ctl_ioctl_lun_enable;
1114	port->lun_disable = ctl_ioctl_lun_disable;
1115	port->targ_lun_arg = &softc->ioctl_info;
1116	port->fe_datamove = ctl_ioctl_datamove;
1117	port->fe_done = ctl_ioctl_done;
1118	port->max_targets = 15;
1119	port->max_target_id = 15;
1120
1121	if (ctl_port_register(&softc->ioctl_info.port,
1122	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1123		printf("ctl: ioctl front end registration failed, will "
1124		       "continue anyway\n");
1125	}
1126
1127#ifdef CTL_IO_DELAY
1128	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1129		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1130		       sizeof(struct callout), CTL_TIMER_BYTES);
1131		return (EINVAL);
1132	}
1133#endif /* CTL_IO_DELAY */
1134
1135	return (0);
1136}
1137
1138void
1139ctl_shutdown(void)
1140{
1141	struct ctl_softc *softc;
1142	struct ctl_lun *lun, *next_lun;
1143	struct ctl_io_pool *pool;
1144
1145	softc = (struct ctl_softc *)control_softc;
1146
1147	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1148		printf("ctl: ioctl front end deregistration failed\n");
1149
1150	mtx_lock(&softc->ctl_lock);
1151
1152	/*
1153	 * Free up each LUN.
1154	 */
1155	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1156		next_lun = STAILQ_NEXT(lun, links);
1157		ctl_free_lun(lun);
1158	}
1159
1160	mtx_unlock(&softc->ctl_lock);
1161
1162	ctl_frontend_deregister(&ioctl_frontend);
1163
1164	/*
1165	 * This will rip the rug out from under any FETDs or anyone else
1166	 * that has a pool allocated.  Since we increment our module
1167	 * refcount any time someone outside the main CTL module allocates
1168	 * a pool, we shouldn't have any problems here.  The user won't be
1169	 * able to unload the CTL module until client modules have
1170	 * successfully unloaded.
1171	 */
1172	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1173		ctl_pool_free(pool);
1174
1175#if 0
1176	ctl_shutdown_thread(softc->work_thread);
1177	mtx_destroy(&softc->queue_lock);
1178#endif
1179
1180	ctl_tpc_shutdown(softc);
1181	mtx_destroy(&softc->pool_lock);
1182	mtx_destroy(&softc->ctl_lock);
1183
1184	destroy_dev(softc->dev);
1185
1186	sysctl_ctx_free(&softc->sysctl_ctx);
1187
1188	free(control_softc, M_DEVBUF);
1189	control_softc = NULL;
1190
1191	if (bootverbose)
1192		printf("ctl: CAM Target Layer unloaded\n");
1193}
1194
1195static int
1196ctl_module_event_handler(module_t mod, int what, void *arg)
1197{
1198
1199	switch (what) {
1200	case MOD_LOAD:
1201		return (ctl_init());
1202	case MOD_UNLOAD:
1203		return (EBUSY);
1204	default:
1205		return (EOPNOTSUPP);
1206	}
1207}
1208
1209/*
1210 * XXX KDM should we do some access checks here?  Bump a reference count to
1211 * prevent a CTL module from being unloaded while someone has it open?
1212 */
1213static int
1214ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1215{
1216	return (0);
1217}
1218
1219static int
1220ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1221{
1222	return (0);
1223}
1224
1225int
1226ctl_port_enable(ctl_port_type port_type)
1227{
1228	struct ctl_softc *softc;
1229	struct ctl_port *port;
1230
1231	if (ctl_is_single == 0) {
1232		union ctl_ha_msg msg_info;
1233		int isc_retval;
1234
1235#if 0
1236		printf("%s: HA mode, synchronizing frontend enable\n",
1237		        __func__);
1238#endif
1239		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1240	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1241		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1242			printf("Sync msg send error retval %d\n", isc_retval);
1243		}
1244		if (!rcv_sync_msg) {
1245			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1246			        sizeof(msg_info), 1);
1247		}
1248#if 0
1249        	printf("CTL:Frontend Enable\n");
1250	} else {
1251		printf("%s: single mode, skipping frontend synchronization\n",
1252		        __func__);
1253#endif
1254	}
1255
1256	softc = control_softc;
1257
1258	STAILQ_FOREACH(port, &softc->port_list, links) {
1259		if (port_type & port->port_type)
1260		{
1261#if 0
1262			printf("port %d\n", port->targ_port);
1263#endif
1264			ctl_port_online(port);
1265		}
1266	}
1267
1268	return (0);
1269}
1270
1271int
1272ctl_port_disable(ctl_port_type port_type)
1273{
1274	struct ctl_softc *softc;
1275	struct ctl_port *port;
1276
1277	softc = control_softc;
1278
1279	STAILQ_FOREACH(port, &softc->port_list, links) {
1280		if (port_type & port->port_type)
1281			ctl_port_offline(port);
1282	}
1283
1284	return (0);
1285}
1286
1287/*
1288 * Returns 0 for success, 1 for failure.
1289 * Currently the only failure mode is if there aren't enough entries
1290 * allocated.  So, in case of a failure, look at num_entries_dropped,
1291 * reallocate and try again.
1292 */
1293int
1294ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1295	      int *num_entries_filled, int *num_entries_dropped,
1296	      ctl_port_type port_type, int no_virtual)
1297{
1298	struct ctl_softc *softc;
1299	struct ctl_port *port;
1300	int entries_dropped, entries_filled;
1301	int retval;
1302	int i;
1303
1304	softc = control_softc;
1305
1306	retval = 0;
1307	entries_filled = 0;
1308	entries_dropped = 0;
1309
1310	i = 0;
1311	mtx_lock(&softc->ctl_lock);
1312	STAILQ_FOREACH(port, &softc->port_list, links) {
1313		struct ctl_port_entry *entry;
1314
1315		if ((port->port_type & port_type) == 0)
1316			continue;
1317
1318		if ((no_virtual != 0)
1319		 && (port->virtual_port != 0))
1320			continue;
1321
1322		if (entries_filled >= num_entries_alloced) {
1323			entries_dropped++;
1324			continue;
1325		}
1326		entry = &entries[i];
1327
1328		entry->port_type = port->port_type;
1329		strlcpy(entry->port_name, port->port_name,
1330			sizeof(entry->port_name));
1331		entry->physical_port = port->physical_port;
1332		entry->virtual_port = port->virtual_port;
1333		entry->wwnn = port->wwnn;
1334		entry->wwpn = port->wwpn;
1335
1336		i++;
1337		entries_filled++;
1338	}
1339
1340	mtx_unlock(&softc->ctl_lock);
1341
1342	if (entries_dropped > 0)
1343		retval = 1;
1344
1345	*num_entries_dropped = entries_dropped;
1346	*num_entries_filled = entries_filled;
1347
1348	return (retval);
1349}
1350
1351static void
1352ctl_ioctl_online(void *arg)
1353{
1354	struct ctl_ioctl_info *ioctl_info;
1355
1356	ioctl_info = (struct ctl_ioctl_info *)arg;
1357
1358	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1359}
1360
1361static void
1362ctl_ioctl_offline(void *arg)
1363{
1364	struct ctl_ioctl_info *ioctl_info;
1365
1366	ioctl_info = (struct ctl_ioctl_info *)arg;
1367
1368	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1369}
1370
1371/*
1372 * Remove an initiator by port number and initiator ID.
1373 * Returns 0 for success, -1 for failure.
1374 */
1375int
1376ctl_remove_initiator(struct ctl_port *port, int iid)
1377{
1378	struct ctl_softc *softc = control_softc;
1379
1380	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1381
1382	if (iid > CTL_MAX_INIT_PER_PORT) {
1383		printf("%s: initiator ID %u > maximun %u!\n",
1384		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1385		return (-1);
1386	}
1387
1388	mtx_lock(&softc->ctl_lock);
1389	port->wwpn_iid[iid].in_use--;
1390	port->wwpn_iid[iid].last_use = time_uptime;
1391	mtx_unlock(&softc->ctl_lock);
1392
1393	return (0);
1394}
1395
1396/*
1397 * Add an initiator to the initiator map.
1398 * Returns iid for success, < 0 for failure.
1399 */
1400int
1401ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1402{
1403	struct ctl_softc *softc = control_softc;
1404	time_t best_time;
1405	int i, best;
1406
1407	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1408
1409	if (iid >= CTL_MAX_INIT_PER_PORT) {
1410		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1411		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1412		free(name, M_CTL);
1413		return (-1);
1414	}
1415
1416	mtx_lock(&softc->ctl_lock);
1417
1418	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1419		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1420			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1421				iid = i;
1422				break;
1423			}
1424			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1425			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1426				iid = i;
1427				break;
1428			}
1429		}
1430	}
1431
1432	if (iid < 0) {
1433		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1434			if (port->wwpn_iid[i].in_use == 0 &&
1435			    port->wwpn_iid[i].wwpn == 0 &&
1436			    port->wwpn_iid[i].name == NULL) {
1437				iid = i;
1438				break;
1439			}
1440		}
1441	}
1442
1443	if (iid < 0) {
1444		best = -1;
1445		best_time = INT32_MAX;
1446		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1447			if (port->wwpn_iid[i].in_use == 0) {
1448				if (port->wwpn_iid[i].last_use < best_time) {
1449					best = i;
1450					best_time = port->wwpn_iid[i].last_use;
1451				}
1452			}
1453		}
1454		iid = best;
1455	}
1456
1457	if (iid < 0) {
1458		mtx_unlock(&softc->ctl_lock);
1459		free(name, M_CTL);
1460		return (-2);
1461	}
1462
1463	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1464		/*
1465		 * This is not an error yet.
1466		 */
1467		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1468#if 0
1469			printf("%s: port %d iid %u WWPN %#jx arrived"
1470			    " again\n", __func__, port->targ_port,
1471			    iid, (uintmax_t)wwpn);
1472#endif
1473			goto take;
1474		}
1475		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1476		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1477#if 0
1478			printf("%s: port %d iid %u name '%s' arrived"
1479			    " again\n", __func__, port->targ_port,
1480			    iid, name);
1481#endif
1482			goto take;
1483		}
1484
1485		/*
1486		 * This is an error, but what do we do about it?  The
1487		 * driver is telling us we have a new WWPN for this
1488		 * initiator ID, so we pretty much need to use it.
1489		 */
1490		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1491		    " but WWPN %#jx '%s' is still at that address\n",
1492		    __func__, port->targ_port, iid, wwpn, name,
1493		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1494		    port->wwpn_iid[iid].name);
1495
1496		/*
1497		 * XXX KDM clear have_ca and ua_pending on each LUN for
1498		 * this initiator.
1499		 */
1500	}
1501take:
1502	free(port->wwpn_iid[iid].name, M_CTL);
1503	port->wwpn_iid[iid].name = name;
1504	port->wwpn_iid[iid].wwpn = wwpn;
1505	port->wwpn_iid[iid].in_use++;
1506	mtx_unlock(&softc->ctl_lock);
1507
1508	return (iid);
1509}
1510
1511static int
1512ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1513{
1514	int len;
1515
1516	switch (port->port_type) {
1517	case CTL_PORT_FC:
1518	{
1519		struct scsi_transportid_fcp *id =
1520		    (struct scsi_transportid_fcp *)buf;
1521		if (port->wwpn_iid[iid].wwpn == 0)
1522			return (0);
1523		memset(id, 0, sizeof(*id));
1524		id->format_protocol = SCSI_PROTO_FC;
1525		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1526		return (sizeof(*id));
1527	}
1528	case CTL_PORT_ISCSI:
1529	{
1530		struct scsi_transportid_iscsi_port *id =
1531		    (struct scsi_transportid_iscsi_port *)buf;
1532		if (port->wwpn_iid[iid].name == NULL)
1533			return (0);
1534		memset(id, 0, 256);
1535		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1536		    SCSI_PROTO_ISCSI;
1537		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1538		len = roundup2(min(len, 252), 4);
1539		scsi_ulto2b(len, id->additional_length);
1540		return (sizeof(*id) + len);
1541	}
1542	case CTL_PORT_SAS:
1543	{
1544		struct scsi_transportid_sas *id =
1545		    (struct scsi_transportid_sas *)buf;
1546		if (port->wwpn_iid[iid].wwpn == 0)
1547			return (0);
1548		memset(id, 0, sizeof(*id));
1549		id->format_protocol = SCSI_PROTO_SAS;
1550		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1551		return (sizeof(*id));
1552	}
1553	default:
1554	{
1555		struct scsi_transportid_spi *id =
1556		    (struct scsi_transportid_spi *)buf;
1557		memset(id, 0, sizeof(*id));
1558		id->format_protocol = SCSI_PROTO_SPI;
1559		scsi_ulto2b(iid, id->scsi_addr);
1560		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1561		return (sizeof(*id));
1562	}
1563	}
1564}
1565
1566static int
1567ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1568{
1569	return (0);
1570}
1571
1572static int
1573ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1574{
1575	return (0);
1576}
1577
1578/*
1579 * Data movement routine for the CTL ioctl frontend port.
1580 */
1581static int
1582ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1583{
1584	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1585	struct ctl_sg_entry ext_entry, kern_entry;
1586	int ext_sglen, ext_sg_entries, kern_sg_entries;
1587	int ext_sg_start, ext_offset;
1588	int len_to_copy, len_copied;
1589	int kern_watermark, ext_watermark;
1590	int ext_sglist_malloced;
1591	int i, j;
1592
1593	ext_sglist_malloced = 0;
1594	ext_sg_start = 0;
1595	ext_offset = 0;
1596
1597	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1598
1599	/*
1600	 * If this flag is set, fake the data transfer.
1601	 */
1602	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1603		ctsio->ext_data_filled = ctsio->ext_data_len;
1604		goto bailout;
1605	}
1606
1607	/*
1608	 * To simplify things here, if we have a single buffer, stick it in
1609	 * a S/G entry and just make it a single entry S/G list.
1610	 */
1611	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1612		int len_seen;
1613
1614		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1615
1616		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1617							   M_WAITOK);
1618		ext_sglist_malloced = 1;
1619		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1620				   ext_sglen) != 0) {
1621			ctl_set_internal_failure(ctsio,
1622						 /*sks_valid*/ 0,
1623						 /*retry_count*/ 0);
1624			goto bailout;
1625		}
1626		ext_sg_entries = ctsio->ext_sg_entries;
1627		len_seen = 0;
1628		for (i = 0; i < ext_sg_entries; i++) {
1629			if ((len_seen + ext_sglist[i].len) >=
1630			     ctsio->ext_data_filled) {
1631				ext_sg_start = i;
1632				ext_offset = ctsio->ext_data_filled - len_seen;
1633				break;
1634			}
1635			len_seen += ext_sglist[i].len;
1636		}
1637	} else {
1638		ext_sglist = &ext_entry;
1639		ext_sglist->addr = ctsio->ext_data_ptr;
1640		ext_sglist->len = ctsio->ext_data_len;
1641		ext_sg_entries = 1;
1642		ext_sg_start = 0;
1643		ext_offset = ctsio->ext_data_filled;
1644	}
1645
1646	if (ctsio->kern_sg_entries > 0) {
1647		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1648		kern_sg_entries = ctsio->kern_sg_entries;
1649	} else {
1650		kern_sglist = &kern_entry;
1651		kern_sglist->addr = ctsio->kern_data_ptr;
1652		kern_sglist->len = ctsio->kern_data_len;
1653		kern_sg_entries = 1;
1654	}
1655
1656
1657	kern_watermark = 0;
1658	ext_watermark = ext_offset;
1659	len_copied = 0;
1660	for (i = ext_sg_start, j = 0;
1661	     i < ext_sg_entries && j < kern_sg_entries;) {
1662		uint8_t *ext_ptr, *kern_ptr;
1663
1664		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1665				      kern_sglist[j].len - kern_watermark);
1666
1667		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1668		ext_ptr = ext_ptr + ext_watermark;
1669		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1670			/*
1671			 * XXX KDM fix this!
1672			 */
1673			panic("need to implement bus address support");
1674#if 0
1675			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1676#endif
1677		} else
1678			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1679		kern_ptr = kern_ptr + kern_watermark;
1680
1681		kern_watermark += len_to_copy;
1682		ext_watermark += len_to_copy;
1683
1684		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1685		     CTL_FLAG_DATA_IN) {
1686			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1687					 "bytes to user\n", len_to_copy));
1688			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1689					 "to %p\n", kern_ptr, ext_ptr));
1690			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1691				ctl_set_internal_failure(ctsio,
1692							 /*sks_valid*/ 0,
1693							 /*retry_count*/ 0);
1694				goto bailout;
1695			}
1696		} else {
1697			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1698					 "bytes from user\n", len_to_copy));
1699			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1700					 "to %p\n", ext_ptr, kern_ptr));
1701			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1702				ctl_set_internal_failure(ctsio,
1703							 /*sks_valid*/ 0,
1704							 /*retry_count*/0);
1705				goto bailout;
1706			}
1707		}
1708
1709		len_copied += len_to_copy;
1710
1711		if (ext_sglist[i].len == ext_watermark) {
1712			i++;
1713			ext_watermark = 0;
1714		}
1715
1716		if (kern_sglist[j].len == kern_watermark) {
1717			j++;
1718			kern_watermark = 0;
1719		}
1720	}
1721
1722	ctsio->ext_data_filled += len_copied;
1723
1724	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1725			 "kern_sg_entries: %d\n", ext_sg_entries,
1726			 kern_sg_entries));
1727	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1728			 "kern_data_len = %d\n", ctsio->ext_data_len,
1729			 ctsio->kern_data_len));
1730
1731
1732	/* XXX KDM set residual?? */
1733bailout:
1734
1735	if (ext_sglist_malloced != 0)
1736		free(ext_sglist, M_CTL);
1737
1738	return (CTL_RETVAL_COMPLETE);
1739}
1740
1741/*
1742 * Serialize a command that went down the "wrong" side, and so was sent to
1743 * this controller for execution.  The logic is a little different than the
1744 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1745 * sent back to the other side, but in the success case, we execute the
1746 * command on this side (XFER mode) or tell the other side to execute it
1747 * (SER_ONLY mode).
1748 */
1749static int
1750ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1751{
1752	struct ctl_softc *ctl_softc;
1753	union ctl_ha_msg msg_info;
1754	struct ctl_lun *lun;
1755	int retval = 0;
1756	uint32_t targ_lun;
1757
1758	ctl_softc = control_softc;
1759
1760	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1761	lun = ctl_softc->ctl_luns[targ_lun];
1762	if (lun==NULL)
1763	{
1764		/*
1765		 * Why isn't LUN defined? The other side wouldn't
1766		 * send a cmd if the LUN is undefined.
1767		 */
1768		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1769
1770		/* "Logical unit not supported" */
1771		ctl_set_sense_data(&msg_info.scsi.sense_data,
1772				   lun,
1773				   /*sense_format*/SSD_TYPE_NONE,
1774				   /*current_error*/ 1,
1775				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1776				   /*asc*/ 0x25,
1777				   /*ascq*/ 0x00,
1778				   SSD_ELEM_NONE);
1779
1780		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1781		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1782		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1783		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1784		msg_info.hdr.serializing_sc = NULL;
1785		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1786	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1787				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1788		}
1789		return(1);
1790
1791	}
1792
1793	mtx_lock(&lun->lun_lock);
1794    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1795
1796	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1797		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1798		 ooa_links))) {
1799	case CTL_ACTION_BLOCK:
1800		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1801		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1802				  blocked_links);
1803		break;
1804	case CTL_ACTION_PASS:
1805	case CTL_ACTION_SKIP:
1806		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1807			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1808			ctl_enqueue_rtr((union ctl_io *)ctsio);
1809		} else {
1810
1811			/* send msg back to other side */
1812			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1813			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1814			msg_info.hdr.msg_type = CTL_MSG_R2R;
1815#if 0
1816			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1817#endif
1818		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1819			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1820			}
1821		}
1822		break;
1823	case CTL_ACTION_OVERLAP:
1824		/* OVERLAPPED COMMANDS ATTEMPTED */
1825		ctl_set_sense_data(&msg_info.scsi.sense_data,
1826				   lun,
1827				   /*sense_format*/SSD_TYPE_NONE,
1828				   /*current_error*/ 1,
1829				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1830				   /*asc*/ 0x4E,
1831				   /*ascq*/ 0x00,
1832				   SSD_ELEM_NONE);
1833
1834		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1835		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1836		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1837		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1838		msg_info.hdr.serializing_sc = NULL;
1839		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1840#if 0
1841		printf("BAD JUJU:Major Bummer Overlap\n");
1842#endif
1843		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1844		retval = 1;
1845		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1846		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1847		}
1848		break;
1849	case CTL_ACTION_OVERLAP_TAG:
1850		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1851		ctl_set_sense_data(&msg_info.scsi.sense_data,
1852				   lun,
1853				   /*sense_format*/SSD_TYPE_NONE,
1854				   /*current_error*/ 1,
1855				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1856				   /*asc*/ 0x4D,
1857				   /*ascq*/ ctsio->tag_num & 0xff,
1858				   SSD_ELEM_NONE);
1859
1860		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1861		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1862		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1863		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1864		msg_info.hdr.serializing_sc = NULL;
1865		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1866#if 0
1867		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1868#endif
1869		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1870		retval = 1;
1871		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1872		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1873		}
1874		break;
1875	case CTL_ACTION_ERROR:
1876	default:
1877		/* "Internal target failure" */
1878		ctl_set_sense_data(&msg_info.scsi.sense_data,
1879				   lun,
1880				   /*sense_format*/SSD_TYPE_NONE,
1881				   /*current_error*/ 1,
1882				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1883				   /*asc*/ 0x44,
1884				   /*ascq*/ 0x00,
1885				   SSD_ELEM_NONE);
1886
1887		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1888		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1889		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1890		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1891		msg_info.hdr.serializing_sc = NULL;
1892		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1893#if 0
1894		printf("BAD JUJU:Major Bummer HW Error\n");
1895#endif
1896		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1897		retval = 1;
1898		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1899		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1900		}
1901		break;
1902	}
1903	mtx_unlock(&lun->lun_lock);
1904	return (retval);
1905}
1906
1907static int
1908ctl_ioctl_submit_wait(union ctl_io *io)
1909{
1910	struct ctl_fe_ioctl_params params;
1911	ctl_fe_ioctl_state last_state;
1912	int done, retval;
1913
1914	retval = 0;
1915
1916	bzero(&params, sizeof(params));
1917
1918	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1919	cv_init(&params.sem, "ctlioccv");
1920	params.state = CTL_IOCTL_INPROG;
1921	last_state = params.state;
1922
1923	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1924
1925	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1926
1927	/* This shouldn't happen */
1928	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1929		return (retval);
1930
1931	done = 0;
1932
1933	do {
1934		mtx_lock(&params.ioctl_mtx);
1935		/*
1936		 * Check the state here, and don't sleep if the state has
1937		 * already changed (i.e. wakeup has already occured, but we
1938		 * weren't waiting yet).
1939		 */
1940		if (params.state == last_state) {
1941			/* XXX KDM cv_wait_sig instead? */
1942			cv_wait(&params.sem, &params.ioctl_mtx);
1943		}
1944		last_state = params.state;
1945
1946		switch (params.state) {
1947		case CTL_IOCTL_INPROG:
1948			/* Why did we wake up? */
1949			/* XXX KDM error here? */
1950			mtx_unlock(&params.ioctl_mtx);
1951			break;
1952		case CTL_IOCTL_DATAMOVE:
1953			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1954
1955			/*
1956			 * change last_state back to INPROG to avoid
1957			 * deadlock on subsequent data moves.
1958			 */
1959			params.state = last_state = CTL_IOCTL_INPROG;
1960
1961			mtx_unlock(&params.ioctl_mtx);
1962			ctl_ioctl_do_datamove(&io->scsiio);
1963			/*
1964			 * Note that in some cases, most notably writes,
1965			 * this will queue the I/O and call us back later.
1966			 * In other cases, generally reads, this routine
1967			 * will immediately call back and wake us up,
1968			 * probably using our own context.
1969			 */
1970			io->scsiio.be_move_done(io);
1971			break;
1972		case CTL_IOCTL_DONE:
1973			mtx_unlock(&params.ioctl_mtx);
1974			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1975			done = 1;
1976			break;
1977		default:
1978			mtx_unlock(&params.ioctl_mtx);
1979			/* XXX KDM error here? */
1980			break;
1981		}
1982	} while (done == 0);
1983
1984	mtx_destroy(&params.ioctl_mtx);
1985	cv_destroy(&params.sem);
1986
1987	return (CTL_RETVAL_COMPLETE);
1988}
1989
1990static void
1991ctl_ioctl_datamove(union ctl_io *io)
1992{
1993	struct ctl_fe_ioctl_params *params;
1994
1995	params = (struct ctl_fe_ioctl_params *)
1996		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1997
1998	mtx_lock(&params->ioctl_mtx);
1999	params->state = CTL_IOCTL_DATAMOVE;
2000	cv_broadcast(&params->sem);
2001	mtx_unlock(&params->ioctl_mtx);
2002}
2003
2004static void
2005ctl_ioctl_done(union ctl_io *io)
2006{
2007	struct ctl_fe_ioctl_params *params;
2008
2009	params = (struct ctl_fe_ioctl_params *)
2010		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2011
2012	mtx_lock(&params->ioctl_mtx);
2013	params->state = CTL_IOCTL_DONE;
2014	cv_broadcast(&params->sem);
2015	mtx_unlock(&params->ioctl_mtx);
2016}
2017
2018static void
2019ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2020{
2021	struct ctl_fe_ioctl_startstop_info *sd_info;
2022
2023	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2024
2025	sd_info->hs_info.status = metatask->status;
2026	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2027	sd_info->hs_info.luns_complete =
2028		metatask->taskinfo.startstop.luns_complete;
2029	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2030
2031	cv_broadcast(&sd_info->sem);
2032}
2033
2034static void
2035ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2036{
2037	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2038
2039	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2040
2041	mtx_lock(fe_bbr_info->lock);
2042	fe_bbr_info->bbr_info->status = metatask->status;
2043	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2044	fe_bbr_info->wakeup_done = 1;
2045	mtx_unlock(fe_bbr_info->lock);
2046
2047	cv_broadcast(&fe_bbr_info->sem);
2048}
2049
2050/*
2051 * Returns 0 for success, errno for failure.
2052 */
2053static int
2054ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2055		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2056{
2057	union ctl_io *io;
2058	int retval;
2059
2060	retval = 0;
2061
2062	mtx_lock(&lun->lun_lock);
2063	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2064	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2065	     ooa_links)) {
2066		struct ctl_ooa_entry *entry;
2067
2068		/*
2069		 * If we've got more than we can fit, just count the
2070		 * remaining entries.
2071		 */
2072		if (*cur_fill_num >= ooa_hdr->alloc_num)
2073			continue;
2074
2075		entry = &kern_entries[*cur_fill_num];
2076
2077		entry->tag_num = io->scsiio.tag_num;
2078		entry->lun_num = lun->lun;
2079#ifdef CTL_TIME_IO
2080		entry->start_bt = io->io_hdr.start_bt;
2081#endif
2082		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2083		entry->cdb_len = io->scsiio.cdb_len;
2084		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2085			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2086
2087		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2088			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2089
2090		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2091			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2092
2093		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2094			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2095
2096		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2097			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2098	}
2099	mtx_unlock(&lun->lun_lock);
2100
2101	return (retval);
2102}
2103
2104static void *
2105ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2106		 size_t error_str_len)
2107{
2108	void *kptr;
2109
2110	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2111
2112	if (copyin(user_addr, kptr, len) != 0) {
2113		snprintf(error_str, error_str_len, "Error copying %d bytes "
2114			 "from user address %p to kernel address %p", len,
2115			 user_addr, kptr);
2116		free(kptr, M_CTL);
2117		return (NULL);
2118	}
2119
2120	return (kptr);
2121}
2122
2123static void
2124ctl_free_args(int num_args, struct ctl_be_arg *args)
2125{
2126	int i;
2127
2128	if (args == NULL)
2129		return;
2130
2131	for (i = 0; i < num_args; i++) {
2132		free(args[i].kname, M_CTL);
2133		free(args[i].kvalue, M_CTL);
2134	}
2135
2136	free(args, M_CTL);
2137}
2138
2139static struct ctl_be_arg *
2140ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2141		char *error_str, size_t error_str_len)
2142{
2143	struct ctl_be_arg *args;
2144	int i;
2145
2146	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2147				error_str, error_str_len);
2148
2149	if (args == NULL)
2150		goto bailout;
2151
2152	for (i = 0; i < num_args; i++) {
2153		args[i].kname = NULL;
2154		args[i].kvalue = NULL;
2155	}
2156
2157	for (i = 0; i < num_args; i++) {
2158		uint8_t *tmpptr;
2159
2160		args[i].kname = ctl_copyin_alloc(args[i].name,
2161			args[i].namelen, error_str, error_str_len);
2162		if (args[i].kname == NULL)
2163			goto bailout;
2164
2165		if (args[i].kname[args[i].namelen - 1] != '\0') {
2166			snprintf(error_str, error_str_len, "Argument %d "
2167				 "name is not NUL-terminated", i);
2168			goto bailout;
2169		}
2170
2171		if (args[i].flags & CTL_BEARG_RD) {
2172			tmpptr = ctl_copyin_alloc(args[i].value,
2173				args[i].vallen, error_str, error_str_len);
2174			if (tmpptr == NULL)
2175				goto bailout;
2176			if ((args[i].flags & CTL_BEARG_ASCII)
2177			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2178				snprintf(error_str, error_str_len, "Argument "
2179				    "%d value is not NUL-terminated", i);
2180				goto bailout;
2181			}
2182			args[i].kvalue = tmpptr;
2183		} else {
2184			args[i].kvalue = malloc(args[i].vallen,
2185			    M_CTL, M_WAITOK | M_ZERO);
2186		}
2187	}
2188
2189	return (args);
2190bailout:
2191
2192	ctl_free_args(num_args, args);
2193
2194	return (NULL);
2195}
2196
2197static void
2198ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2199{
2200	int i;
2201
2202	for (i = 0; i < num_args; i++) {
2203		if (args[i].flags & CTL_BEARG_WR)
2204			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2205	}
2206}
2207
2208/*
2209 * Escape characters that are illegal or not recommended in XML.
2210 */
2211int
2212ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2213{
2214	int retval;
2215
2216	retval = 0;
2217
2218	for (; *str; str++) {
2219		switch (*str) {
2220		case '&':
2221			retval = sbuf_printf(sb, "&amp;");
2222			break;
2223		case '>':
2224			retval = sbuf_printf(sb, "&gt;");
2225			break;
2226		case '<':
2227			retval = sbuf_printf(sb, "&lt;");
2228			break;
2229		default:
2230			retval = sbuf_putc(sb, *str);
2231			break;
2232		}
2233
2234		if (retval != 0)
2235			break;
2236
2237	}
2238
2239	return (retval);
2240}
2241
2242static int
2243ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2244	  struct thread *td)
2245{
2246	struct ctl_softc *softc;
2247	int retval;
2248
2249	softc = control_softc;
2250
2251	retval = 0;
2252
2253	switch (cmd) {
2254	case CTL_IO: {
2255		union ctl_io *io;
2256		void *pool_tmp;
2257
2258		/*
2259		 * If we haven't been "enabled", don't allow any SCSI I/O
2260		 * to this FETD.
2261		 */
2262		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2263			retval = EPERM;
2264			break;
2265		}
2266
2267		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2268		if (io == NULL) {
2269			printf("ctl_ioctl: can't allocate ctl_io!\n");
2270			retval = ENOSPC;
2271			break;
2272		}
2273
2274		/*
2275		 * Need to save the pool reference so it doesn't get
2276		 * spammed by the user's ctl_io.
2277		 */
2278		pool_tmp = io->io_hdr.pool;
2279
2280		memcpy(io, (void *)addr, sizeof(*io));
2281
2282		io->io_hdr.pool = pool_tmp;
2283		/*
2284		 * No status yet, so make sure the status is set properly.
2285		 */
2286		io->io_hdr.status = CTL_STATUS_NONE;
2287
2288		/*
2289		 * The user sets the initiator ID, target and LUN IDs.
2290		 */
2291		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2292		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2293		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2294		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2295			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2296
2297		retval = ctl_ioctl_submit_wait(io);
2298
2299		if (retval != 0) {
2300			ctl_free_io(io);
2301			break;
2302		}
2303
2304		memcpy((void *)addr, io, sizeof(*io));
2305
2306		/* return this to our pool */
2307		ctl_free_io(io);
2308
2309		break;
2310	}
2311	case CTL_ENABLE_PORT:
2312	case CTL_DISABLE_PORT:
2313	case CTL_SET_PORT_WWNS: {
2314		struct ctl_port *port;
2315		struct ctl_port_entry *entry;
2316
2317		entry = (struct ctl_port_entry *)addr;
2318
2319		mtx_lock(&softc->ctl_lock);
2320		STAILQ_FOREACH(port, &softc->port_list, links) {
2321			int action, done;
2322
2323			action = 0;
2324			done = 0;
2325
2326			if ((entry->port_type == CTL_PORT_NONE)
2327			 && (entry->targ_port == port->targ_port)) {
2328				/*
2329				 * If the user only wants to enable or
2330				 * disable or set WWNs on a specific port,
2331				 * do the operation and we're done.
2332				 */
2333				action = 1;
2334				done = 1;
2335			} else if (entry->port_type & port->port_type) {
2336				/*
2337				 * Compare the user's type mask with the
2338				 * particular frontend type to see if we
2339				 * have a match.
2340				 */
2341				action = 1;
2342				done = 0;
2343
2344				/*
2345				 * Make sure the user isn't trying to set
2346				 * WWNs on multiple ports at the same time.
2347				 */
2348				if (cmd == CTL_SET_PORT_WWNS) {
2349					printf("%s: Can't set WWNs on "
2350					       "multiple ports\n", __func__);
2351					retval = EINVAL;
2352					break;
2353				}
2354			}
2355			if (action != 0) {
2356				/*
2357				 * XXX KDM we have to drop the lock here,
2358				 * because the online/offline operations
2359				 * can potentially block.  We need to
2360				 * reference count the frontends so they
2361				 * can't go away,
2362				 */
2363				mtx_unlock(&softc->ctl_lock);
2364
2365				if (cmd == CTL_ENABLE_PORT) {
2366					struct ctl_lun *lun;
2367
2368					STAILQ_FOREACH(lun, &softc->lun_list,
2369						       links) {
2370						port->lun_enable(port->targ_lun_arg,
2371						    lun->target,
2372						    lun->lun);
2373					}
2374
2375					ctl_port_online(port);
2376				} else if (cmd == CTL_DISABLE_PORT) {
2377					struct ctl_lun *lun;
2378
2379					ctl_port_offline(port);
2380
2381					STAILQ_FOREACH(lun, &softc->lun_list,
2382						       links) {
2383						port->lun_disable(
2384						    port->targ_lun_arg,
2385						    lun->target,
2386						    lun->lun);
2387					}
2388				}
2389
2390				mtx_lock(&softc->ctl_lock);
2391
2392				if (cmd == CTL_SET_PORT_WWNS)
2393					ctl_port_set_wwns(port,
2394					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2395					    1 : 0, entry->wwnn,
2396					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2397					    1 : 0, entry->wwpn);
2398			}
2399			if (done != 0)
2400				break;
2401		}
2402		mtx_unlock(&softc->ctl_lock);
2403		break;
2404	}
2405	case CTL_GET_PORT_LIST: {
2406		struct ctl_port *port;
2407		struct ctl_port_list *list;
2408		int i;
2409
2410		list = (struct ctl_port_list *)addr;
2411
2412		if (list->alloc_len != (list->alloc_num *
2413		    sizeof(struct ctl_port_entry))) {
2414			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2415			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2416			       "%zu\n", __func__, list->alloc_len,
2417			       list->alloc_num, sizeof(struct ctl_port_entry));
2418			retval = EINVAL;
2419			break;
2420		}
2421		list->fill_len = 0;
2422		list->fill_num = 0;
2423		list->dropped_num = 0;
2424		i = 0;
2425		mtx_lock(&softc->ctl_lock);
2426		STAILQ_FOREACH(port, &softc->port_list, links) {
2427			struct ctl_port_entry entry, *list_entry;
2428
2429			if (list->fill_num >= list->alloc_num) {
2430				list->dropped_num++;
2431				continue;
2432			}
2433
2434			entry.port_type = port->port_type;
2435			strlcpy(entry.port_name, port->port_name,
2436				sizeof(entry.port_name));
2437			entry.targ_port = port->targ_port;
2438			entry.physical_port = port->physical_port;
2439			entry.virtual_port = port->virtual_port;
2440			entry.wwnn = port->wwnn;
2441			entry.wwpn = port->wwpn;
2442			if (port->status & CTL_PORT_STATUS_ONLINE)
2443				entry.online = 1;
2444			else
2445				entry.online = 0;
2446
2447			list_entry = &list->entries[i];
2448
2449			retval = copyout(&entry, list_entry, sizeof(entry));
2450			if (retval != 0) {
2451				printf("%s: CTL_GET_PORT_LIST: copyout "
2452				       "returned %d\n", __func__, retval);
2453				break;
2454			}
2455			i++;
2456			list->fill_num++;
2457			list->fill_len += sizeof(entry);
2458		}
2459		mtx_unlock(&softc->ctl_lock);
2460
2461		/*
2462		 * If this is non-zero, we had a copyout fault, so there's
2463		 * probably no point in attempting to set the status inside
2464		 * the structure.
2465		 */
2466		if (retval != 0)
2467			break;
2468
2469		if (list->dropped_num > 0)
2470			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2471		else
2472			list->status = CTL_PORT_LIST_OK;
2473		break;
2474	}
2475	case CTL_DUMP_OOA: {
2476		struct ctl_lun *lun;
2477		union ctl_io *io;
2478		char printbuf[128];
2479		struct sbuf sb;
2480
2481		mtx_lock(&softc->ctl_lock);
2482		printf("Dumping OOA queues:\n");
2483		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2484			mtx_lock(&lun->lun_lock);
2485			for (io = (union ctl_io *)TAILQ_FIRST(
2486			     &lun->ooa_queue); io != NULL;
2487			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2488			     ooa_links)) {
2489				sbuf_new(&sb, printbuf, sizeof(printbuf),
2490					 SBUF_FIXEDLEN);
2491				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2492					    (intmax_t)lun->lun,
2493					    io->scsiio.tag_num,
2494					    (io->io_hdr.flags &
2495					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2496					    (io->io_hdr.flags &
2497					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2498					    (io->io_hdr.flags &
2499					    CTL_FLAG_ABORT) ? " ABORT" : "",
2500			                    (io->io_hdr.flags &
2501		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2502				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2503				sbuf_finish(&sb);
2504				printf("%s\n", sbuf_data(&sb));
2505			}
2506			mtx_unlock(&lun->lun_lock);
2507		}
2508		printf("OOA queues dump done\n");
2509		mtx_unlock(&softc->ctl_lock);
2510		break;
2511	}
2512	case CTL_GET_OOA: {
2513		struct ctl_lun *lun;
2514		struct ctl_ooa *ooa_hdr;
2515		struct ctl_ooa_entry *entries;
2516		uint32_t cur_fill_num;
2517
2518		ooa_hdr = (struct ctl_ooa *)addr;
2519
2520		if ((ooa_hdr->alloc_len == 0)
2521		 || (ooa_hdr->alloc_num == 0)) {
2522			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2523			       "must be non-zero\n", __func__,
2524			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2525			retval = EINVAL;
2526			break;
2527		}
2528
2529		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2530		    sizeof(struct ctl_ooa_entry))) {
2531			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2532			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2533			       __func__, ooa_hdr->alloc_len,
2534			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2535			retval = EINVAL;
2536			break;
2537		}
2538
2539		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2540		if (entries == NULL) {
2541			printf("%s: could not allocate %d bytes for OOA "
2542			       "dump\n", __func__, ooa_hdr->alloc_len);
2543			retval = ENOMEM;
2544			break;
2545		}
2546
2547		mtx_lock(&softc->ctl_lock);
2548		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2549		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2550		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2551			mtx_unlock(&softc->ctl_lock);
2552			free(entries, M_CTL);
2553			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2554			       __func__, (uintmax_t)ooa_hdr->lun_num);
2555			retval = EINVAL;
2556			break;
2557		}
2558
2559		cur_fill_num = 0;
2560
2561		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2562			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2563				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2564					ooa_hdr, entries);
2565				if (retval != 0)
2566					break;
2567			}
2568			if (retval != 0) {
2569				mtx_unlock(&softc->ctl_lock);
2570				free(entries, M_CTL);
2571				break;
2572			}
2573		} else {
2574			lun = softc->ctl_luns[ooa_hdr->lun_num];
2575
2576			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2577						    entries);
2578		}
2579		mtx_unlock(&softc->ctl_lock);
2580
2581		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2582		ooa_hdr->fill_len = ooa_hdr->fill_num *
2583			sizeof(struct ctl_ooa_entry);
2584		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2585		if (retval != 0) {
2586			printf("%s: error copying out %d bytes for OOA dump\n",
2587			       __func__, ooa_hdr->fill_len);
2588		}
2589
2590		getbintime(&ooa_hdr->cur_bt);
2591
2592		if (cur_fill_num > ooa_hdr->alloc_num) {
2593			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2594			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2595		} else {
2596			ooa_hdr->dropped_num = 0;
2597			ooa_hdr->status = CTL_OOA_OK;
2598		}
2599
2600		free(entries, M_CTL);
2601		break;
2602	}
2603	case CTL_CHECK_OOA: {
2604		union ctl_io *io;
2605		struct ctl_lun *lun;
2606		struct ctl_ooa_info *ooa_info;
2607
2608
2609		ooa_info = (struct ctl_ooa_info *)addr;
2610
2611		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2612			ooa_info->status = CTL_OOA_INVALID_LUN;
2613			break;
2614		}
2615		mtx_lock(&softc->ctl_lock);
2616		lun = softc->ctl_luns[ooa_info->lun_id];
2617		if (lun == NULL) {
2618			mtx_unlock(&softc->ctl_lock);
2619			ooa_info->status = CTL_OOA_INVALID_LUN;
2620			break;
2621		}
2622		mtx_lock(&lun->lun_lock);
2623		mtx_unlock(&softc->ctl_lock);
2624		ooa_info->num_entries = 0;
2625		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2626		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2627		     &io->io_hdr, ooa_links)) {
2628			ooa_info->num_entries++;
2629		}
2630		mtx_unlock(&lun->lun_lock);
2631
2632		ooa_info->status = CTL_OOA_SUCCESS;
2633
2634		break;
2635	}
2636	case CTL_HARD_START:
2637	case CTL_HARD_STOP: {
2638		struct ctl_fe_ioctl_startstop_info ss_info;
2639		struct cfi_metatask *metatask;
2640		struct mtx hs_mtx;
2641
2642		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2643
2644		cv_init(&ss_info.sem, "hard start/stop cv" );
2645
2646		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2647		if (metatask == NULL) {
2648			retval = ENOMEM;
2649			mtx_destroy(&hs_mtx);
2650			break;
2651		}
2652
2653		if (cmd == CTL_HARD_START)
2654			metatask->tasktype = CFI_TASK_STARTUP;
2655		else
2656			metatask->tasktype = CFI_TASK_SHUTDOWN;
2657
2658		metatask->callback = ctl_ioctl_hard_startstop_callback;
2659		metatask->callback_arg = &ss_info;
2660
2661		cfi_action(metatask);
2662
2663		/* Wait for the callback */
2664		mtx_lock(&hs_mtx);
2665		cv_wait_sig(&ss_info.sem, &hs_mtx);
2666		mtx_unlock(&hs_mtx);
2667
2668		/*
2669		 * All information has been copied from the metatask by the
2670		 * time cv_broadcast() is called, so we free the metatask here.
2671		 */
2672		cfi_free_metatask(metatask);
2673
2674		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2675
2676		mtx_destroy(&hs_mtx);
2677		break;
2678	}
2679	case CTL_BBRREAD: {
2680		struct ctl_bbrread_info *bbr_info;
2681		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2682		struct mtx bbr_mtx;
2683		struct cfi_metatask *metatask;
2684
2685		bbr_info = (struct ctl_bbrread_info *)addr;
2686
2687		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2688
2689		bzero(&bbr_mtx, sizeof(bbr_mtx));
2690		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2691
2692		fe_bbr_info.bbr_info = bbr_info;
2693		fe_bbr_info.lock = &bbr_mtx;
2694
2695		cv_init(&fe_bbr_info.sem, "BBR read cv");
2696		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2697
2698		if (metatask == NULL) {
2699			mtx_destroy(&bbr_mtx);
2700			cv_destroy(&fe_bbr_info.sem);
2701			retval = ENOMEM;
2702			break;
2703		}
2704		metatask->tasktype = CFI_TASK_BBRREAD;
2705		metatask->callback = ctl_ioctl_bbrread_callback;
2706		metatask->callback_arg = &fe_bbr_info;
2707		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2708		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2709		metatask->taskinfo.bbrread.len = bbr_info->len;
2710
2711		cfi_action(metatask);
2712
2713		mtx_lock(&bbr_mtx);
2714		while (fe_bbr_info.wakeup_done == 0)
2715			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2716		mtx_unlock(&bbr_mtx);
2717
2718		bbr_info->status = metatask->status;
2719		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2720		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2721		memcpy(&bbr_info->sense_data,
2722		       &metatask->taskinfo.bbrread.sense_data,
2723		       ctl_min(sizeof(bbr_info->sense_data),
2724			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2725
2726		cfi_free_metatask(metatask);
2727
2728		mtx_destroy(&bbr_mtx);
2729		cv_destroy(&fe_bbr_info.sem);
2730
2731		break;
2732	}
2733	case CTL_DELAY_IO: {
2734		struct ctl_io_delay_info *delay_info;
2735#ifdef CTL_IO_DELAY
2736		struct ctl_lun *lun;
2737#endif /* CTL_IO_DELAY */
2738
2739		delay_info = (struct ctl_io_delay_info *)addr;
2740
2741#ifdef CTL_IO_DELAY
2742		mtx_lock(&softc->ctl_lock);
2743
2744		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2745		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2746			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2747		} else {
2748			lun = softc->ctl_luns[delay_info->lun_id];
2749			mtx_lock(&lun->lun_lock);
2750
2751			delay_info->status = CTL_DELAY_STATUS_OK;
2752
2753			switch (delay_info->delay_type) {
2754			case CTL_DELAY_TYPE_CONT:
2755				break;
2756			case CTL_DELAY_TYPE_ONESHOT:
2757				break;
2758			default:
2759				delay_info->status =
2760					CTL_DELAY_STATUS_INVALID_TYPE;
2761				break;
2762			}
2763
2764			switch (delay_info->delay_loc) {
2765			case CTL_DELAY_LOC_DATAMOVE:
2766				lun->delay_info.datamove_type =
2767					delay_info->delay_type;
2768				lun->delay_info.datamove_delay =
2769					delay_info->delay_secs;
2770				break;
2771			case CTL_DELAY_LOC_DONE:
2772				lun->delay_info.done_type =
2773					delay_info->delay_type;
2774				lun->delay_info.done_delay =
2775					delay_info->delay_secs;
2776				break;
2777			default:
2778				delay_info->status =
2779					CTL_DELAY_STATUS_INVALID_LOC;
2780				break;
2781			}
2782			mtx_unlock(&lun->lun_lock);
2783		}
2784
2785		mtx_unlock(&softc->ctl_lock);
2786#else
2787		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2788#endif /* CTL_IO_DELAY */
2789		break;
2790	}
2791	case CTL_REALSYNC_SET: {
2792		int *syncstate;
2793
2794		syncstate = (int *)addr;
2795
2796		mtx_lock(&softc->ctl_lock);
2797		switch (*syncstate) {
2798		case 0:
2799			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2800			break;
2801		case 1:
2802			softc->flags |= CTL_FLAG_REAL_SYNC;
2803			break;
2804		default:
2805			retval = EINVAL;
2806			break;
2807		}
2808		mtx_unlock(&softc->ctl_lock);
2809		break;
2810	}
2811	case CTL_REALSYNC_GET: {
2812		int *syncstate;
2813
2814		syncstate = (int*)addr;
2815
2816		mtx_lock(&softc->ctl_lock);
2817		if (softc->flags & CTL_FLAG_REAL_SYNC)
2818			*syncstate = 1;
2819		else
2820			*syncstate = 0;
2821		mtx_unlock(&softc->ctl_lock);
2822
2823		break;
2824	}
2825	case CTL_SETSYNC:
2826	case CTL_GETSYNC: {
2827		struct ctl_sync_info *sync_info;
2828		struct ctl_lun *lun;
2829
2830		sync_info = (struct ctl_sync_info *)addr;
2831
2832		mtx_lock(&softc->ctl_lock);
2833		lun = softc->ctl_luns[sync_info->lun_id];
2834		if (lun == NULL) {
2835			mtx_unlock(&softc->ctl_lock);
2836			sync_info->status = CTL_GS_SYNC_NO_LUN;
2837		}
2838		/*
2839		 * Get or set the sync interval.  We're not bounds checking
2840		 * in the set case, hopefully the user won't do something
2841		 * silly.
2842		 */
2843		mtx_lock(&lun->lun_lock);
2844		mtx_unlock(&softc->ctl_lock);
2845		if (cmd == CTL_GETSYNC)
2846			sync_info->sync_interval = lun->sync_interval;
2847		else
2848			lun->sync_interval = sync_info->sync_interval;
2849		mtx_unlock(&lun->lun_lock);
2850
2851		sync_info->status = CTL_GS_SYNC_OK;
2852
2853		break;
2854	}
2855	case CTL_GETSTATS: {
2856		struct ctl_stats *stats;
2857		struct ctl_lun *lun;
2858		int i;
2859
2860		stats = (struct ctl_stats *)addr;
2861
2862		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2863		     stats->alloc_len) {
2864			stats->status = CTL_SS_NEED_MORE_SPACE;
2865			stats->num_luns = softc->num_luns;
2866			break;
2867		}
2868		/*
2869		 * XXX KDM no locking here.  If the LUN list changes,
2870		 * things can blow up.
2871		 */
2872		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2873		     i++, lun = STAILQ_NEXT(lun, links)) {
2874			retval = copyout(&lun->stats, &stats->lun_stats[i],
2875					 sizeof(lun->stats));
2876			if (retval != 0)
2877				break;
2878		}
2879		stats->num_luns = softc->num_luns;
2880		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2881				 softc->num_luns;
2882		stats->status = CTL_SS_OK;
2883#ifdef CTL_TIME_IO
2884		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2885#else
2886		stats->flags = CTL_STATS_FLAG_NONE;
2887#endif
2888		getnanouptime(&stats->timestamp);
2889		break;
2890	}
2891	case CTL_ERROR_INJECT: {
2892		struct ctl_error_desc *err_desc, *new_err_desc;
2893		struct ctl_lun *lun;
2894
2895		err_desc = (struct ctl_error_desc *)addr;
2896
2897		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2898				      M_WAITOK | M_ZERO);
2899		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2900
2901		mtx_lock(&softc->ctl_lock);
2902		lun = softc->ctl_luns[err_desc->lun_id];
2903		if (lun == NULL) {
2904			mtx_unlock(&softc->ctl_lock);
2905			free(new_err_desc, M_CTL);
2906			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2907			       __func__, (uintmax_t)err_desc->lun_id);
2908			retval = EINVAL;
2909			break;
2910		}
2911		mtx_lock(&lun->lun_lock);
2912		mtx_unlock(&softc->ctl_lock);
2913
2914		/*
2915		 * We could do some checking here to verify the validity
2916		 * of the request, but given the complexity of error
2917		 * injection requests, the checking logic would be fairly
2918		 * complex.
2919		 *
2920		 * For now, if the request is invalid, it just won't get
2921		 * executed and might get deleted.
2922		 */
2923		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2924
2925		/*
2926		 * XXX KDM check to make sure the serial number is unique,
2927		 * in case we somehow manage to wrap.  That shouldn't
2928		 * happen for a very long time, but it's the right thing to
2929		 * do.
2930		 */
2931		new_err_desc->serial = lun->error_serial;
2932		err_desc->serial = lun->error_serial;
2933		lun->error_serial++;
2934
2935		mtx_unlock(&lun->lun_lock);
2936		break;
2937	}
2938	case CTL_ERROR_INJECT_DELETE: {
2939		struct ctl_error_desc *delete_desc, *desc, *desc2;
2940		struct ctl_lun *lun;
2941		int delete_done;
2942
2943		delete_desc = (struct ctl_error_desc *)addr;
2944		delete_done = 0;
2945
2946		mtx_lock(&softc->ctl_lock);
2947		lun = softc->ctl_luns[delete_desc->lun_id];
2948		if (lun == NULL) {
2949			mtx_unlock(&softc->ctl_lock);
2950			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2951			       __func__, (uintmax_t)delete_desc->lun_id);
2952			retval = EINVAL;
2953			break;
2954		}
2955		mtx_lock(&lun->lun_lock);
2956		mtx_unlock(&softc->ctl_lock);
2957		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2958			if (desc->serial != delete_desc->serial)
2959				continue;
2960
2961			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2962				      links);
2963			free(desc, M_CTL);
2964			delete_done = 1;
2965		}
2966		mtx_unlock(&lun->lun_lock);
2967		if (delete_done == 0) {
2968			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2969			       "error serial %ju on LUN %u\n", __func__,
2970			       delete_desc->serial, delete_desc->lun_id);
2971			retval = EINVAL;
2972			break;
2973		}
2974		break;
2975	}
2976	case CTL_DUMP_STRUCTS: {
2977		int i, j, k, idx;
2978		struct ctl_port *port;
2979		struct ctl_frontend *fe;
2980
2981		mtx_lock(&softc->ctl_lock);
2982		printf("CTL Persistent Reservation information start:\n");
2983		for (i = 0; i < CTL_MAX_LUNS; i++) {
2984			struct ctl_lun *lun;
2985
2986			lun = softc->ctl_luns[i];
2987
2988			if ((lun == NULL)
2989			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2990				continue;
2991
2992			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2993				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2994					idx = j * CTL_MAX_INIT_PER_PORT + k;
2995					if (lun->pr_keys[idx] == 0)
2996						continue;
2997					printf("  LUN %d port %d iid %d key "
2998					       "%#jx\n", i, j, k,
2999					       (uintmax_t)lun->pr_keys[idx]);
3000				}
3001			}
3002		}
3003		printf("CTL Persistent Reservation information end\n");
3004		printf("CTL Ports:\n");
3005		STAILQ_FOREACH(port, &softc->port_list, links) {
3006			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3007			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3008			       port->frontend->name, port->port_type,
3009			       port->physical_port, port->virtual_port,
3010			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3011			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3012				if (port->wwpn_iid[j].in_use == 0 &&
3013				    port->wwpn_iid[j].wwpn == 0 &&
3014				    port->wwpn_iid[j].name == NULL)
3015					continue;
3016
3017				printf("    iid %u use %d WWPN %#jx '%s'\n",
3018				    j, port->wwpn_iid[j].in_use,
3019				    (uintmax_t)port->wwpn_iid[j].wwpn,
3020				    port->wwpn_iid[j].name);
3021			}
3022		}
3023		printf("CTL Port information end\n");
3024		mtx_unlock(&softc->ctl_lock);
3025		/*
3026		 * XXX KDM calling this without a lock.  We'd likely want
3027		 * to drop the lock before calling the frontend's dump
3028		 * routine anyway.
3029		 */
3030		printf("CTL Frontends:\n");
3031		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3032			printf("  Frontend '%s'\n", fe->name);
3033			if (fe->fe_dump != NULL)
3034				fe->fe_dump();
3035		}
3036		printf("CTL Frontend information end\n");
3037		break;
3038	}
3039	case CTL_LUN_REQ: {
3040		struct ctl_lun_req *lun_req;
3041		struct ctl_backend_driver *backend;
3042
3043		lun_req = (struct ctl_lun_req *)addr;
3044
3045		backend = ctl_backend_find(lun_req->backend);
3046		if (backend == NULL) {
3047			lun_req->status = CTL_LUN_ERROR;
3048			snprintf(lun_req->error_str,
3049				 sizeof(lun_req->error_str),
3050				 "Backend \"%s\" not found.",
3051				 lun_req->backend);
3052			break;
3053		}
3054		if (lun_req->num_be_args > 0) {
3055			lun_req->kern_be_args = ctl_copyin_args(
3056				lun_req->num_be_args,
3057				lun_req->be_args,
3058				lun_req->error_str,
3059				sizeof(lun_req->error_str));
3060			if (lun_req->kern_be_args == NULL) {
3061				lun_req->status = CTL_LUN_ERROR;
3062				break;
3063			}
3064		}
3065
3066		retval = backend->ioctl(dev, cmd, addr, flag, td);
3067
3068		if (lun_req->num_be_args > 0) {
3069			ctl_copyout_args(lun_req->num_be_args,
3070				      lun_req->kern_be_args);
3071			ctl_free_args(lun_req->num_be_args,
3072				      lun_req->kern_be_args);
3073		}
3074		break;
3075	}
3076	case CTL_LUN_LIST: {
3077		struct sbuf *sb;
3078		struct ctl_lun *lun;
3079		struct ctl_lun_list *list;
3080		struct ctl_option *opt;
3081
3082		list = (struct ctl_lun_list *)addr;
3083
3084		/*
3085		 * Allocate a fixed length sbuf here, based on the length
3086		 * of the user's buffer.  We could allocate an auto-extending
3087		 * buffer, and then tell the user how much larger our
3088		 * amount of data is than his buffer, but that presents
3089		 * some problems:
3090		 *
3091		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3092		 *     we can't hold a lock while calling them with an
3093		 *     auto-extending buffer.
3094 		 *
3095		 * 2.  There is not currently a LUN reference counting
3096		 *     mechanism, outside of outstanding transactions on
3097		 *     the LUN's OOA queue.  So a LUN could go away on us
3098		 *     while we're getting the LUN number, backend-specific
3099		 *     information, etc.  Thus, given the way things
3100		 *     currently work, we need to hold the CTL lock while
3101		 *     grabbing LUN information.
3102		 *
3103		 * So, from the user's standpoint, the best thing to do is
3104		 * allocate what he thinks is a reasonable buffer length,
3105		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3106		 * double the buffer length and try again.  (And repeat
3107		 * that until he succeeds.)
3108		 */
3109		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3110		if (sb == NULL) {
3111			list->status = CTL_LUN_LIST_ERROR;
3112			snprintf(list->error_str, sizeof(list->error_str),
3113				 "Unable to allocate %d bytes for LUN list",
3114				 list->alloc_len);
3115			break;
3116		}
3117
3118		sbuf_printf(sb, "<ctllunlist>\n");
3119
3120		mtx_lock(&softc->ctl_lock);
3121		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3122			mtx_lock(&lun->lun_lock);
3123			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3124					     (uintmax_t)lun->lun);
3125
3126			/*
3127			 * Bail out as soon as we see that we've overfilled
3128			 * the buffer.
3129			 */
3130			if (retval != 0)
3131				break;
3132
3133			retval = sbuf_printf(sb, "\t<backend_type>%s"
3134					     "</backend_type>\n",
3135					     (lun->backend == NULL) ?  "none" :
3136					     lun->backend->name);
3137
3138			if (retval != 0)
3139				break;
3140
3141			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3142					     lun->be_lun->lun_type);
3143
3144			if (retval != 0)
3145				break;
3146
3147			if (lun->backend == NULL) {
3148				retval = sbuf_printf(sb, "</lun>\n");
3149				if (retval != 0)
3150					break;
3151				continue;
3152			}
3153
3154			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3155					     (lun->be_lun->maxlba > 0) ?
3156					     lun->be_lun->maxlba + 1 : 0);
3157
3158			if (retval != 0)
3159				break;
3160
3161			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3162					     lun->be_lun->blocksize);
3163
3164			if (retval != 0)
3165				break;
3166
3167			retval = sbuf_printf(sb, "\t<serial_number>");
3168
3169			if (retval != 0)
3170				break;
3171
3172			retval = ctl_sbuf_printf_esc(sb,
3173						     lun->be_lun->serial_num);
3174
3175			if (retval != 0)
3176				break;
3177
3178			retval = sbuf_printf(sb, "</serial_number>\n");
3179
3180			if (retval != 0)
3181				break;
3182
3183			retval = sbuf_printf(sb, "\t<device_id>");
3184
3185			if (retval != 0)
3186				break;
3187
3188			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3189
3190			if (retval != 0)
3191				break;
3192
3193			retval = sbuf_printf(sb, "</device_id>\n");
3194
3195			if (retval != 0)
3196				break;
3197
3198			if (lun->backend->lun_info != NULL) {
3199				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3200				if (retval != 0)
3201					break;
3202			}
3203			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3204				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3205				    opt->name, opt->value, opt->name);
3206				if (retval != 0)
3207					break;
3208			}
3209
3210			retval = sbuf_printf(sb, "</lun>\n");
3211
3212			if (retval != 0)
3213				break;
3214			mtx_unlock(&lun->lun_lock);
3215		}
3216		if (lun != NULL)
3217			mtx_unlock(&lun->lun_lock);
3218		mtx_unlock(&softc->ctl_lock);
3219
3220		if ((retval != 0)
3221		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3222			retval = 0;
3223			sbuf_delete(sb);
3224			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3225			snprintf(list->error_str, sizeof(list->error_str),
3226				 "Out of space, %d bytes is too small",
3227				 list->alloc_len);
3228			break;
3229		}
3230
3231		sbuf_finish(sb);
3232
3233		retval = copyout(sbuf_data(sb), list->lun_xml,
3234				 sbuf_len(sb) + 1);
3235
3236		list->fill_len = sbuf_len(sb) + 1;
3237		list->status = CTL_LUN_LIST_OK;
3238		sbuf_delete(sb);
3239		break;
3240	}
3241	case CTL_ISCSI: {
3242		struct ctl_iscsi *ci;
3243		struct ctl_frontend *fe;
3244
3245		ci = (struct ctl_iscsi *)addr;
3246
3247		fe = ctl_frontend_find("iscsi");
3248		if (fe == NULL) {
3249			ci->status = CTL_ISCSI_ERROR;
3250			snprintf(ci->error_str, sizeof(ci->error_str),
3251			    "Frontend \"iscsi\" not found.");
3252			break;
3253		}
3254
3255		retval = fe->ioctl(dev, cmd, addr, flag, td);
3256		break;
3257	}
3258	case CTL_PORT_REQ: {
3259		struct ctl_req *req;
3260		struct ctl_frontend *fe;
3261
3262		req = (struct ctl_req *)addr;
3263
3264		fe = ctl_frontend_find(req->driver);
3265		if (fe == NULL) {
3266			req->status = CTL_LUN_ERROR;
3267			snprintf(req->error_str, sizeof(req->error_str),
3268			    "Frontend \"%s\" not found.", req->driver);
3269			break;
3270		}
3271		if (req->num_args > 0) {
3272			req->kern_args = ctl_copyin_args(req->num_args,
3273			    req->args, req->error_str, sizeof(req->error_str));
3274			if (req->kern_args == NULL) {
3275				req->status = CTL_LUN_ERROR;
3276				break;
3277			}
3278		}
3279
3280		retval = fe->ioctl(dev, cmd, addr, flag, td);
3281
3282		if (req->num_args > 0) {
3283			ctl_copyout_args(req->num_args, req->kern_args);
3284			ctl_free_args(req->num_args, req->kern_args);
3285		}
3286		break;
3287	}
3288	case CTL_PORT_LIST: {
3289		struct sbuf *sb;
3290		struct ctl_port *port;
3291		struct ctl_lun_list *list;
3292		struct ctl_option *opt;
3293
3294		list = (struct ctl_lun_list *)addr;
3295
3296		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3297		if (sb == NULL) {
3298			list->status = CTL_LUN_LIST_ERROR;
3299			snprintf(list->error_str, sizeof(list->error_str),
3300				 "Unable to allocate %d bytes for LUN list",
3301				 list->alloc_len);
3302			break;
3303		}
3304
3305		sbuf_printf(sb, "<ctlportlist>\n");
3306
3307		mtx_lock(&softc->ctl_lock);
3308		STAILQ_FOREACH(port, &softc->port_list, links) {
3309			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3310					     (uintmax_t)port->targ_port);
3311
3312			/*
3313			 * Bail out as soon as we see that we've overfilled
3314			 * the buffer.
3315			 */
3316			if (retval != 0)
3317				break;
3318
3319			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3320			    "</frontend_type>\n", port->frontend->name);
3321			if (retval != 0)
3322				break;
3323
3324			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3325					     port->port_type);
3326			if (retval != 0)
3327				break;
3328
3329			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3330			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3331			if (retval != 0)
3332				break;
3333
3334			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3335			    port->port_name);
3336			if (retval != 0)
3337				break;
3338
3339			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3340			    port->physical_port);
3341			if (retval != 0)
3342				break;
3343
3344			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3345			    port->virtual_port);
3346			if (retval != 0)
3347				break;
3348
3349			retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3350			    (uintmax_t)port->wwnn);
3351			if (retval != 0)
3352				break;
3353
3354			retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3355			    (uintmax_t)port->wwpn);
3356			if (retval != 0)
3357				break;
3358
3359			if (port->port_info != NULL) {
3360				retval = port->port_info(port->onoff_arg, sb);
3361				if (retval != 0)
3362					break;
3363			}
3364			STAILQ_FOREACH(opt, &port->options, links) {
3365				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3366				    opt->name, opt->value, opt->name);
3367				if (retval != 0)
3368					break;
3369			}
3370
3371			retval = sbuf_printf(sb, "</targ_port>\n");
3372			if (retval != 0)
3373				break;
3374		}
3375		mtx_unlock(&softc->ctl_lock);
3376
3377		if ((retval != 0)
3378		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3379			retval = 0;
3380			sbuf_delete(sb);
3381			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3382			snprintf(list->error_str, sizeof(list->error_str),
3383				 "Out of space, %d bytes is too small",
3384				 list->alloc_len);
3385			break;
3386		}
3387
3388		sbuf_finish(sb);
3389
3390		retval = copyout(sbuf_data(sb), list->lun_xml,
3391				 sbuf_len(sb) + 1);
3392
3393		list->fill_len = sbuf_len(sb) + 1;
3394		list->status = CTL_LUN_LIST_OK;
3395		sbuf_delete(sb);
3396		break;
3397	}
3398	default: {
3399		/* XXX KDM should we fix this? */
3400#if 0
3401		struct ctl_backend_driver *backend;
3402		unsigned int type;
3403		int found;
3404
3405		found = 0;
3406
3407		/*
3408		 * We encode the backend type as the ioctl type for backend
3409		 * ioctls.  So parse it out here, and then search for a
3410		 * backend of this type.
3411		 */
3412		type = _IOC_TYPE(cmd);
3413
3414		STAILQ_FOREACH(backend, &softc->be_list, links) {
3415			if (backend->type == type) {
3416				found = 1;
3417				break;
3418			}
3419		}
3420		if (found == 0) {
3421			printf("ctl: unknown ioctl command %#lx or backend "
3422			       "%d\n", cmd, type);
3423			retval = EINVAL;
3424			break;
3425		}
3426		retval = backend->ioctl(dev, cmd, addr, flag, td);
3427#endif
3428		retval = ENOTTY;
3429		break;
3430	}
3431	}
3432	return (retval);
3433}
3434
3435uint32_t
3436ctl_get_initindex(struct ctl_nexus *nexus)
3437{
3438	if (nexus->targ_port < CTL_MAX_PORTS)
3439		return (nexus->initid.id +
3440			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3441	else
3442		return (nexus->initid.id +
3443		       ((nexus->targ_port - CTL_MAX_PORTS) *
3444			CTL_MAX_INIT_PER_PORT));
3445}
3446
3447uint32_t
3448ctl_get_resindex(struct ctl_nexus *nexus)
3449{
3450	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3451}
3452
3453uint32_t
3454ctl_port_idx(int port_num)
3455{
3456	if (port_num < CTL_MAX_PORTS)
3457		return(port_num);
3458	else
3459		return(port_num - CTL_MAX_PORTS);
3460}
3461
3462static uint32_t
3463ctl_map_lun(int port_num, uint32_t lun_id)
3464{
3465	struct ctl_port *port;
3466
3467	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3468	if (port == NULL)
3469		return (UINT32_MAX);
3470	if (port->lun_map == NULL)
3471		return (lun_id);
3472	return (port->lun_map(port->targ_lun_arg, lun_id));
3473}
3474
3475static uint32_t
3476ctl_map_lun_back(int port_num, uint32_t lun_id)
3477{
3478	struct ctl_port *port;
3479	uint32_t i;
3480
3481	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3482	if (port->lun_map == NULL)
3483		return (lun_id);
3484	for (i = 0; i < CTL_MAX_LUNS; i++) {
3485		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3486			return (i);
3487	}
3488	return (UINT32_MAX);
3489}
3490
3491/*
3492 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3493 * that are a power of 2.
3494 */
3495int
3496ctl_ffz(uint32_t *mask, uint32_t size)
3497{
3498	uint32_t num_chunks, num_pieces;
3499	int i, j;
3500
3501	num_chunks = (size >> 5);
3502	if (num_chunks == 0)
3503		num_chunks++;
3504	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3505
3506	for (i = 0; i < num_chunks; i++) {
3507		for (j = 0; j < num_pieces; j++) {
3508			if ((mask[i] & (1 << j)) == 0)
3509				return ((i << 5) + j);
3510		}
3511	}
3512
3513	return (-1);
3514}
3515
3516int
3517ctl_set_mask(uint32_t *mask, uint32_t bit)
3518{
3519	uint32_t chunk, piece;
3520
3521	chunk = bit >> 5;
3522	piece = bit % (sizeof(uint32_t) * 8);
3523
3524	if ((mask[chunk] & (1 << piece)) != 0)
3525		return (-1);
3526	else
3527		mask[chunk] |= (1 << piece);
3528
3529	return (0);
3530}
3531
3532int
3533ctl_clear_mask(uint32_t *mask, uint32_t bit)
3534{
3535	uint32_t chunk, piece;
3536
3537	chunk = bit >> 5;
3538	piece = bit % (sizeof(uint32_t) * 8);
3539
3540	if ((mask[chunk] & (1 << piece)) == 0)
3541		return (-1);
3542	else
3543		mask[chunk] &= ~(1 << piece);
3544
3545	return (0);
3546}
3547
3548int
3549ctl_is_set(uint32_t *mask, uint32_t bit)
3550{
3551	uint32_t chunk, piece;
3552
3553	chunk = bit >> 5;
3554	piece = bit % (sizeof(uint32_t) * 8);
3555
3556	if ((mask[chunk] & (1 << piece)) == 0)
3557		return (0);
3558	else
3559		return (1);
3560}
3561
3562#ifdef unused
3563/*
3564 * The bus, target and lun are optional, they can be filled in later.
3565 * can_wait is used to determine whether we can wait on the malloc or not.
3566 */
3567union ctl_io*
3568ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3569	      uint32_t targ_lun, int can_wait)
3570{
3571	union ctl_io *io;
3572
3573	if (can_wait)
3574		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3575	else
3576		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3577
3578	if (io != NULL) {
3579		io->io_hdr.io_type = io_type;
3580		io->io_hdr.targ_port = targ_port;
3581		/*
3582		 * XXX KDM this needs to change/go away.  We need to move
3583		 * to a preallocated pool of ctl_scsiio structures.
3584		 */
3585		io->io_hdr.nexus.targ_target.id = targ_target;
3586		io->io_hdr.nexus.targ_lun = targ_lun;
3587	}
3588
3589	return (io);
3590}
3591
3592void
3593ctl_kfree_io(union ctl_io *io)
3594{
3595	free(io, M_CTL);
3596}
3597#endif /* unused */
3598
3599/*
3600 * ctl_softc, pool_type, total_ctl_io are passed in.
3601 * npool is passed out.
3602 */
3603int
3604ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3605		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3606{
3607	uint32_t i;
3608	union ctl_io *cur_io, *next_io;
3609	struct ctl_io_pool *pool;
3610	int retval;
3611
3612	retval = 0;
3613
3614	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3615					    M_NOWAIT | M_ZERO);
3616	if (pool == NULL) {
3617		retval = ENOMEM;
3618		goto bailout;
3619	}
3620
3621	pool->type = pool_type;
3622	pool->ctl_softc = ctl_softc;
3623
3624	mtx_lock(&ctl_softc->pool_lock);
3625	pool->id = ctl_softc->cur_pool_id++;
3626	mtx_unlock(&ctl_softc->pool_lock);
3627
3628	pool->flags = CTL_POOL_FLAG_NONE;
3629	pool->refcount = 1;		/* Reference for validity. */
3630	STAILQ_INIT(&pool->free_queue);
3631
3632	/*
3633	 * XXX KDM other options here:
3634	 * - allocate a page at a time
3635	 * - allocate one big chunk of memory.
3636	 * Page allocation might work well, but would take a little more
3637	 * tracking.
3638	 */
3639	for (i = 0; i < total_ctl_io; i++) {
3640		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3641						M_NOWAIT);
3642		if (cur_io == NULL) {
3643			retval = ENOMEM;
3644			break;
3645		}
3646		cur_io->io_hdr.pool = pool;
3647		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3648		pool->total_ctl_io++;
3649		pool->free_ctl_io++;
3650	}
3651
3652	if (retval != 0) {
3653		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3654		     cur_io != NULL; cur_io = next_io) {
3655			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3656							      links);
3657			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3658				      ctl_io_hdr, links);
3659			free(cur_io, M_CTLIO);
3660		}
3661
3662		free(pool, M_CTL);
3663		goto bailout;
3664	}
3665	mtx_lock(&ctl_softc->pool_lock);
3666	ctl_softc->num_pools++;
3667	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3668	/*
3669	 * Increment our usage count if this is an external consumer, so we
3670	 * can't get unloaded until the external consumer (most likely a
3671	 * FETD) unloads and frees his pool.
3672	 *
3673	 * XXX KDM will this increment the caller's module use count, or
3674	 * mine?
3675	 */
3676#if 0
3677	if ((pool_type != CTL_POOL_EMERGENCY)
3678	 && (pool_type != CTL_POOL_INTERNAL)
3679	 && (pool_type != CTL_POOL_4OTHERSC))
3680		MOD_INC_USE_COUNT;
3681#endif
3682
3683	mtx_unlock(&ctl_softc->pool_lock);
3684
3685	*npool = pool;
3686
3687bailout:
3688
3689	return (retval);
3690}
3691
3692static int
3693ctl_pool_acquire(struct ctl_io_pool *pool)
3694{
3695
3696	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3697
3698	if (pool->flags & CTL_POOL_FLAG_INVALID)
3699		return (EINVAL);
3700
3701	pool->refcount++;
3702
3703	return (0);
3704}
3705
3706static void
3707ctl_pool_release(struct ctl_io_pool *pool)
3708{
3709	struct ctl_softc *ctl_softc = pool->ctl_softc;
3710	union ctl_io *io;
3711
3712	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3713
3714	if (--pool->refcount != 0)
3715		return;
3716
3717	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3718		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3719			      links);
3720		free(io, M_CTLIO);
3721	}
3722
3723	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3724	ctl_softc->num_pools--;
3725
3726	/*
3727	 * XXX KDM will this decrement the caller's usage count or mine?
3728	 */
3729#if 0
3730	if ((pool->type != CTL_POOL_EMERGENCY)
3731	 && (pool->type != CTL_POOL_INTERNAL)
3732	 && (pool->type != CTL_POOL_4OTHERSC))
3733		MOD_DEC_USE_COUNT;
3734#endif
3735
3736	free(pool, M_CTL);
3737}
3738
3739void
3740ctl_pool_free(struct ctl_io_pool *pool)
3741{
3742	struct ctl_softc *ctl_softc;
3743
3744	if (pool == NULL)
3745		return;
3746
3747	ctl_softc = pool->ctl_softc;
3748	mtx_lock(&ctl_softc->pool_lock);
3749	pool->flags |= CTL_POOL_FLAG_INVALID;
3750	ctl_pool_release(pool);
3751	mtx_unlock(&ctl_softc->pool_lock);
3752}
3753
3754/*
3755 * This routine does not block (except for spinlocks of course).
3756 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3757 * possible.
3758 */
3759union ctl_io *
3760ctl_alloc_io(void *pool_ref)
3761{
3762	union ctl_io *io;
3763	struct ctl_softc *ctl_softc;
3764	struct ctl_io_pool *pool, *npool;
3765	struct ctl_io_pool *emergency_pool;
3766
3767	pool = (struct ctl_io_pool *)pool_ref;
3768
3769	if (pool == NULL) {
3770		printf("%s: pool is NULL\n", __func__);
3771		return (NULL);
3772	}
3773
3774	emergency_pool = NULL;
3775
3776	ctl_softc = pool->ctl_softc;
3777
3778	mtx_lock(&ctl_softc->pool_lock);
3779	/*
3780	 * First, try to get the io structure from the user's pool.
3781	 */
3782	if (ctl_pool_acquire(pool) == 0) {
3783		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3784		if (io != NULL) {
3785			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3786			pool->total_allocated++;
3787			pool->free_ctl_io--;
3788			mtx_unlock(&ctl_softc->pool_lock);
3789			return (io);
3790		} else
3791			ctl_pool_release(pool);
3792	}
3793	/*
3794	 * If he doesn't have any io structures left, search for an
3795	 * emergency pool and grab one from there.
3796	 */
3797	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3798		if (npool->type != CTL_POOL_EMERGENCY)
3799			continue;
3800
3801		if (ctl_pool_acquire(npool) != 0)
3802			continue;
3803
3804		emergency_pool = npool;
3805
3806		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3807		if (io != NULL) {
3808			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3809			npool->total_allocated++;
3810			npool->free_ctl_io--;
3811			mtx_unlock(&ctl_softc->pool_lock);
3812			return (io);
3813		} else
3814			ctl_pool_release(npool);
3815	}
3816
3817	/* Drop the spinlock before we malloc */
3818	mtx_unlock(&ctl_softc->pool_lock);
3819
3820	/*
3821	 * The emergency pool (if it exists) didn't have one, so try an
3822	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3823	 */
3824	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3825	if (io != NULL) {
3826		/*
3827		 * If the emergency pool exists but is empty, add this
3828		 * ctl_io to its list when it gets freed.
3829		 */
3830		if (emergency_pool != NULL) {
3831			mtx_lock(&ctl_softc->pool_lock);
3832			if (ctl_pool_acquire(emergency_pool) == 0) {
3833				io->io_hdr.pool = emergency_pool;
3834				emergency_pool->total_ctl_io++;
3835				/*
3836				 * Need to bump this, otherwise
3837				 * total_allocated and total_freed won't
3838				 * match when we no longer have anything
3839				 * outstanding.
3840				 */
3841				emergency_pool->total_allocated++;
3842			}
3843			mtx_unlock(&ctl_softc->pool_lock);
3844		} else
3845			io->io_hdr.pool = NULL;
3846	}
3847
3848	return (io);
3849}
3850
3851void
3852ctl_free_io(union ctl_io *io)
3853{
3854	if (io == NULL)
3855		return;
3856
3857	/*
3858	 * If this ctl_io has a pool, return it to that pool.
3859	 */
3860	if (io->io_hdr.pool != NULL) {
3861		struct ctl_io_pool *pool;
3862
3863		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3864		mtx_lock(&pool->ctl_softc->pool_lock);
3865		io->io_hdr.io_type = 0xff;
3866		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3867		pool->total_freed++;
3868		pool->free_ctl_io++;
3869		ctl_pool_release(pool);
3870		mtx_unlock(&pool->ctl_softc->pool_lock);
3871	} else {
3872		/*
3873		 * Otherwise, just free it.  We probably malloced it and
3874		 * the emergency pool wasn't available.
3875		 */
3876		free(io, M_CTLIO);
3877	}
3878
3879}
3880
3881void
3882ctl_zero_io(union ctl_io *io)
3883{
3884	void *pool_ref;
3885
3886	if (io == NULL)
3887		return;
3888
3889	/*
3890	 * May need to preserve linked list pointers at some point too.
3891	 */
3892	pool_ref = io->io_hdr.pool;
3893
3894	memset(io, 0, sizeof(*io));
3895
3896	io->io_hdr.pool = pool_ref;
3897}
3898
3899/*
3900 * This routine is currently used for internal copies of ctl_ios that need
3901 * to persist for some reason after we've already returned status to the
3902 * FETD.  (Thus the flag set.)
3903 *
3904 * XXX XXX
3905 * Note that this makes a blind copy of all fields in the ctl_io, except
3906 * for the pool reference.  This includes any memory that has been
3907 * allocated!  That memory will no longer be valid after done has been
3908 * called, so this would be VERY DANGEROUS for command that actually does
3909 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3910 * start and stop commands, which don't transfer any data, so this is not a
3911 * problem.  If it is used for anything else, the caller would also need to
3912 * allocate data buffer space and this routine would need to be modified to
3913 * copy the data buffer(s) as well.
3914 */
3915void
3916ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3917{
3918	void *pool_ref;
3919
3920	if ((src == NULL)
3921	 || (dest == NULL))
3922		return;
3923
3924	/*
3925	 * May need to preserve linked list pointers at some point too.
3926	 */
3927	pool_ref = dest->io_hdr.pool;
3928
3929	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3930
3931	dest->io_hdr.pool = pool_ref;
3932	/*
3933	 * We need to know that this is an internal copy, and doesn't need
3934	 * to get passed back to the FETD that allocated it.
3935	 */
3936	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3937}
3938
3939#ifdef NEEDTOPORT
3940static void
3941ctl_update_power_subpage(struct copan_power_subpage *page)
3942{
3943	int num_luns, num_partitions, config_type;
3944	struct ctl_softc *softc;
3945	cs_BOOL_t aor_present, shelf_50pct_power;
3946	cs_raidset_personality_t rs_type;
3947	int max_active_luns;
3948
3949	softc = control_softc;
3950
3951	/* subtract out the processor LUN */
3952	num_luns = softc->num_luns - 1;
3953	/*
3954	 * Default to 7 LUNs active, which was the only number we allowed
3955	 * in the past.
3956	 */
3957	max_active_luns = 7;
3958
3959	num_partitions = config_GetRsPartitionInfo();
3960	config_type = config_GetConfigType();
3961	shelf_50pct_power = config_GetShelfPowerMode();
3962	aor_present = config_IsAorRsPresent();
3963
3964	rs_type = ddb_GetRsRaidType(1);
3965	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3966	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3967		EPRINT(0, "Unsupported RS type %d!", rs_type);
3968	}
3969
3970
3971	page->total_luns = num_luns;
3972
3973	switch (config_type) {
3974	case 40:
3975		/*
3976		 * In a 40 drive configuration, it doesn't matter what DC
3977		 * cards we have, whether we have AOR enabled or not,
3978		 * partitioning or not, or what type of RAIDset we have.
3979		 * In that scenario, we can power up every LUN we present
3980		 * to the user.
3981		 */
3982		max_active_luns = num_luns;
3983
3984		break;
3985	case 64:
3986		if (shelf_50pct_power == CS_FALSE) {
3987			/* 25% power */
3988			if (aor_present == CS_TRUE) {
3989				if (rs_type ==
3990				     CS_RAIDSET_PERSONALITY_RAID5) {
3991					max_active_luns = 7;
3992				} else if (rs_type ==
3993					 CS_RAIDSET_PERSONALITY_RAID1){
3994					max_active_luns = 14;
3995				} else {
3996					/* XXX KDM now what?? */
3997				}
3998			} else {
3999				if (rs_type ==
4000				     CS_RAIDSET_PERSONALITY_RAID5) {
4001					max_active_luns = 8;
4002				} else if (rs_type ==
4003					 CS_RAIDSET_PERSONALITY_RAID1){
4004					max_active_luns = 16;
4005				} else {
4006					/* XXX KDM now what?? */
4007				}
4008			}
4009		} else {
4010			/* 50% power */
4011			/*
4012			 * With 50% power in a 64 drive configuration, we
4013			 * can power all LUNs we present.
4014			 */
4015			max_active_luns = num_luns;
4016		}
4017		break;
4018	case 112:
4019		if (shelf_50pct_power == CS_FALSE) {
4020			/* 25% power */
4021			if (aor_present == CS_TRUE) {
4022				if (rs_type ==
4023				     CS_RAIDSET_PERSONALITY_RAID5) {
4024					max_active_luns = 7;
4025				} else if (rs_type ==
4026					 CS_RAIDSET_PERSONALITY_RAID1){
4027					max_active_luns = 14;
4028				} else {
4029					/* XXX KDM now what?? */
4030				}
4031			} else {
4032				if (rs_type ==
4033				     CS_RAIDSET_PERSONALITY_RAID5) {
4034					max_active_luns = 8;
4035				} else if (rs_type ==
4036					 CS_RAIDSET_PERSONALITY_RAID1){
4037					max_active_luns = 16;
4038				} else {
4039					/* XXX KDM now what?? */
4040				}
4041			}
4042		} else {
4043			/* 50% power */
4044			if (aor_present == CS_TRUE) {
4045				if (rs_type ==
4046				     CS_RAIDSET_PERSONALITY_RAID5) {
4047					max_active_luns = 14;
4048				} else if (rs_type ==
4049					 CS_RAIDSET_PERSONALITY_RAID1){
4050					/*
4051					 * We're assuming here that disk
4052					 * caching is enabled, and so we're
4053					 * able to power up half of each
4054					 * LUN, and cache all writes.
4055					 */
4056					max_active_luns = num_luns;
4057				} else {
4058					/* XXX KDM now what?? */
4059				}
4060			} else {
4061				if (rs_type ==
4062				     CS_RAIDSET_PERSONALITY_RAID5) {
4063					max_active_luns = 15;
4064				} else if (rs_type ==
4065					 CS_RAIDSET_PERSONALITY_RAID1){
4066					max_active_luns = 30;
4067				} else {
4068					/* XXX KDM now what?? */
4069				}
4070			}
4071		}
4072		break;
4073	default:
4074		/*
4075		 * In this case, we have an unknown configuration, so we
4076		 * just use the default from above.
4077		 */
4078		break;
4079	}
4080
4081	page->max_active_luns = max_active_luns;
4082#if 0
4083	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4084	       page->total_luns, page->max_active_luns);
4085#endif
4086}
4087#endif /* NEEDTOPORT */
4088
4089/*
4090 * This routine could be used in the future to load default and/or saved
4091 * mode page parameters for a particuar lun.
4092 */
4093static int
4094ctl_init_page_index(struct ctl_lun *lun)
4095{
4096	int i;
4097	struct ctl_page_index *page_index;
4098	struct ctl_softc *softc;
4099	const char *value;
4100
4101	memcpy(&lun->mode_pages.index, page_index_template,
4102	       sizeof(page_index_template));
4103
4104	softc = lun->ctl_softc;
4105
4106	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4107
4108		page_index = &lun->mode_pages.index[i];
4109		/*
4110		 * If this is a disk-only mode page, there's no point in
4111		 * setting it up.  For some pages, we have to have some
4112		 * basic information about the disk in order to calculate the
4113		 * mode page data.
4114		 */
4115		if ((lun->be_lun->lun_type != T_DIRECT)
4116		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4117			continue;
4118
4119		switch (page_index->page_code & SMPH_PC_MASK) {
4120		case SMS_FORMAT_DEVICE_PAGE: {
4121			struct scsi_format_page *format_page;
4122
4123			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4124				panic("subpage is incorrect!");
4125
4126			/*
4127			 * Sectors per track are set above.  Bytes per
4128			 * sector need to be set here on a per-LUN basis.
4129			 */
4130			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4131			       &format_page_default,
4132			       sizeof(format_page_default));
4133			memcpy(&lun->mode_pages.format_page[
4134			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4135			       sizeof(format_page_changeable));
4136			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4137			       &format_page_default,
4138			       sizeof(format_page_default));
4139			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4140			       &format_page_default,
4141			       sizeof(format_page_default));
4142
4143			format_page = &lun->mode_pages.format_page[
4144				CTL_PAGE_CURRENT];
4145			scsi_ulto2b(lun->be_lun->blocksize,
4146				    format_page->bytes_per_sector);
4147
4148			format_page = &lun->mode_pages.format_page[
4149				CTL_PAGE_DEFAULT];
4150			scsi_ulto2b(lun->be_lun->blocksize,
4151				    format_page->bytes_per_sector);
4152
4153			format_page = &lun->mode_pages.format_page[
4154				CTL_PAGE_SAVED];
4155			scsi_ulto2b(lun->be_lun->blocksize,
4156				    format_page->bytes_per_sector);
4157
4158			page_index->page_data =
4159				(uint8_t *)lun->mode_pages.format_page;
4160			break;
4161		}
4162		case SMS_RIGID_DISK_PAGE: {
4163			struct scsi_rigid_disk_page *rigid_disk_page;
4164			uint32_t sectors_per_cylinder;
4165			uint64_t cylinders;
4166#ifndef	__XSCALE__
4167			int shift;
4168#endif /* !__XSCALE__ */
4169
4170			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4171				panic("invalid subpage value %d",
4172				      page_index->subpage);
4173
4174			/*
4175			 * Rotation rate and sectors per track are set
4176			 * above.  We calculate the cylinders here based on
4177			 * capacity.  Due to the number of heads and
4178			 * sectors per track we're using, smaller arrays
4179			 * may turn out to have 0 cylinders.  Linux and
4180			 * FreeBSD don't pay attention to these mode pages
4181			 * to figure out capacity, but Solaris does.  It
4182			 * seems to deal with 0 cylinders just fine, and
4183			 * works out a fake geometry based on the capacity.
4184			 */
4185			memcpy(&lun->mode_pages.rigid_disk_page[
4186			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4187			       sizeof(rigid_disk_page_default));
4188			memcpy(&lun->mode_pages.rigid_disk_page[
4189			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4190			       sizeof(rigid_disk_page_changeable));
4191			memcpy(&lun->mode_pages.rigid_disk_page[
4192			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4193			       sizeof(rigid_disk_page_default));
4194			memcpy(&lun->mode_pages.rigid_disk_page[
4195			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4196			       sizeof(rigid_disk_page_default));
4197
4198			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4199				CTL_DEFAULT_HEADS;
4200
4201			/*
4202			 * The divide method here will be more accurate,
4203			 * probably, but results in floating point being
4204			 * used in the kernel on i386 (__udivdi3()).  On the
4205			 * XScale, though, __udivdi3() is implemented in
4206			 * software.
4207			 *
4208			 * The shift method for cylinder calculation is
4209			 * accurate if sectors_per_cylinder is a power of
4210			 * 2.  Otherwise it might be slightly off -- you
4211			 * might have a bit of a truncation problem.
4212			 */
4213#ifdef	__XSCALE__
4214			cylinders = (lun->be_lun->maxlba + 1) /
4215				sectors_per_cylinder;
4216#else
4217			for (shift = 31; shift > 0; shift--) {
4218				if (sectors_per_cylinder & (1 << shift))
4219					break;
4220			}
4221			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4222#endif
4223
4224			/*
4225			 * We've basically got 3 bytes, or 24 bits for the
4226			 * cylinder size in the mode page.  If we're over,
4227			 * just round down to 2^24.
4228			 */
4229			if (cylinders > 0xffffff)
4230				cylinders = 0xffffff;
4231
4232			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4233				CTL_PAGE_CURRENT];
4234			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4235
4236			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4237				CTL_PAGE_DEFAULT];
4238			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4239
4240			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4241				CTL_PAGE_SAVED];
4242			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4243
4244			page_index->page_data =
4245				(uint8_t *)lun->mode_pages.rigid_disk_page;
4246			break;
4247		}
4248		case SMS_CACHING_PAGE: {
4249			struct scsi_caching_page *caching_page;
4250
4251			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4252				panic("invalid subpage value %d",
4253				      page_index->subpage);
4254			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4255			       &caching_page_default,
4256			       sizeof(caching_page_default));
4257			memcpy(&lun->mode_pages.caching_page[
4258			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4259			       sizeof(caching_page_changeable));
4260			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4261			       &caching_page_default,
4262			       sizeof(caching_page_default));
4263			caching_page = &lun->mode_pages.caching_page[
4264			    CTL_PAGE_SAVED];
4265			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4266			if (value != NULL && strcmp(value, "off") == 0)
4267				caching_page->flags1 &= ~SCP_WCE;
4268			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4269			if (value != NULL && strcmp(value, "off") == 0)
4270				caching_page->flags1 |= SCP_RCD;
4271			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4272			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4273			       sizeof(caching_page_default));
4274			page_index->page_data =
4275				(uint8_t *)lun->mode_pages.caching_page;
4276			break;
4277		}
4278		case SMS_CONTROL_MODE_PAGE: {
4279			struct scsi_control_page *control_page;
4280
4281			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4282				panic("invalid subpage value %d",
4283				      page_index->subpage);
4284
4285			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4286			       &control_page_default,
4287			       sizeof(control_page_default));
4288			memcpy(&lun->mode_pages.control_page[
4289			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4290			       sizeof(control_page_changeable));
4291			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4292			       &control_page_default,
4293			       sizeof(control_page_default));
4294			control_page = &lun->mode_pages.control_page[
4295			    CTL_PAGE_SAVED];
4296			value = ctl_get_opt(&lun->be_lun->options, "reordering");
4297			if (value != NULL && strcmp(value, "unrestricted") == 0) {
4298				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4299				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4300			}
4301			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4302			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4303			       sizeof(control_page_default));
4304			page_index->page_data =
4305				(uint8_t *)lun->mode_pages.control_page;
4306			break;
4307
4308		}
4309		case SMS_VENDOR_SPECIFIC_PAGE:{
4310			switch (page_index->subpage) {
4311			case PWR_SUBPAGE_CODE: {
4312				struct copan_power_subpage *current_page,
4313							   *saved_page;
4314
4315				memcpy(&lun->mode_pages.power_subpage[
4316				       CTL_PAGE_CURRENT],
4317				       &power_page_default,
4318				       sizeof(power_page_default));
4319				memcpy(&lun->mode_pages.power_subpage[
4320				       CTL_PAGE_CHANGEABLE],
4321				       &power_page_changeable,
4322				       sizeof(power_page_changeable));
4323				memcpy(&lun->mode_pages.power_subpage[
4324				       CTL_PAGE_DEFAULT],
4325				       &power_page_default,
4326				       sizeof(power_page_default));
4327				memcpy(&lun->mode_pages.power_subpage[
4328				       CTL_PAGE_SAVED],
4329				       &power_page_default,
4330				       sizeof(power_page_default));
4331				page_index->page_data =
4332				    (uint8_t *)lun->mode_pages.power_subpage;
4333
4334				current_page = (struct copan_power_subpage *)
4335					(page_index->page_data +
4336					 (page_index->page_len *
4337					  CTL_PAGE_CURRENT));
4338			        saved_page = (struct copan_power_subpage *)
4339				        (page_index->page_data +
4340					 (page_index->page_len *
4341					  CTL_PAGE_SAVED));
4342				break;
4343			}
4344			case APS_SUBPAGE_CODE: {
4345				struct copan_aps_subpage *current_page,
4346							 *saved_page;
4347
4348				// This gets set multiple times but
4349				// it should always be the same. It's
4350				// only done during init so who cares.
4351				index_to_aps_page = i;
4352
4353				memcpy(&lun->mode_pages.aps_subpage[
4354				       CTL_PAGE_CURRENT],
4355				       &aps_page_default,
4356				       sizeof(aps_page_default));
4357				memcpy(&lun->mode_pages.aps_subpage[
4358				       CTL_PAGE_CHANGEABLE],
4359				       &aps_page_changeable,
4360				       sizeof(aps_page_changeable));
4361				memcpy(&lun->mode_pages.aps_subpage[
4362				       CTL_PAGE_DEFAULT],
4363				       &aps_page_default,
4364				       sizeof(aps_page_default));
4365				memcpy(&lun->mode_pages.aps_subpage[
4366				       CTL_PAGE_SAVED],
4367				       &aps_page_default,
4368				       sizeof(aps_page_default));
4369				page_index->page_data =
4370					(uint8_t *)lun->mode_pages.aps_subpage;
4371
4372				current_page = (struct copan_aps_subpage *)
4373					(page_index->page_data +
4374					 (page_index->page_len *
4375					  CTL_PAGE_CURRENT));
4376				saved_page = (struct copan_aps_subpage *)
4377					(page_index->page_data +
4378					 (page_index->page_len *
4379					  CTL_PAGE_SAVED));
4380				break;
4381			}
4382			case DBGCNF_SUBPAGE_CODE: {
4383				struct copan_debugconf_subpage *current_page,
4384							       *saved_page;
4385
4386				memcpy(&lun->mode_pages.debugconf_subpage[
4387				       CTL_PAGE_CURRENT],
4388				       &debugconf_page_default,
4389				       sizeof(debugconf_page_default));
4390				memcpy(&lun->mode_pages.debugconf_subpage[
4391				       CTL_PAGE_CHANGEABLE],
4392				       &debugconf_page_changeable,
4393				       sizeof(debugconf_page_changeable));
4394				memcpy(&lun->mode_pages.debugconf_subpage[
4395				       CTL_PAGE_DEFAULT],
4396				       &debugconf_page_default,
4397				       sizeof(debugconf_page_default));
4398				memcpy(&lun->mode_pages.debugconf_subpage[
4399				       CTL_PAGE_SAVED],
4400				       &debugconf_page_default,
4401				       sizeof(debugconf_page_default));
4402				page_index->page_data =
4403					(uint8_t *)lun->mode_pages.debugconf_subpage;
4404
4405				current_page = (struct copan_debugconf_subpage *)
4406					(page_index->page_data +
4407					 (page_index->page_len *
4408					  CTL_PAGE_CURRENT));
4409				saved_page = (struct copan_debugconf_subpage *)
4410					(page_index->page_data +
4411					 (page_index->page_len *
4412					  CTL_PAGE_SAVED));
4413				break;
4414			}
4415			default:
4416				panic("invalid subpage value %d",
4417				      page_index->subpage);
4418				break;
4419			}
4420   			break;
4421		}
4422		default:
4423			panic("invalid page value %d",
4424			      page_index->page_code & SMPH_PC_MASK);
4425			break;
4426    	}
4427	}
4428
4429	return (CTL_RETVAL_COMPLETE);
4430}
4431
4432/*
4433 * LUN allocation.
4434 *
4435 * Requirements:
4436 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4437 *   wants us to allocate the LUN and he can block.
4438 * - ctl_softc is always set
4439 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4440 *
4441 * Returns 0 for success, non-zero (errno) for failure.
4442 */
4443static int
4444ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4445	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4446{
4447	struct ctl_lun *nlun, *lun;
4448	struct ctl_port *port;
4449	struct scsi_vpd_id_descriptor *desc;
4450	struct scsi_vpd_id_t10 *t10id;
4451	const char *eui, *naa, *scsiname, *vendor, *value;
4452	int lun_number, i, lun_malloced;
4453	int devidlen, idlen1, idlen2 = 0, len;
4454
4455	if (be_lun == NULL)
4456		return (EINVAL);
4457
4458	/*
4459	 * We currently only support Direct Access or Processor LUN types.
4460	 */
4461	switch (be_lun->lun_type) {
4462	case T_DIRECT:
4463		break;
4464	case T_PROCESSOR:
4465		break;
4466	case T_SEQUENTIAL:
4467	case T_CHANGER:
4468	default:
4469		be_lun->lun_config_status(be_lun->be_lun,
4470					  CTL_LUN_CONFIG_FAILURE);
4471		break;
4472	}
4473	if (ctl_lun == NULL) {
4474		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4475		lun_malloced = 1;
4476	} else {
4477		lun_malloced = 0;
4478		lun = ctl_lun;
4479	}
4480
4481	memset(lun, 0, sizeof(*lun));
4482	if (lun_malloced)
4483		lun->flags = CTL_LUN_MALLOCED;
4484
4485	/* Generate LUN ID. */
4486	devidlen = max(CTL_DEVID_MIN_LEN,
4487	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4488	idlen1 = sizeof(*t10id) + devidlen;
4489	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4490	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4491	if (scsiname != NULL) {
4492		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4493		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4494	}
4495	eui = ctl_get_opt(&be_lun->options, "eui");
4496	if (eui != NULL) {
4497		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4498	}
4499	naa = ctl_get_opt(&be_lun->options, "naa");
4500	if (naa != NULL) {
4501		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4502	}
4503	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4504	    M_CTL, M_WAITOK | M_ZERO);
4505	lun->lun_devid->len = len;
4506	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4507	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4508	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4509	desc->length = idlen1;
4510	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4511	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4512	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4513		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4514	} else {
4515		strncpy(t10id->vendor, vendor,
4516		    min(sizeof(t10id->vendor), strlen(vendor)));
4517	}
4518	strncpy((char *)t10id->vendor_spec_id,
4519	    (char *)be_lun->device_id, devidlen);
4520	if (scsiname != NULL) {
4521		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4522		    desc->length);
4523		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4524		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4525		    SVPD_ID_TYPE_SCSI_NAME;
4526		desc->length = idlen2;
4527		strlcpy(desc->identifier, scsiname, idlen2);
4528	}
4529	if (eui != NULL) {
4530		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4531		    desc->length);
4532		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4533		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4534		    SVPD_ID_TYPE_EUI64;
4535		desc->length = 8;
4536		scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4537	}
4538	if (naa != NULL) {
4539		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4540		    desc->length);
4541		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4542		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4543		    SVPD_ID_TYPE_NAA;
4544		desc->length = 8;
4545		scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4546	}
4547
4548	mtx_lock(&ctl_softc->ctl_lock);
4549	/*
4550	 * See if the caller requested a particular LUN number.  If so, see
4551	 * if it is available.  Otherwise, allocate the first available LUN.
4552	 */
4553	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4554		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4555		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4556			mtx_unlock(&ctl_softc->ctl_lock);
4557			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4558				printf("ctl: requested LUN ID %d is higher "
4559				       "than CTL_MAX_LUNS - 1 (%d)\n",
4560				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4561			} else {
4562				/*
4563				 * XXX KDM return an error, or just assign
4564				 * another LUN ID in this case??
4565				 */
4566				printf("ctl: requested LUN ID %d is already "
4567				       "in use\n", be_lun->req_lun_id);
4568			}
4569			if (lun->flags & CTL_LUN_MALLOCED)
4570				free(lun, M_CTL);
4571			be_lun->lun_config_status(be_lun->be_lun,
4572						  CTL_LUN_CONFIG_FAILURE);
4573			return (ENOSPC);
4574		}
4575		lun_number = be_lun->req_lun_id;
4576	} else {
4577		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4578		if (lun_number == -1) {
4579			mtx_unlock(&ctl_softc->ctl_lock);
4580			printf("ctl: can't allocate LUN on target %ju, out of "
4581			       "LUNs\n", (uintmax_t)target_id.id);
4582			if (lun->flags & CTL_LUN_MALLOCED)
4583				free(lun, M_CTL);
4584			be_lun->lun_config_status(be_lun->be_lun,
4585						  CTL_LUN_CONFIG_FAILURE);
4586			return (ENOSPC);
4587		}
4588	}
4589	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4590
4591	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4592	lun->target = target_id;
4593	lun->lun = lun_number;
4594	lun->be_lun = be_lun;
4595	/*
4596	 * The processor LUN is always enabled.  Disk LUNs come on line
4597	 * disabled, and must be enabled by the backend.
4598	 */
4599	lun->flags |= CTL_LUN_DISABLED;
4600	lun->backend = be_lun->be;
4601	be_lun->ctl_lun = lun;
4602	be_lun->lun_id = lun_number;
4603	atomic_add_int(&be_lun->be->num_luns, 1);
4604	if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4605		lun->flags |= CTL_LUN_OFFLINE;
4606
4607	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4608		lun->flags |= CTL_LUN_STOPPED;
4609
4610	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4611		lun->flags |= CTL_LUN_INOPERABLE;
4612
4613	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4614		lun->flags |= CTL_LUN_PRIMARY_SC;
4615
4616	value = ctl_get_opt(&be_lun->options, "readonly");
4617	if (value != NULL && strcmp(value, "on") == 0)
4618		lun->flags |= CTL_LUN_READONLY;
4619
4620	lun->ctl_softc = ctl_softc;
4621	TAILQ_INIT(&lun->ooa_queue);
4622	TAILQ_INIT(&lun->blocked_queue);
4623	STAILQ_INIT(&lun->error_list);
4624	ctl_tpc_lun_init(lun);
4625
4626	/*
4627	 * Initialize the mode page index.
4628	 */
4629	ctl_init_page_index(lun);
4630
4631	/*
4632	 * Set the poweron UA for all initiators on this LUN only.
4633	 */
4634	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4635		lun->pending_ua[i] = CTL_UA_POWERON;
4636
4637	/*
4638	 * Now, before we insert this lun on the lun list, set the lun
4639	 * inventory changed UA for all other luns.
4640	 */
4641	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4642		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4643			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4644		}
4645	}
4646
4647	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4648
4649	ctl_softc->ctl_luns[lun_number] = lun;
4650
4651	ctl_softc->num_luns++;
4652
4653	/* Setup statistics gathering */
4654	lun->stats.device_type = be_lun->lun_type;
4655	lun->stats.lun_number = lun_number;
4656	if (lun->stats.device_type == T_DIRECT)
4657		lun->stats.blocksize = be_lun->blocksize;
4658	else
4659		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4660	for (i = 0;i < CTL_MAX_PORTS;i++)
4661		lun->stats.ports[i].targ_port = i;
4662
4663	mtx_unlock(&ctl_softc->ctl_lock);
4664
4665	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4666
4667	/*
4668	 * Run through each registered FETD and bring it online if it isn't
4669	 * already.  Enable the target ID if it hasn't been enabled, and
4670	 * enable this particular LUN.
4671	 */
4672	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4673		int retval;
4674
4675		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4676		if (retval != 0) {
4677			printf("ctl_alloc_lun: FETD %s port %d returned error "
4678			       "%d for lun_enable on target %ju lun %d\n",
4679			       port->port_name, port->targ_port, retval,
4680			       (uintmax_t)target_id.id, lun_number);
4681		} else
4682			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4683	}
4684	return (0);
4685}
4686
4687/*
4688 * Delete a LUN.
4689 * Assumptions:
4690 * - LUN has already been marked invalid and any pending I/O has been taken
4691 *   care of.
4692 */
4693static int
4694ctl_free_lun(struct ctl_lun *lun)
4695{
4696	struct ctl_softc *softc;
4697#if 0
4698	struct ctl_port *port;
4699#endif
4700	struct ctl_lun *nlun;
4701	int i;
4702
4703	softc = lun->ctl_softc;
4704
4705	mtx_assert(&softc->ctl_lock, MA_OWNED);
4706
4707	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4708
4709	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4710
4711	softc->ctl_luns[lun->lun] = NULL;
4712
4713	if (!TAILQ_EMPTY(&lun->ooa_queue))
4714		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4715
4716	softc->num_luns--;
4717
4718	/*
4719	 * XXX KDM this scheme only works for a single target/multiple LUN
4720	 * setup.  It needs to be revamped for a multiple target scheme.
4721	 *
4722	 * XXX KDM this results in port->lun_disable() getting called twice,
4723	 * once when ctl_disable_lun() is called, and a second time here.
4724	 * We really need to re-think the LUN disable semantics.  There
4725	 * should probably be several steps/levels to LUN removal:
4726	 *  - disable
4727	 *  - invalidate
4728	 *  - free
4729 	 *
4730	 * Right now we only have a disable method when communicating to
4731	 * the front end ports, at least for individual LUNs.
4732	 */
4733#if 0
4734	STAILQ_FOREACH(port, &softc->port_list, links) {
4735		int retval;
4736
4737		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4738					 lun->lun);
4739		if (retval != 0) {
4740			printf("ctl_free_lun: FETD %s port %d returned error "
4741			       "%d for lun_disable on target %ju lun %jd\n",
4742			       port->port_name, port->targ_port, retval,
4743			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4744		}
4745
4746		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4747			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4748
4749			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4750			if (retval != 0) {
4751				printf("ctl_free_lun: FETD %s port %d "
4752				       "returned error %d for targ_disable on "
4753				       "target %ju\n", port->port_name,
4754				       port->targ_port, retval,
4755				       (uintmax_t)lun->target.id);
4756			} else
4757				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4758
4759			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4760				continue;
4761
4762#if 0
4763			port->port_offline(port->onoff_arg);
4764			port->status &= ~CTL_PORT_STATUS_ONLINE;
4765#endif
4766		}
4767	}
4768#endif
4769
4770	/*
4771	 * Tell the backend to free resources, if this LUN has a backend.
4772	 */
4773	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4774	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4775
4776	ctl_tpc_lun_shutdown(lun);
4777	mtx_destroy(&lun->lun_lock);
4778	free(lun->lun_devid, M_CTL);
4779	if (lun->flags & CTL_LUN_MALLOCED)
4780		free(lun, M_CTL);
4781
4782	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4783		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4784			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4785		}
4786	}
4787
4788	return (0);
4789}
4790
4791static void
4792ctl_create_lun(struct ctl_be_lun *be_lun)
4793{
4794	struct ctl_softc *ctl_softc;
4795
4796	ctl_softc = control_softc;
4797
4798	/*
4799	 * ctl_alloc_lun() should handle all potential failure cases.
4800	 */
4801	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4802}
4803
4804int
4805ctl_add_lun(struct ctl_be_lun *be_lun)
4806{
4807	struct ctl_softc *ctl_softc = control_softc;
4808
4809	mtx_lock(&ctl_softc->ctl_lock);
4810	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4811	mtx_unlock(&ctl_softc->ctl_lock);
4812	wakeup(&ctl_softc->pending_lun_queue);
4813
4814	return (0);
4815}
4816
4817int
4818ctl_enable_lun(struct ctl_be_lun *be_lun)
4819{
4820	struct ctl_softc *ctl_softc;
4821	struct ctl_port *port, *nport;
4822	struct ctl_lun *lun;
4823	int retval;
4824
4825	ctl_softc = control_softc;
4826
4827	lun = (struct ctl_lun *)be_lun->ctl_lun;
4828
4829	mtx_lock(&ctl_softc->ctl_lock);
4830	mtx_lock(&lun->lun_lock);
4831	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4832		/*
4833		 * eh?  Why did we get called if the LUN is already
4834		 * enabled?
4835		 */
4836		mtx_unlock(&lun->lun_lock);
4837		mtx_unlock(&ctl_softc->ctl_lock);
4838		return (0);
4839	}
4840	lun->flags &= ~CTL_LUN_DISABLED;
4841	mtx_unlock(&lun->lun_lock);
4842
4843	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4844		nport = STAILQ_NEXT(port, links);
4845
4846		/*
4847		 * Drop the lock while we call the FETD's enable routine.
4848		 * This can lead to a callback into CTL (at least in the
4849		 * case of the internal initiator frontend.
4850		 */
4851		mtx_unlock(&ctl_softc->ctl_lock);
4852		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4853		mtx_lock(&ctl_softc->ctl_lock);
4854		if (retval != 0) {
4855			printf("%s: FETD %s port %d returned error "
4856			       "%d for lun_enable on target %ju lun %jd\n",
4857			       __func__, port->port_name, port->targ_port, retval,
4858			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4859		}
4860#if 0
4861		 else {
4862            /* NOTE:  TODO:  why does lun enable affect port status? */
4863			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4864		}
4865#endif
4866	}
4867
4868	mtx_unlock(&ctl_softc->ctl_lock);
4869
4870	return (0);
4871}
4872
4873int
4874ctl_disable_lun(struct ctl_be_lun *be_lun)
4875{
4876	struct ctl_softc *ctl_softc;
4877	struct ctl_port *port;
4878	struct ctl_lun *lun;
4879	int retval;
4880
4881	ctl_softc = control_softc;
4882
4883	lun = (struct ctl_lun *)be_lun->ctl_lun;
4884
4885	mtx_lock(&ctl_softc->ctl_lock);
4886	mtx_lock(&lun->lun_lock);
4887	if (lun->flags & CTL_LUN_DISABLED) {
4888		mtx_unlock(&lun->lun_lock);
4889		mtx_unlock(&ctl_softc->ctl_lock);
4890		return (0);
4891	}
4892	lun->flags |= CTL_LUN_DISABLED;
4893	mtx_unlock(&lun->lun_lock);
4894
4895	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4896		mtx_unlock(&ctl_softc->ctl_lock);
4897		/*
4898		 * Drop the lock before we call the frontend's disable
4899		 * routine, to avoid lock order reversals.
4900		 *
4901		 * XXX KDM what happens if the frontend list changes while
4902		 * we're traversing it?  It's unlikely, but should be handled.
4903		 */
4904		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4905					 lun->lun);
4906		mtx_lock(&ctl_softc->ctl_lock);
4907		if (retval != 0) {
4908			printf("ctl_alloc_lun: FETD %s port %d returned error "
4909			       "%d for lun_disable on target %ju lun %jd\n",
4910			       port->port_name, port->targ_port, retval,
4911			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4912		}
4913	}
4914
4915	mtx_unlock(&ctl_softc->ctl_lock);
4916
4917	return (0);
4918}
4919
4920int
4921ctl_start_lun(struct ctl_be_lun *be_lun)
4922{
4923	struct ctl_softc *ctl_softc;
4924	struct ctl_lun *lun;
4925
4926	ctl_softc = control_softc;
4927
4928	lun = (struct ctl_lun *)be_lun->ctl_lun;
4929
4930	mtx_lock(&lun->lun_lock);
4931	lun->flags &= ~CTL_LUN_STOPPED;
4932	mtx_unlock(&lun->lun_lock);
4933
4934	return (0);
4935}
4936
4937int
4938ctl_stop_lun(struct ctl_be_lun *be_lun)
4939{
4940	struct ctl_softc *ctl_softc;
4941	struct ctl_lun *lun;
4942
4943	ctl_softc = control_softc;
4944
4945	lun = (struct ctl_lun *)be_lun->ctl_lun;
4946
4947	mtx_lock(&lun->lun_lock);
4948	lun->flags |= CTL_LUN_STOPPED;
4949	mtx_unlock(&lun->lun_lock);
4950
4951	return (0);
4952}
4953
4954int
4955ctl_lun_offline(struct ctl_be_lun *be_lun)
4956{
4957	struct ctl_softc *ctl_softc;
4958	struct ctl_lun *lun;
4959
4960	ctl_softc = control_softc;
4961
4962	lun = (struct ctl_lun *)be_lun->ctl_lun;
4963
4964	mtx_lock(&lun->lun_lock);
4965	lun->flags |= CTL_LUN_OFFLINE;
4966	mtx_unlock(&lun->lun_lock);
4967
4968	return (0);
4969}
4970
4971int
4972ctl_lun_online(struct ctl_be_lun *be_lun)
4973{
4974	struct ctl_softc *ctl_softc;
4975	struct ctl_lun *lun;
4976
4977	ctl_softc = control_softc;
4978
4979	lun = (struct ctl_lun *)be_lun->ctl_lun;
4980
4981	mtx_lock(&lun->lun_lock);
4982	lun->flags &= ~CTL_LUN_OFFLINE;
4983	mtx_unlock(&lun->lun_lock);
4984
4985	return (0);
4986}
4987
4988int
4989ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4990{
4991	struct ctl_softc *ctl_softc;
4992	struct ctl_lun *lun;
4993
4994	ctl_softc = control_softc;
4995
4996	lun = (struct ctl_lun *)be_lun->ctl_lun;
4997
4998	mtx_lock(&lun->lun_lock);
4999
5000	/*
5001	 * The LUN needs to be disabled before it can be marked invalid.
5002	 */
5003	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
5004		mtx_unlock(&lun->lun_lock);
5005		return (-1);
5006	}
5007	/*
5008	 * Mark the LUN invalid.
5009	 */
5010	lun->flags |= CTL_LUN_INVALID;
5011
5012	/*
5013	 * If there is nothing in the OOA queue, go ahead and free the LUN.
5014	 * If we have something in the OOA queue, we'll free it when the
5015	 * last I/O completes.
5016	 */
5017	if (TAILQ_EMPTY(&lun->ooa_queue)) {
5018		mtx_unlock(&lun->lun_lock);
5019		mtx_lock(&ctl_softc->ctl_lock);
5020		ctl_free_lun(lun);
5021		mtx_unlock(&ctl_softc->ctl_lock);
5022	} else
5023		mtx_unlock(&lun->lun_lock);
5024
5025	return (0);
5026}
5027
5028int
5029ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5030{
5031	struct ctl_softc *ctl_softc;
5032	struct ctl_lun *lun;
5033
5034	ctl_softc = control_softc;
5035	lun = (struct ctl_lun *)be_lun->ctl_lun;
5036
5037	mtx_lock(&lun->lun_lock);
5038	lun->flags |= CTL_LUN_INOPERABLE;
5039	mtx_unlock(&lun->lun_lock);
5040
5041	return (0);
5042}
5043
5044int
5045ctl_lun_operable(struct ctl_be_lun *be_lun)
5046{
5047	struct ctl_softc *ctl_softc;
5048	struct ctl_lun *lun;
5049
5050	ctl_softc = control_softc;
5051	lun = (struct ctl_lun *)be_lun->ctl_lun;
5052
5053	mtx_lock(&lun->lun_lock);
5054	lun->flags &= ~CTL_LUN_INOPERABLE;
5055	mtx_unlock(&lun->lun_lock);
5056
5057	return (0);
5058}
5059
5060int
5061ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5062		   int lock)
5063{
5064	struct ctl_softc *softc;
5065	struct ctl_lun *lun;
5066	struct copan_aps_subpage *current_sp;
5067	struct ctl_page_index *page_index;
5068	int i;
5069
5070	softc = control_softc;
5071
5072	mtx_lock(&softc->ctl_lock);
5073
5074	lun = (struct ctl_lun *)be_lun->ctl_lun;
5075	mtx_lock(&lun->lun_lock);
5076
5077	page_index = NULL;
5078	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5079		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5080		     APS_PAGE_CODE)
5081			continue;
5082
5083		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5084			continue;
5085		page_index = &lun->mode_pages.index[i];
5086	}
5087
5088	if (page_index == NULL) {
5089		mtx_unlock(&lun->lun_lock);
5090		mtx_unlock(&softc->ctl_lock);
5091		printf("%s: APS subpage not found for lun %ju!\n", __func__,
5092		       (uintmax_t)lun->lun);
5093		return (1);
5094	}
5095#if 0
5096	if ((softc->aps_locked_lun != 0)
5097	 && (softc->aps_locked_lun != lun->lun)) {
5098		printf("%s: attempt to lock LUN %llu when %llu is already "
5099		       "locked\n");
5100		mtx_unlock(&lun->lun_lock);
5101		mtx_unlock(&softc->ctl_lock);
5102		return (1);
5103	}
5104#endif
5105
5106	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5107		(page_index->page_len * CTL_PAGE_CURRENT));
5108
5109	if (lock != 0) {
5110		current_sp->lock_active = APS_LOCK_ACTIVE;
5111		softc->aps_locked_lun = lun->lun;
5112	} else {
5113		current_sp->lock_active = 0;
5114		softc->aps_locked_lun = 0;
5115	}
5116
5117
5118	/*
5119	 * If we're in HA mode, try to send the lock message to the other
5120	 * side.
5121	 */
5122	if (ctl_is_single == 0) {
5123		int isc_retval;
5124		union ctl_ha_msg lock_msg;
5125
5126		lock_msg.hdr.nexus = *nexus;
5127		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5128		if (lock != 0)
5129			lock_msg.aps.lock_flag = 1;
5130		else
5131			lock_msg.aps.lock_flag = 0;
5132		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5133					 sizeof(lock_msg), 0);
5134		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5135			printf("%s: APS (lock=%d) error returned from "
5136			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5137			mtx_unlock(&lun->lun_lock);
5138			mtx_unlock(&softc->ctl_lock);
5139			return (1);
5140		}
5141	}
5142
5143	mtx_unlock(&lun->lun_lock);
5144	mtx_unlock(&softc->ctl_lock);
5145
5146	return (0);
5147}
5148
5149void
5150ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5151{
5152	struct ctl_lun *lun;
5153	struct ctl_softc *softc;
5154	int i;
5155
5156	softc = control_softc;
5157
5158	lun = (struct ctl_lun *)be_lun->ctl_lun;
5159
5160	mtx_lock(&lun->lun_lock);
5161
5162	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5163		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5164
5165	mtx_unlock(&lun->lun_lock);
5166}
5167
5168/*
5169 * Backend "memory move is complete" callback for requests that never
5170 * make it down to say RAIDCore's configuration code.
5171 */
5172int
5173ctl_config_move_done(union ctl_io *io)
5174{
5175	int retval;
5176
5177	retval = CTL_RETVAL_COMPLETE;
5178
5179
5180	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5181	/*
5182	 * XXX KDM this shouldn't happen, but what if it does?
5183	 */
5184	if (io->io_hdr.io_type != CTL_IO_SCSI)
5185		panic("I/O type isn't CTL_IO_SCSI!");
5186
5187	if ((io->io_hdr.port_status == 0)
5188	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5189	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5190		io->io_hdr.status = CTL_SUCCESS;
5191	else if ((io->io_hdr.port_status != 0)
5192	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5193	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5194		/*
5195		 * For hardware error sense keys, the sense key
5196		 * specific value is defined to be a retry count,
5197		 * but we use it to pass back an internal FETD
5198		 * error code.  XXX KDM  Hopefully the FETD is only
5199		 * using 16 bits for an error code, since that's
5200		 * all the space we have in the sks field.
5201		 */
5202		ctl_set_internal_failure(&io->scsiio,
5203					 /*sks_valid*/ 1,
5204					 /*retry_count*/
5205					 io->io_hdr.port_status);
5206		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5207			free(io->scsiio.kern_data_ptr, M_CTL);
5208		ctl_done(io);
5209		goto bailout;
5210	}
5211
5212	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5213	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5214	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5215		/*
5216		 * XXX KDM just assuming a single pointer here, and not a
5217		 * S/G list.  If we start using S/G lists for config data,
5218		 * we'll need to know how to clean them up here as well.
5219		 */
5220		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5221			free(io->scsiio.kern_data_ptr, M_CTL);
5222		/* Hopefully the user has already set the status... */
5223		ctl_done(io);
5224	} else {
5225		/*
5226		 * XXX KDM now we need to continue data movement.  Some
5227		 * options:
5228		 * - call ctl_scsiio() again?  We don't do this for data
5229		 *   writes, because for those at least we know ahead of
5230		 *   time where the write will go and how long it is.  For
5231		 *   config writes, though, that information is largely
5232		 *   contained within the write itself, thus we need to
5233		 *   parse out the data again.
5234		 *
5235		 * - Call some other function once the data is in?
5236		 */
5237
5238		/*
5239		 * XXX KDM call ctl_scsiio() again for now, and check flag
5240		 * bits to see whether we're allocated or not.
5241		 */
5242		retval = ctl_scsiio(&io->scsiio);
5243	}
5244bailout:
5245	return (retval);
5246}
5247
5248/*
5249 * This gets called by a backend driver when it is done with a
5250 * data_submit method.
5251 */
5252void
5253ctl_data_submit_done(union ctl_io *io)
5254{
5255	/*
5256	 * If the IO_CONT flag is set, we need to call the supplied
5257	 * function to continue processing the I/O, instead of completing
5258	 * the I/O just yet.
5259	 *
5260	 * If there is an error, though, we don't want to keep processing.
5261	 * Instead, just send status back to the initiator.
5262	 */
5263	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5264	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5265	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5266	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5267		io->scsiio.io_cont(io);
5268		return;
5269	}
5270	ctl_done(io);
5271}
5272
5273/*
5274 * This gets called by a backend driver when it is done with a
5275 * configuration write.
5276 */
5277void
5278ctl_config_write_done(union ctl_io *io)
5279{
5280	uint8_t *buf;
5281
5282	/*
5283	 * If the IO_CONT flag is set, we need to call the supplied
5284	 * function to continue processing the I/O, instead of completing
5285	 * the I/O just yet.
5286	 *
5287	 * If there is an error, though, we don't want to keep processing.
5288	 * Instead, just send status back to the initiator.
5289	 */
5290	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5291	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5292	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5293	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5294		io->scsiio.io_cont(io);
5295		return;
5296	}
5297	/*
5298	 * Since a configuration write can be done for commands that actually
5299	 * have data allocated, like write buffer, and commands that have
5300	 * no data, like start/stop unit, we need to check here.
5301	 */
5302	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5303		buf = io->scsiio.kern_data_ptr;
5304	else
5305		buf = NULL;
5306	ctl_done(io);
5307	if (buf)
5308		free(buf, M_CTL);
5309}
5310
5311/*
5312 * SCSI release command.
5313 */
5314int
5315ctl_scsi_release(struct ctl_scsiio *ctsio)
5316{
5317	int length, longid, thirdparty_id, resv_id;
5318	struct ctl_softc *ctl_softc;
5319	struct ctl_lun *lun;
5320	uint32_t residx;
5321
5322	length = 0;
5323	resv_id = 0;
5324
5325	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5326
5327	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5328	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5329	ctl_softc = control_softc;
5330
5331	switch (ctsio->cdb[0]) {
5332	case RELEASE_10: {
5333		struct scsi_release_10 *cdb;
5334
5335		cdb = (struct scsi_release_10 *)ctsio->cdb;
5336
5337		if (cdb->byte2 & SR10_LONGID)
5338			longid = 1;
5339		else
5340			thirdparty_id = cdb->thirdparty_id;
5341
5342		resv_id = cdb->resv_id;
5343		length = scsi_2btoul(cdb->length);
5344		break;
5345	}
5346	}
5347
5348
5349	/*
5350	 * XXX KDM right now, we only support LUN reservation.  We don't
5351	 * support 3rd party reservations, or extent reservations, which
5352	 * might actually need the parameter list.  If we've gotten this
5353	 * far, we've got a LUN reservation.  Anything else got kicked out
5354	 * above.  So, according to SPC, ignore the length.
5355	 */
5356	length = 0;
5357
5358	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5359	 && (length > 0)) {
5360		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5361		ctsio->kern_data_len = length;
5362		ctsio->kern_total_len = length;
5363		ctsio->kern_data_resid = 0;
5364		ctsio->kern_rel_offset = 0;
5365		ctsio->kern_sg_entries = 0;
5366		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5367		ctsio->be_move_done = ctl_config_move_done;
5368		ctl_datamove((union ctl_io *)ctsio);
5369
5370		return (CTL_RETVAL_COMPLETE);
5371	}
5372
5373	if (length > 0)
5374		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5375
5376	mtx_lock(&lun->lun_lock);
5377
5378	/*
5379	 * According to SPC, it is not an error for an intiator to attempt
5380	 * to release a reservation on a LUN that isn't reserved, or that
5381	 * is reserved by another initiator.  The reservation can only be
5382	 * released, though, by the initiator who made it or by one of
5383	 * several reset type events.
5384	 */
5385	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5386			lun->flags &= ~CTL_LUN_RESERVED;
5387
5388	mtx_unlock(&lun->lun_lock);
5389
5390	ctsio->scsi_status = SCSI_STATUS_OK;
5391	ctsio->io_hdr.status = CTL_SUCCESS;
5392
5393	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5394		free(ctsio->kern_data_ptr, M_CTL);
5395		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5396	}
5397
5398	ctl_done((union ctl_io *)ctsio);
5399	return (CTL_RETVAL_COMPLETE);
5400}
5401
5402int
5403ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5404{
5405	int extent, thirdparty, longid;
5406	int resv_id, length;
5407	uint64_t thirdparty_id;
5408	struct ctl_softc *ctl_softc;
5409	struct ctl_lun *lun;
5410	uint32_t residx;
5411
5412	extent = 0;
5413	thirdparty = 0;
5414	longid = 0;
5415	resv_id = 0;
5416	length = 0;
5417	thirdparty_id = 0;
5418
5419	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5420
5421	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5422	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5423	ctl_softc = control_softc;
5424
5425	switch (ctsio->cdb[0]) {
5426	case RESERVE_10: {
5427		struct scsi_reserve_10 *cdb;
5428
5429		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5430
5431		if (cdb->byte2 & SR10_LONGID)
5432			longid = 1;
5433		else
5434			thirdparty_id = cdb->thirdparty_id;
5435
5436		resv_id = cdb->resv_id;
5437		length = scsi_2btoul(cdb->length);
5438		break;
5439	}
5440	}
5441
5442	/*
5443	 * XXX KDM right now, we only support LUN reservation.  We don't
5444	 * support 3rd party reservations, or extent reservations, which
5445	 * might actually need the parameter list.  If we've gotten this
5446	 * far, we've got a LUN reservation.  Anything else got kicked out
5447	 * above.  So, according to SPC, ignore the length.
5448	 */
5449	length = 0;
5450
5451	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5452	 && (length > 0)) {
5453		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5454		ctsio->kern_data_len = length;
5455		ctsio->kern_total_len = length;
5456		ctsio->kern_data_resid = 0;
5457		ctsio->kern_rel_offset = 0;
5458		ctsio->kern_sg_entries = 0;
5459		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5460		ctsio->be_move_done = ctl_config_move_done;
5461		ctl_datamove((union ctl_io *)ctsio);
5462
5463		return (CTL_RETVAL_COMPLETE);
5464	}
5465
5466	if (length > 0)
5467		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5468
5469	mtx_lock(&lun->lun_lock);
5470	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5471		ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5472		ctsio->io_hdr.status = CTL_SCSI_ERROR;
5473		goto bailout;
5474	}
5475
5476	lun->flags |= CTL_LUN_RESERVED;
5477	lun->res_idx = residx;
5478
5479	ctsio->scsi_status = SCSI_STATUS_OK;
5480	ctsio->io_hdr.status = CTL_SUCCESS;
5481
5482bailout:
5483	mtx_unlock(&lun->lun_lock);
5484
5485	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5486		free(ctsio->kern_data_ptr, M_CTL);
5487		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5488	}
5489
5490	ctl_done((union ctl_io *)ctsio);
5491	return (CTL_RETVAL_COMPLETE);
5492}
5493
5494int
5495ctl_start_stop(struct ctl_scsiio *ctsio)
5496{
5497	struct scsi_start_stop_unit *cdb;
5498	struct ctl_lun *lun;
5499	struct ctl_softc *ctl_softc;
5500	int retval;
5501
5502	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5503
5504	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5505	ctl_softc = control_softc;
5506	retval = 0;
5507
5508	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5509
5510	/*
5511	 * XXX KDM
5512	 * We don't support the immediate bit on a stop unit.  In order to
5513	 * do that, we would need to code up a way to know that a stop is
5514	 * pending, and hold off any new commands until it completes, one
5515	 * way or another.  Then we could accept or reject those commands
5516	 * depending on its status.  We would almost need to do the reverse
5517	 * of what we do below for an immediate start -- return the copy of
5518	 * the ctl_io to the FETD with status to send to the host (and to
5519	 * free the copy!) and then free the original I/O once the stop
5520	 * actually completes.  That way, the OOA queue mechanism can work
5521	 * to block commands that shouldn't proceed.  Another alternative
5522	 * would be to put the copy in the queue in place of the original,
5523	 * and return the original back to the caller.  That could be
5524	 * slightly safer..
5525	 */
5526	if ((cdb->byte2 & SSS_IMMED)
5527	 && ((cdb->how & SSS_START) == 0)) {
5528		ctl_set_invalid_field(ctsio,
5529				      /*sks_valid*/ 1,
5530				      /*command*/ 1,
5531				      /*field*/ 1,
5532				      /*bit_valid*/ 1,
5533				      /*bit*/ 0);
5534		ctl_done((union ctl_io *)ctsio);
5535		return (CTL_RETVAL_COMPLETE);
5536	}
5537
5538	if ((lun->flags & CTL_LUN_PR_RESERVED)
5539	 && ((cdb->how & SSS_START)==0)) {
5540		uint32_t residx;
5541
5542		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5543		if (lun->pr_keys[residx] == 0
5544		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5545
5546			ctl_set_reservation_conflict(ctsio);
5547			ctl_done((union ctl_io *)ctsio);
5548			return (CTL_RETVAL_COMPLETE);
5549		}
5550	}
5551
5552	/*
5553	 * If there is no backend on this device, we can't start or stop
5554	 * it.  In theory we shouldn't get any start/stop commands in the
5555	 * first place at this level if the LUN doesn't have a backend.
5556	 * That should get stopped by the command decode code.
5557	 */
5558	if (lun->backend == NULL) {
5559		ctl_set_invalid_opcode(ctsio);
5560		ctl_done((union ctl_io *)ctsio);
5561		return (CTL_RETVAL_COMPLETE);
5562	}
5563
5564	/*
5565	 * XXX KDM Copan-specific offline behavior.
5566	 * Figure out a reasonable way to port this?
5567	 */
5568#ifdef NEEDTOPORT
5569	mtx_lock(&lun->lun_lock);
5570
5571	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5572	 && (lun->flags & CTL_LUN_OFFLINE)) {
5573		/*
5574		 * If the LUN is offline, and the on/offline bit isn't set,
5575		 * reject the start or stop.  Otherwise, let it through.
5576		 */
5577		mtx_unlock(&lun->lun_lock);
5578		ctl_set_lun_not_ready(ctsio);
5579		ctl_done((union ctl_io *)ctsio);
5580	} else {
5581		mtx_unlock(&lun->lun_lock);
5582#endif /* NEEDTOPORT */
5583		/*
5584		 * This could be a start or a stop when we're online,
5585		 * or a stop/offline or start/online.  A start or stop when
5586		 * we're offline is covered in the case above.
5587		 */
5588		/*
5589		 * In the non-immediate case, we send the request to
5590		 * the backend and return status to the user when
5591		 * it is done.
5592		 *
5593		 * In the immediate case, we allocate a new ctl_io
5594		 * to hold a copy of the request, and send that to
5595		 * the backend.  We then set good status on the
5596		 * user's request and return it immediately.
5597		 */
5598		if (cdb->byte2 & SSS_IMMED) {
5599			union ctl_io *new_io;
5600
5601			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5602			if (new_io == NULL) {
5603				ctl_set_busy(ctsio);
5604				ctl_done((union ctl_io *)ctsio);
5605			} else {
5606				ctl_copy_io((union ctl_io *)ctsio,
5607					    new_io);
5608				retval = lun->backend->config_write(new_io);
5609				ctl_set_success(ctsio);
5610				ctl_done((union ctl_io *)ctsio);
5611			}
5612		} else {
5613			retval = lun->backend->config_write(
5614				(union ctl_io *)ctsio);
5615		}
5616#ifdef NEEDTOPORT
5617	}
5618#endif
5619	return (retval);
5620}
5621
5622/*
5623 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5624 * we don't really do anything with the LBA and length fields if the user
5625 * passes them in.  Instead we'll just flush out the cache for the entire
5626 * LUN.
5627 */
5628int
5629ctl_sync_cache(struct ctl_scsiio *ctsio)
5630{
5631	struct ctl_lun *lun;
5632	struct ctl_softc *ctl_softc;
5633	uint64_t starting_lba;
5634	uint32_t block_count;
5635	int retval;
5636
5637	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5638
5639	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5640	ctl_softc = control_softc;
5641	retval = 0;
5642
5643	switch (ctsio->cdb[0]) {
5644	case SYNCHRONIZE_CACHE: {
5645		struct scsi_sync_cache *cdb;
5646		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5647
5648		starting_lba = scsi_4btoul(cdb->begin_lba);
5649		block_count = scsi_2btoul(cdb->lb_count);
5650		break;
5651	}
5652	case SYNCHRONIZE_CACHE_16: {
5653		struct scsi_sync_cache_16 *cdb;
5654		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5655
5656		starting_lba = scsi_8btou64(cdb->begin_lba);
5657		block_count = scsi_4btoul(cdb->lb_count);
5658		break;
5659	}
5660	default:
5661		ctl_set_invalid_opcode(ctsio);
5662		ctl_done((union ctl_io *)ctsio);
5663		goto bailout;
5664		break; /* NOTREACHED */
5665	}
5666
5667	/*
5668	 * We check the LBA and length, but don't do anything with them.
5669	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5670	 * get flushed.  This check will just help satisfy anyone who wants
5671	 * to see an error for an out of range LBA.
5672	 */
5673	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5674		ctl_set_lba_out_of_range(ctsio);
5675		ctl_done((union ctl_io *)ctsio);
5676		goto bailout;
5677	}
5678
5679	/*
5680	 * If this LUN has no backend, we can't flush the cache anyway.
5681	 */
5682	if (lun->backend == NULL) {
5683		ctl_set_invalid_opcode(ctsio);
5684		ctl_done((union ctl_io *)ctsio);
5685		goto bailout;
5686	}
5687
5688	/*
5689	 * Check to see whether we're configured to send the SYNCHRONIZE
5690	 * CACHE command directly to the back end.
5691	 */
5692	mtx_lock(&lun->lun_lock);
5693	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5694	 && (++(lun->sync_count) >= lun->sync_interval)) {
5695		lun->sync_count = 0;
5696		mtx_unlock(&lun->lun_lock);
5697		retval = lun->backend->config_write((union ctl_io *)ctsio);
5698	} else {
5699		mtx_unlock(&lun->lun_lock);
5700		ctl_set_success(ctsio);
5701		ctl_done((union ctl_io *)ctsio);
5702	}
5703
5704bailout:
5705
5706	return (retval);
5707}
5708
5709int
5710ctl_format(struct ctl_scsiio *ctsio)
5711{
5712	struct scsi_format *cdb;
5713	struct ctl_lun *lun;
5714	struct ctl_softc *ctl_softc;
5715	int length, defect_list_len;
5716
5717	CTL_DEBUG_PRINT(("ctl_format\n"));
5718
5719	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5720	ctl_softc = control_softc;
5721
5722	cdb = (struct scsi_format *)ctsio->cdb;
5723
5724	length = 0;
5725	if (cdb->byte2 & SF_FMTDATA) {
5726		if (cdb->byte2 & SF_LONGLIST)
5727			length = sizeof(struct scsi_format_header_long);
5728		else
5729			length = sizeof(struct scsi_format_header_short);
5730	}
5731
5732	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5733	 && (length > 0)) {
5734		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5735		ctsio->kern_data_len = length;
5736		ctsio->kern_total_len = length;
5737		ctsio->kern_data_resid = 0;
5738		ctsio->kern_rel_offset = 0;
5739		ctsio->kern_sg_entries = 0;
5740		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5741		ctsio->be_move_done = ctl_config_move_done;
5742		ctl_datamove((union ctl_io *)ctsio);
5743
5744		return (CTL_RETVAL_COMPLETE);
5745	}
5746
5747	defect_list_len = 0;
5748
5749	if (cdb->byte2 & SF_FMTDATA) {
5750		if (cdb->byte2 & SF_LONGLIST) {
5751			struct scsi_format_header_long *header;
5752
5753			header = (struct scsi_format_header_long *)
5754				ctsio->kern_data_ptr;
5755
5756			defect_list_len = scsi_4btoul(header->defect_list_len);
5757			if (defect_list_len != 0) {
5758				ctl_set_invalid_field(ctsio,
5759						      /*sks_valid*/ 1,
5760						      /*command*/ 0,
5761						      /*field*/ 2,
5762						      /*bit_valid*/ 0,
5763						      /*bit*/ 0);
5764				goto bailout;
5765			}
5766		} else {
5767			struct scsi_format_header_short *header;
5768
5769			header = (struct scsi_format_header_short *)
5770				ctsio->kern_data_ptr;
5771
5772			defect_list_len = scsi_2btoul(header->defect_list_len);
5773			if (defect_list_len != 0) {
5774				ctl_set_invalid_field(ctsio,
5775						      /*sks_valid*/ 1,
5776						      /*command*/ 0,
5777						      /*field*/ 2,
5778						      /*bit_valid*/ 0,
5779						      /*bit*/ 0);
5780				goto bailout;
5781			}
5782		}
5783	}
5784
5785	/*
5786	 * The format command will clear out the "Medium format corrupted"
5787	 * status if set by the configuration code.  That status is really
5788	 * just a way to notify the host that we have lost the media, and
5789	 * get them to issue a command that will basically make them think
5790	 * they're blowing away the media.
5791	 */
5792	mtx_lock(&lun->lun_lock);
5793	lun->flags &= ~CTL_LUN_INOPERABLE;
5794	mtx_unlock(&lun->lun_lock);
5795
5796	ctsio->scsi_status = SCSI_STATUS_OK;
5797	ctsio->io_hdr.status = CTL_SUCCESS;
5798bailout:
5799
5800	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5801		free(ctsio->kern_data_ptr, M_CTL);
5802		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5803	}
5804
5805	ctl_done((union ctl_io *)ctsio);
5806	return (CTL_RETVAL_COMPLETE);
5807}
5808
5809int
5810ctl_read_buffer(struct ctl_scsiio *ctsio)
5811{
5812	struct scsi_read_buffer *cdb;
5813	struct ctl_lun *lun;
5814	int buffer_offset, len;
5815	static uint8_t descr[4];
5816	static uint8_t echo_descr[4] = { 0 };
5817
5818	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5819
5820	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5821	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5822
5823	if (lun->flags & CTL_LUN_PR_RESERVED) {
5824		uint32_t residx;
5825
5826		/*
5827		 * XXX KDM need a lock here.
5828		 */
5829		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5830		if ((lun->res_type == SPR_TYPE_EX_AC
5831		  && residx != lun->pr_res_idx)
5832		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5833		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5834		  && lun->pr_keys[residx] == 0)) {
5835			ctl_set_reservation_conflict(ctsio);
5836			ctl_done((union ctl_io *)ctsio);
5837			return (CTL_RETVAL_COMPLETE);
5838	        }
5839	}
5840
5841	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5842	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5843	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5844		ctl_set_invalid_field(ctsio,
5845				      /*sks_valid*/ 1,
5846				      /*command*/ 1,
5847				      /*field*/ 1,
5848				      /*bit_valid*/ 1,
5849				      /*bit*/ 4);
5850		ctl_done((union ctl_io *)ctsio);
5851		return (CTL_RETVAL_COMPLETE);
5852	}
5853
5854	len = scsi_3btoul(cdb->length);
5855	buffer_offset = scsi_3btoul(cdb->offset);
5856
5857	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5858		ctl_set_invalid_field(ctsio,
5859				      /*sks_valid*/ 1,
5860				      /*command*/ 1,
5861				      /*field*/ 6,
5862				      /*bit_valid*/ 0,
5863				      /*bit*/ 0);
5864		ctl_done((union ctl_io *)ctsio);
5865		return (CTL_RETVAL_COMPLETE);
5866	}
5867
5868	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5869		descr[0] = 0;
5870		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5871		ctsio->kern_data_ptr = descr;
5872		len = min(len, sizeof(descr));
5873	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5874		ctsio->kern_data_ptr = echo_descr;
5875		len = min(len, sizeof(echo_descr));
5876	} else
5877		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5878	ctsio->kern_data_len = len;
5879	ctsio->kern_total_len = len;
5880	ctsio->kern_data_resid = 0;
5881	ctsio->kern_rel_offset = 0;
5882	ctsio->kern_sg_entries = 0;
5883	ctsio->be_move_done = ctl_config_move_done;
5884	ctl_datamove((union ctl_io *)ctsio);
5885
5886	return (CTL_RETVAL_COMPLETE);
5887}
5888
5889int
5890ctl_write_buffer(struct ctl_scsiio *ctsio)
5891{
5892	struct scsi_write_buffer *cdb;
5893	struct ctl_lun *lun;
5894	int buffer_offset, len;
5895
5896	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5897
5898	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5899	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5900
5901	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5902		ctl_set_invalid_field(ctsio,
5903				      /*sks_valid*/ 1,
5904				      /*command*/ 1,
5905				      /*field*/ 1,
5906				      /*bit_valid*/ 1,
5907				      /*bit*/ 4);
5908		ctl_done((union ctl_io *)ctsio);
5909		return (CTL_RETVAL_COMPLETE);
5910	}
5911
5912	len = scsi_3btoul(cdb->length);
5913	buffer_offset = scsi_3btoul(cdb->offset);
5914
5915	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5916		ctl_set_invalid_field(ctsio,
5917				      /*sks_valid*/ 1,
5918				      /*command*/ 1,
5919				      /*field*/ 6,
5920				      /*bit_valid*/ 0,
5921				      /*bit*/ 0);
5922		ctl_done((union ctl_io *)ctsio);
5923		return (CTL_RETVAL_COMPLETE);
5924	}
5925
5926	/*
5927	 * If we've got a kernel request that hasn't been malloced yet,
5928	 * malloc it and tell the caller the data buffer is here.
5929	 */
5930	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5931		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5932		ctsio->kern_data_len = len;
5933		ctsio->kern_total_len = len;
5934		ctsio->kern_data_resid = 0;
5935		ctsio->kern_rel_offset = 0;
5936		ctsio->kern_sg_entries = 0;
5937		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5938		ctsio->be_move_done = ctl_config_move_done;
5939		ctl_datamove((union ctl_io *)ctsio);
5940
5941		return (CTL_RETVAL_COMPLETE);
5942	}
5943
5944	ctl_done((union ctl_io *)ctsio);
5945
5946	return (CTL_RETVAL_COMPLETE);
5947}
5948
5949int
5950ctl_write_same(struct ctl_scsiio *ctsio)
5951{
5952	struct ctl_lun *lun;
5953	struct ctl_lba_len_flags *lbalen;
5954	uint64_t lba;
5955	uint32_t num_blocks;
5956	int len, retval;
5957	uint8_t byte2;
5958
5959	retval = CTL_RETVAL_COMPLETE;
5960
5961	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5962
5963	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5964
5965	switch (ctsio->cdb[0]) {
5966	case WRITE_SAME_10: {
5967		struct scsi_write_same_10 *cdb;
5968
5969		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5970
5971		lba = scsi_4btoul(cdb->addr);
5972		num_blocks = scsi_2btoul(cdb->length);
5973		byte2 = cdb->byte2;
5974		break;
5975	}
5976	case WRITE_SAME_16: {
5977		struct scsi_write_same_16 *cdb;
5978
5979		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5980
5981		lba = scsi_8btou64(cdb->addr);
5982		num_blocks = scsi_4btoul(cdb->length);
5983		byte2 = cdb->byte2;
5984		break;
5985	}
5986	default:
5987		/*
5988		 * We got a command we don't support.  This shouldn't
5989		 * happen, commands should be filtered out above us.
5990		 */
5991		ctl_set_invalid_opcode(ctsio);
5992		ctl_done((union ctl_io *)ctsio);
5993
5994		return (CTL_RETVAL_COMPLETE);
5995		break; /* NOTREACHED */
5996	}
5997
5998	/* NDOB and ANCHOR flags can be used only together with UNMAP */
5999	if ((byte2 & SWS_UNMAP) == 0 &&
6000	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
6001		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
6002		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
6003		ctl_done((union ctl_io *)ctsio);
6004		return (CTL_RETVAL_COMPLETE);
6005	}
6006
6007	/*
6008	 * The first check is to make sure we're in bounds, the second
6009	 * check is to catch wrap-around problems.  If the lba + num blocks
6010	 * is less than the lba, then we've wrapped around and the block
6011	 * range is invalid anyway.
6012	 */
6013	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6014	 || ((lba + num_blocks) < lba)) {
6015		ctl_set_lba_out_of_range(ctsio);
6016		ctl_done((union ctl_io *)ctsio);
6017		return (CTL_RETVAL_COMPLETE);
6018	}
6019
6020	/* Zero number of blocks means "to the last logical block" */
6021	if (num_blocks == 0) {
6022		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
6023			ctl_set_invalid_field(ctsio,
6024					      /*sks_valid*/ 0,
6025					      /*command*/ 1,
6026					      /*field*/ 0,
6027					      /*bit_valid*/ 0,
6028					      /*bit*/ 0);
6029			ctl_done((union ctl_io *)ctsio);
6030			return (CTL_RETVAL_COMPLETE);
6031		}
6032		num_blocks = (lun->be_lun->maxlba + 1) - lba;
6033	}
6034
6035	len = lun->be_lun->blocksize;
6036
6037	/*
6038	 * If we've got a kernel request that hasn't been malloced yet,
6039	 * malloc it and tell the caller the data buffer is here.
6040	 */
6041	if ((byte2 & SWS_NDOB) == 0 &&
6042	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6043		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6044		ctsio->kern_data_len = len;
6045		ctsio->kern_total_len = len;
6046		ctsio->kern_data_resid = 0;
6047		ctsio->kern_rel_offset = 0;
6048		ctsio->kern_sg_entries = 0;
6049		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6050		ctsio->be_move_done = ctl_config_move_done;
6051		ctl_datamove((union ctl_io *)ctsio);
6052
6053		return (CTL_RETVAL_COMPLETE);
6054	}
6055
6056	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6057	lbalen->lba = lba;
6058	lbalen->len = num_blocks;
6059	lbalen->flags = byte2;
6060	retval = lun->backend->config_write((union ctl_io *)ctsio);
6061
6062	return (retval);
6063}
6064
6065int
6066ctl_unmap(struct ctl_scsiio *ctsio)
6067{
6068	struct ctl_lun *lun;
6069	struct scsi_unmap *cdb;
6070	struct ctl_ptr_len_flags *ptrlen;
6071	struct scsi_unmap_header *hdr;
6072	struct scsi_unmap_desc *buf, *end, *endnz, *range;
6073	uint64_t lba;
6074	uint32_t num_blocks;
6075	int len, retval;
6076	uint8_t byte2;
6077
6078	retval = CTL_RETVAL_COMPLETE;
6079
6080	CTL_DEBUG_PRINT(("ctl_unmap\n"));
6081
6082	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6083	cdb = (struct scsi_unmap *)ctsio->cdb;
6084
6085	len = scsi_2btoul(cdb->length);
6086	byte2 = cdb->byte2;
6087
6088	/*
6089	 * If we've got a kernel request that hasn't been malloced yet,
6090	 * malloc it and tell the caller the data buffer is here.
6091	 */
6092	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6093		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6094		ctsio->kern_data_len = len;
6095		ctsio->kern_total_len = len;
6096		ctsio->kern_data_resid = 0;
6097		ctsio->kern_rel_offset = 0;
6098		ctsio->kern_sg_entries = 0;
6099		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6100		ctsio->be_move_done = ctl_config_move_done;
6101		ctl_datamove((union ctl_io *)ctsio);
6102
6103		return (CTL_RETVAL_COMPLETE);
6104	}
6105
6106	len = ctsio->kern_total_len - ctsio->kern_data_resid;
6107	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6108	if (len < sizeof (*hdr) ||
6109	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6110	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6111	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6112		ctl_set_invalid_field(ctsio,
6113				      /*sks_valid*/ 0,
6114				      /*command*/ 0,
6115				      /*field*/ 0,
6116				      /*bit_valid*/ 0,
6117				      /*bit*/ 0);
6118		ctl_done((union ctl_io *)ctsio);
6119		return (CTL_RETVAL_COMPLETE);
6120	}
6121	len = scsi_2btoul(hdr->desc_length);
6122	buf = (struct scsi_unmap_desc *)(hdr + 1);
6123	end = buf + len / sizeof(*buf);
6124
6125	endnz = buf;
6126	for (range = buf; range < end; range++) {
6127		lba = scsi_8btou64(range->lba);
6128		num_blocks = scsi_4btoul(range->length);
6129		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6130		 || ((lba + num_blocks) < lba)) {
6131			ctl_set_lba_out_of_range(ctsio);
6132			ctl_done((union ctl_io *)ctsio);
6133			return (CTL_RETVAL_COMPLETE);
6134		}
6135		if (num_blocks != 0)
6136			endnz = range + 1;
6137	}
6138
6139	/*
6140	 * Block backend can not handle zero last range.
6141	 * Filter it out and return if there is nothing left.
6142	 */
6143	len = (uint8_t *)endnz - (uint8_t *)buf;
6144	if (len == 0) {
6145		ctl_set_success(ctsio);
6146		ctl_done((union ctl_io *)ctsio);
6147		return (CTL_RETVAL_COMPLETE);
6148	}
6149
6150	mtx_lock(&lun->lun_lock);
6151	ptrlen = (struct ctl_ptr_len_flags *)
6152	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6153	ptrlen->ptr = (void *)buf;
6154	ptrlen->len = len;
6155	ptrlen->flags = byte2;
6156	ctl_check_blocked(lun);
6157	mtx_unlock(&lun->lun_lock);
6158
6159	retval = lun->backend->config_write((union ctl_io *)ctsio);
6160	return (retval);
6161}
6162
6163/*
6164 * Note that this function currently doesn't actually do anything inside
6165 * CTL to enforce things if the DQue bit is turned on.
6166 *
6167 * Also note that this function can't be used in the default case, because
6168 * the DQue bit isn't set in the changeable mask for the control mode page
6169 * anyway.  This is just here as an example for how to implement a page
6170 * handler, and a placeholder in case we want to allow the user to turn
6171 * tagged queueing on and off.
6172 *
6173 * The D_SENSE bit handling is functional, however, and will turn
6174 * descriptor sense on and off for a given LUN.
6175 */
6176int
6177ctl_control_page_handler(struct ctl_scsiio *ctsio,
6178			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6179{
6180	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6181	struct ctl_lun *lun;
6182	struct ctl_softc *softc;
6183	int set_ua;
6184	uint32_t initidx;
6185
6186	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6187	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6188	set_ua = 0;
6189
6190	user_cp = (struct scsi_control_page *)page_ptr;
6191	current_cp = (struct scsi_control_page *)
6192		(page_index->page_data + (page_index->page_len *
6193		CTL_PAGE_CURRENT));
6194	saved_cp = (struct scsi_control_page *)
6195		(page_index->page_data + (page_index->page_len *
6196		CTL_PAGE_SAVED));
6197
6198	softc = control_softc;
6199
6200	mtx_lock(&lun->lun_lock);
6201	if (((current_cp->rlec & SCP_DSENSE) == 0)
6202	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6203		/*
6204		 * Descriptor sense is currently turned off and the user
6205		 * wants to turn it on.
6206		 */
6207		current_cp->rlec |= SCP_DSENSE;
6208		saved_cp->rlec |= SCP_DSENSE;
6209		lun->flags |= CTL_LUN_SENSE_DESC;
6210		set_ua = 1;
6211	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6212		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6213		/*
6214		 * Descriptor sense is currently turned on, and the user
6215		 * wants to turn it off.
6216		 */
6217		current_cp->rlec &= ~SCP_DSENSE;
6218		saved_cp->rlec &= ~SCP_DSENSE;
6219		lun->flags &= ~CTL_LUN_SENSE_DESC;
6220		set_ua = 1;
6221	}
6222	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6223	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6224		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6225		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6226		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6227		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6228		set_ua = 1;
6229	}
6230	if ((current_cp->eca_and_aen & SCP_SWP) !=
6231	    (user_cp->eca_and_aen & SCP_SWP)) {
6232		current_cp->eca_and_aen &= ~SCP_SWP;
6233		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6234		saved_cp->eca_and_aen &= ~SCP_SWP;
6235		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6236		set_ua = 1;
6237	}
6238	if (set_ua != 0) {
6239		int i;
6240		/*
6241		 * Let other initiators know that the mode
6242		 * parameters for this LUN have changed.
6243		 */
6244		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6245			if (i == initidx)
6246				continue;
6247
6248			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6249		}
6250	}
6251	mtx_unlock(&lun->lun_lock);
6252
6253	return (0);
6254}
6255
6256int
6257ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6258		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6259{
6260	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6261	struct ctl_lun *lun;
6262	int set_ua;
6263	uint32_t initidx;
6264
6265	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6266	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6267	set_ua = 0;
6268
6269	user_cp = (struct scsi_caching_page *)page_ptr;
6270	current_cp = (struct scsi_caching_page *)
6271		(page_index->page_data + (page_index->page_len *
6272		CTL_PAGE_CURRENT));
6273	saved_cp = (struct scsi_caching_page *)
6274		(page_index->page_data + (page_index->page_len *
6275		CTL_PAGE_SAVED));
6276
6277	mtx_lock(&lun->lun_lock);
6278	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6279	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6280		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6281		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6282		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6283		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6284		set_ua = 1;
6285	}
6286	if (set_ua != 0) {
6287		int i;
6288		/*
6289		 * Let other initiators know that the mode
6290		 * parameters for this LUN have changed.
6291		 */
6292		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6293			if (i == initidx)
6294				continue;
6295
6296			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6297		}
6298	}
6299	mtx_unlock(&lun->lun_lock);
6300
6301	return (0);
6302}
6303
6304int
6305ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6306		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6307{
6308	return (0);
6309}
6310
6311int
6312ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6313			   struct ctl_page_index *page_index, int pc)
6314{
6315	struct copan_power_subpage *page;
6316
6317	page = (struct copan_power_subpage *)page_index->page_data +
6318		(page_index->page_len * pc);
6319
6320	switch (pc) {
6321	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6322		/*
6323		 * We don't update the changable bits for this page.
6324		 */
6325		break;
6326	case SMS_PAGE_CTRL_CURRENT >> 6:
6327	case SMS_PAGE_CTRL_DEFAULT >> 6:
6328	case SMS_PAGE_CTRL_SAVED >> 6:
6329#ifdef NEEDTOPORT
6330		ctl_update_power_subpage(page);
6331#endif
6332		break;
6333	default:
6334#ifdef NEEDTOPORT
6335		EPRINT(0, "Invalid PC %d!!", pc);
6336#endif
6337		break;
6338	}
6339	return (0);
6340}
6341
6342
6343int
6344ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6345		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6346{
6347	struct copan_aps_subpage *user_sp;
6348	struct copan_aps_subpage *current_sp;
6349	union ctl_modepage_info *modepage_info;
6350	struct ctl_softc *softc;
6351	struct ctl_lun *lun;
6352	int retval;
6353
6354	retval = CTL_RETVAL_COMPLETE;
6355	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6356		     (page_index->page_len * CTL_PAGE_CURRENT));
6357	softc = control_softc;
6358	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6359
6360	user_sp = (struct copan_aps_subpage *)page_ptr;
6361
6362	modepage_info = (union ctl_modepage_info *)
6363		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6364
6365	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6366	modepage_info->header.subpage = page_index->subpage;
6367	modepage_info->aps.lock_active = user_sp->lock_active;
6368
6369	mtx_lock(&softc->ctl_lock);
6370
6371	/*
6372	 * If there is a request to lock the LUN and another LUN is locked
6373	 * this is an error. If the requested LUN is already locked ignore
6374	 * the request. If no LUN is locked attempt to lock it.
6375	 * if there is a request to unlock the LUN and the LUN is currently
6376	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6377	 * if another LUN is locked or no LUN is locked.
6378	 */
6379	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6380		if (softc->aps_locked_lun == lun->lun) {
6381			/*
6382			 * This LUN is already locked, so we're done.
6383			 */
6384			retval = CTL_RETVAL_COMPLETE;
6385		} else if (softc->aps_locked_lun == 0) {
6386			/*
6387			 * No one has the lock, pass the request to the
6388			 * backend.
6389			 */
6390			retval = lun->backend->config_write(
6391				(union ctl_io *)ctsio);
6392		} else {
6393			/*
6394			 * Someone else has the lock, throw out the request.
6395			 */
6396			ctl_set_already_locked(ctsio);
6397			free(ctsio->kern_data_ptr, M_CTL);
6398			ctl_done((union ctl_io *)ctsio);
6399
6400			/*
6401			 * Set the return value so that ctl_do_mode_select()
6402			 * won't try to complete the command.  We already
6403			 * completed it here.
6404			 */
6405			retval = CTL_RETVAL_ERROR;
6406		}
6407	} else if (softc->aps_locked_lun == lun->lun) {
6408		/*
6409		 * This LUN is locked, so pass the unlock request to the
6410		 * backend.
6411		 */
6412		retval = lun->backend->config_write((union ctl_io *)ctsio);
6413	}
6414	mtx_unlock(&softc->ctl_lock);
6415
6416	return (retval);
6417}
6418
6419int
6420ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6421				struct ctl_page_index *page_index,
6422				uint8_t *page_ptr)
6423{
6424	uint8_t *c;
6425	int i;
6426
6427	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6428	ctl_time_io_secs =
6429		(c[0] << 8) |
6430		(c[1] << 0) |
6431		0;
6432	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6433	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6434	printf("page data:");
6435	for (i=0; i<8; i++)
6436		printf(" %.2x",page_ptr[i]);
6437	printf("\n");
6438	return (0);
6439}
6440
6441int
6442ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6443			       struct ctl_page_index *page_index,
6444			       int pc)
6445{
6446	struct copan_debugconf_subpage *page;
6447
6448	page = (struct copan_debugconf_subpage *)page_index->page_data +
6449		(page_index->page_len * pc);
6450
6451	switch (pc) {
6452	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6453	case SMS_PAGE_CTRL_DEFAULT >> 6:
6454	case SMS_PAGE_CTRL_SAVED >> 6:
6455		/*
6456		 * We don't update the changable or default bits for this page.
6457		 */
6458		break;
6459	case SMS_PAGE_CTRL_CURRENT >> 6:
6460		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6461		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6462		break;
6463	default:
6464#ifdef NEEDTOPORT
6465		EPRINT(0, "Invalid PC %d!!", pc);
6466#endif /* NEEDTOPORT */
6467		break;
6468	}
6469	return (0);
6470}
6471
6472
6473static int
6474ctl_do_mode_select(union ctl_io *io)
6475{
6476	struct scsi_mode_page_header *page_header;
6477	struct ctl_page_index *page_index;
6478	struct ctl_scsiio *ctsio;
6479	int control_dev, page_len;
6480	int page_len_offset, page_len_size;
6481	union ctl_modepage_info *modepage_info;
6482	struct ctl_lun *lun;
6483	int *len_left, *len_used;
6484	int retval, i;
6485
6486	ctsio = &io->scsiio;
6487	page_index = NULL;
6488	page_len = 0;
6489	retval = CTL_RETVAL_COMPLETE;
6490
6491	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6492
6493	if (lun->be_lun->lun_type != T_DIRECT)
6494		control_dev = 1;
6495	else
6496		control_dev = 0;
6497
6498	modepage_info = (union ctl_modepage_info *)
6499		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6500	len_left = &modepage_info->header.len_left;
6501	len_used = &modepage_info->header.len_used;
6502
6503do_next_page:
6504
6505	page_header = (struct scsi_mode_page_header *)
6506		(ctsio->kern_data_ptr + *len_used);
6507
6508	if (*len_left == 0) {
6509		free(ctsio->kern_data_ptr, M_CTL);
6510		ctl_set_success(ctsio);
6511		ctl_done((union ctl_io *)ctsio);
6512		return (CTL_RETVAL_COMPLETE);
6513	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6514
6515		free(ctsio->kern_data_ptr, M_CTL);
6516		ctl_set_param_len_error(ctsio);
6517		ctl_done((union ctl_io *)ctsio);
6518		return (CTL_RETVAL_COMPLETE);
6519
6520	} else if ((page_header->page_code & SMPH_SPF)
6521		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6522
6523		free(ctsio->kern_data_ptr, M_CTL);
6524		ctl_set_param_len_error(ctsio);
6525		ctl_done((union ctl_io *)ctsio);
6526		return (CTL_RETVAL_COMPLETE);
6527	}
6528
6529
6530	/*
6531	 * XXX KDM should we do something with the block descriptor?
6532	 */
6533	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6534
6535		if ((control_dev != 0)
6536		 && (lun->mode_pages.index[i].page_flags &
6537		     CTL_PAGE_FLAG_DISK_ONLY))
6538			continue;
6539
6540		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6541		    (page_header->page_code & SMPH_PC_MASK))
6542			continue;
6543
6544		/*
6545		 * If neither page has a subpage code, then we've got a
6546		 * match.
6547		 */
6548		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6549		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6550			page_index = &lun->mode_pages.index[i];
6551			page_len = page_header->page_length;
6552			break;
6553		}
6554
6555		/*
6556		 * If both pages have subpages, then the subpage numbers
6557		 * have to match.
6558		 */
6559		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6560		  && (page_header->page_code & SMPH_SPF)) {
6561			struct scsi_mode_page_header_sp *sph;
6562
6563			sph = (struct scsi_mode_page_header_sp *)page_header;
6564
6565			if (lun->mode_pages.index[i].subpage ==
6566			    sph->subpage) {
6567				page_index = &lun->mode_pages.index[i];
6568				page_len = scsi_2btoul(sph->page_length);
6569				break;
6570			}
6571		}
6572	}
6573
6574	/*
6575	 * If we couldn't find the page, or if we don't have a mode select
6576	 * handler for it, send back an error to the user.
6577	 */
6578	if ((page_index == NULL)
6579	 || (page_index->select_handler == NULL)) {
6580		ctl_set_invalid_field(ctsio,
6581				      /*sks_valid*/ 1,
6582				      /*command*/ 0,
6583				      /*field*/ *len_used,
6584				      /*bit_valid*/ 0,
6585				      /*bit*/ 0);
6586		free(ctsio->kern_data_ptr, M_CTL);
6587		ctl_done((union ctl_io *)ctsio);
6588		return (CTL_RETVAL_COMPLETE);
6589	}
6590
6591	if (page_index->page_code & SMPH_SPF) {
6592		page_len_offset = 2;
6593		page_len_size = 2;
6594	} else {
6595		page_len_size = 1;
6596		page_len_offset = 1;
6597	}
6598
6599	/*
6600	 * If the length the initiator gives us isn't the one we specify in
6601	 * the mode page header, or if they didn't specify enough data in
6602	 * the CDB to avoid truncating this page, kick out the request.
6603	 */
6604	if ((page_len != (page_index->page_len - page_len_offset -
6605			  page_len_size))
6606	 || (*len_left < page_index->page_len)) {
6607
6608
6609		ctl_set_invalid_field(ctsio,
6610				      /*sks_valid*/ 1,
6611				      /*command*/ 0,
6612				      /*field*/ *len_used + page_len_offset,
6613				      /*bit_valid*/ 0,
6614				      /*bit*/ 0);
6615		free(ctsio->kern_data_ptr, M_CTL);
6616		ctl_done((union ctl_io *)ctsio);
6617		return (CTL_RETVAL_COMPLETE);
6618	}
6619
6620	/*
6621	 * Run through the mode page, checking to make sure that the bits
6622	 * the user changed are actually legal for him to change.
6623	 */
6624	for (i = 0; i < page_index->page_len; i++) {
6625		uint8_t *user_byte, *change_mask, *current_byte;
6626		int bad_bit;
6627		int j;
6628
6629		user_byte = (uint8_t *)page_header + i;
6630		change_mask = page_index->page_data +
6631			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6632		current_byte = page_index->page_data +
6633			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6634
6635		/*
6636		 * Check to see whether the user set any bits in this byte
6637		 * that he is not allowed to set.
6638		 */
6639		if ((*user_byte & ~(*change_mask)) ==
6640		    (*current_byte & ~(*change_mask)))
6641			continue;
6642
6643		/*
6644		 * Go through bit by bit to determine which one is illegal.
6645		 */
6646		bad_bit = 0;
6647		for (j = 7; j >= 0; j--) {
6648			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6649			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6650				bad_bit = i;
6651				break;
6652			}
6653		}
6654		ctl_set_invalid_field(ctsio,
6655				      /*sks_valid*/ 1,
6656				      /*command*/ 0,
6657				      /*field*/ *len_used + i,
6658				      /*bit_valid*/ 1,
6659				      /*bit*/ bad_bit);
6660		free(ctsio->kern_data_ptr, M_CTL);
6661		ctl_done((union ctl_io *)ctsio);
6662		return (CTL_RETVAL_COMPLETE);
6663	}
6664
6665	/*
6666	 * Decrement these before we call the page handler, since we may
6667	 * end up getting called back one way or another before the handler
6668	 * returns to this context.
6669	 */
6670	*len_left -= page_index->page_len;
6671	*len_used += page_index->page_len;
6672
6673	retval = page_index->select_handler(ctsio, page_index,
6674					    (uint8_t *)page_header);
6675
6676	/*
6677	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6678	 * wait until this queued command completes to finish processing
6679	 * the mode page.  If it returns anything other than
6680	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6681	 * already set the sense information, freed the data pointer, and
6682	 * completed the io for us.
6683	 */
6684	if (retval != CTL_RETVAL_COMPLETE)
6685		goto bailout_no_done;
6686
6687	/*
6688	 * If the initiator sent us more than one page, parse the next one.
6689	 */
6690	if (*len_left > 0)
6691		goto do_next_page;
6692
6693	ctl_set_success(ctsio);
6694	free(ctsio->kern_data_ptr, M_CTL);
6695	ctl_done((union ctl_io *)ctsio);
6696
6697bailout_no_done:
6698
6699	return (CTL_RETVAL_COMPLETE);
6700
6701}
6702
6703int
6704ctl_mode_select(struct ctl_scsiio *ctsio)
6705{
6706	int param_len, pf, sp;
6707	int header_size, bd_len;
6708	int len_left, len_used;
6709	struct ctl_page_index *page_index;
6710	struct ctl_lun *lun;
6711	int control_dev, page_len;
6712	union ctl_modepage_info *modepage_info;
6713	int retval;
6714
6715	pf = 0;
6716	sp = 0;
6717	page_len = 0;
6718	len_used = 0;
6719	len_left = 0;
6720	retval = 0;
6721	bd_len = 0;
6722	page_index = NULL;
6723
6724	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6725
6726	if (lun->be_lun->lun_type != T_DIRECT)
6727		control_dev = 1;
6728	else
6729		control_dev = 0;
6730
6731	switch (ctsio->cdb[0]) {
6732	case MODE_SELECT_6: {
6733		struct scsi_mode_select_6 *cdb;
6734
6735		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6736
6737		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6738		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6739
6740		param_len = cdb->length;
6741		header_size = sizeof(struct scsi_mode_header_6);
6742		break;
6743	}
6744	case MODE_SELECT_10: {
6745		struct scsi_mode_select_10 *cdb;
6746
6747		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6748
6749		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6750		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6751
6752		param_len = scsi_2btoul(cdb->length);
6753		header_size = sizeof(struct scsi_mode_header_10);
6754		break;
6755	}
6756	default:
6757		ctl_set_invalid_opcode(ctsio);
6758		ctl_done((union ctl_io *)ctsio);
6759		return (CTL_RETVAL_COMPLETE);
6760		break; /* NOTREACHED */
6761	}
6762
6763	/*
6764	 * From SPC-3:
6765	 * "A parameter list length of zero indicates that the Data-Out Buffer
6766	 * shall be empty. This condition shall not be considered as an error."
6767	 */
6768	if (param_len == 0) {
6769		ctl_set_success(ctsio);
6770		ctl_done((union ctl_io *)ctsio);
6771		return (CTL_RETVAL_COMPLETE);
6772	}
6773
6774	/*
6775	 * Since we'll hit this the first time through, prior to
6776	 * allocation, we don't need to free a data buffer here.
6777	 */
6778	if (param_len < header_size) {
6779		ctl_set_param_len_error(ctsio);
6780		ctl_done((union ctl_io *)ctsio);
6781		return (CTL_RETVAL_COMPLETE);
6782	}
6783
6784	/*
6785	 * Allocate the data buffer and grab the user's data.  In theory,
6786	 * we shouldn't have to sanity check the parameter list length here
6787	 * because the maximum size is 64K.  We should be able to malloc
6788	 * that much without too many problems.
6789	 */
6790	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6791		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6792		ctsio->kern_data_len = param_len;
6793		ctsio->kern_total_len = param_len;
6794		ctsio->kern_data_resid = 0;
6795		ctsio->kern_rel_offset = 0;
6796		ctsio->kern_sg_entries = 0;
6797		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6798		ctsio->be_move_done = ctl_config_move_done;
6799		ctl_datamove((union ctl_io *)ctsio);
6800
6801		return (CTL_RETVAL_COMPLETE);
6802	}
6803
6804	switch (ctsio->cdb[0]) {
6805	case MODE_SELECT_6: {
6806		struct scsi_mode_header_6 *mh6;
6807
6808		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6809		bd_len = mh6->blk_desc_len;
6810		break;
6811	}
6812	case MODE_SELECT_10: {
6813		struct scsi_mode_header_10 *mh10;
6814
6815		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6816		bd_len = scsi_2btoul(mh10->blk_desc_len);
6817		break;
6818	}
6819	default:
6820		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6821		break;
6822	}
6823
6824	if (param_len < (header_size + bd_len)) {
6825		free(ctsio->kern_data_ptr, M_CTL);
6826		ctl_set_param_len_error(ctsio);
6827		ctl_done((union ctl_io *)ctsio);
6828		return (CTL_RETVAL_COMPLETE);
6829	}
6830
6831	/*
6832	 * Set the IO_CONT flag, so that if this I/O gets passed to
6833	 * ctl_config_write_done(), it'll get passed back to
6834	 * ctl_do_mode_select() for further processing, or completion if
6835	 * we're all done.
6836	 */
6837	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6838	ctsio->io_cont = ctl_do_mode_select;
6839
6840	modepage_info = (union ctl_modepage_info *)
6841		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6842
6843	memset(modepage_info, 0, sizeof(*modepage_info));
6844
6845	len_left = param_len - header_size - bd_len;
6846	len_used = header_size + bd_len;
6847
6848	modepage_info->header.len_left = len_left;
6849	modepage_info->header.len_used = len_used;
6850
6851	return (ctl_do_mode_select((union ctl_io *)ctsio));
6852}
6853
6854int
6855ctl_mode_sense(struct ctl_scsiio *ctsio)
6856{
6857	struct ctl_lun *lun;
6858	int pc, page_code, dbd, llba, subpage;
6859	int alloc_len, page_len, header_len, total_len;
6860	struct scsi_mode_block_descr *block_desc;
6861	struct ctl_page_index *page_index;
6862	int control_dev;
6863
6864	dbd = 0;
6865	llba = 0;
6866	block_desc = NULL;
6867	page_index = NULL;
6868
6869	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6870
6871	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6872
6873	if (lun->be_lun->lun_type != T_DIRECT)
6874		control_dev = 1;
6875	else
6876		control_dev = 0;
6877
6878	if (lun->flags & CTL_LUN_PR_RESERVED) {
6879		uint32_t residx;
6880
6881		/*
6882		 * XXX KDM need a lock here.
6883		 */
6884		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6885		if ((lun->res_type == SPR_TYPE_EX_AC
6886		  && residx != lun->pr_res_idx)
6887		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6888		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6889		  && lun->pr_keys[residx] == 0)) {
6890			ctl_set_reservation_conflict(ctsio);
6891			ctl_done((union ctl_io *)ctsio);
6892			return (CTL_RETVAL_COMPLETE);
6893		}
6894	}
6895
6896	switch (ctsio->cdb[0]) {
6897	case MODE_SENSE_6: {
6898		struct scsi_mode_sense_6 *cdb;
6899
6900		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6901
6902		header_len = sizeof(struct scsi_mode_hdr_6);
6903		if (cdb->byte2 & SMS_DBD)
6904			dbd = 1;
6905		else
6906			header_len += sizeof(struct scsi_mode_block_descr);
6907
6908		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6909		page_code = cdb->page & SMS_PAGE_CODE;
6910		subpage = cdb->subpage;
6911		alloc_len = cdb->length;
6912		break;
6913	}
6914	case MODE_SENSE_10: {
6915		struct scsi_mode_sense_10 *cdb;
6916
6917		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6918
6919		header_len = sizeof(struct scsi_mode_hdr_10);
6920
6921		if (cdb->byte2 & SMS_DBD)
6922			dbd = 1;
6923		else
6924			header_len += sizeof(struct scsi_mode_block_descr);
6925		if (cdb->byte2 & SMS10_LLBAA)
6926			llba = 1;
6927		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6928		page_code = cdb->page & SMS_PAGE_CODE;
6929		subpage = cdb->subpage;
6930		alloc_len = scsi_2btoul(cdb->length);
6931		break;
6932	}
6933	default:
6934		ctl_set_invalid_opcode(ctsio);
6935		ctl_done((union ctl_io *)ctsio);
6936		return (CTL_RETVAL_COMPLETE);
6937		break; /* NOTREACHED */
6938	}
6939
6940	/*
6941	 * We have to make a first pass through to calculate the size of
6942	 * the pages that match the user's query.  Then we allocate enough
6943	 * memory to hold it, and actually copy the data into the buffer.
6944	 */
6945	switch (page_code) {
6946	case SMS_ALL_PAGES_PAGE: {
6947		int i;
6948
6949		page_len = 0;
6950
6951		/*
6952		 * At the moment, values other than 0 and 0xff here are
6953		 * reserved according to SPC-3.
6954		 */
6955		if ((subpage != SMS_SUBPAGE_PAGE_0)
6956		 && (subpage != SMS_SUBPAGE_ALL)) {
6957			ctl_set_invalid_field(ctsio,
6958					      /*sks_valid*/ 1,
6959					      /*command*/ 1,
6960					      /*field*/ 3,
6961					      /*bit_valid*/ 0,
6962					      /*bit*/ 0);
6963			ctl_done((union ctl_io *)ctsio);
6964			return (CTL_RETVAL_COMPLETE);
6965		}
6966
6967		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6968			if ((control_dev != 0)
6969			 && (lun->mode_pages.index[i].page_flags &
6970			     CTL_PAGE_FLAG_DISK_ONLY))
6971				continue;
6972
6973			/*
6974			 * We don't use this subpage if the user didn't
6975			 * request all subpages.
6976			 */
6977			if ((lun->mode_pages.index[i].subpage != 0)
6978			 && (subpage == SMS_SUBPAGE_PAGE_0))
6979				continue;
6980
6981#if 0
6982			printf("found page %#x len %d\n",
6983			       lun->mode_pages.index[i].page_code &
6984			       SMPH_PC_MASK,
6985			       lun->mode_pages.index[i].page_len);
6986#endif
6987			page_len += lun->mode_pages.index[i].page_len;
6988		}
6989		break;
6990	}
6991	default: {
6992		int i;
6993
6994		page_len = 0;
6995
6996		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6997			/* Look for the right page code */
6998			if ((lun->mode_pages.index[i].page_code &
6999			     SMPH_PC_MASK) != page_code)
7000				continue;
7001
7002			/* Look for the right subpage or the subpage wildcard*/
7003			if ((lun->mode_pages.index[i].subpage != subpage)
7004			 && (subpage != SMS_SUBPAGE_ALL))
7005				continue;
7006
7007			/* Make sure the page is supported for this dev type */
7008			if ((control_dev != 0)
7009			 && (lun->mode_pages.index[i].page_flags &
7010			     CTL_PAGE_FLAG_DISK_ONLY))
7011				continue;
7012
7013#if 0
7014			printf("found page %#x len %d\n",
7015			       lun->mode_pages.index[i].page_code &
7016			       SMPH_PC_MASK,
7017			       lun->mode_pages.index[i].page_len);
7018#endif
7019
7020			page_len += lun->mode_pages.index[i].page_len;
7021		}
7022
7023		if (page_len == 0) {
7024			ctl_set_invalid_field(ctsio,
7025					      /*sks_valid*/ 1,
7026					      /*command*/ 1,
7027					      /*field*/ 2,
7028					      /*bit_valid*/ 1,
7029					      /*bit*/ 5);
7030			ctl_done((union ctl_io *)ctsio);
7031			return (CTL_RETVAL_COMPLETE);
7032		}
7033		break;
7034	}
7035	}
7036
7037	total_len = header_len + page_len;
7038#if 0
7039	printf("header_len = %d, page_len = %d, total_len = %d\n",
7040	       header_len, page_len, total_len);
7041#endif
7042
7043	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7044	ctsio->kern_sg_entries = 0;
7045	ctsio->kern_data_resid = 0;
7046	ctsio->kern_rel_offset = 0;
7047	if (total_len < alloc_len) {
7048		ctsio->residual = alloc_len - total_len;
7049		ctsio->kern_data_len = total_len;
7050		ctsio->kern_total_len = total_len;
7051	} else {
7052		ctsio->residual = 0;
7053		ctsio->kern_data_len = alloc_len;
7054		ctsio->kern_total_len = alloc_len;
7055	}
7056
7057	switch (ctsio->cdb[0]) {
7058	case MODE_SENSE_6: {
7059		struct scsi_mode_hdr_6 *header;
7060
7061		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7062
7063		header->datalen = ctl_min(total_len - 1, 254);
7064		if (control_dev == 0) {
7065			header->dev_specific = 0x10; /* DPOFUA */
7066			if ((lun->flags & CTL_LUN_READONLY) ||
7067			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
7068			    .eca_and_aen & SCP_SWP) != 0)
7069				    header->dev_specific |= 0x80; /* WP */
7070		}
7071		if (dbd)
7072			header->block_descr_len = 0;
7073		else
7074			header->block_descr_len =
7075				sizeof(struct scsi_mode_block_descr);
7076		block_desc = (struct scsi_mode_block_descr *)&header[1];
7077		break;
7078	}
7079	case MODE_SENSE_10: {
7080		struct scsi_mode_hdr_10 *header;
7081		int datalen;
7082
7083		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7084
7085		datalen = ctl_min(total_len - 2, 65533);
7086		scsi_ulto2b(datalen, header->datalen);
7087		if (control_dev == 0) {
7088			header->dev_specific = 0x10; /* DPOFUA */
7089			if ((lun->flags & CTL_LUN_READONLY) ||
7090			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
7091			    .eca_and_aen & SCP_SWP) != 0)
7092				    header->dev_specific |= 0x80; /* WP */
7093		}
7094		if (dbd)
7095			scsi_ulto2b(0, header->block_descr_len);
7096		else
7097			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7098				    header->block_descr_len);
7099		block_desc = (struct scsi_mode_block_descr *)&header[1];
7100		break;
7101	}
7102	default:
7103		panic("invalid CDB type %#x", ctsio->cdb[0]);
7104		break; /* NOTREACHED */
7105	}
7106
7107	/*
7108	 * If we've got a disk, use its blocksize in the block
7109	 * descriptor.  Otherwise, just set it to 0.
7110	 */
7111	if (dbd == 0) {
7112		if (control_dev == 0)
7113			scsi_ulto3b(lun->be_lun->blocksize,
7114				    block_desc->block_len);
7115		else
7116			scsi_ulto3b(0, block_desc->block_len);
7117	}
7118
7119	switch (page_code) {
7120	case SMS_ALL_PAGES_PAGE: {
7121		int i, data_used;
7122
7123		data_used = header_len;
7124		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7125			struct ctl_page_index *page_index;
7126
7127			page_index = &lun->mode_pages.index[i];
7128
7129			if ((control_dev != 0)
7130			 && (page_index->page_flags &
7131			    CTL_PAGE_FLAG_DISK_ONLY))
7132				continue;
7133
7134			/*
7135			 * We don't use this subpage if the user didn't
7136			 * request all subpages.  We already checked (above)
7137			 * to make sure the user only specified a subpage
7138			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7139			 */
7140			if ((page_index->subpage != 0)
7141			 && (subpage == SMS_SUBPAGE_PAGE_0))
7142				continue;
7143
7144			/*
7145			 * Call the handler, if it exists, to update the
7146			 * page to the latest values.
7147			 */
7148			if (page_index->sense_handler != NULL)
7149				page_index->sense_handler(ctsio, page_index,pc);
7150
7151			memcpy(ctsio->kern_data_ptr + data_used,
7152			       page_index->page_data +
7153			       (page_index->page_len * pc),
7154			       page_index->page_len);
7155			data_used += page_index->page_len;
7156		}
7157		break;
7158	}
7159	default: {
7160		int i, data_used;
7161
7162		data_used = header_len;
7163
7164		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7165			struct ctl_page_index *page_index;
7166
7167			page_index = &lun->mode_pages.index[i];
7168
7169			/* Look for the right page code */
7170			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7171				continue;
7172
7173			/* Look for the right subpage or the subpage wildcard*/
7174			if ((page_index->subpage != subpage)
7175			 && (subpage != SMS_SUBPAGE_ALL))
7176				continue;
7177
7178			/* Make sure the page is supported for this dev type */
7179			if ((control_dev != 0)
7180			 && (page_index->page_flags &
7181			     CTL_PAGE_FLAG_DISK_ONLY))
7182				continue;
7183
7184			/*
7185			 * Call the handler, if it exists, to update the
7186			 * page to the latest values.
7187			 */
7188			if (page_index->sense_handler != NULL)
7189				page_index->sense_handler(ctsio, page_index,pc);
7190
7191			memcpy(ctsio->kern_data_ptr + data_used,
7192			       page_index->page_data +
7193			       (page_index->page_len * pc),
7194			       page_index->page_len);
7195			data_used += page_index->page_len;
7196		}
7197		break;
7198	}
7199	}
7200
7201	ctsio->scsi_status = SCSI_STATUS_OK;
7202
7203	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7204	ctsio->be_move_done = ctl_config_move_done;
7205	ctl_datamove((union ctl_io *)ctsio);
7206
7207	return (CTL_RETVAL_COMPLETE);
7208}
7209
7210int
7211ctl_read_capacity(struct ctl_scsiio *ctsio)
7212{
7213	struct scsi_read_capacity *cdb;
7214	struct scsi_read_capacity_data *data;
7215	struct ctl_lun *lun;
7216	uint32_t lba;
7217
7218	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7219
7220	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7221
7222	lba = scsi_4btoul(cdb->addr);
7223	if (((cdb->pmi & SRC_PMI) == 0)
7224	 && (lba != 0)) {
7225		ctl_set_invalid_field(/*ctsio*/ ctsio,
7226				      /*sks_valid*/ 1,
7227				      /*command*/ 1,
7228				      /*field*/ 2,
7229				      /*bit_valid*/ 0,
7230				      /*bit*/ 0);
7231		ctl_done((union ctl_io *)ctsio);
7232		return (CTL_RETVAL_COMPLETE);
7233	}
7234
7235	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7236
7237	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7238	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7239	ctsio->residual = 0;
7240	ctsio->kern_data_len = sizeof(*data);
7241	ctsio->kern_total_len = sizeof(*data);
7242	ctsio->kern_data_resid = 0;
7243	ctsio->kern_rel_offset = 0;
7244	ctsio->kern_sg_entries = 0;
7245
7246	/*
7247	 * If the maximum LBA is greater than 0xfffffffe, the user must
7248	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7249	 * serivce action set.
7250	 */
7251	if (lun->be_lun->maxlba > 0xfffffffe)
7252		scsi_ulto4b(0xffffffff, data->addr);
7253	else
7254		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7255
7256	/*
7257	 * XXX KDM this may not be 512 bytes...
7258	 */
7259	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7260
7261	ctsio->scsi_status = SCSI_STATUS_OK;
7262
7263	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7264	ctsio->be_move_done = ctl_config_move_done;
7265	ctl_datamove((union ctl_io *)ctsio);
7266
7267	return (CTL_RETVAL_COMPLETE);
7268}
7269
7270int
7271ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7272{
7273	struct scsi_read_capacity_16 *cdb;
7274	struct scsi_read_capacity_data_long *data;
7275	struct ctl_lun *lun;
7276	uint64_t lba;
7277	uint32_t alloc_len;
7278
7279	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7280
7281	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7282
7283	alloc_len = scsi_4btoul(cdb->alloc_len);
7284	lba = scsi_8btou64(cdb->addr);
7285
7286	if ((cdb->reladr & SRC16_PMI)
7287	 && (lba != 0)) {
7288		ctl_set_invalid_field(/*ctsio*/ ctsio,
7289				      /*sks_valid*/ 1,
7290				      /*command*/ 1,
7291				      /*field*/ 2,
7292				      /*bit_valid*/ 0,
7293				      /*bit*/ 0);
7294		ctl_done((union ctl_io *)ctsio);
7295		return (CTL_RETVAL_COMPLETE);
7296	}
7297
7298	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7299
7300	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7301	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7302
7303	if (sizeof(*data) < alloc_len) {
7304		ctsio->residual = alloc_len - sizeof(*data);
7305		ctsio->kern_data_len = sizeof(*data);
7306		ctsio->kern_total_len = sizeof(*data);
7307	} else {
7308		ctsio->residual = 0;
7309		ctsio->kern_data_len = alloc_len;
7310		ctsio->kern_total_len = alloc_len;
7311	}
7312	ctsio->kern_data_resid = 0;
7313	ctsio->kern_rel_offset = 0;
7314	ctsio->kern_sg_entries = 0;
7315
7316	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7317	/* XXX KDM this may not be 512 bytes... */
7318	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7319	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7320	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7321	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7322		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7323
7324	ctsio->scsi_status = SCSI_STATUS_OK;
7325
7326	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7327	ctsio->be_move_done = ctl_config_move_done;
7328	ctl_datamove((union ctl_io *)ctsio);
7329
7330	return (CTL_RETVAL_COMPLETE);
7331}
7332
7333int
7334ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7335{
7336	struct scsi_maintenance_in *cdb;
7337	int retval;
7338	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7339	int num_target_port_groups, num_target_ports, single;
7340	struct ctl_lun *lun;
7341	struct ctl_softc *softc;
7342	struct ctl_port *port;
7343	struct scsi_target_group_data *rtg_ptr;
7344	struct scsi_target_group_data_extended *rtg_ext_ptr;
7345	struct scsi_target_port_group_descriptor *tpg_desc;
7346
7347	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7348
7349	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7350	softc = control_softc;
7351	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7352
7353	retval = CTL_RETVAL_COMPLETE;
7354
7355	switch (cdb->byte2 & STG_PDF_MASK) {
7356	case STG_PDF_LENGTH:
7357		ext = 0;
7358		break;
7359	case STG_PDF_EXTENDED:
7360		ext = 1;
7361		break;
7362	default:
7363		ctl_set_invalid_field(/*ctsio*/ ctsio,
7364				      /*sks_valid*/ 1,
7365				      /*command*/ 1,
7366				      /*field*/ 2,
7367				      /*bit_valid*/ 1,
7368				      /*bit*/ 5);
7369		ctl_done((union ctl_io *)ctsio);
7370		return(retval);
7371	}
7372
7373	single = ctl_is_single;
7374	if (single)
7375		num_target_port_groups = 1;
7376	else
7377		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7378	num_target_ports = 0;
7379	mtx_lock(&softc->ctl_lock);
7380	STAILQ_FOREACH(port, &softc->port_list, links) {
7381		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7382			continue;
7383		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7384			continue;
7385		num_target_ports++;
7386	}
7387	mtx_unlock(&softc->ctl_lock);
7388
7389	if (ext)
7390		total_len = sizeof(struct scsi_target_group_data_extended);
7391	else
7392		total_len = sizeof(struct scsi_target_group_data);
7393	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7394		num_target_port_groups +
7395	    sizeof(struct scsi_target_port_descriptor) *
7396		num_target_ports * num_target_port_groups;
7397
7398	alloc_len = scsi_4btoul(cdb->length);
7399
7400	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7401
7402	ctsio->kern_sg_entries = 0;
7403
7404	if (total_len < alloc_len) {
7405		ctsio->residual = alloc_len - total_len;
7406		ctsio->kern_data_len = total_len;
7407		ctsio->kern_total_len = total_len;
7408	} else {
7409		ctsio->residual = 0;
7410		ctsio->kern_data_len = alloc_len;
7411		ctsio->kern_total_len = alloc_len;
7412	}
7413	ctsio->kern_data_resid = 0;
7414	ctsio->kern_rel_offset = 0;
7415
7416	if (ext) {
7417		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7418		    ctsio->kern_data_ptr;
7419		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7420		rtg_ext_ptr->format_type = 0x10;
7421		rtg_ext_ptr->implicit_transition_time = 0;
7422		tpg_desc = &rtg_ext_ptr->groups[0];
7423	} else {
7424		rtg_ptr = (struct scsi_target_group_data *)
7425		    ctsio->kern_data_ptr;
7426		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7427		tpg_desc = &rtg_ptr->groups[0];
7428	}
7429
7430	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7431	mtx_lock(&softc->ctl_lock);
7432	for (g = 0; g < num_target_port_groups; g++) {
7433		if (g == pg)
7434			tpg_desc->pref_state = TPG_PRIMARY |
7435			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7436		else
7437			tpg_desc->pref_state =
7438			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7439		tpg_desc->support = TPG_AO_SUP;
7440		if (!single)
7441			tpg_desc->support |= TPG_AN_SUP;
7442		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7443		tpg_desc->status = TPG_IMPLICIT;
7444		pc = 0;
7445		STAILQ_FOREACH(port, &softc->port_list, links) {
7446			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7447				continue;
7448			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7449			    CTL_MAX_LUNS)
7450				continue;
7451			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7452			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7453			    relative_target_port_identifier);
7454			pc++;
7455		}
7456		tpg_desc->target_port_count = pc;
7457		tpg_desc = (struct scsi_target_port_group_descriptor *)
7458		    &tpg_desc->descriptors[pc];
7459	}
7460	mtx_unlock(&softc->ctl_lock);
7461
7462	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7463	ctsio->be_move_done = ctl_config_move_done;
7464
7465	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7466			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7467			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7468			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7469			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7470
7471	ctl_datamove((union ctl_io *)ctsio);
7472	return(retval);
7473}
7474
7475int
7476ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7477{
7478	struct ctl_lun *lun;
7479	struct scsi_report_supported_opcodes *cdb;
7480	const struct ctl_cmd_entry *entry, *sentry;
7481	struct scsi_report_supported_opcodes_all *all;
7482	struct scsi_report_supported_opcodes_descr *descr;
7483	struct scsi_report_supported_opcodes_one *one;
7484	int retval;
7485	int alloc_len, total_len;
7486	int opcode, service_action, i, j, num;
7487
7488	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7489
7490	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7491	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7492
7493	retval = CTL_RETVAL_COMPLETE;
7494
7495	opcode = cdb->requested_opcode;
7496	service_action = scsi_2btoul(cdb->requested_service_action);
7497	switch (cdb->options & RSO_OPTIONS_MASK) {
7498	case RSO_OPTIONS_ALL:
7499		num = 0;
7500		for (i = 0; i < 256; i++) {
7501			entry = &ctl_cmd_table[i];
7502			if (entry->flags & CTL_CMD_FLAG_SA5) {
7503				for (j = 0; j < 32; j++) {
7504					sentry = &((const struct ctl_cmd_entry *)
7505					    entry->execute)[j];
7506					if (ctl_cmd_applicable(
7507					    lun->be_lun->lun_type, sentry))
7508						num++;
7509				}
7510			} else {
7511				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7512				    entry))
7513					num++;
7514			}
7515		}
7516		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7517		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7518		break;
7519	case RSO_OPTIONS_OC:
7520		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7521			ctl_set_invalid_field(/*ctsio*/ ctsio,
7522					      /*sks_valid*/ 1,
7523					      /*command*/ 1,
7524					      /*field*/ 2,
7525					      /*bit_valid*/ 1,
7526					      /*bit*/ 2);
7527			ctl_done((union ctl_io *)ctsio);
7528			return (CTL_RETVAL_COMPLETE);
7529		}
7530		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7531		break;
7532	case RSO_OPTIONS_OC_SA:
7533		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7534		    service_action >= 32) {
7535			ctl_set_invalid_field(/*ctsio*/ ctsio,
7536					      /*sks_valid*/ 1,
7537					      /*command*/ 1,
7538					      /*field*/ 2,
7539					      /*bit_valid*/ 1,
7540					      /*bit*/ 2);
7541			ctl_done((union ctl_io *)ctsio);
7542			return (CTL_RETVAL_COMPLETE);
7543		}
7544		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7545		break;
7546	default:
7547		ctl_set_invalid_field(/*ctsio*/ ctsio,
7548				      /*sks_valid*/ 1,
7549				      /*command*/ 1,
7550				      /*field*/ 2,
7551				      /*bit_valid*/ 1,
7552				      /*bit*/ 2);
7553		ctl_done((union ctl_io *)ctsio);
7554		return (CTL_RETVAL_COMPLETE);
7555	}
7556
7557	alloc_len = scsi_4btoul(cdb->length);
7558
7559	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7560
7561	ctsio->kern_sg_entries = 0;
7562
7563	if (total_len < alloc_len) {
7564		ctsio->residual = alloc_len - total_len;
7565		ctsio->kern_data_len = total_len;
7566		ctsio->kern_total_len = total_len;
7567	} else {
7568		ctsio->residual = 0;
7569		ctsio->kern_data_len = alloc_len;
7570		ctsio->kern_total_len = alloc_len;
7571	}
7572	ctsio->kern_data_resid = 0;
7573	ctsio->kern_rel_offset = 0;
7574
7575	switch (cdb->options & RSO_OPTIONS_MASK) {
7576	case RSO_OPTIONS_ALL:
7577		all = (struct scsi_report_supported_opcodes_all *)
7578		    ctsio->kern_data_ptr;
7579		num = 0;
7580		for (i = 0; i < 256; i++) {
7581			entry = &ctl_cmd_table[i];
7582			if (entry->flags & CTL_CMD_FLAG_SA5) {
7583				for (j = 0; j < 32; j++) {
7584					sentry = &((const struct ctl_cmd_entry *)
7585					    entry->execute)[j];
7586					if (!ctl_cmd_applicable(
7587					    lun->be_lun->lun_type, sentry))
7588						continue;
7589					descr = &all->descr[num++];
7590					descr->opcode = i;
7591					scsi_ulto2b(j, descr->service_action);
7592					descr->flags = RSO_SERVACTV;
7593					scsi_ulto2b(sentry->length,
7594					    descr->cdb_length);
7595				}
7596			} else {
7597				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7598				    entry))
7599					continue;
7600				descr = &all->descr[num++];
7601				descr->opcode = i;
7602				scsi_ulto2b(0, descr->service_action);
7603				descr->flags = 0;
7604				scsi_ulto2b(entry->length, descr->cdb_length);
7605			}
7606		}
7607		scsi_ulto4b(
7608		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7609		    all->length);
7610		break;
7611	case RSO_OPTIONS_OC:
7612		one = (struct scsi_report_supported_opcodes_one *)
7613		    ctsio->kern_data_ptr;
7614		entry = &ctl_cmd_table[opcode];
7615		goto fill_one;
7616	case RSO_OPTIONS_OC_SA:
7617		one = (struct scsi_report_supported_opcodes_one *)
7618		    ctsio->kern_data_ptr;
7619		entry = &ctl_cmd_table[opcode];
7620		entry = &((const struct ctl_cmd_entry *)
7621		    entry->execute)[service_action];
7622fill_one:
7623		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7624			one->support = 3;
7625			scsi_ulto2b(entry->length, one->cdb_length);
7626			one->cdb_usage[0] = opcode;
7627			memcpy(&one->cdb_usage[1], entry->usage,
7628			    entry->length - 1);
7629		} else
7630			one->support = 1;
7631		break;
7632	}
7633
7634	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7635	ctsio->be_move_done = ctl_config_move_done;
7636
7637	ctl_datamove((union ctl_io *)ctsio);
7638	return(retval);
7639}
7640
7641int
7642ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7643{
7644	struct ctl_lun *lun;
7645	struct scsi_report_supported_tmf *cdb;
7646	struct scsi_report_supported_tmf_data *data;
7647	int retval;
7648	int alloc_len, total_len;
7649
7650	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7651
7652	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7653	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7654
7655	retval = CTL_RETVAL_COMPLETE;
7656
7657	total_len = sizeof(struct scsi_report_supported_tmf_data);
7658	alloc_len = scsi_4btoul(cdb->length);
7659
7660	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7661
7662	ctsio->kern_sg_entries = 0;
7663
7664	if (total_len < alloc_len) {
7665		ctsio->residual = alloc_len - total_len;
7666		ctsio->kern_data_len = total_len;
7667		ctsio->kern_total_len = total_len;
7668	} else {
7669		ctsio->residual = 0;
7670		ctsio->kern_data_len = alloc_len;
7671		ctsio->kern_total_len = alloc_len;
7672	}
7673	ctsio->kern_data_resid = 0;
7674	ctsio->kern_rel_offset = 0;
7675
7676	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7677	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7678	data->byte2 |= RST_ITNRS;
7679
7680	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7681	ctsio->be_move_done = ctl_config_move_done;
7682
7683	ctl_datamove((union ctl_io *)ctsio);
7684	return (retval);
7685}
7686
7687int
7688ctl_report_timestamp(struct ctl_scsiio *ctsio)
7689{
7690	struct ctl_lun *lun;
7691	struct scsi_report_timestamp *cdb;
7692	struct scsi_report_timestamp_data *data;
7693	struct timeval tv;
7694	int64_t timestamp;
7695	int retval;
7696	int alloc_len, total_len;
7697
7698	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7699
7700	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7701	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7702
7703	retval = CTL_RETVAL_COMPLETE;
7704
7705	total_len = sizeof(struct scsi_report_timestamp_data);
7706	alloc_len = scsi_4btoul(cdb->length);
7707
7708	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7709
7710	ctsio->kern_sg_entries = 0;
7711
7712	if (total_len < alloc_len) {
7713		ctsio->residual = alloc_len - total_len;
7714		ctsio->kern_data_len = total_len;
7715		ctsio->kern_total_len = total_len;
7716	} else {
7717		ctsio->residual = 0;
7718		ctsio->kern_data_len = alloc_len;
7719		ctsio->kern_total_len = alloc_len;
7720	}
7721	ctsio->kern_data_resid = 0;
7722	ctsio->kern_rel_offset = 0;
7723
7724	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7725	scsi_ulto2b(sizeof(*data) - 2, data->length);
7726	data->origin = RTS_ORIG_OUTSIDE;
7727	getmicrotime(&tv);
7728	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7729	scsi_ulto4b(timestamp >> 16, data->timestamp);
7730	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7731
7732	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7733	ctsio->be_move_done = ctl_config_move_done;
7734
7735	ctl_datamove((union ctl_io *)ctsio);
7736	return (retval);
7737}
7738
7739int
7740ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7741{
7742	struct scsi_per_res_in *cdb;
7743	int alloc_len, total_len = 0;
7744	/* struct scsi_per_res_in_rsrv in_data; */
7745	struct ctl_lun *lun;
7746	struct ctl_softc *softc;
7747
7748	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7749
7750	softc = control_softc;
7751
7752	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7753
7754	alloc_len = scsi_2btoul(cdb->length);
7755
7756	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7757
7758retry:
7759	mtx_lock(&lun->lun_lock);
7760	switch (cdb->action) {
7761	case SPRI_RK: /* read keys */
7762		total_len = sizeof(struct scsi_per_res_in_keys) +
7763			lun->pr_key_count *
7764			sizeof(struct scsi_per_res_key);
7765		break;
7766	case SPRI_RR: /* read reservation */
7767		if (lun->flags & CTL_LUN_PR_RESERVED)
7768			total_len = sizeof(struct scsi_per_res_in_rsrv);
7769		else
7770			total_len = sizeof(struct scsi_per_res_in_header);
7771		break;
7772	case SPRI_RC: /* report capabilities */
7773		total_len = sizeof(struct scsi_per_res_cap);
7774		break;
7775	case SPRI_RS: /* read full status */
7776		total_len = sizeof(struct scsi_per_res_in_header) +
7777		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7778		    lun->pr_key_count;
7779		break;
7780	default:
7781		panic("Invalid PR type %x", cdb->action);
7782	}
7783	mtx_unlock(&lun->lun_lock);
7784
7785	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7786
7787	if (total_len < alloc_len) {
7788		ctsio->residual = alloc_len - total_len;
7789		ctsio->kern_data_len = total_len;
7790		ctsio->kern_total_len = total_len;
7791	} else {
7792		ctsio->residual = 0;
7793		ctsio->kern_data_len = alloc_len;
7794		ctsio->kern_total_len = alloc_len;
7795	}
7796
7797	ctsio->kern_data_resid = 0;
7798	ctsio->kern_rel_offset = 0;
7799	ctsio->kern_sg_entries = 0;
7800
7801	mtx_lock(&lun->lun_lock);
7802	switch (cdb->action) {
7803	case SPRI_RK: { // read keys
7804        struct scsi_per_res_in_keys *res_keys;
7805		int i, key_count;
7806
7807		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7808
7809		/*
7810		 * We had to drop the lock to allocate our buffer, which
7811		 * leaves time for someone to come in with another
7812		 * persistent reservation.  (That is unlikely, though,
7813		 * since this should be the only persistent reservation
7814		 * command active right now.)
7815		 */
7816		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7817		    (lun->pr_key_count *
7818		     sizeof(struct scsi_per_res_key)))){
7819			mtx_unlock(&lun->lun_lock);
7820			free(ctsio->kern_data_ptr, M_CTL);
7821			printf("%s: reservation length changed, retrying\n",
7822			       __func__);
7823			goto retry;
7824		}
7825
7826		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7827
7828		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7829			     lun->pr_key_count, res_keys->header.length);
7830
7831		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7832			if (lun->pr_keys[i] == 0)
7833				continue;
7834
7835			/*
7836			 * We used lun->pr_key_count to calculate the
7837			 * size to allocate.  If it turns out the number of
7838			 * initiators with the registered flag set is
7839			 * larger than that (i.e. they haven't been kept in
7840			 * sync), we've got a problem.
7841			 */
7842			if (key_count >= lun->pr_key_count) {
7843#ifdef NEEDTOPORT
7844				csevent_log(CSC_CTL | CSC_SHELF_SW |
7845					    CTL_PR_ERROR,
7846					    csevent_LogType_Fault,
7847					    csevent_AlertLevel_Yellow,
7848					    csevent_FRU_ShelfController,
7849					    csevent_FRU_Firmware,
7850				        csevent_FRU_Unknown,
7851					    "registered keys %d >= key "
7852					    "count %d", key_count,
7853					    lun->pr_key_count);
7854#endif
7855				key_count++;
7856				continue;
7857			}
7858			scsi_u64to8b(lun->pr_keys[i],
7859			    res_keys->keys[key_count].key);
7860			key_count++;
7861		}
7862		break;
7863	}
7864	case SPRI_RR: { // read reservation
7865		struct scsi_per_res_in_rsrv *res;
7866		int tmp_len, header_only;
7867
7868		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7869
7870		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7871
7872		if (lun->flags & CTL_LUN_PR_RESERVED)
7873		{
7874			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7875			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7876				    res->header.length);
7877			header_only = 0;
7878		} else {
7879			tmp_len = sizeof(struct scsi_per_res_in_header);
7880			scsi_ulto4b(0, res->header.length);
7881			header_only = 1;
7882		}
7883
7884		/*
7885		 * We had to drop the lock to allocate our buffer, which
7886		 * leaves time for someone to come in with another
7887		 * persistent reservation.  (That is unlikely, though,
7888		 * since this should be the only persistent reservation
7889		 * command active right now.)
7890		 */
7891		if (tmp_len != total_len) {
7892			mtx_unlock(&lun->lun_lock);
7893			free(ctsio->kern_data_ptr, M_CTL);
7894			printf("%s: reservation status changed, retrying\n",
7895			       __func__);
7896			goto retry;
7897		}
7898
7899		/*
7900		 * No reservation held, so we're done.
7901		 */
7902		if (header_only != 0)
7903			break;
7904
7905		/*
7906		 * If the registration is an All Registrants type, the key
7907		 * is 0, since it doesn't really matter.
7908		 */
7909		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7910			scsi_u64to8b(lun->pr_keys[lun->pr_res_idx],
7911			    res->data.reservation);
7912		}
7913		res->data.scopetype = lun->res_type;
7914		break;
7915	}
7916	case SPRI_RC:     //report capabilities
7917	{
7918		struct scsi_per_res_cap *res_cap;
7919		uint16_t type_mask;
7920
7921		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7922		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7923		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7924		type_mask = SPRI_TM_WR_EX_AR |
7925			    SPRI_TM_EX_AC_RO |
7926			    SPRI_TM_WR_EX_RO |
7927			    SPRI_TM_EX_AC |
7928			    SPRI_TM_WR_EX |
7929			    SPRI_TM_EX_AC_AR;
7930		scsi_ulto2b(type_mask, res_cap->type_mask);
7931		break;
7932	}
7933	case SPRI_RS: { // read full status
7934		struct scsi_per_res_in_full *res_status;
7935		struct scsi_per_res_in_full_desc *res_desc;
7936		struct ctl_port *port;
7937		int i, len;
7938
7939		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7940
7941		/*
7942		 * We had to drop the lock to allocate our buffer, which
7943		 * leaves time for someone to come in with another
7944		 * persistent reservation.  (That is unlikely, though,
7945		 * since this should be the only persistent reservation
7946		 * command active right now.)
7947		 */
7948		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7949		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7950		     lun->pr_key_count)){
7951			mtx_unlock(&lun->lun_lock);
7952			free(ctsio->kern_data_ptr, M_CTL);
7953			printf("%s: reservation length changed, retrying\n",
7954			       __func__);
7955			goto retry;
7956		}
7957
7958		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7959
7960		res_desc = &res_status->desc[0];
7961		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7962			if (lun->pr_keys[i] == 0)
7963				continue;
7964
7965			scsi_u64to8b(lun->pr_keys[i], res_desc->res_key.key);
7966			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7967			    (lun->pr_res_idx == i ||
7968			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7969				res_desc->flags = SPRI_FULL_R_HOLDER;
7970				res_desc->scopetype = lun->res_type;
7971			}
7972			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7973			    res_desc->rel_trgt_port_id);
7974			len = 0;
7975			port = softc->ctl_ports[
7976			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7977			if (port != NULL)
7978				len = ctl_create_iid(port,
7979				    i % CTL_MAX_INIT_PER_PORT,
7980				    res_desc->transport_id);
7981			scsi_ulto4b(len, res_desc->additional_length);
7982			res_desc = (struct scsi_per_res_in_full_desc *)
7983			    &res_desc->transport_id[len];
7984		}
7985		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7986		    res_status->header.length);
7987		break;
7988	}
7989	default:
7990		/*
7991		 * This is a bug, because we just checked for this above,
7992		 * and should have returned an error.
7993		 */
7994		panic("Invalid PR type %x", cdb->action);
7995		break; /* NOTREACHED */
7996	}
7997	mtx_unlock(&lun->lun_lock);
7998
7999	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8000	ctsio->be_move_done = ctl_config_move_done;
8001
8002	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
8003			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
8004			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
8005			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
8006			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
8007
8008	ctl_datamove((union ctl_io *)ctsio);
8009
8010	return (CTL_RETVAL_COMPLETE);
8011}
8012
8013/*
8014 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
8015 * it should return.
8016 */
8017static int
8018ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
8019		uint64_t sa_res_key, uint8_t type, uint32_t residx,
8020		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
8021		struct scsi_per_res_out_parms* param)
8022{
8023	union ctl_ha_msg persis_io;
8024	int retval, i;
8025	int isc_retval;
8026
8027	retval = 0;
8028
8029	mtx_lock(&lun->lun_lock);
8030	if (sa_res_key == 0) {
8031		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8032			/* validate scope and type */
8033			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8034			     SPR_LU_SCOPE) {
8035				mtx_unlock(&lun->lun_lock);
8036				ctl_set_invalid_field(/*ctsio*/ ctsio,
8037						      /*sks_valid*/ 1,
8038						      /*command*/ 1,
8039						      /*field*/ 2,
8040						      /*bit_valid*/ 1,
8041						      /*bit*/ 4);
8042				ctl_done((union ctl_io *)ctsio);
8043				return (1);
8044			}
8045
8046		        if (type>8 || type==2 || type==4 || type==0) {
8047				mtx_unlock(&lun->lun_lock);
8048				ctl_set_invalid_field(/*ctsio*/ ctsio,
8049       	           				      /*sks_valid*/ 1,
8050						      /*command*/ 1,
8051						      /*field*/ 2,
8052						      /*bit_valid*/ 1,
8053						      /*bit*/ 0);
8054				ctl_done((union ctl_io *)ctsio);
8055				return (1);
8056		        }
8057
8058			/*
8059			 * Unregister everybody else and build UA for
8060			 * them
8061			 */
8062			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8063				if (i == residx || lun->pr_keys[i] == 0)
8064					continue;
8065
8066				if (!persis_offset
8067				 && i <CTL_MAX_INITIATORS)
8068					lun->pending_ua[i] |=
8069						CTL_UA_REG_PREEMPT;
8070				else if (persis_offset
8071				      && i >= persis_offset)
8072					lun->pending_ua[i-persis_offset] |=
8073						CTL_UA_REG_PREEMPT;
8074				lun->pr_keys[i] = 0;
8075			}
8076			lun->pr_key_count = 1;
8077			lun->res_type = type;
8078			if (lun->res_type != SPR_TYPE_WR_EX_AR
8079			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8080				lun->pr_res_idx = residx;
8081
8082			/* send msg to other side */
8083			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8084			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8085			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8086			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8087			persis_io.pr.pr_info.res_type = type;
8088			memcpy(persis_io.pr.pr_info.sa_res_key,
8089			       param->serv_act_res_key,
8090			       sizeof(param->serv_act_res_key));
8091			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8092			     &persis_io, sizeof(persis_io), 0)) >
8093			     CTL_HA_STATUS_SUCCESS) {
8094				printf("CTL:Persis Out error returned "
8095				       "from ctl_ha_msg_send %d\n",
8096				       isc_retval);
8097			}
8098		} else {
8099			/* not all registrants */
8100			mtx_unlock(&lun->lun_lock);
8101			free(ctsio->kern_data_ptr, M_CTL);
8102			ctl_set_invalid_field(ctsio,
8103					      /*sks_valid*/ 1,
8104					      /*command*/ 0,
8105					      /*field*/ 8,
8106					      /*bit_valid*/ 0,
8107					      /*bit*/ 0);
8108			ctl_done((union ctl_io *)ctsio);
8109			return (1);
8110		}
8111	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8112		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8113		int found = 0;
8114
8115		if (res_key == sa_res_key) {
8116			/* special case */
8117			/*
8118			 * The spec implies this is not good but doesn't
8119			 * say what to do. There are two choices either
8120			 * generate a res conflict or check condition
8121			 * with illegal field in parameter data. Since
8122			 * that is what is done when the sa_res_key is
8123			 * zero I'll take that approach since this has
8124			 * to do with the sa_res_key.
8125			 */
8126			mtx_unlock(&lun->lun_lock);
8127			free(ctsio->kern_data_ptr, M_CTL);
8128			ctl_set_invalid_field(ctsio,
8129					      /*sks_valid*/ 1,
8130					      /*command*/ 0,
8131					      /*field*/ 8,
8132					      /*bit_valid*/ 0,
8133					      /*bit*/ 0);
8134			ctl_done((union ctl_io *)ctsio);
8135			return (1);
8136		}
8137
8138		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8139			if (lun->pr_keys[i] != sa_res_key)
8140				continue;
8141
8142			found = 1;
8143			lun->pr_keys[i] = 0;
8144			lun->pr_key_count--;
8145
8146			if (!persis_offset && i < CTL_MAX_INITIATORS)
8147				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8148			else if (persis_offset && i >= persis_offset)
8149				lun->pending_ua[i-persis_offset] |=
8150					CTL_UA_REG_PREEMPT;
8151		}
8152		if (!found) {
8153			mtx_unlock(&lun->lun_lock);
8154			free(ctsio->kern_data_ptr, M_CTL);
8155			ctl_set_reservation_conflict(ctsio);
8156			ctl_done((union ctl_io *)ctsio);
8157			return (CTL_RETVAL_COMPLETE);
8158		}
8159		/* send msg to other side */
8160		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8161		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8162		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8163		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8164		persis_io.pr.pr_info.res_type = type;
8165		memcpy(persis_io.pr.pr_info.sa_res_key,
8166		       param->serv_act_res_key,
8167		       sizeof(param->serv_act_res_key));
8168		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8169		     &persis_io, sizeof(persis_io), 0)) >
8170		     CTL_HA_STATUS_SUCCESS) {
8171			printf("CTL:Persis Out error returned from "
8172			       "ctl_ha_msg_send %d\n", isc_retval);
8173		}
8174	} else {
8175		/* Reserved but not all registrants */
8176		/* sa_res_key is res holder */
8177		if (sa_res_key == lun->pr_keys[lun->pr_res_idx]) {
8178			/* validate scope and type */
8179			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8180			     SPR_LU_SCOPE) {
8181				mtx_unlock(&lun->lun_lock);
8182				ctl_set_invalid_field(/*ctsio*/ ctsio,
8183						      /*sks_valid*/ 1,
8184						      /*command*/ 1,
8185						      /*field*/ 2,
8186						      /*bit_valid*/ 1,
8187						      /*bit*/ 4);
8188				ctl_done((union ctl_io *)ctsio);
8189				return (1);
8190			}
8191
8192			if (type>8 || type==2 || type==4 || type==0) {
8193				mtx_unlock(&lun->lun_lock);
8194				ctl_set_invalid_field(/*ctsio*/ ctsio,
8195						      /*sks_valid*/ 1,
8196						      /*command*/ 1,
8197						      /*field*/ 2,
8198						      /*bit_valid*/ 1,
8199						      /*bit*/ 0);
8200				ctl_done((union ctl_io *)ctsio);
8201				return (1);
8202			}
8203
8204			/*
8205			 * Do the following:
8206			 * if sa_res_key != res_key remove all
8207			 * registrants w/sa_res_key and generate UA
8208			 * for these registrants(Registrations
8209			 * Preempted) if it wasn't an exclusive
8210			 * reservation generate UA(Reservations
8211			 * Preempted) for all other registered nexuses
8212			 * if the type has changed. Establish the new
8213			 * reservation and holder. If res_key and
8214			 * sa_res_key are the same do the above
8215			 * except don't unregister the res holder.
8216			 */
8217
8218			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8219				if (i == residx || lun->pr_keys[i] == 0)
8220					continue;
8221
8222				if (sa_res_key == lun->pr_keys[i]) {
8223					lun->pr_keys[i] = 0;
8224					lun->pr_key_count--;
8225
8226					if (!persis_offset
8227					 && i < CTL_MAX_INITIATORS)
8228						lun->pending_ua[i] |=
8229							CTL_UA_REG_PREEMPT;
8230					else if (persis_offset
8231					      && i >= persis_offset)
8232						lun->pending_ua[i-persis_offset] |=
8233						  CTL_UA_REG_PREEMPT;
8234				} else if (type != lun->res_type
8235					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8236					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8237						if (!persis_offset
8238						 && i < CTL_MAX_INITIATORS)
8239							lun->pending_ua[i] |=
8240							CTL_UA_RES_RELEASE;
8241						else if (persis_offset
8242						      && i >= persis_offset)
8243							lun->pending_ua[
8244							i-persis_offset] |=
8245							CTL_UA_RES_RELEASE;
8246				}
8247			}
8248			lun->res_type = type;
8249			if (lun->res_type != SPR_TYPE_WR_EX_AR
8250			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8251				lun->pr_res_idx = residx;
8252			else
8253				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8254
8255			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8256			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8257			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8258			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8259			persis_io.pr.pr_info.res_type = type;
8260			memcpy(persis_io.pr.pr_info.sa_res_key,
8261			       param->serv_act_res_key,
8262			       sizeof(param->serv_act_res_key));
8263			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8264			     &persis_io, sizeof(persis_io), 0)) >
8265			     CTL_HA_STATUS_SUCCESS) {
8266				printf("CTL:Persis Out error returned "
8267				       "from ctl_ha_msg_send %d\n",
8268				       isc_retval);
8269			}
8270		} else {
8271			/*
8272			 * sa_res_key is not the res holder just
8273			 * remove registrants
8274			 */
8275			int found=0;
8276
8277			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8278				if (sa_res_key != lun->pr_keys[i])
8279					continue;
8280
8281				found = 1;
8282				lun->pr_keys[i] = 0;
8283				lun->pr_key_count--;
8284
8285				if (!persis_offset
8286				 && i < CTL_MAX_INITIATORS)
8287					lun->pending_ua[i] |=
8288						CTL_UA_REG_PREEMPT;
8289				else if (persis_offset
8290				      && i >= persis_offset)
8291					lun->pending_ua[i-persis_offset] |=
8292						CTL_UA_REG_PREEMPT;
8293			}
8294
8295			if (!found) {
8296				mtx_unlock(&lun->lun_lock);
8297				free(ctsio->kern_data_ptr, M_CTL);
8298				ctl_set_reservation_conflict(ctsio);
8299				ctl_done((union ctl_io *)ctsio);
8300		        	return (1);
8301			}
8302			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8303			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8304			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8305			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8306			persis_io.pr.pr_info.res_type = type;
8307			memcpy(persis_io.pr.pr_info.sa_res_key,
8308			       param->serv_act_res_key,
8309			       sizeof(param->serv_act_res_key));
8310			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8311			     &persis_io, sizeof(persis_io), 0)) >
8312			     CTL_HA_STATUS_SUCCESS) {
8313				printf("CTL:Persis Out error returned "
8314				       "from ctl_ha_msg_send %d\n",
8315				isc_retval);
8316			}
8317		}
8318	}
8319
8320	lun->PRGeneration++;
8321	mtx_unlock(&lun->lun_lock);
8322
8323	return (retval);
8324}
8325
8326static void
8327ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8328{
8329	uint64_t sa_res_key;
8330	int i;
8331
8332	sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8333
8334	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8335	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8336	 || sa_res_key != lun->pr_keys[lun->pr_res_idx]) {
8337		if (sa_res_key == 0) {
8338			/*
8339			 * Unregister everybody else and build UA for
8340			 * them
8341			 */
8342			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8343				if (i == msg->pr.pr_info.residx ||
8344				    lun->pr_keys[i] == 0)
8345					continue;
8346
8347				if (!persis_offset
8348				 && i < CTL_MAX_INITIATORS)
8349					lun->pending_ua[i] |=
8350						CTL_UA_REG_PREEMPT;
8351				else if (persis_offset && i >= persis_offset)
8352					lun->pending_ua[i - persis_offset] |=
8353						CTL_UA_REG_PREEMPT;
8354				lun->pr_keys[i] = 0;
8355			}
8356
8357			lun->pr_key_count = 1;
8358			lun->res_type = msg->pr.pr_info.res_type;
8359			if (lun->res_type != SPR_TYPE_WR_EX_AR
8360			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8361				lun->pr_res_idx = msg->pr.pr_info.residx;
8362		} else {
8363		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8364				if (sa_res_key == lun->pr_keys[i])
8365					continue;
8366
8367				lun->pr_keys[i] = 0;
8368				lun->pr_key_count--;
8369
8370				if (!persis_offset
8371				 && i < persis_offset)
8372					lun->pending_ua[i] |=
8373						CTL_UA_REG_PREEMPT;
8374				else if (persis_offset
8375				      && i >= persis_offset)
8376					lun->pending_ua[i - persis_offset] |=
8377						CTL_UA_REG_PREEMPT;
8378			}
8379		}
8380	} else {
8381		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8382			if (i == msg->pr.pr_info.residx ||
8383			    lun->pr_keys[i] == 0)
8384				continue;
8385
8386			if (sa_res_key == lun->pr_keys[i]) {
8387				lun->pr_keys[i] = 0;
8388				lun->pr_key_count--;
8389				if (!persis_offset
8390				 && i < CTL_MAX_INITIATORS)
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			} else if (msg->pr.pr_info.res_type != lun->res_type
8398				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8399				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8400					if (!persis_offset
8401					 && i < persis_offset)
8402						lun->pending_ua[i] |=
8403							CTL_UA_RES_RELEASE;
8404					else if (persis_offset
8405					      && i >= persis_offset)
8406					lun->pending_ua[i - persis_offset] |=
8407						CTL_UA_RES_RELEASE;
8408			}
8409		}
8410		lun->res_type = msg->pr.pr_info.res_type;
8411		if (lun->res_type != SPR_TYPE_WR_EX_AR
8412		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8413			lun->pr_res_idx = msg->pr.pr_info.residx;
8414		else
8415			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8416	}
8417	lun->PRGeneration++;
8418
8419}
8420
8421
8422int
8423ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8424{
8425	int retval;
8426	int isc_retval;
8427	u_int32_t param_len;
8428	struct scsi_per_res_out *cdb;
8429	struct ctl_lun *lun;
8430	struct scsi_per_res_out_parms* param;
8431	struct ctl_softc *softc;
8432	uint32_t residx;
8433	uint64_t res_key, sa_res_key;
8434	uint8_t type;
8435	union ctl_ha_msg persis_io;
8436	int    i;
8437
8438	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8439
8440	retval = CTL_RETVAL_COMPLETE;
8441
8442	softc = control_softc;
8443
8444	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8445	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8446
8447	/*
8448	 * We only support whole-LUN scope.  The scope & type are ignored for
8449	 * register, register and ignore existing key and clear.
8450	 * We sometimes ignore scope and type on preempts too!!
8451	 * Verify reservation type here as well.
8452	 */
8453	type = cdb->scope_type & SPR_TYPE_MASK;
8454	if ((cdb->action == SPRO_RESERVE)
8455	 || (cdb->action == SPRO_RELEASE)) {
8456		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8457			ctl_set_invalid_field(/*ctsio*/ ctsio,
8458					      /*sks_valid*/ 1,
8459					      /*command*/ 1,
8460					      /*field*/ 2,
8461					      /*bit_valid*/ 1,
8462					      /*bit*/ 4);
8463			ctl_done((union ctl_io *)ctsio);
8464			return (CTL_RETVAL_COMPLETE);
8465		}
8466
8467		if (type>8 || type==2 || type==4 || type==0) {
8468			ctl_set_invalid_field(/*ctsio*/ ctsio,
8469					      /*sks_valid*/ 1,
8470					      /*command*/ 1,
8471					      /*field*/ 2,
8472					      /*bit_valid*/ 1,
8473					      /*bit*/ 0);
8474			ctl_done((union ctl_io *)ctsio);
8475			return (CTL_RETVAL_COMPLETE);
8476		}
8477	}
8478
8479	param_len = scsi_4btoul(cdb->length);
8480
8481	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8482		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8483		ctsio->kern_data_len = param_len;
8484		ctsio->kern_total_len = param_len;
8485		ctsio->kern_data_resid = 0;
8486		ctsio->kern_rel_offset = 0;
8487		ctsio->kern_sg_entries = 0;
8488		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8489		ctsio->be_move_done = ctl_config_move_done;
8490		ctl_datamove((union ctl_io *)ctsio);
8491
8492		return (CTL_RETVAL_COMPLETE);
8493	}
8494
8495	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8496
8497	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8498	res_key = scsi_8btou64(param->res_key.key);
8499	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8500
8501	/*
8502	 * Validate the reservation key here except for SPRO_REG_IGNO
8503	 * This must be done for all other service actions
8504	 */
8505	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8506		mtx_lock(&lun->lun_lock);
8507		if (lun->pr_keys[residx] != 0) {
8508		    if (res_key != lun->pr_keys[residx]) {
8509				/*
8510				 * The current key passed in doesn't match
8511				 * the one the initiator previously
8512				 * registered.
8513				 */
8514				mtx_unlock(&lun->lun_lock);
8515				free(ctsio->kern_data_ptr, M_CTL);
8516				ctl_set_reservation_conflict(ctsio);
8517				ctl_done((union ctl_io *)ctsio);
8518				return (CTL_RETVAL_COMPLETE);
8519			}
8520		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8521			/*
8522			 * We are not registered
8523			 */
8524			mtx_unlock(&lun->lun_lock);
8525			free(ctsio->kern_data_ptr, M_CTL);
8526			ctl_set_reservation_conflict(ctsio);
8527			ctl_done((union ctl_io *)ctsio);
8528			return (CTL_RETVAL_COMPLETE);
8529		} else if (res_key != 0) {
8530			/*
8531			 * We are not registered and trying to register but
8532			 * the register key isn't zero.
8533			 */
8534			mtx_unlock(&lun->lun_lock);
8535			free(ctsio->kern_data_ptr, M_CTL);
8536			ctl_set_reservation_conflict(ctsio);
8537			ctl_done((union ctl_io *)ctsio);
8538			return (CTL_RETVAL_COMPLETE);
8539		}
8540		mtx_unlock(&lun->lun_lock);
8541	}
8542
8543	switch (cdb->action & SPRO_ACTION_MASK) {
8544	case SPRO_REGISTER:
8545	case SPRO_REG_IGNO: {
8546
8547#if 0
8548		printf("Registration received\n");
8549#endif
8550
8551		/*
8552		 * We don't support any of these options, as we report in
8553		 * the read capabilities request (see
8554		 * ctl_persistent_reserve_in(), above).
8555		 */
8556		if ((param->flags & SPR_SPEC_I_PT)
8557		 || (param->flags & SPR_ALL_TG_PT)
8558		 || (param->flags & SPR_APTPL)) {
8559			int bit_ptr;
8560
8561			if (param->flags & SPR_APTPL)
8562				bit_ptr = 0;
8563			else if (param->flags & SPR_ALL_TG_PT)
8564				bit_ptr = 2;
8565			else /* SPR_SPEC_I_PT */
8566				bit_ptr = 3;
8567
8568			free(ctsio->kern_data_ptr, M_CTL);
8569			ctl_set_invalid_field(ctsio,
8570					      /*sks_valid*/ 1,
8571					      /*command*/ 0,
8572					      /*field*/ 20,
8573					      /*bit_valid*/ 1,
8574					      /*bit*/ bit_ptr);
8575			ctl_done((union ctl_io *)ctsio);
8576			return (CTL_RETVAL_COMPLETE);
8577		}
8578
8579		mtx_lock(&lun->lun_lock);
8580
8581		/*
8582		 * The initiator wants to clear the
8583		 * key/unregister.
8584		 */
8585		if (sa_res_key == 0) {
8586			if ((res_key == 0
8587			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8588			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8589			  && lun->pr_keys[residx] == 0)) {
8590				mtx_unlock(&lun->lun_lock);
8591				goto done;
8592			}
8593
8594			lun->pr_keys[residx] = 0;
8595			lun->pr_key_count--;
8596
8597			if (residx == lun->pr_res_idx) {
8598				lun->flags &= ~CTL_LUN_PR_RESERVED;
8599				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8600
8601				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8602				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8603				 && lun->pr_key_count) {
8604					/*
8605					 * If the reservation is a registrants
8606					 * only type we need to generate a UA
8607					 * for other registered inits.  The
8608					 * sense code should be RESERVATIONS
8609					 * RELEASED
8610					 */
8611
8612					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8613						if (lun->pr_keys[
8614						    i + persis_offset] == 0)
8615							continue;
8616						lun->pending_ua[i] |=
8617							CTL_UA_RES_RELEASE;
8618					}
8619				}
8620				lun->res_type = 0;
8621			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8622				if (lun->pr_key_count==0) {
8623					lun->flags &= ~CTL_LUN_PR_RESERVED;
8624					lun->res_type = 0;
8625					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8626				}
8627			}
8628			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8629			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8630			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8631			persis_io.pr.pr_info.residx = residx;
8632			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8633			     &persis_io, sizeof(persis_io), 0 )) >
8634			     CTL_HA_STATUS_SUCCESS) {
8635				printf("CTL:Persis Out error returned from "
8636				       "ctl_ha_msg_send %d\n", isc_retval);
8637			}
8638		} else /* sa_res_key != 0 */ {
8639
8640			/*
8641			 * If we aren't registered currently then increment
8642			 * the key count and set the registered flag.
8643			 */
8644			if (lun->pr_keys[residx] == 0)
8645				lun->pr_key_count++;
8646			lun->pr_keys[residx] = sa_res_key;
8647
8648			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8649			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8650			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8651			persis_io.pr.pr_info.residx = residx;
8652			memcpy(persis_io.pr.pr_info.sa_res_key,
8653			       param->serv_act_res_key,
8654			       sizeof(param->serv_act_res_key));
8655			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8656			     &persis_io, sizeof(persis_io), 0)) >
8657			     CTL_HA_STATUS_SUCCESS) {
8658				printf("CTL:Persis Out error returned from "
8659				       "ctl_ha_msg_send %d\n", isc_retval);
8660			}
8661		}
8662		lun->PRGeneration++;
8663		mtx_unlock(&lun->lun_lock);
8664
8665		break;
8666	}
8667	case SPRO_RESERVE:
8668#if 0
8669                printf("Reserve executed type %d\n", type);
8670#endif
8671		mtx_lock(&lun->lun_lock);
8672		if (lun->flags & CTL_LUN_PR_RESERVED) {
8673			/*
8674			 * if this isn't the reservation holder and it's
8675			 * not a "all registrants" type or if the type is
8676			 * different then we have a conflict
8677			 */
8678			if ((lun->pr_res_idx != residx
8679			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8680			 || lun->res_type != type) {
8681				mtx_unlock(&lun->lun_lock);
8682				free(ctsio->kern_data_ptr, M_CTL);
8683				ctl_set_reservation_conflict(ctsio);
8684				ctl_done((union ctl_io *)ctsio);
8685				return (CTL_RETVAL_COMPLETE);
8686			}
8687			mtx_unlock(&lun->lun_lock);
8688		} else /* create a reservation */ {
8689			/*
8690			 * If it's not an "all registrants" type record
8691			 * reservation holder
8692			 */
8693			if (type != SPR_TYPE_WR_EX_AR
8694			 && type != SPR_TYPE_EX_AC_AR)
8695				lun->pr_res_idx = residx; /* Res holder */
8696			else
8697				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8698
8699			lun->flags |= CTL_LUN_PR_RESERVED;
8700			lun->res_type = type;
8701
8702			mtx_unlock(&lun->lun_lock);
8703
8704			/* send msg to other side */
8705			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8706			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8707			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8708			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8709			persis_io.pr.pr_info.res_type = type;
8710			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8711			     &persis_io, sizeof(persis_io), 0)) >
8712			     CTL_HA_STATUS_SUCCESS) {
8713				printf("CTL:Persis Out error returned from "
8714				       "ctl_ha_msg_send %d\n", isc_retval);
8715			}
8716		}
8717		break;
8718
8719	case SPRO_RELEASE:
8720		mtx_lock(&lun->lun_lock);
8721		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8722			/* No reservation exists return good status */
8723			mtx_unlock(&lun->lun_lock);
8724			goto done;
8725		}
8726		/*
8727		 * Is this nexus a reservation holder?
8728		 */
8729		if (lun->pr_res_idx != residx
8730		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8731			/*
8732			 * not a res holder return good status but
8733			 * do nothing
8734			 */
8735			mtx_unlock(&lun->lun_lock);
8736			goto done;
8737		}
8738
8739		if (lun->res_type != type) {
8740			mtx_unlock(&lun->lun_lock);
8741			free(ctsio->kern_data_ptr, M_CTL);
8742			ctl_set_illegal_pr_release(ctsio);
8743			ctl_done((union ctl_io *)ctsio);
8744			return (CTL_RETVAL_COMPLETE);
8745		}
8746
8747		/* okay to release */
8748		lun->flags &= ~CTL_LUN_PR_RESERVED;
8749		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8750		lun->res_type = 0;
8751
8752		/*
8753		 * if this isn't an exclusive access
8754		 * res generate UA for all other
8755		 * registrants.
8756		 */
8757		if (type != SPR_TYPE_EX_AC
8758		 && type != SPR_TYPE_WR_EX) {
8759			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8760				if (i == residx ||
8761				    lun->pr_keys[i + persis_offset] == 0)
8762					continue;
8763				lun->pending_ua[i] |= CTL_UA_RES_RELEASE;
8764			}
8765		}
8766		mtx_unlock(&lun->lun_lock);
8767		/* Send msg to other side */
8768		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8769		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8770		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8771		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8772		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8773			printf("CTL:Persis Out error returned from "
8774			       "ctl_ha_msg_send %d\n", isc_retval);
8775		}
8776		break;
8777
8778	case SPRO_CLEAR:
8779		/* send msg to other side */
8780
8781		mtx_lock(&lun->lun_lock);
8782		lun->flags &= ~CTL_LUN_PR_RESERVED;
8783		lun->res_type = 0;
8784		lun->pr_key_count = 0;
8785		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8786
8787		lun->pr_keys[residx] = 0;
8788
8789		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8790			if (lun->pr_keys[i] != 0) {
8791				if (!persis_offset && i < CTL_MAX_INITIATORS)
8792					lun->pending_ua[i] |=
8793						CTL_UA_RES_PREEMPT;
8794				else if (persis_offset && i >= persis_offset)
8795					lun->pending_ua[i-persis_offset] |=
8796					    CTL_UA_RES_PREEMPT;
8797
8798				lun->pr_keys[i] = 0;
8799			}
8800		lun->PRGeneration++;
8801		mtx_unlock(&lun->lun_lock);
8802		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8803		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8804		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8805		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8806		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8807			printf("CTL:Persis Out error returned from "
8808			       "ctl_ha_msg_send %d\n", isc_retval);
8809		}
8810		break;
8811
8812	case SPRO_PREEMPT: {
8813		int nretval;
8814
8815		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8816					  residx, ctsio, cdb, param);
8817		if (nretval != 0)
8818			return (CTL_RETVAL_COMPLETE);
8819		break;
8820	}
8821	default:
8822		panic("Invalid PR type %x", cdb->action);
8823	}
8824
8825done:
8826	free(ctsio->kern_data_ptr, M_CTL);
8827	ctl_set_success(ctsio);
8828	ctl_done((union ctl_io *)ctsio);
8829
8830	return (retval);
8831}
8832
8833/*
8834 * This routine is for handling a message from the other SC pertaining to
8835 * persistent reserve out. All the error checking will have been done
8836 * so only perorming the action need be done here to keep the two
8837 * in sync.
8838 */
8839static void
8840ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8841{
8842	struct ctl_lun *lun;
8843	struct ctl_softc *softc;
8844	int i;
8845	uint32_t targ_lun;
8846
8847	softc = control_softc;
8848
8849	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8850	lun = softc->ctl_luns[targ_lun];
8851	mtx_lock(&lun->lun_lock);
8852	switch(msg->pr.pr_info.action) {
8853	case CTL_PR_REG_KEY:
8854		if (lun->pr_keys[msg->pr.pr_info.residx] == 0)
8855			lun->pr_key_count++;
8856		lun->pr_keys[msg->pr.pr_info.residx] =
8857		    scsi_8btou64(msg->pr.pr_info.sa_res_key);
8858		lun->PRGeneration++;
8859		break;
8860
8861	case CTL_PR_UNREG_KEY:
8862		lun->pr_keys[msg->pr.pr_info.residx] = 0;
8863		lun->pr_key_count--;
8864
8865		/* XXX Need to see if the reservation has been released */
8866		/* if so do we need to generate UA? */
8867		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8868			lun->flags &= ~CTL_LUN_PR_RESERVED;
8869			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8870
8871			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8872			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8873			 && lun->pr_key_count) {
8874				/*
8875				 * If the reservation is a registrants
8876				 * only type we need to generate a UA
8877				 * for other registered inits.  The
8878				 * sense code should be RESERVATIONS
8879				 * RELEASED
8880				 */
8881
8882				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8883					if (lun->pr_keys[i+
8884					    persis_offset] == 0)
8885						continue;
8886
8887					lun->pending_ua[i] |=
8888						CTL_UA_RES_RELEASE;
8889				}
8890			}
8891			lun->res_type = 0;
8892		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8893			if (lun->pr_key_count==0) {
8894				lun->flags &= ~CTL_LUN_PR_RESERVED;
8895				lun->res_type = 0;
8896				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8897			}
8898		}
8899		lun->PRGeneration++;
8900		break;
8901
8902	case CTL_PR_RESERVE:
8903		lun->flags |= CTL_LUN_PR_RESERVED;
8904		lun->res_type = msg->pr.pr_info.res_type;
8905		lun->pr_res_idx = msg->pr.pr_info.residx;
8906
8907		break;
8908
8909	case CTL_PR_RELEASE:
8910		/*
8911		 * if this isn't an exclusive access res generate UA for all
8912		 * other registrants.
8913		 */
8914		if (lun->res_type != SPR_TYPE_EX_AC
8915		 && lun->res_type != SPR_TYPE_WR_EX) {
8916			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8917				if (lun->pr_keys[i+persis_offset] != 0)
8918					lun->pending_ua[i] |=
8919						CTL_UA_RES_RELEASE;
8920		}
8921
8922		lun->flags &= ~CTL_LUN_PR_RESERVED;
8923		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8924		lun->res_type = 0;
8925		break;
8926
8927	case CTL_PR_PREEMPT:
8928		ctl_pro_preempt_other(lun, msg);
8929		break;
8930	case CTL_PR_CLEAR:
8931		lun->flags &= ~CTL_LUN_PR_RESERVED;
8932		lun->res_type = 0;
8933		lun->pr_key_count = 0;
8934		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8935
8936		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8937			if (lun->pr_keys[i] == 0)
8938				continue;
8939			if (!persis_offset
8940			 && i < CTL_MAX_INITIATORS)
8941				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
8942			else if (persis_offset
8943			      && i >= persis_offset)
8944				lun->pending_ua[i-persis_offset] |=
8945					CTL_UA_RES_PREEMPT;
8946			lun->pr_keys[i] = 0;
8947		}
8948		lun->PRGeneration++;
8949		break;
8950	}
8951
8952	mtx_unlock(&lun->lun_lock);
8953}
8954
8955int
8956ctl_read_write(struct ctl_scsiio *ctsio)
8957{
8958	struct ctl_lun *lun;
8959	struct ctl_lba_len_flags *lbalen;
8960	uint64_t lba;
8961	uint32_t num_blocks;
8962	int flags, retval;
8963	int isread;
8964
8965	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8966
8967	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8968
8969	flags = 0;
8970	retval = CTL_RETVAL_COMPLETE;
8971
8972	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8973	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8974	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8975		uint32_t residx;
8976
8977		/*
8978		 * XXX KDM need a lock here.
8979		 */
8980		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8981		if ((lun->res_type == SPR_TYPE_EX_AC
8982		  && residx != lun->pr_res_idx)
8983		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8984		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8985		  && lun->pr_keys[residx] == 0)) {
8986			ctl_set_reservation_conflict(ctsio);
8987			ctl_done((union ctl_io *)ctsio);
8988			return (CTL_RETVAL_COMPLETE);
8989	        }
8990	}
8991
8992	switch (ctsio->cdb[0]) {
8993	case READ_6:
8994	case WRITE_6: {
8995		struct scsi_rw_6 *cdb;
8996
8997		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8998
8999		lba = scsi_3btoul(cdb->addr);
9000		/* only 5 bits are valid in the most significant address byte */
9001		lba &= 0x1fffff;
9002		num_blocks = cdb->length;
9003		/*
9004		 * This is correct according to SBC-2.
9005		 */
9006		if (num_blocks == 0)
9007			num_blocks = 256;
9008		break;
9009	}
9010	case READ_10:
9011	case WRITE_10: {
9012		struct scsi_rw_10 *cdb;
9013
9014		cdb = (struct scsi_rw_10 *)ctsio->cdb;
9015		if (cdb->byte2 & SRW10_FUA)
9016			flags |= CTL_LLF_FUA;
9017		if (cdb->byte2 & SRW10_DPO)
9018			flags |= CTL_LLF_DPO;
9019		lba = scsi_4btoul(cdb->addr);
9020		num_blocks = scsi_2btoul(cdb->length);
9021		break;
9022	}
9023	case WRITE_VERIFY_10: {
9024		struct scsi_write_verify_10 *cdb;
9025
9026		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9027		flags |= CTL_LLF_FUA;
9028		if (cdb->byte2 & SWV_DPO)
9029			flags |= CTL_LLF_DPO;
9030		lba = scsi_4btoul(cdb->addr);
9031		num_blocks = scsi_2btoul(cdb->length);
9032		break;
9033	}
9034	case READ_12:
9035	case WRITE_12: {
9036		struct scsi_rw_12 *cdb;
9037
9038		cdb = (struct scsi_rw_12 *)ctsio->cdb;
9039		if (cdb->byte2 & SRW12_FUA)
9040			flags |= CTL_LLF_FUA;
9041		if (cdb->byte2 & SRW12_DPO)
9042			flags |= CTL_LLF_DPO;
9043		lba = scsi_4btoul(cdb->addr);
9044		num_blocks = scsi_4btoul(cdb->length);
9045		break;
9046	}
9047	case WRITE_VERIFY_12: {
9048		struct scsi_write_verify_12 *cdb;
9049
9050		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9051		flags |= CTL_LLF_FUA;
9052		if (cdb->byte2 & SWV_DPO)
9053			flags |= CTL_LLF_DPO;
9054		lba = scsi_4btoul(cdb->addr);
9055		num_blocks = scsi_4btoul(cdb->length);
9056		break;
9057	}
9058	case READ_16:
9059	case WRITE_16: {
9060		struct scsi_rw_16 *cdb;
9061
9062		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9063		if (cdb->byte2 & SRW12_FUA)
9064			flags |= CTL_LLF_FUA;
9065		if (cdb->byte2 & SRW12_DPO)
9066			flags |= CTL_LLF_DPO;
9067		lba = scsi_8btou64(cdb->addr);
9068		num_blocks = scsi_4btoul(cdb->length);
9069		break;
9070	}
9071	case WRITE_ATOMIC_16: {
9072		struct scsi_rw_16 *cdb;
9073
9074		if (lun->be_lun->atomicblock == 0) {
9075			ctl_set_invalid_opcode(ctsio);
9076			ctl_done((union ctl_io *)ctsio);
9077			return (CTL_RETVAL_COMPLETE);
9078		}
9079
9080		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9081		if (cdb->byte2 & SRW12_FUA)
9082			flags |= CTL_LLF_FUA;
9083		if (cdb->byte2 & SRW12_DPO)
9084			flags |= CTL_LLF_DPO;
9085		lba = scsi_8btou64(cdb->addr);
9086		num_blocks = scsi_4btoul(cdb->length);
9087		if (num_blocks > lun->be_lun->atomicblock) {
9088			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
9089			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
9090			    /*bit*/ 0);
9091			ctl_done((union ctl_io *)ctsio);
9092			return (CTL_RETVAL_COMPLETE);
9093		}
9094		break;
9095	}
9096	case WRITE_VERIFY_16: {
9097		struct scsi_write_verify_16 *cdb;
9098
9099		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9100		flags |= CTL_LLF_FUA;
9101		if (cdb->byte2 & SWV_DPO)
9102			flags |= CTL_LLF_DPO;
9103		lba = scsi_8btou64(cdb->addr);
9104		num_blocks = scsi_4btoul(cdb->length);
9105		break;
9106	}
9107	default:
9108		/*
9109		 * We got a command we don't support.  This shouldn't
9110		 * happen, commands should be filtered out above us.
9111		 */
9112		ctl_set_invalid_opcode(ctsio);
9113		ctl_done((union ctl_io *)ctsio);
9114
9115		return (CTL_RETVAL_COMPLETE);
9116		break; /* NOTREACHED */
9117	}
9118
9119	/*
9120	 * The first check is to make sure we're in bounds, the second
9121	 * check is to catch wrap-around problems.  If the lba + num blocks
9122	 * is less than the lba, then we've wrapped around and the block
9123	 * range is invalid anyway.
9124	 */
9125	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9126	 || ((lba + num_blocks) < lba)) {
9127		ctl_set_lba_out_of_range(ctsio);
9128		ctl_done((union ctl_io *)ctsio);
9129		return (CTL_RETVAL_COMPLETE);
9130	}
9131
9132	/*
9133	 * According to SBC-3, a transfer length of 0 is not an error.
9134	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9135	 * translates to 256 blocks for those commands.
9136	 */
9137	if (num_blocks == 0) {
9138		ctl_set_success(ctsio);
9139		ctl_done((union ctl_io *)ctsio);
9140		return (CTL_RETVAL_COMPLETE);
9141	}
9142
9143	/* Set FUA and/or DPO if caches are disabled. */
9144	if (isread) {
9145		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9146		    SCP_RCD) != 0)
9147			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9148	} else {
9149		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9150		    SCP_WCE) == 0)
9151			flags |= CTL_LLF_FUA;
9152	}
9153
9154	lbalen = (struct ctl_lba_len_flags *)
9155	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9156	lbalen->lba = lba;
9157	lbalen->len = num_blocks;
9158	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9159
9160	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9161	ctsio->kern_rel_offset = 0;
9162
9163	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9164
9165	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9166
9167	return (retval);
9168}
9169
9170static int
9171ctl_cnw_cont(union ctl_io *io)
9172{
9173	struct ctl_scsiio *ctsio;
9174	struct ctl_lun *lun;
9175	struct ctl_lba_len_flags *lbalen;
9176	int retval;
9177
9178	ctsio = &io->scsiio;
9179	ctsio->io_hdr.status = CTL_STATUS_NONE;
9180	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9181	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9182	lbalen = (struct ctl_lba_len_flags *)
9183	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9184	lbalen->flags &= ~CTL_LLF_COMPARE;
9185	lbalen->flags |= CTL_LLF_WRITE;
9186
9187	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9188	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9189	return (retval);
9190}
9191
9192int
9193ctl_cnw(struct ctl_scsiio *ctsio)
9194{
9195	struct ctl_lun *lun;
9196	struct ctl_lba_len_flags *lbalen;
9197	uint64_t lba;
9198	uint32_t num_blocks;
9199	int flags, retval;
9200
9201	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9202
9203	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9204
9205	flags = 0;
9206	retval = CTL_RETVAL_COMPLETE;
9207
9208	switch (ctsio->cdb[0]) {
9209	case COMPARE_AND_WRITE: {
9210		struct scsi_compare_and_write *cdb;
9211
9212		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9213		if (cdb->byte2 & SRW10_FUA)
9214			flags |= CTL_LLF_FUA;
9215		if (cdb->byte2 & SRW10_DPO)
9216			flags |= CTL_LLF_DPO;
9217		lba = scsi_8btou64(cdb->addr);
9218		num_blocks = cdb->length;
9219		break;
9220	}
9221	default:
9222		/*
9223		 * We got a command we don't support.  This shouldn't
9224		 * happen, commands should be filtered out above us.
9225		 */
9226		ctl_set_invalid_opcode(ctsio);
9227		ctl_done((union ctl_io *)ctsio);
9228
9229		return (CTL_RETVAL_COMPLETE);
9230		break; /* NOTREACHED */
9231	}
9232
9233	/*
9234	 * The first check is to make sure we're in bounds, the second
9235	 * check is to catch wrap-around problems.  If the lba + num blocks
9236	 * is less than the lba, then we've wrapped around and the block
9237	 * range is invalid anyway.
9238	 */
9239	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9240	 || ((lba + num_blocks) < lba)) {
9241		ctl_set_lba_out_of_range(ctsio);
9242		ctl_done((union ctl_io *)ctsio);
9243		return (CTL_RETVAL_COMPLETE);
9244	}
9245
9246	/*
9247	 * According to SBC-3, a transfer length of 0 is not an error.
9248	 */
9249	if (num_blocks == 0) {
9250		ctl_set_success(ctsio);
9251		ctl_done((union ctl_io *)ctsio);
9252		return (CTL_RETVAL_COMPLETE);
9253	}
9254
9255	/* Set FUA if write cache is disabled. */
9256	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9257	    SCP_WCE) == 0)
9258		flags |= CTL_LLF_FUA;
9259
9260	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9261	ctsio->kern_rel_offset = 0;
9262
9263	/*
9264	 * Set the IO_CONT flag, so that if this I/O gets passed to
9265	 * ctl_data_submit_done(), it'll get passed back to
9266	 * ctl_ctl_cnw_cont() for further processing.
9267	 */
9268	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9269	ctsio->io_cont = ctl_cnw_cont;
9270
9271	lbalen = (struct ctl_lba_len_flags *)
9272	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9273	lbalen->lba = lba;
9274	lbalen->len = num_blocks;
9275	lbalen->flags = CTL_LLF_COMPARE | flags;
9276
9277	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9278	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9279	return (retval);
9280}
9281
9282int
9283ctl_verify(struct ctl_scsiio *ctsio)
9284{
9285	struct ctl_lun *lun;
9286	struct ctl_lba_len_flags *lbalen;
9287	uint64_t lba;
9288	uint32_t num_blocks;
9289	int bytchk, flags;
9290	int retval;
9291
9292	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9293
9294	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9295
9296	bytchk = 0;
9297	flags = CTL_LLF_FUA;
9298	retval = CTL_RETVAL_COMPLETE;
9299
9300	switch (ctsio->cdb[0]) {
9301	case VERIFY_10: {
9302		struct scsi_verify_10 *cdb;
9303
9304		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9305		if (cdb->byte2 & SVFY_BYTCHK)
9306			bytchk = 1;
9307		if (cdb->byte2 & SVFY_DPO)
9308			flags |= CTL_LLF_DPO;
9309		lba = scsi_4btoul(cdb->addr);
9310		num_blocks = scsi_2btoul(cdb->length);
9311		break;
9312	}
9313	case VERIFY_12: {
9314		struct scsi_verify_12 *cdb;
9315
9316		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9317		if (cdb->byte2 & SVFY_BYTCHK)
9318			bytchk = 1;
9319		if (cdb->byte2 & SVFY_DPO)
9320			flags |= CTL_LLF_DPO;
9321		lba = scsi_4btoul(cdb->addr);
9322		num_blocks = scsi_4btoul(cdb->length);
9323		break;
9324	}
9325	case VERIFY_16: {
9326		struct scsi_rw_16 *cdb;
9327
9328		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9329		if (cdb->byte2 & SVFY_BYTCHK)
9330			bytchk = 1;
9331		if (cdb->byte2 & SVFY_DPO)
9332			flags |= CTL_LLF_DPO;
9333		lba = scsi_8btou64(cdb->addr);
9334		num_blocks = scsi_4btoul(cdb->length);
9335		break;
9336	}
9337	default:
9338		/*
9339		 * We got a command we don't support.  This shouldn't
9340		 * happen, commands should be filtered out above us.
9341		 */
9342		ctl_set_invalid_opcode(ctsio);
9343		ctl_done((union ctl_io *)ctsio);
9344		return (CTL_RETVAL_COMPLETE);
9345	}
9346
9347	/*
9348	 * The first check is to make sure we're in bounds, the second
9349	 * check is to catch wrap-around problems.  If the lba + num blocks
9350	 * is less than the lba, then we've wrapped around and the block
9351	 * range is invalid anyway.
9352	 */
9353	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9354	 || ((lba + num_blocks) < lba)) {
9355		ctl_set_lba_out_of_range(ctsio);
9356		ctl_done((union ctl_io *)ctsio);
9357		return (CTL_RETVAL_COMPLETE);
9358	}
9359
9360	/*
9361	 * According to SBC-3, a transfer length of 0 is not an error.
9362	 */
9363	if (num_blocks == 0) {
9364		ctl_set_success(ctsio);
9365		ctl_done((union ctl_io *)ctsio);
9366		return (CTL_RETVAL_COMPLETE);
9367	}
9368
9369	lbalen = (struct ctl_lba_len_flags *)
9370	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9371	lbalen->lba = lba;
9372	lbalen->len = num_blocks;
9373	if (bytchk) {
9374		lbalen->flags = CTL_LLF_COMPARE | flags;
9375		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9376	} else {
9377		lbalen->flags = CTL_LLF_VERIFY | flags;
9378		ctsio->kern_total_len = 0;
9379	}
9380	ctsio->kern_rel_offset = 0;
9381
9382	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9383	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9384	return (retval);
9385}
9386
9387int
9388ctl_report_luns(struct ctl_scsiio *ctsio)
9389{
9390	struct scsi_report_luns *cdb;
9391	struct scsi_report_luns_data *lun_data;
9392	struct ctl_lun *lun, *request_lun;
9393	int num_luns, retval;
9394	uint32_t alloc_len, lun_datalen;
9395	int num_filled, well_known;
9396	uint32_t initidx, targ_lun_id, lun_id;
9397
9398	retval = CTL_RETVAL_COMPLETE;
9399	well_known = 0;
9400
9401	cdb = (struct scsi_report_luns *)ctsio->cdb;
9402
9403	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9404
9405	mtx_lock(&control_softc->ctl_lock);
9406	num_luns = control_softc->num_luns;
9407	mtx_unlock(&control_softc->ctl_lock);
9408
9409	switch (cdb->select_report) {
9410	case RPL_REPORT_DEFAULT:
9411	case RPL_REPORT_ALL:
9412		break;
9413	case RPL_REPORT_WELLKNOWN:
9414		well_known = 1;
9415		num_luns = 0;
9416		break;
9417	default:
9418		ctl_set_invalid_field(ctsio,
9419				      /*sks_valid*/ 1,
9420				      /*command*/ 1,
9421				      /*field*/ 2,
9422				      /*bit_valid*/ 0,
9423				      /*bit*/ 0);
9424		ctl_done((union ctl_io *)ctsio);
9425		return (retval);
9426		break; /* NOTREACHED */
9427	}
9428
9429	alloc_len = scsi_4btoul(cdb->length);
9430	/*
9431	 * The initiator has to allocate at least 16 bytes for this request,
9432	 * so he can at least get the header and the first LUN.  Otherwise
9433	 * we reject the request (per SPC-3 rev 14, section 6.21).
9434	 */
9435	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9436	    sizeof(struct scsi_report_luns_lundata))) {
9437		ctl_set_invalid_field(ctsio,
9438				      /*sks_valid*/ 1,
9439				      /*command*/ 1,
9440				      /*field*/ 6,
9441				      /*bit_valid*/ 0,
9442				      /*bit*/ 0);
9443		ctl_done((union ctl_io *)ctsio);
9444		return (retval);
9445	}
9446
9447	request_lun = (struct ctl_lun *)
9448		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9449
9450	lun_datalen = sizeof(*lun_data) +
9451		(num_luns * sizeof(struct scsi_report_luns_lundata));
9452
9453	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9454	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9455	ctsio->kern_sg_entries = 0;
9456
9457	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9458
9459	mtx_lock(&control_softc->ctl_lock);
9460	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9461		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9462		if (lun_id >= CTL_MAX_LUNS)
9463			continue;
9464		lun = control_softc->ctl_luns[lun_id];
9465		if (lun == NULL)
9466			continue;
9467
9468		if (targ_lun_id <= 0xff) {
9469			/*
9470			 * Peripheral addressing method, bus number 0.
9471			 */
9472			lun_data->luns[num_filled].lundata[0] =
9473				RPL_LUNDATA_ATYP_PERIPH;
9474			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9475			num_filled++;
9476		} else if (targ_lun_id <= 0x3fff) {
9477			/*
9478			 * Flat addressing method.
9479			 */
9480			lun_data->luns[num_filled].lundata[0] =
9481				RPL_LUNDATA_ATYP_FLAT |
9482				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9483#ifdef OLDCTLHEADERS
9484				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9485				(targ_lun_id & SRLD_BUS_LUN_MASK);
9486#endif
9487			lun_data->luns[num_filled].lundata[1] =
9488#ifdef OLDCTLHEADERS
9489				targ_lun_id >> SRLD_BUS_LUN_BITS;
9490#endif
9491				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9492			num_filled++;
9493		} else {
9494			printf("ctl_report_luns: bogus LUN number %jd, "
9495			       "skipping\n", (intmax_t)targ_lun_id);
9496		}
9497		/*
9498		 * According to SPC-3, rev 14 section 6.21:
9499		 *
9500		 * "The execution of a REPORT LUNS command to any valid and
9501		 * installed logical unit shall clear the REPORTED LUNS DATA
9502		 * HAS CHANGED unit attention condition for all logical
9503		 * units of that target with respect to the requesting
9504		 * initiator. A valid and installed logical unit is one
9505		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9506		 * INQUIRY data (see 6.4.2)."
9507		 *
9508		 * If request_lun is NULL, the LUN this report luns command
9509		 * was issued to is either disabled or doesn't exist. In that
9510		 * case, we shouldn't clear any pending lun change unit
9511		 * attention.
9512		 */
9513		if (request_lun != NULL) {
9514			mtx_lock(&lun->lun_lock);
9515			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9516			mtx_unlock(&lun->lun_lock);
9517		}
9518	}
9519	mtx_unlock(&control_softc->ctl_lock);
9520
9521	/*
9522	 * It's quite possible that we've returned fewer LUNs than we allocated
9523	 * space for.  Trim it.
9524	 */
9525	lun_datalen = sizeof(*lun_data) +
9526		(num_filled * sizeof(struct scsi_report_luns_lundata));
9527
9528	if (lun_datalen < alloc_len) {
9529		ctsio->residual = alloc_len - lun_datalen;
9530		ctsio->kern_data_len = lun_datalen;
9531		ctsio->kern_total_len = lun_datalen;
9532	} else {
9533		ctsio->residual = 0;
9534		ctsio->kern_data_len = alloc_len;
9535		ctsio->kern_total_len = alloc_len;
9536	}
9537	ctsio->kern_data_resid = 0;
9538	ctsio->kern_rel_offset = 0;
9539	ctsio->kern_sg_entries = 0;
9540
9541	/*
9542	 * We set this to the actual data length, regardless of how much
9543	 * space we actually have to return results.  If the user looks at
9544	 * this value, he'll know whether or not he allocated enough space
9545	 * and reissue the command if necessary.  We don't support well
9546	 * known logical units, so if the user asks for that, return none.
9547	 */
9548	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9549
9550	/*
9551	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9552	 * this request.
9553	 */
9554	ctsio->scsi_status = SCSI_STATUS_OK;
9555
9556	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9557	ctsio->be_move_done = ctl_config_move_done;
9558	ctl_datamove((union ctl_io *)ctsio);
9559
9560	return (retval);
9561}
9562
9563int
9564ctl_request_sense(struct ctl_scsiio *ctsio)
9565{
9566	struct scsi_request_sense *cdb;
9567	struct scsi_sense_data *sense_ptr;
9568	struct ctl_lun *lun;
9569	uint32_t initidx;
9570	int have_error;
9571	scsi_sense_data_type sense_format;
9572
9573	cdb = (struct scsi_request_sense *)ctsio->cdb;
9574
9575	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9576
9577	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9578
9579	/*
9580	 * Determine which sense format the user wants.
9581	 */
9582	if (cdb->byte2 & SRS_DESC)
9583		sense_format = SSD_TYPE_DESC;
9584	else
9585		sense_format = SSD_TYPE_FIXED;
9586
9587	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9588	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9589	ctsio->kern_sg_entries = 0;
9590
9591	/*
9592	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9593	 * larger than the largest allowed value for the length field in the
9594	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9595	 */
9596	ctsio->residual = 0;
9597	ctsio->kern_data_len = cdb->length;
9598	ctsio->kern_total_len = cdb->length;
9599
9600	ctsio->kern_data_resid = 0;
9601	ctsio->kern_rel_offset = 0;
9602	ctsio->kern_sg_entries = 0;
9603
9604	/*
9605	 * If we don't have a LUN, we don't have any pending sense.
9606	 */
9607	if (lun == NULL)
9608		goto no_sense;
9609
9610	have_error = 0;
9611	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9612	/*
9613	 * Check for pending sense, and then for pending unit attentions.
9614	 * Pending sense gets returned first, then pending unit attentions.
9615	 */
9616	mtx_lock(&lun->lun_lock);
9617#ifdef CTL_WITH_CA
9618	if (ctl_is_set(lun->have_ca, initidx)) {
9619		scsi_sense_data_type stored_format;
9620
9621		/*
9622		 * Check to see which sense format was used for the stored
9623		 * sense data.
9624		 */
9625		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9626
9627		/*
9628		 * If the user requested a different sense format than the
9629		 * one we stored, then we need to convert it to the other
9630		 * format.  If we're going from descriptor to fixed format
9631		 * sense data, we may lose things in translation, depending
9632		 * on what options were used.
9633		 *
9634		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9635		 * for some reason we'll just copy it out as-is.
9636		 */
9637		if ((stored_format == SSD_TYPE_FIXED)
9638		 && (sense_format == SSD_TYPE_DESC))
9639			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9640			    &lun->pending_sense[initidx],
9641			    (struct scsi_sense_data_desc *)sense_ptr);
9642		else if ((stored_format == SSD_TYPE_DESC)
9643		      && (sense_format == SSD_TYPE_FIXED))
9644			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9645			    &lun->pending_sense[initidx],
9646			    (struct scsi_sense_data_fixed *)sense_ptr);
9647		else
9648			memcpy(sense_ptr, &lun->pending_sense[initidx],
9649			       ctl_min(sizeof(*sense_ptr),
9650			       sizeof(lun->pending_sense[initidx])));
9651
9652		ctl_clear_mask(lun->have_ca, initidx);
9653		have_error = 1;
9654	} else
9655#endif
9656	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9657		ctl_ua_type ua_type;
9658
9659		ua_type = ctl_build_ua(&lun->pending_ua[initidx],
9660				       sense_ptr, sense_format);
9661		if (ua_type != CTL_UA_NONE)
9662			have_error = 1;
9663	}
9664	mtx_unlock(&lun->lun_lock);
9665
9666	/*
9667	 * We already have a pending error, return it.
9668	 */
9669	if (have_error != 0) {
9670		/*
9671		 * We report the SCSI status as OK, since the status of the
9672		 * request sense command itself is OK.
9673		 */
9674		ctsio->scsi_status = SCSI_STATUS_OK;
9675
9676		/*
9677		 * We report 0 for the sense length, because we aren't doing
9678		 * autosense in this case.  We're reporting sense as
9679		 * parameter data.
9680		 */
9681		ctsio->sense_len = 0;
9682		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9683		ctsio->be_move_done = ctl_config_move_done;
9684		ctl_datamove((union ctl_io *)ctsio);
9685
9686		return (CTL_RETVAL_COMPLETE);
9687	}
9688
9689no_sense:
9690
9691	/*
9692	 * No sense information to report, so we report that everything is
9693	 * okay.
9694	 */
9695	ctl_set_sense_data(sense_ptr,
9696			   lun,
9697			   sense_format,
9698			   /*current_error*/ 1,
9699			   /*sense_key*/ SSD_KEY_NO_SENSE,
9700			   /*asc*/ 0x00,
9701			   /*ascq*/ 0x00,
9702			   SSD_ELEM_NONE);
9703
9704	ctsio->scsi_status = SCSI_STATUS_OK;
9705
9706	/*
9707	 * We report 0 for the sense length, because we aren't doing
9708	 * autosense in this case.  We're reporting sense as parameter data.
9709	 */
9710	ctsio->sense_len = 0;
9711	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9712	ctsio->be_move_done = ctl_config_move_done;
9713	ctl_datamove((union ctl_io *)ctsio);
9714
9715	return (CTL_RETVAL_COMPLETE);
9716}
9717
9718int
9719ctl_tur(struct ctl_scsiio *ctsio)
9720{
9721	struct ctl_lun *lun;
9722
9723	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9724
9725	CTL_DEBUG_PRINT(("ctl_tur\n"));
9726
9727	if (lun == NULL)
9728		return (EINVAL);
9729
9730	ctsio->scsi_status = SCSI_STATUS_OK;
9731	ctsio->io_hdr.status = CTL_SUCCESS;
9732
9733	ctl_done((union ctl_io *)ctsio);
9734
9735	return (CTL_RETVAL_COMPLETE);
9736}
9737
9738#ifdef notyet
9739static int
9740ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9741{
9742
9743}
9744#endif
9745
9746static int
9747ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9748{
9749	struct scsi_vpd_supported_pages *pages;
9750	int sup_page_size;
9751	struct ctl_lun *lun;
9752
9753	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9754
9755	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9756	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9757	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9758	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9759	ctsio->kern_sg_entries = 0;
9760
9761	if (sup_page_size < alloc_len) {
9762		ctsio->residual = alloc_len - sup_page_size;
9763		ctsio->kern_data_len = sup_page_size;
9764		ctsio->kern_total_len = sup_page_size;
9765	} else {
9766		ctsio->residual = 0;
9767		ctsio->kern_data_len = alloc_len;
9768		ctsio->kern_total_len = alloc_len;
9769	}
9770	ctsio->kern_data_resid = 0;
9771	ctsio->kern_rel_offset = 0;
9772	ctsio->kern_sg_entries = 0;
9773
9774	/*
9775	 * The control device is always connected.  The disk device, on the
9776	 * other hand, may not be online all the time.  Need to change this
9777	 * to figure out whether the disk device is actually online or not.
9778	 */
9779	if (lun != NULL)
9780		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9781				lun->be_lun->lun_type;
9782	else
9783		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9784
9785	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9786	/* Supported VPD pages */
9787	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9788	/* Serial Number */
9789	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9790	/* Device Identification */
9791	pages->page_list[2] = SVPD_DEVICE_ID;
9792	/* Extended INQUIRY Data */
9793	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9794	/* Mode Page Policy */
9795	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9796	/* SCSI Ports */
9797	pages->page_list[5] = SVPD_SCSI_PORTS;
9798	/* Third-party Copy */
9799	pages->page_list[6] = SVPD_SCSI_TPC;
9800	/* Block limits */
9801	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9802	/* Block Device Characteristics */
9803	pages->page_list[8] = SVPD_BDC;
9804	/* Logical Block Provisioning */
9805	pages->page_list[9] = SVPD_LBP;
9806
9807	ctsio->scsi_status = SCSI_STATUS_OK;
9808
9809	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9810	ctsio->be_move_done = ctl_config_move_done;
9811	ctl_datamove((union ctl_io *)ctsio);
9812
9813	return (CTL_RETVAL_COMPLETE);
9814}
9815
9816static int
9817ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9818{
9819	struct scsi_vpd_unit_serial_number *sn_ptr;
9820	struct ctl_lun *lun;
9821
9822	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9823
9824	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9825	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9826	ctsio->kern_sg_entries = 0;
9827
9828	if (sizeof(*sn_ptr) < alloc_len) {
9829		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9830		ctsio->kern_data_len = sizeof(*sn_ptr);
9831		ctsio->kern_total_len = sizeof(*sn_ptr);
9832	} else {
9833		ctsio->residual = 0;
9834		ctsio->kern_data_len = alloc_len;
9835		ctsio->kern_total_len = alloc_len;
9836	}
9837	ctsio->kern_data_resid = 0;
9838	ctsio->kern_rel_offset = 0;
9839	ctsio->kern_sg_entries = 0;
9840
9841	/*
9842	 * The control device is always connected.  The disk device, on the
9843	 * other hand, may not be online all the time.  Need to change this
9844	 * to figure out whether the disk device is actually online or not.
9845	 */
9846	if (lun != NULL)
9847		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9848				  lun->be_lun->lun_type;
9849	else
9850		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9851
9852	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9853	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9854	/*
9855	 * If we don't have a LUN, we just leave the serial number as
9856	 * all spaces.
9857	 */
9858	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9859	if (lun != NULL) {
9860		strncpy((char *)sn_ptr->serial_num,
9861			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9862	}
9863	ctsio->scsi_status = SCSI_STATUS_OK;
9864
9865	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9866	ctsio->be_move_done = ctl_config_move_done;
9867	ctl_datamove((union ctl_io *)ctsio);
9868
9869	return (CTL_RETVAL_COMPLETE);
9870}
9871
9872
9873static int
9874ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9875{
9876	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9877	struct ctl_lun *lun;
9878	int data_len;
9879
9880	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9881
9882	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9883	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9884	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9885	ctsio->kern_sg_entries = 0;
9886
9887	if (data_len < alloc_len) {
9888		ctsio->residual = alloc_len - data_len;
9889		ctsio->kern_data_len = data_len;
9890		ctsio->kern_total_len = data_len;
9891	} else {
9892		ctsio->residual = 0;
9893		ctsio->kern_data_len = alloc_len;
9894		ctsio->kern_total_len = alloc_len;
9895	}
9896	ctsio->kern_data_resid = 0;
9897	ctsio->kern_rel_offset = 0;
9898	ctsio->kern_sg_entries = 0;
9899
9900	/*
9901	 * The control device is always connected.  The disk device, on the
9902	 * other hand, may not be online all the time.
9903	 */
9904	if (lun != NULL)
9905		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9906				     lun->be_lun->lun_type;
9907	else
9908		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9909	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9910	eid_ptr->page_length = data_len - 4;
9911	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9912	eid_ptr->flags3 = SVPD_EID_V_SUP;
9913
9914	ctsio->scsi_status = SCSI_STATUS_OK;
9915	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9916	ctsio->be_move_done = ctl_config_move_done;
9917	ctl_datamove((union ctl_io *)ctsio);
9918
9919	return (CTL_RETVAL_COMPLETE);
9920}
9921
9922static int
9923ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9924{
9925	struct scsi_vpd_mode_page_policy *mpp_ptr;
9926	struct ctl_lun *lun;
9927	int data_len;
9928
9929	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9930
9931	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9932	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9933
9934	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9935	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9936	ctsio->kern_sg_entries = 0;
9937
9938	if (data_len < alloc_len) {
9939		ctsio->residual = alloc_len - data_len;
9940		ctsio->kern_data_len = data_len;
9941		ctsio->kern_total_len = data_len;
9942	} else {
9943		ctsio->residual = 0;
9944		ctsio->kern_data_len = alloc_len;
9945		ctsio->kern_total_len = alloc_len;
9946	}
9947	ctsio->kern_data_resid = 0;
9948	ctsio->kern_rel_offset = 0;
9949	ctsio->kern_sg_entries = 0;
9950
9951	/*
9952	 * The control device is always connected.  The disk device, on the
9953	 * other hand, may not be online all the time.
9954	 */
9955	if (lun != NULL)
9956		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9957				     lun->be_lun->lun_type;
9958	else
9959		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9960	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9961	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9962	mpp_ptr->descr[0].page_code = 0x3f;
9963	mpp_ptr->descr[0].subpage_code = 0xff;
9964	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9965
9966	ctsio->scsi_status = SCSI_STATUS_OK;
9967	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9968	ctsio->be_move_done = ctl_config_move_done;
9969	ctl_datamove((union ctl_io *)ctsio);
9970
9971	return (CTL_RETVAL_COMPLETE);
9972}
9973
9974static int
9975ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9976{
9977	struct scsi_vpd_device_id *devid_ptr;
9978	struct scsi_vpd_id_descriptor *desc;
9979	struct ctl_softc *ctl_softc;
9980	struct ctl_lun *lun;
9981	struct ctl_port *port;
9982	int data_len;
9983	uint8_t proto;
9984
9985	ctl_softc = control_softc;
9986
9987	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9988	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9989
9990	data_len = sizeof(struct scsi_vpd_device_id) +
9991	    sizeof(struct scsi_vpd_id_descriptor) +
9992		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9993	    sizeof(struct scsi_vpd_id_descriptor) +
9994		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9995	if (lun && lun->lun_devid)
9996		data_len += lun->lun_devid->len;
9997	if (port->port_devid)
9998		data_len += port->port_devid->len;
9999	if (port->target_devid)
10000		data_len += port->target_devid->len;
10001
10002	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10003	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
10004	ctsio->kern_sg_entries = 0;
10005
10006	if (data_len < alloc_len) {
10007		ctsio->residual = alloc_len - data_len;
10008		ctsio->kern_data_len = data_len;
10009		ctsio->kern_total_len = data_len;
10010	} else {
10011		ctsio->residual = 0;
10012		ctsio->kern_data_len = alloc_len;
10013		ctsio->kern_total_len = alloc_len;
10014	}
10015	ctsio->kern_data_resid = 0;
10016	ctsio->kern_rel_offset = 0;
10017	ctsio->kern_sg_entries = 0;
10018
10019	/*
10020	 * The control device is always connected.  The disk device, on the
10021	 * other hand, may not be online all the time.
10022	 */
10023	if (lun != NULL)
10024		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10025				     lun->be_lun->lun_type;
10026	else
10027		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10028	devid_ptr->page_code = SVPD_DEVICE_ID;
10029	scsi_ulto2b(data_len - 4, devid_ptr->length);
10030
10031	if (port->port_type == CTL_PORT_FC)
10032		proto = SCSI_PROTO_FC << 4;
10033	else if (port->port_type == CTL_PORT_ISCSI)
10034		proto = SCSI_PROTO_ISCSI << 4;
10035	else
10036		proto = SCSI_PROTO_SPI << 4;
10037	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
10038
10039	/*
10040	 * We're using a LUN association here.  i.e., this device ID is a
10041	 * per-LUN identifier.
10042	 */
10043	if (lun && lun->lun_devid) {
10044		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
10045		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10046		    lun->lun_devid->len);
10047	}
10048
10049	/*
10050	 * This is for the WWPN which is a port association.
10051	 */
10052	if (port->port_devid) {
10053		memcpy(desc, port->port_devid->data, port->port_devid->len);
10054		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10055		    port->port_devid->len);
10056	}
10057
10058	/*
10059	 * This is for the Relative Target Port(type 4h) identifier
10060	 */
10061	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10062	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10063	    SVPD_ID_TYPE_RELTARG;
10064	desc->length = 4;
10065	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
10066	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10067	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
10068
10069	/*
10070	 * This is for the Target Port Group(type 5h) identifier
10071	 */
10072	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10073	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10074	    SVPD_ID_TYPE_TPORTGRP;
10075	desc->length = 4;
10076	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
10077	    &desc->identifier[2]);
10078	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10079	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10080
10081	/*
10082	 * This is for the Target identifier
10083	 */
10084	if (port->target_devid) {
10085		memcpy(desc, port->target_devid->data, port->target_devid->len);
10086	}
10087
10088	ctsio->scsi_status = SCSI_STATUS_OK;
10089	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10090	ctsio->be_move_done = ctl_config_move_done;
10091	ctl_datamove((union ctl_io *)ctsio);
10092
10093	return (CTL_RETVAL_COMPLETE);
10094}
10095
10096static int
10097ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10098{
10099	struct ctl_softc *softc = control_softc;
10100	struct scsi_vpd_scsi_ports *sp;
10101	struct scsi_vpd_port_designation *pd;
10102	struct scsi_vpd_port_designation_cont *pdc;
10103	struct ctl_lun *lun;
10104	struct ctl_port *port;
10105	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10106	int num_target_port_groups, single;
10107
10108	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10109
10110	single = ctl_is_single;
10111	if (single)
10112		num_target_port_groups = 1;
10113	else
10114		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10115	num_target_ports = 0;
10116	iid_len = 0;
10117	id_len = 0;
10118	mtx_lock(&softc->ctl_lock);
10119	STAILQ_FOREACH(port, &softc->port_list, links) {
10120		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10121			continue;
10122		if (lun != NULL &&
10123		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10124		    CTL_MAX_LUNS)
10125			continue;
10126		num_target_ports++;
10127		if (port->init_devid)
10128			iid_len += port->init_devid->len;
10129		if (port->port_devid)
10130			id_len += port->port_devid->len;
10131	}
10132	mtx_unlock(&softc->ctl_lock);
10133
10134	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10135	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10136	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10137	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10138	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10139	ctsio->kern_sg_entries = 0;
10140
10141	if (data_len < alloc_len) {
10142		ctsio->residual = alloc_len - data_len;
10143		ctsio->kern_data_len = data_len;
10144		ctsio->kern_total_len = data_len;
10145	} else {
10146		ctsio->residual = 0;
10147		ctsio->kern_data_len = alloc_len;
10148		ctsio->kern_total_len = alloc_len;
10149	}
10150	ctsio->kern_data_resid = 0;
10151	ctsio->kern_rel_offset = 0;
10152	ctsio->kern_sg_entries = 0;
10153
10154	/*
10155	 * The control device is always connected.  The disk device, on the
10156	 * other hand, may not be online all the time.  Need to change this
10157	 * to figure out whether the disk device is actually online or not.
10158	 */
10159	if (lun != NULL)
10160		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10161				  lun->be_lun->lun_type;
10162	else
10163		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10164
10165	sp->page_code = SVPD_SCSI_PORTS;
10166	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10167	    sp->page_length);
10168	pd = &sp->design[0];
10169
10170	mtx_lock(&softc->ctl_lock);
10171	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10172		pg = 0;
10173	else
10174		pg = 1;
10175	for (g = 0; g < num_target_port_groups; g++) {
10176		STAILQ_FOREACH(port, &softc->port_list, links) {
10177			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10178				continue;
10179			if (lun != NULL &&
10180			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10181			    CTL_MAX_LUNS)
10182				continue;
10183			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10184			scsi_ulto2b(p, pd->relative_port_id);
10185			if (port->init_devid && g == pg) {
10186				iid_len = port->init_devid->len;
10187				memcpy(pd->initiator_transportid,
10188				    port->init_devid->data, port->init_devid->len);
10189			} else
10190				iid_len = 0;
10191			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10192			pdc = (struct scsi_vpd_port_designation_cont *)
10193			    (&pd->initiator_transportid[iid_len]);
10194			if (port->port_devid && g == pg) {
10195				id_len = port->port_devid->len;
10196				memcpy(pdc->target_port_descriptors,
10197				    port->port_devid->data, port->port_devid->len);
10198			} else
10199				id_len = 0;
10200			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10201			pd = (struct scsi_vpd_port_designation *)
10202			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10203		}
10204	}
10205	mtx_unlock(&softc->ctl_lock);
10206
10207	ctsio->scsi_status = SCSI_STATUS_OK;
10208	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10209	ctsio->be_move_done = ctl_config_move_done;
10210	ctl_datamove((union ctl_io *)ctsio);
10211
10212	return (CTL_RETVAL_COMPLETE);
10213}
10214
10215static int
10216ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10217{
10218	struct scsi_vpd_block_limits *bl_ptr;
10219	struct ctl_lun *lun;
10220	int bs;
10221
10222	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10223
10224	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10225	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10226	ctsio->kern_sg_entries = 0;
10227
10228	if (sizeof(*bl_ptr) < alloc_len) {
10229		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10230		ctsio->kern_data_len = sizeof(*bl_ptr);
10231		ctsio->kern_total_len = sizeof(*bl_ptr);
10232	} else {
10233		ctsio->residual = 0;
10234		ctsio->kern_data_len = alloc_len;
10235		ctsio->kern_total_len = alloc_len;
10236	}
10237	ctsio->kern_data_resid = 0;
10238	ctsio->kern_rel_offset = 0;
10239	ctsio->kern_sg_entries = 0;
10240
10241	/*
10242	 * The control device is always connected.  The disk device, on the
10243	 * other hand, may not be online all the time.  Need to change this
10244	 * to figure out whether the disk device is actually online or not.
10245	 */
10246	if (lun != NULL)
10247		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10248				  lun->be_lun->lun_type;
10249	else
10250		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10251
10252	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10253	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
10254	bl_ptr->max_cmp_write_len = 0xff;
10255	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10256	if (lun != NULL) {
10257		bs = lun->be_lun->blocksize;
10258		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10259		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10260			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10261			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10262			if (lun->be_lun->pblockexp != 0) {
10263				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10264				    bl_ptr->opt_unmap_grain);
10265				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10266				    bl_ptr->unmap_grain_align);
10267			}
10268		}
10269		scsi_ulto4b(lun->be_lun->atomicblock,
10270		    bl_ptr->max_atomic_transfer_length);
10271		scsi_ulto4b(0, bl_ptr->atomic_alignment);
10272		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10273	}
10274	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10275
10276	ctsio->scsi_status = SCSI_STATUS_OK;
10277	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10278	ctsio->be_move_done = ctl_config_move_done;
10279	ctl_datamove((union ctl_io *)ctsio);
10280
10281	return (CTL_RETVAL_COMPLETE);
10282}
10283
10284static int
10285ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10286{
10287	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10288	struct ctl_lun *lun;
10289
10290	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10291
10292	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10293	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10294	ctsio->kern_sg_entries = 0;
10295
10296	if (sizeof(*bdc_ptr) < alloc_len) {
10297		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10298		ctsio->kern_data_len = sizeof(*bdc_ptr);
10299		ctsio->kern_total_len = sizeof(*bdc_ptr);
10300	} else {
10301		ctsio->residual = 0;
10302		ctsio->kern_data_len = alloc_len;
10303		ctsio->kern_total_len = alloc_len;
10304	}
10305	ctsio->kern_data_resid = 0;
10306	ctsio->kern_rel_offset = 0;
10307	ctsio->kern_sg_entries = 0;
10308
10309	/*
10310	 * The control device is always connected.  The disk device, on the
10311	 * other hand, may not be online all the time.  Need to change this
10312	 * to figure out whether the disk device is actually online or not.
10313	 */
10314	if (lun != NULL)
10315		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10316				  lun->be_lun->lun_type;
10317	else
10318		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10319	bdc_ptr->page_code = SVPD_BDC;
10320	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10321	scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10322	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10323
10324	ctsio->scsi_status = SCSI_STATUS_OK;
10325	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10326	ctsio->be_move_done = ctl_config_move_done;
10327	ctl_datamove((union ctl_io *)ctsio);
10328
10329	return (CTL_RETVAL_COMPLETE);
10330}
10331
10332static int
10333ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10334{
10335	struct scsi_vpd_logical_block_prov *lbp_ptr;
10336	struct ctl_lun *lun;
10337
10338	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10339
10340	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10341	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10342	ctsio->kern_sg_entries = 0;
10343
10344	if (sizeof(*lbp_ptr) < alloc_len) {
10345		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10346		ctsio->kern_data_len = sizeof(*lbp_ptr);
10347		ctsio->kern_total_len = sizeof(*lbp_ptr);
10348	} else {
10349		ctsio->residual = 0;
10350		ctsio->kern_data_len = alloc_len;
10351		ctsio->kern_total_len = alloc_len;
10352	}
10353	ctsio->kern_data_resid = 0;
10354	ctsio->kern_rel_offset = 0;
10355	ctsio->kern_sg_entries = 0;
10356
10357	/*
10358	 * The control device is always connected.  The disk device, on the
10359	 * other hand, may not be online all the time.  Need to change this
10360	 * to figure out whether the disk device is actually online or not.
10361	 */
10362	if (lun != NULL)
10363		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10364				  lun->be_lun->lun_type;
10365	else
10366		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10367
10368	lbp_ptr->page_code = SVPD_LBP;
10369	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10370	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10371		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10372		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10373		lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10374	}
10375
10376	ctsio->scsi_status = SCSI_STATUS_OK;
10377	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10378	ctsio->be_move_done = ctl_config_move_done;
10379	ctl_datamove((union ctl_io *)ctsio);
10380
10381	return (CTL_RETVAL_COMPLETE);
10382}
10383
10384static int
10385ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10386{
10387	struct scsi_inquiry *cdb;
10388	struct ctl_lun *lun;
10389	int alloc_len, retval;
10390
10391	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10392	cdb = (struct scsi_inquiry *)ctsio->cdb;
10393
10394	retval = CTL_RETVAL_COMPLETE;
10395
10396	alloc_len = scsi_2btoul(cdb->length);
10397
10398	switch (cdb->page_code) {
10399	case SVPD_SUPPORTED_PAGES:
10400		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10401		break;
10402	case SVPD_UNIT_SERIAL_NUMBER:
10403		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10404		break;
10405	case SVPD_DEVICE_ID:
10406		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10407		break;
10408	case SVPD_EXTENDED_INQUIRY_DATA:
10409		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10410		break;
10411	case SVPD_MODE_PAGE_POLICY:
10412		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10413		break;
10414	case SVPD_SCSI_PORTS:
10415		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10416		break;
10417	case SVPD_SCSI_TPC:
10418		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10419		break;
10420	case SVPD_BLOCK_LIMITS:
10421		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10422		break;
10423	case SVPD_BDC:
10424		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10425		break;
10426	case SVPD_LBP:
10427		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10428		break;
10429	default:
10430		ctl_set_invalid_field(ctsio,
10431				      /*sks_valid*/ 1,
10432				      /*command*/ 1,
10433				      /*field*/ 2,
10434				      /*bit_valid*/ 0,
10435				      /*bit*/ 0);
10436		ctl_done((union ctl_io *)ctsio);
10437		retval = CTL_RETVAL_COMPLETE;
10438		break;
10439	}
10440
10441	return (retval);
10442}
10443
10444static int
10445ctl_inquiry_std(struct ctl_scsiio *ctsio)
10446{
10447	struct scsi_inquiry_data *inq_ptr;
10448	struct scsi_inquiry *cdb;
10449	struct ctl_softc *ctl_softc;
10450	struct ctl_lun *lun;
10451	char *val;
10452	uint32_t alloc_len, data_len;
10453	ctl_port_type port_type;
10454
10455	ctl_softc = control_softc;
10456
10457	/*
10458	 * Figure out whether we're talking to a Fibre Channel port or not.
10459	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10460	 * SCSI front ends.
10461	 */
10462	port_type = ctl_softc->ctl_ports[
10463	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10464	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10465		port_type = CTL_PORT_SCSI;
10466
10467	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10468	cdb = (struct scsi_inquiry *)ctsio->cdb;
10469	alloc_len = scsi_2btoul(cdb->length);
10470
10471	/*
10472	 * We malloc the full inquiry data size here and fill it
10473	 * in.  If the user only asks for less, we'll give him
10474	 * that much.
10475	 */
10476	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10477	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10478	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10479	ctsio->kern_sg_entries = 0;
10480	ctsio->kern_data_resid = 0;
10481	ctsio->kern_rel_offset = 0;
10482
10483	if (data_len < alloc_len) {
10484		ctsio->residual = alloc_len - data_len;
10485		ctsio->kern_data_len = data_len;
10486		ctsio->kern_total_len = data_len;
10487	} else {
10488		ctsio->residual = 0;
10489		ctsio->kern_data_len = alloc_len;
10490		ctsio->kern_total_len = alloc_len;
10491	}
10492
10493	/*
10494	 * If we have a LUN configured, report it as connected.  Otherwise,
10495	 * report that it is offline or no device is supported, depending
10496	 * on the value of inquiry_pq_no_lun.
10497	 *
10498	 * According to the spec (SPC-4 r34), the peripheral qualifier
10499	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10500	 *
10501	 * "A peripheral device having the specified peripheral device type
10502	 * is not connected to this logical unit. However, the device
10503	 * server is capable of supporting the specified peripheral device
10504	 * type on this logical unit."
10505	 *
10506	 * According to the same spec, the peripheral qualifier
10507	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10508	 *
10509	 * "The device server is not capable of supporting a peripheral
10510	 * device on this logical unit. For this peripheral qualifier the
10511	 * peripheral device type shall be set to 1Fh. All other peripheral
10512	 * device type values are reserved for this peripheral qualifier."
10513	 *
10514	 * Given the text, it would seem that we probably want to report that
10515	 * the LUN is offline here.  There is no LUN connected, but we can
10516	 * support a LUN at the given LUN number.
10517	 *
10518	 * In the real world, though, it sounds like things are a little
10519	 * different:
10520	 *
10521	 * - Linux, when presented with a LUN with the offline peripheral
10522	 *   qualifier, will create an sg driver instance for it.  So when
10523	 *   you attach it to CTL, you wind up with a ton of sg driver
10524	 *   instances.  (One for every LUN that Linux bothered to probe.)
10525	 *   Linux does this despite the fact that it issues a REPORT LUNs
10526	 *   to LUN 0 to get the inventory of supported LUNs.
10527	 *
10528	 * - There is other anecdotal evidence (from Emulex folks) about
10529	 *   arrays that use the offline peripheral qualifier for LUNs that
10530	 *   are on the "passive" path in an active/passive array.
10531	 *
10532	 * So the solution is provide a hopefully reasonable default
10533	 * (return bad/no LUN) and allow the user to change the behavior
10534	 * with a tunable/sysctl variable.
10535	 */
10536	if (lun != NULL)
10537		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10538				  lun->be_lun->lun_type;
10539	else if (ctl_softc->inquiry_pq_no_lun == 0)
10540		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10541	else
10542		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10543
10544	/* RMB in byte 2 is 0 */
10545	inq_ptr->version = SCSI_REV_SPC4;
10546
10547	/*
10548	 * According to SAM-3, even if a device only supports a single
10549	 * level of LUN addressing, it should still set the HISUP bit:
10550	 *
10551	 * 4.9.1 Logical unit numbers overview
10552	 *
10553	 * All logical unit number formats described in this standard are
10554	 * hierarchical in structure even when only a single level in that
10555	 * hierarchy is used. The HISUP bit shall be set to one in the
10556	 * standard INQUIRY data (see SPC-2) when any logical unit number
10557	 * format described in this standard is used.  Non-hierarchical
10558	 * formats are outside the scope of this standard.
10559	 *
10560	 * Therefore we set the HiSup bit here.
10561	 *
10562	 * The reponse format is 2, per SPC-3.
10563	 */
10564	inq_ptr->response_format = SID_HiSup | 2;
10565
10566	inq_ptr->additional_length = data_len -
10567	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10568	CTL_DEBUG_PRINT(("additional_length = %d\n",
10569			 inq_ptr->additional_length));
10570
10571	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10572	/* 16 bit addressing */
10573	if (port_type == CTL_PORT_SCSI)
10574		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10575	/* XXX set the SID_MultiP bit here if we're actually going to
10576	   respond on multiple ports */
10577	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10578
10579	/* 16 bit data bus, synchronous transfers */
10580	if (port_type == CTL_PORT_SCSI)
10581		inq_ptr->flags = SID_WBus16 | SID_Sync;
10582	/*
10583	 * XXX KDM do we want to support tagged queueing on the control
10584	 * device at all?
10585	 */
10586	if ((lun == NULL)
10587	 || (lun->be_lun->lun_type != T_PROCESSOR))
10588		inq_ptr->flags |= SID_CmdQue;
10589	/*
10590	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10591	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10592	 * name and 4 bytes for the revision.
10593	 */
10594	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10595	    "vendor")) == NULL) {
10596		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10597	} else {
10598		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10599		strncpy(inq_ptr->vendor, val,
10600		    min(sizeof(inq_ptr->vendor), strlen(val)));
10601	}
10602	if (lun == NULL) {
10603		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10604		    sizeof(inq_ptr->product));
10605	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10606		switch (lun->be_lun->lun_type) {
10607		case T_DIRECT:
10608			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10609			    sizeof(inq_ptr->product));
10610			break;
10611		case T_PROCESSOR:
10612			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10613			    sizeof(inq_ptr->product));
10614			break;
10615		default:
10616			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10617			    sizeof(inq_ptr->product));
10618			break;
10619		}
10620	} else {
10621		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10622		strncpy(inq_ptr->product, val,
10623		    min(sizeof(inq_ptr->product), strlen(val)));
10624	}
10625
10626	/*
10627	 * XXX make this a macro somewhere so it automatically gets
10628	 * incremented when we make changes.
10629	 */
10630	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10631	    "revision")) == NULL) {
10632		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10633	} else {
10634		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10635		strncpy(inq_ptr->revision, val,
10636		    min(sizeof(inq_ptr->revision), strlen(val)));
10637	}
10638
10639	/*
10640	 * For parallel SCSI, we support double transition and single
10641	 * transition clocking.  We also support QAS (Quick Arbitration
10642	 * and Selection) and Information Unit transfers on both the
10643	 * control and array devices.
10644	 */
10645	if (port_type == CTL_PORT_SCSI)
10646		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10647				    SID_SPI_IUS;
10648
10649	/* SAM-5 (no version claimed) */
10650	scsi_ulto2b(0x00A0, inq_ptr->version1);
10651	/* SPC-4 (no version claimed) */
10652	scsi_ulto2b(0x0460, inq_ptr->version2);
10653	if (port_type == CTL_PORT_FC) {
10654		/* FCP-2 ANSI INCITS.350:2003 */
10655		scsi_ulto2b(0x0917, inq_ptr->version3);
10656	} else if (port_type == CTL_PORT_SCSI) {
10657		/* SPI-4 ANSI INCITS.362:200x */
10658		scsi_ulto2b(0x0B56, inq_ptr->version3);
10659	} else if (port_type == CTL_PORT_ISCSI) {
10660		/* iSCSI (no version claimed) */
10661		scsi_ulto2b(0x0960, inq_ptr->version3);
10662	} else if (port_type == CTL_PORT_SAS) {
10663		/* SAS (no version claimed) */
10664		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10665	}
10666
10667	if (lun == NULL) {
10668		/* SBC-4 (no version claimed) */
10669		scsi_ulto2b(0x0600, inq_ptr->version4);
10670	} else {
10671		switch (lun->be_lun->lun_type) {
10672		case T_DIRECT:
10673			/* SBC-4 (no version claimed) */
10674			scsi_ulto2b(0x0600, inq_ptr->version4);
10675			break;
10676		case T_PROCESSOR:
10677		default:
10678			break;
10679		}
10680	}
10681
10682	ctsio->scsi_status = SCSI_STATUS_OK;
10683	if (ctsio->kern_data_len > 0) {
10684		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10685		ctsio->be_move_done = ctl_config_move_done;
10686		ctl_datamove((union ctl_io *)ctsio);
10687	} else {
10688		ctsio->io_hdr.status = CTL_SUCCESS;
10689		ctl_done((union ctl_io *)ctsio);
10690	}
10691
10692	return (CTL_RETVAL_COMPLETE);
10693}
10694
10695int
10696ctl_inquiry(struct ctl_scsiio *ctsio)
10697{
10698	struct scsi_inquiry *cdb;
10699	int retval;
10700
10701	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10702
10703	cdb = (struct scsi_inquiry *)ctsio->cdb;
10704	if (cdb->byte2 & SI_EVPD)
10705		retval = ctl_inquiry_evpd(ctsio);
10706	else if (cdb->page_code == 0)
10707		retval = ctl_inquiry_std(ctsio);
10708	else {
10709		ctl_set_invalid_field(ctsio,
10710				      /*sks_valid*/ 1,
10711				      /*command*/ 1,
10712				      /*field*/ 2,
10713				      /*bit_valid*/ 0,
10714				      /*bit*/ 0);
10715		ctl_done((union ctl_io *)ctsio);
10716		return (CTL_RETVAL_COMPLETE);
10717	}
10718
10719	return (retval);
10720}
10721
10722/*
10723 * For known CDB types, parse the LBA and length.
10724 */
10725static int
10726ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10727{
10728	if (io->io_hdr.io_type != CTL_IO_SCSI)
10729		return (1);
10730
10731	switch (io->scsiio.cdb[0]) {
10732	case COMPARE_AND_WRITE: {
10733		struct scsi_compare_and_write *cdb;
10734
10735		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10736
10737		*lba = scsi_8btou64(cdb->addr);
10738		*len = cdb->length;
10739		break;
10740	}
10741	case READ_6:
10742	case WRITE_6: {
10743		struct scsi_rw_6 *cdb;
10744
10745		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10746
10747		*lba = scsi_3btoul(cdb->addr);
10748		/* only 5 bits are valid in the most significant address byte */
10749		*lba &= 0x1fffff;
10750		*len = cdb->length;
10751		break;
10752	}
10753	case READ_10:
10754	case WRITE_10: {
10755		struct scsi_rw_10 *cdb;
10756
10757		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10758
10759		*lba = scsi_4btoul(cdb->addr);
10760		*len = scsi_2btoul(cdb->length);
10761		break;
10762	}
10763	case WRITE_VERIFY_10: {
10764		struct scsi_write_verify_10 *cdb;
10765
10766		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10767
10768		*lba = scsi_4btoul(cdb->addr);
10769		*len = scsi_2btoul(cdb->length);
10770		break;
10771	}
10772	case READ_12:
10773	case WRITE_12: {
10774		struct scsi_rw_12 *cdb;
10775
10776		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10777
10778		*lba = scsi_4btoul(cdb->addr);
10779		*len = scsi_4btoul(cdb->length);
10780		break;
10781	}
10782	case WRITE_VERIFY_12: {
10783		struct scsi_write_verify_12 *cdb;
10784
10785		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10786
10787		*lba = scsi_4btoul(cdb->addr);
10788		*len = scsi_4btoul(cdb->length);
10789		break;
10790	}
10791	case READ_16:
10792	case WRITE_16:
10793	case WRITE_ATOMIC_16: {
10794		struct scsi_rw_16 *cdb;
10795
10796		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10797
10798		*lba = scsi_8btou64(cdb->addr);
10799		*len = scsi_4btoul(cdb->length);
10800		break;
10801	}
10802	case WRITE_VERIFY_16: {
10803		struct scsi_write_verify_16 *cdb;
10804
10805		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10806
10807		*lba = scsi_8btou64(cdb->addr);
10808		*len = scsi_4btoul(cdb->length);
10809		break;
10810	}
10811	case WRITE_SAME_10: {
10812		struct scsi_write_same_10 *cdb;
10813
10814		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10815
10816		*lba = scsi_4btoul(cdb->addr);
10817		*len = scsi_2btoul(cdb->length);
10818		break;
10819	}
10820	case WRITE_SAME_16: {
10821		struct scsi_write_same_16 *cdb;
10822
10823		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10824
10825		*lba = scsi_8btou64(cdb->addr);
10826		*len = scsi_4btoul(cdb->length);
10827		break;
10828	}
10829	case VERIFY_10: {
10830		struct scsi_verify_10 *cdb;
10831
10832		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10833
10834		*lba = scsi_4btoul(cdb->addr);
10835		*len = scsi_2btoul(cdb->length);
10836		break;
10837	}
10838	case VERIFY_12: {
10839		struct scsi_verify_12 *cdb;
10840
10841		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10842
10843		*lba = scsi_4btoul(cdb->addr);
10844		*len = scsi_4btoul(cdb->length);
10845		break;
10846	}
10847	case VERIFY_16: {
10848		struct scsi_verify_16 *cdb;
10849
10850		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10851
10852		*lba = scsi_8btou64(cdb->addr);
10853		*len = scsi_4btoul(cdb->length);
10854		break;
10855	}
10856	case UNMAP: {
10857		*lba = 0;
10858		*len = UINT64_MAX;
10859		break;
10860	}
10861	default:
10862		return (1);
10863		break; /* NOTREACHED */
10864	}
10865
10866	return (0);
10867}
10868
10869static ctl_action
10870ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10871{
10872	uint64_t endlba1, endlba2;
10873
10874	endlba1 = lba1 + len1 - 1;
10875	endlba2 = lba2 + len2 - 1;
10876
10877	if ((endlba1 < lba2)
10878	 || (endlba2 < lba1))
10879		return (CTL_ACTION_PASS);
10880	else
10881		return (CTL_ACTION_BLOCK);
10882}
10883
10884static int
10885ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10886{
10887	struct ctl_ptr_len_flags *ptrlen;
10888	struct scsi_unmap_desc *buf, *end, *range;
10889	uint64_t lba;
10890	uint32_t len;
10891
10892	/* If not UNMAP -- go other way. */
10893	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10894	    io->scsiio.cdb[0] != UNMAP)
10895		return (CTL_ACTION_ERROR);
10896
10897	/* If UNMAP without data -- block and wait for data. */
10898	ptrlen = (struct ctl_ptr_len_flags *)
10899	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10900	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10901	    ptrlen->ptr == NULL)
10902		return (CTL_ACTION_BLOCK);
10903
10904	/* UNMAP with data -- check for collision. */
10905	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10906	end = buf + ptrlen->len / sizeof(*buf);
10907	for (range = buf; range < end; range++) {
10908		lba = scsi_8btou64(range->lba);
10909		len = scsi_4btoul(range->length);
10910		if ((lba < lba2 + len2) && (lba + len > lba2))
10911			return (CTL_ACTION_BLOCK);
10912	}
10913	return (CTL_ACTION_PASS);
10914}
10915
10916static ctl_action
10917ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10918{
10919	uint64_t lba1, lba2;
10920	uint64_t len1, len2;
10921	int retval;
10922
10923	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10924		return (CTL_ACTION_ERROR);
10925
10926	retval = ctl_extent_check_unmap(io2, lba1, len1);
10927	if (retval != CTL_ACTION_ERROR)
10928		return (retval);
10929
10930	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10931		return (CTL_ACTION_ERROR);
10932
10933	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10934}
10935
10936static ctl_action
10937ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10938    union ctl_io *ooa_io)
10939{
10940	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10941	ctl_serialize_action *serialize_row;
10942
10943	/*
10944	 * The initiator attempted multiple untagged commands at the same
10945	 * time.  Can't do that.
10946	 */
10947	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10948	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10949	 && ((pending_io->io_hdr.nexus.targ_port ==
10950	      ooa_io->io_hdr.nexus.targ_port)
10951	  && (pending_io->io_hdr.nexus.initid.id ==
10952	      ooa_io->io_hdr.nexus.initid.id))
10953	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10954		return (CTL_ACTION_OVERLAP);
10955
10956	/*
10957	 * The initiator attempted to send multiple tagged commands with
10958	 * the same ID.  (It's fine if different initiators have the same
10959	 * tag ID.)
10960	 *
10961	 * Even if all of those conditions are true, we don't kill the I/O
10962	 * if the command ahead of us has been aborted.  We won't end up
10963	 * sending it to the FETD, and it's perfectly legal to resend a
10964	 * command with the same tag number as long as the previous
10965	 * instance of this tag number has been aborted somehow.
10966	 */
10967	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10968	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10969	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10970	 && ((pending_io->io_hdr.nexus.targ_port ==
10971	      ooa_io->io_hdr.nexus.targ_port)
10972	  && (pending_io->io_hdr.nexus.initid.id ==
10973	      ooa_io->io_hdr.nexus.initid.id))
10974	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10975		return (CTL_ACTION_OVERLAP_TAG);
10976
10977	/*
10978	 * If we get a head of queue tag, SAM-3 says that we should
10979	 * immediately execute it.
10980	 *
10981	 * What happens if this command would normally block for some other
10982	 * reason?  e.g. a request sense with a head of queue tag
10983	 * immediately after a write.  Normally that would block, but this
10984	 * will result in its getting executed immediately...
10985	 *
10986	 * We currently return "pass" instead of "skip", so we'll end up
10987	 * going through the rest of the queue to check for overlapped tags.
10988	 *
10989	 * XXX KDM check for other types of blockage first??
10990	 */
10991	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10992		return (CTL_ACTION_PASS);
10993
10994	/*
10995	 * Ordered tags have to block until all items ahead of them
10996	 * have completed.  If we get called with an ordered tag, we always
10997	 * block, if something else is ahead of us in the queue.
10998	 */
10999	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11000		return (CTL_ACTION_BLOCK);
11001
11002	/*
11003	 * Simple tags get blocked until all head of queue and ordered tags
11004	 * ahead of them have completed.  I'm lumping untagged commands in
11005	 * with simple tags here.  XXX KDM is that the right thing to do?
11006	 */
11007	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11008	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11009	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11010	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11011		return (CTL_ACTION_BLOCK);
11012
11013	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11014	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11015
11016	serialize_row = ctl_serialize_table[ooa_entry->seridx];
11017
11018	switch (serialize_row[pending_entry->seridx]) {
11019	case CTL_SER_BLOCK:
11020		return (CTL_ACTION_BLOCK);
11021	case CTL_SER_EXTENT:
11022		return (ctl_extent_check(pending_io, ooa_io));
11023	case CTL_SER_EXTENTOPT:
11024		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11025		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11026			return (ctl_extent_check(pending_io, ooa_io));
11027		/* FALLTHROUGH */
11028	case CTL_SER_PASS:
11029		return (CTL_ACTION_PASS);
11030	case CTL_SER_BLOCKOPT:
11031		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11032		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11033			return (CTL_ACTION_BLOCK);
11034		return (CTL_ACTION_PASS);
11035	case CTL_SER_SKIP:
11036		return (CTL_ACTION_SKIP);
11037	default:
11038		panic("invalid serialization value %d",
11039		      serialize_row[pending_entry->seridx]);
11040	}
11041
11042	return (CTL_ACTION_ERROR);
11043}
11044
11045/*
11046 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11047 * Assumptions:
11048 * - pending_io is generally either incoming, or on the blocked queue
11049 * - starting I/O is the I/O we want to start the check with.
11050 */
11051static ctl_action
11052ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11053	      union ctl_io *starting_io)
11054{
11055	union ctl_io *ooa_io;
11056	ctl_action action;
11057
11058	mtx_assert(&lun->lun_lock, MA_OWNED);
11059
11060	/*
11061	 * Run back along the OOA queue, starting with the current
11062	 * blocked I/O and going through every I/O before it on the
11063	 * queue.  If starting_io is NULL, we'll just end up returning
11064	 * CTL_ACTION_PASS.
11065	 */
11066	for (ooa_io = starting_io; ooa_io != NULL;
11067	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11068	     ooa_links)){
11069
11070		/*
11071		 * This routine just checks to see whether
11072		 * cur_blocked is blocked by ooa_io, which is ahead
11073		 * of it in the queue.  It doesn't queue/dequeue
11074		 * cur_blocked.
11075		 */
11076		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11077		switch (action) {
11078		case CTL_ACTION_BLOCK:
11079		case CTL_ACTION_OVERLAP:
11080		case CTL_ACTION_OVERLAP_TAG:
11081		case CTL_ACTION_SKIP:
11082		case CTL_ACTION_ERROR:
11083			return (action);
11084			break; /* NOTREACHED */
11085		case CTL_ACTION_PASS:
11086			break;
11087		default:
11088			panic("invalid action %d", action);
11089			break;  /* NOTREACHED */
11090		}
11091	}
11092
11093	return (CTL_ACTION_PASS);
11094}
11095
11096/*
11097 * Assumptions:
11098 * - An I/O has just completed, and has been removed from the per-LUN OOA
11099 *   queue, so some items on the blocked queue may now be unblocked.
11100 */
11101static int
11102ctl_check_blocked(struct ctl_lun *lun)
11103{
11104	union ctl_io *cur_blocked, *next_blocked;
11105
11106	mtx_assert(&lun->lun_lock, MA_OWNED);
11107
11108	/*
11109	 * Run forward from the head of the blocked queue, checking each
11110	 * entry against the I/Os prior to it on the OOA queue to see if
11111	 * there is still any blockage.
11112	 *
11113	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11114	 * with our removing a variable on it while it is traversing the
11115	 * list.
11116	 */
11117	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11118	     cur_blocked != NULL; cur_blocked = next_blocked) {
11119		union ctl_io *prev_ooa;
11120		ctl_action action;
11121
11122		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11123							  blocked_links);
11124
11125		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11126						      ctl_ooaq, ooa_links);
11127
11128		/*
11129		 * If cur_blocked happens to be the first item in the OOA
11130		 * queue now, prev_ooa will be NULL, and the action
11131		 * returned will just be CTL_ACTION_PASS.
11132		 */
11133		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11134
11135		switch (action) {
11136		case CTL_ACTION_BLOCK:
11137			/* Nothing to do here, still blocked */
11138			break;
11139		case CTL_ACTION_OVERLAP:
11140		case CTL_ACTION_OVERLAP_TAG:
11141			/*
11142			 * This shouldn't happen!  In theory we've already
11143			 * checked this command for overlap...
11144			 */
11145			break;
11146		case CTL_ACTION_PASS:
11147		case CTL_ACTION_SKIP: {
11148			struct ctl_softc *softc;
11149			const struct ctl_cmd_entry *entry;
11150			uint32_t initidx;
11151			int isc_retval;
11152
11153			/*
11154			 * The skip case shouldn't happen, this transaction
11155			 * should have never made it onto the blocked queue.
11156			 */
11157			/*
11158			 * This I/O is no longer blocked, we can remove it
11159			 * from the blocked queue.  Since this is a TAILQ
11160			 * (doubly linked list), we can do O(1) removals
11161			 * from any place on the list.
11162			 */
11163			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11164				     blocked_links);
11165			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11166
11167			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11168				/*
11169				 * Need to send IO back to original side to
11170				 * run
11171				 */
11172				union ctl_ha_msg msg_info;
11173
11174				msg_info.hdr.original_sc =
11175					cur_blocked->io_hdr.original_sc;
11176				msg_info.hdr.serializing_sc = cur_blocked;
11177				msg_info.hdr.msg_type = CTL_MSG_R2R;
11178				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11179				     &msg_info, sizeof(msg_info), 0)) >
11180				     CTL_HA_STATUS_SUCCESS) {
11181					printf("CTL:Check Blocked error from "
11182					       "ctl_ha_msg_send %d\n",
11183					       isc_retval);
11184				}
11185				break;
11186			}
11187			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11188			softc = control_softc;
11189
11190			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11191
11192			/*
11193			 * Check this I/O for LUN state changes that may
11194			 * have happened while this command was blocked.
11195			 * The LUN state may have been changed by a command
11196			 * ahead of us in the queue, so we need to re-check
11197			 * for any states that can be caused by SCSI
11198			 * commands.
11199			 */
11200			if (ctl_scsiio_lun_check(softc, lun, entry,
11201						 &cur_blocked->scsiio) == 0) {
11202				cur_blocked->io_hdr.flags |=
11203				                      CTL_FLAG_IS_WAS_ON_RTR;
11204				ctl_enqueue_rtr(cur_blocked);
11205			} else
11206				ctl_done(cur_blocked);
11207			break;
11208		}
11209		default:
11210			/*
11211			 * This probably shouldn't happen -- we shouldn't
11212			 * get CTL_ACTION_ERROR, or anything else.
11213			 */
11214			break;
11215		}
11216	}
11217
11218	return (CTL_RETVAL_COMPLETE);
11219}
11220
11221/*
11222 * This routine (with one exception) checks LUN flags that can be set by
11223 * commands ahead of us in the OOA queue.  These flags have to be checked
11224 * when a command initially comes in, and when we pull a command off the
11225 * blocked queue and are preparing to execute it.  The reason we have to
11226 * check these flags for commands on the blocked queue is that the LUN
11227 * state may have been changed by a command ahead of us while we're on the
11228 * blocked queue.
11229 *
11230 * Ordering is somewhat important with these checks, so please pay
11231 * careful attention to the placement of any new checks.
11232 */
11233static int
11234ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11235    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11236{
11237	int retval;
11238	uint32_t residx;
11239
11240	retval = 0;
11241
11242	mtx_assert(&lun->lun_lock, MA_OWNED);
11243
11244	/*
11245	 * If this shelf is a secondary shelf controller, we have to reject
11246	 * any media access commands.
11247	 */
11248#if 0
11249	/* No longer needed for HA */
11250	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11251	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11252		ctl_set_lun_standby(ctsio);
11253		retval = 1;
11254		goto bailout;
11255	}
11256#endif
11257
11258	if (entry->pattern & CTL_LUN_PAT_WRITE) {
11259		if (lun->flags & CTL_LUN_READONLY) {
11260			ctl_set_sense(ctsio, /*current_error*/ 1,
11261			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11262			    /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
11263			retval = 1;
11264			goto bailout;
11265		}
11266		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11267		    .eca_and_aen & SCP_SWP) != 0) {
11268			ctl_set_sense(ctsio, /*current_error*/ 1,
11269			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11270			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11271			retval = 1;
11272			goto bailout;
11273		}
11274	}
11275
11276	/*
11277	 * Check for a reservation conflict.  If this command isn't allowed
11278	 * even on reserved LUNs, and if this initiator isn't the one who
11279	 * reserved us, reject the command with a reservation conflict.
11280	 */
11281	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11282	if ((lun->flags & CTL_LUN_RESERVED)
11283	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11284		if (lun->res_idx != residx) {
11285			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11286			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11287			retval = 1;
11288			goto bailout;
11289		}
11290	}
11291
11292	if ((lun->flags & CTL_LUN_PR_RESERVED)
11293	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11294		/*
11295		 * if we aren't registered or it's a res holder type
11296		 * reservation and this isn't the res holder then set a
11297		 * conflict.
11298		 * NOTE: Commands which might be allowed on write exclusive
11299		 * type reservations are checked in the particular command
11300		 * for a conflict. Read and SSU are the only ones.
11301		 */
11302		if (lun->pr_keys[residx] == 0
11303		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11304			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11305			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11306			retval = 1;
11307			goto bailout;
11308		}
11309
11310	}
11311
11312	if ((lun->flags & CTL_LUN_OFFLINE)
11313	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11314		ctl_set_lun_not_ready(ctsio);
11315		retval = 1;
11316		goto bailout;
11317	}
11318
11319	/*
11320	 * If the LUN is stopped, see if this particular command is allowed
11321	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11322	 */
11323	if ((lun->flags & CTL_LUN_STOPPED)
11324	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11325		/* "Logical unit not ready, initializing cmd. required" */
11326		ctl_set_lun_stopped(ctsio);
11327		retval = 1;
11328		goto bailout;
11329	}
11330
11331	if ((lun->flags & CTL_LUN_INOPERABLE)
11332	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11333		/* "Medium format corrupted" */
11334		ctl_set_medium_format_corrupted(ctsio);
11335		retval = 1;
11336		goto bailout;
11337	}
11338
11339bailout:
11340	return (retval);
11341
11342}
11343
11344static void
11345ctl_failover_io(union ctl_io *io, int have_lock)
11346{
11347	ctl_set_busy(&io->scsiio);
11348	ctl_done(io);
11349}
11350
11351static void
11352ctl_failover(void)
11353{
11354	struct ctl_lun *lun;
11355	struct ctl_softc *ctl_softc;
11356	union ctl_io *next_io, *pending_io;
11357	union ctl_io *io;
11358	int lun_idx;
11359	int i;
11360
11361	ctl_softc = control_softc;
11362
11363	mtx_lock(&ctl_softc->ctl_lock);
11364	/*
11365	 * Remove any cmds from the other SC from the rtr queue.  These
11366	 * will obviously only be for LUNs for which we're the primary.
11367	 * We can't send status or get/send data for these commands.
11368	 * Since they haven't been executed yet, we can just remove them.
11369	 * We'll either abort them or delete them below, depending on
11370	 * which HA mode we're in.
11371	 */
11372#ifdef notyet
11373	mtx_lock(&ctl_softc->queue_lock);
11374	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11375	     io != NULL; io = next_io) {
11376		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11377		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11378			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11379				      ctl_io_hdr, links);
11380	}
11381	mtx_unlock(&ctl_softc->queue_lock);
11382#endif
11383
11384	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11385		lun = ctl_softc->ctl_luns[lun_idx];
11386		if (lun==NULL)
11387			continue;
11388
11389		/*
11390		 * Processor LUNs are primary on both sides.
11391		 * XXX will this always be true?
11392		 */
11393		if (lun->be_lun->lun_type == T_PROCESSOR)
11394			continue;
11395
11396		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11397		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11398			printf("FAILOVER: primary lun %d\n", lun_idx);
11399		        /*
11400			 * Remove all commands from the other SC. First from the
11401			 * blocked queue then from the ooa queue. Once we have
11402			 * removed them. Call ctl_check_blocked to see if there
11403			 * is anything that can run.
11404			 */
11405			for (io = (union ctl_io *)TAILQ_FIRST(
11406			     &lun->blocked_queue); io != NULL; io = next_io) {
11407
11408		        	next_io = (union ctl_io *)TAILQ_NEXT(
11409				    &io->io_hdr, blocked_links);
11410
11411				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11412					TAILQ_REMOVE(&lun->blocked_queue,
11413						     &io->io_hdr,blocked_links);
11414					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11415					TAILQ_REMOVE(&lun->ooa_queue,
11416						     &io->io_hdr, ooa_links);
11417
11418					ctl_free_io(io);
11419				}
11420			}
11421
11422			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11423	     		     io != NULL; io = next_io) {
11424
11425		        	next_io = (union ctl_io *)TAILQ_NEXT(
11426				    &io->io_hdr, ooa_links);
11427
11428				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11429
11430					TAILQ_REMOVE(&lun->ooa_queue,
11431						&io->io_hdr,
11432					     	ooa_links);
11433
11434					ctl_free_io(io);
11435				}
11436			}
11437			ctl_check_blocked(lun);
11438		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11439			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11440
11441			printf("FAILOVER: primary lun %d\n", lun_idx);
11442			/*
11443			 * Abort all commands from the other SC.  We can't
11444			 * send status back for them now.  These should get
11445			 * cleaned up when they are completed or come out
11446			 * for a datamove operation.
11447			 */
11448			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11449	     		     io != NULL; io = next_io) {
11450		        	next_io = (union ctl_io *)TAILQ_NEXT(
11451					&io->io_hdr, ooa_links);
11452
11453				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11454					io->io_hdr.flags |= CTL_FLAG_ABORT;
11455			}
11456		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11457			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11458
11459			printf("FAILOVER: secondary lun %d\n", lun_idx);
11460
11461			lun->flags |= CTL_LUN_PRIMARY_SC;
11462
11463			/*
11464			 * We send all I/O that was sent to this controller
11465			 * and redirected to the other side back with
11466			 * busy status, and have the initiator retry it.
11467			 * Figuring out how much data has been transferred,
11468			 * etc. and picking up where we left off would be
11469			 * very tricky.
11470			 *
11471			 * XXX KDM need to remove I/O from the blocked
11472			 * queue as well!
11473			 */
11474			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11475			     &lun->ooa_queue); pending_io != NULL;
11476			     pending_io = next_io) {
11477
11478				next_io =  (union ctl_io *)TAILQ_NEXT(
11479					&pending_io->io_hdr, ooa_links);
11480
11481				pending_io->io_hdr.flags &=
11482					~CTL_FLAG_SENT_2OTHER_SC;
11483
11484				if (pending_io->io_hdr.flags &
11485				    CTL_FLAG_IO_ACTIVE) {
11486					pending_io->io_hdr.flags |=
11487						CTL_FLAG_FAILOVER;
11488				} else {
11489					ctl_set_busy(&pending_io->scsiio);
11490					ctl_done(pending_io);
11491				}
11492			}
11493
11494			/*
11495			 * Build Unit Attention
11496			 */
11497			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11498				lun->pending_ua[i] |=
11499				                     CTL_UA_ASYM_ACC_CHANGE;
11500			}
11501		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11502			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11503			printf("FAILOVER: secondary lun %d\n", lun_idx);
11504			/*
11505			 * if the first io on the OOA is not on the RtR queue
11506			 * add it.
11507			 */
11508			lun->flags |= CTL_LUN_PRIMARY_SC;
11509
11510			pending_io = (union ctl_io *)TAILQ_FIRST(
11511			    &lun->ooa_queue);
11512			if (pending_io==NULL) {
11513				printf("Nothing on OOA queue\n");
11514				continue;
11515			}
11516
11517			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11518			if ((pending_io->io_hdr.flags &
11519			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11520				pending_io->io_hdr.flags |=
11521				    CTL_FLAG_IS_WAS_ON_RTR;
11522				ctl_enqueue_rtr(pending_io);
11523			}
11524#if 0
11525			else
11526			{
11527				printf("Tag 0x%04x is running\n",
11528				      pending_io->scsiio.tag_num);
11529			}
11530#endif
11531
11532			next_io = (union ctl_io *)TAILQ_NEXT(
11533			    &pending_io->io_hdr, ooa_links);
11534			for (pending_io=next_io; pending_io != NULL;
11535			     pending_io = next_io) {
11536				pending_io->io_hdr.flags &=
11537				    ~CTL_FLAG_SENT_2OTHER_SC;
11538				next_io = (union ctl_io *)TAILQ_NEXT(
11539					&pending_io->io_hdr, ooa_links);
11540				if (pending_io->io_hdr.flags &
11541				    CTL_FLAG_IS_WAS_ON_RTR) {
11542#if 0
11543				        printf("Tag 0x%04x is running\n",
11544				      		pending_io->scsiio.tag_num);
11545#endif
11546					continue;
11547				}
11548
11549				switch (ctl_check_ooa(lun, pending_io,
11550			            (union ctl_io *)TAILQ_PREV(
11551				    &pending_io->io_hdr, ctl_ooaq,
11552				    ooa_links))) {
11553
11554				case CTL_ACTION_BLOCK:
11555					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11556							  &pending_io->io_hdr,
11557							  blocked_links);
11558					pending_io->io_hdr.flags |=
11559					    CTL_FLAG_BLOCKED;
11560					break;
11561				case CTL_ACTION_PASS:
11562				case CTL_ACTION_SKIP:
11563					pending_io->io_hdr.flags |=
11564					    CTL_FLAG_IS_WAS_ON_RTR;
11565					ctl_enqueue_rtr(pending_io);
11566					break;
11567				case CTL_ACTION_OVERLAP:
11568					ctl_set_overlapped_cmd(
11569					    (struct ctl_scsiio *)pending_io);
11570					ctl_done(pending_io);
11571					break;
11572				case CTL_ACTION_OVERLAP_TAG:
11573					ctl_set_overlapped_tag(
11574					    (struct ctl_scsiio *)pending_io,
11575					    pending_io->scsiio.tag_num & 0xff);
11576					ctl_done(pending_io);
11577					break;
11578				case CTL_ACTION_ERROR:
11579				default:
11580					ctl_set_internal_failure(
11581						(struct ctl_scsiio *)pending_io,
11582						0,  // sks_valid
11583						0); //retry count
11584					ctl_done(pending_io);
11585					break;
11586				}
11587			}
11588
11589			/*
11590			 * Build Unit Attention
11591			 */
11592			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11593				lun->pending_ua[i] |=
11594				                     CTL_UA_ASYM_ACC_CHANGE;
11595			}
11596		} else {
11597			panic("Unhandled HA mode failover, LUN flags = %#x, "
11598			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11599		}
11600	}
11601	ctl_pause_rtr = 0;
11602	mtx_unlock(&ctl_softc->ctl_lock);
11603}
11604
11605static int
11606ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11607{
11608	struct ctl_lun *lun;
11609	const struct ctl_cmd_entry *entry;
11610	uint32_t initidx, targ_lun;
11611	int retval;
11612
11613	retval = 0;
11614
11615	lun = NULL;
11616
11617	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11618	if ((targ_lun < CTL_MAX_LUNS)
11619	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11620		lun = ctl_softc->ctl_luns[targ_lun];
11621		/*
11622		 * If the LUN is invalid, pretend that it doesn't exist.
11623		 * It will go away as soon as all pending I/O has been
11624		 * completed.
11625		 */
11626		if (lun->flags & CTL_LUN_DISABLED) {
11627			lun = NULL;
11628		} else {
11629			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11630			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11631				lun->be_lun;
11632			if (lun->be_lun->lun_type == T_PROCESSOR) {
11633				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11634			}
11635
11636			/*
11637			 * Every I/O goes into the OOA queue for a
11638			 * particular LUN, and stays there until completion.
11639			 */
11640			mtx_lock(&lun->lun_lock);
11641			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11642			    ooa_links);
11643		}
11644	} else {
11645		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11646		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11647	}
11648
11649	/* Get command entry and return error if it is unsuppotyed. */
11650	entry = ctl_validate_command(ctsio);
11651	if (entry == NULL) {
11652		if (lun)
11653			mtx_unlock(&lun->lun_lock);
11654		return (retval);
11655	}
11656
11657	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11658	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11659
11660	/*
11661	 * Check to see whether we can send this command to LUNs that don't
11662	 * exist.  This should pretty much only be the case for inquiry
11663	 * and request sense.  Further checks, below, really require having
11664	 * a LUN, so we can't really check the command anymore.  Just put
11665	 * it on the rtr queue.
11666	 */
11667	if (lun == NULL) {
11668		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11669			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11670			ctl_enqueue_rtr((union ctl_io *)ctsio);
11671			return (retval);
11672		}
11673
11674		ctl_set_unsupported_lun(ctsio);
11675		ctl_done((union ctl_io *)ctsio);
11676		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11677		return (retval);
11678	} else {
11679		/*
11680		 * Make sure we support this particular command on this LUN.
11681		 * e.g., we don't support writes to the control LUN.
11682		 */
11683		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11684			mtx_unlock(&lun->lun_lock);
11685			ctl_set_invalid_opcode(ctsio);
11686			ctl_done((union ctl_io *)ctsio);
11687			return (retval);
11688		}
11689	}
11690
11691	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11692
11693#ifdef CTL_WITH_CA
11694	/*
11695	 * If we've got a request sense, it'll clear the contingent
11696	 * allegiance condition.  Otherwise, if we have a CA condition for
11697	 * this initiator, clear it, because it sent down a command other
11698	 * than request sense.
11699	 */
11700	if ((ctsio->cdb[0] != REQUEST_SENSE)
11701	 && (ctl_is_set(lun->have_ca, initidx)))
11702		ctl_clear_mask(lun->have_ca, initidx);
11703#endif
11704
11705	/*
11706	 * If the command has this flag set, it handles its own unit
11707	 * attention reporting, we shouldn't do anything.  Otherwise we
11708	 * check for any pending unit attentions, and send them back to the
11709	 * initiator.  We only do this when a command initially comes in,
11710	 * not when we pull it off the blocked queue.
11711	 *
11712	 * According to SAM-3, section 5.3.2, the order that things get
11713	 * presented back to the host is basically unit attentions caused
11714	 * by some sort of reset event, busy status, reservation conflicts
11715	 * or task set full, and finally any other status.
11716	 *
11717	 * One issue here is that some of the unit attentions we report
11718	 * don't fall into the "reset" category (e.g. "reported luns data
11719	 * has changed").  So reporting it here, before the reservation
11720	 * check, may be technically wrong.  I guess the only thing to do
11721	 * would be to check for and report the reset events here, and then
11722	 * check for the other unit attention types after we check for a
11723	 * reservation conflict.
11724	 *
11725	 * XXX KDM need to fix this
11726	 */
11727	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11728		ctl_ua_type ua_type;
11729
11730		if (lun->pending_ua[initidx] != CTL_UA_NONE) {
11731			scsi_sense_data_type sense_format;
11732
11733			if (lun != NULL)
11734				sense_format = (lun->flags &
11735				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11736				    SSD_TYPE_FIXED;
11737			else
11738				sense_format = SSD_TYPE_FIXED;
11739
11740			ua_type = ctl_build_ua(&lun->pending_ua[initidx],
11741			    &ctsio->sense_data, sense_format);
11742			if (ua_type != CTL_UA_NONE) {
11743				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11744				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11745						       CTL_AUTOSENSE;
11746				ctsio->sense_len = SSD_FULL_SIZE;
11747				mtx_unlock(&lun->lun_lock);
11748				ctl_done((union ctl_io *)ctsio);
11749				return (retval);
11750			}
11751		}
11752	}
11753
11754
11755	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11756		mtx_unlock(&lun->lun_lock);
11757		ctl_done((union ctl_io *)ctsio);
11758		return (retval);
11759	}
11760
11761	/*
11762	 * XXX CHD this is where we want to send IO to other side if
11763	 * this LUN is secondary on this SC. We will need to make a copy
11764	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11765	 * the copy we send as FROM_OTHER.
11766	 * We also need to stuff the address of the original IO so we can
11767	 * find it easily. Something similar will need be done on the other
11768	 * side so when we are done we can find the copy.
11769	 */
11770	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11771		union ctl_ha_msg msg_info;
11772		int isc_retval;
11773
11774		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11775
11776		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11777		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11778#if 0
11779		printf("1. ctsio %p\n", ctsio);
11780#endif
11781		msg_info.hdr.serializing_sc = NULL;
11782		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11783		msg_info.scsi.tag_num = ctsio->tag_num;
11784		msg_info.scsi.tag_type = ctsio->tag_type;
11785		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11786
11787		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11788
11789		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11790		    (void *)&msg_info, sizeof(msg_info), 0)) >
11791		    CTL_HA_STATUS_SUCCESS) {
11792			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11793			       isc_retval);
11794			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11795		} else {
11796#if 0
11797			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11798#endif
11799		}
11800
11801		/*
11802		 * XXX KDM this I/O is off the incoming queue, but hasn't
11803		 * been inserted on any other queue.  We may need to come
11804		 * up with a holding queue while we wait for serialization
11805		 * so that we have an idea of what we're waiting for from
11806		 * the other side.
11807		 */
11808		mtx_unlock(&lun->lun_lock);
11809		return (retval);
11810	}
11811
11812	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11813			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11814			      ctl_ooaq, ooa_links))) {
11815	case CTL_ACTION_BLOCK:
11816		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11817		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11818				  blocked_links);
11819		mtx_unlock(&lun->lun_lock);
11820		return (retval);
11821	case CTL_ACTION_PASS:
11822	case CTL_ACTION_SKIP:
11823		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11824		mtx_unlock(&lun->lun_lock);
11825		ctl_enqueue_rtr((union ctl_io *)ctsio);
11826		break;
11827	case CTL_ACTION_OVERLAP:
11828		mtx_unlock(&lun->lun_lock);
11829		ctl_set_overlapped_cmd(ctsio);
11830		ctl_done((union ctl_io *)ctsio);
11831		break;
11832	case CTL_ACTION_OVERLAP_TAG:
11833		mtx_unlock(&lun->lun_lock);
11834		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11835		ctl_done((union ctl_io *)ctsio);
11836		break;
11837	case CTL_ACTION_ERROR:
11838	default:
11839		mtx_unlock(&lun->lun_lock);
11840		ctl_set_internal_failure(ctsio,
11841					 /*sks_valid*/ 0,
11842					 /*retry_count*/ 0);
11843		ctl_done((union ctl_io *)ctsio);
11844		break;
11845	}
11846	return (retval);
11847}
11848
11849const struct ctl_cmd_entry *
11850ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11851{
11852	const struct ctl_cmd_entry *entry;
11853	int service_action;
11854
11855	entry = &ctl_cmd_table[ctsio->cdb[0]];
11856	if (sa)
11857		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11858	if (entry->flags & CTL_CMD_FLAG_SA5) {
11859		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11860		entry = &((const struct ctl_cmd_entry *)
11861		    entry->execute)[service_action];
11862	}
11863	return (entry);
11864}
11865
11866const struct ctl_cmd_entry *
11867ctl_validate_command(struct ctl_scsiio *ctsio)
11868{
11869	const struct ctl_cmd_entry *entry;
11870	int i, sa;
11871	uint8_t diff;
11872
11873	entry = ctl_get_cmd_entry(ctsio, &sa);
11874	if (entry->execute == NULL) {
11875		if (sa)
11876			ctl_set_invalid_field(ctsio,
11877					      /*sks_valid*/ 1,
11878					      /*command*/ 1,
11879					      /*field*/ 1,
11880					      /*bit_valid*/ 1,
11881					      /*bit*/ 4);
11882		else
11883			ctl_set_invalid_opcode(ctsio);
11884		ctl_done((union ctl_io *)ctsio);
11885		return (NULL);
11886	}
11887	KASSERT(entry->length > 0,
11888	    ("Not defined length for command 0x%02x/0x%02x",
11889	     ctsio->cdb[0], ctsio->cdb[1]));
11890	for (i = 1; i < entry->length; i++) {
11891		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11892		if (diff == 0)
11893			continue;
11894		ctl_set_invalid_field(ctsio,
11895				      /*sks_valid*/ 1,
11896				      /*command*/ 1,
11897				      /*field*/ i,
11898				      /*bit_valid*/ 1,
11899				      /*bit*/ fls(diff) - 1);
11900		ctl_done((union ctl_io *)ctsio);
11901		return (NULL);
11902	}
11903	return (entry);
11904}
11905
11906static int
11907ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11908{
11909
11910	switch (lun_type) {
11911	case T_PROCESSOR:
11912		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11913		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11914			return (0);
11915		break;
11916	case T_DIRECT:
11917		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11918		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11919			return (0);
11920		break;
11921	default:
11922		return (0);
11923	}
11924	return (1);
11925}
11926
11927static int
11928ctl_scsiio(struct ctl_scsiio *ctsio)
11929{
11930	int retval;
11931	const struct ctl_cmd_entry *entry;
11932
11933	retval = CTL_RETVAL_COMPLETE;
11934
11935	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11936
11937	entry = ctl_get_cmd_entry(ctsio, NULL);
11938
11939	/*
11940	 * If this I/O has been aborted, just send it straight to
11941	 * ctl_done() without executing it.
11942	 */
11943	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11944		ctl_done((union ctl_io *)ctsio);
11945		goto bailout;
11946	}
11947
11948	/*
11949	 * All the checks should have been handled by ctl_scsiio_precheck().
11950	 * We should be clear now to just execute the I/O.
11951	 */
11952	retval = entry->execute(ctsio);
11953
11954bailout:
11955	return (retval);
11956}
11957
11958/*
11959 * Since we only implement one target right now, a bus reset simply resets
11960 * our single target.
11961 */
11962static int
11963ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11964{
11965	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11966}
11967
11968static int
11969ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11970		 ctl_ua_type ua_type)
11971{
11972	struct ctl_lun *lun;
11973	int retval;
11974
11975	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11976		union ctl_ha_msg msg_info;
11977
11978		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11979		msg_info.hdr.nexus = io->io_hdr.nexus;
11980		if (ua_type==CTL_UA_TARG_RESET)
11981			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11982		else
11983			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11984		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11985		msg_info.hdr.original_sc = NULL;
11986		msg_info.hdr.serializing_sc = NULL;
11987		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11988		    (void *)&msg_info, sizeof(msg_info), 0)) {
11989		}
11990	}
11991	retval = 0;
11992
11993	mtx_lock(&ctl_softc->ctl_lock);
11994	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11995		retval += ctl_lun_reset(lun, io, ua_type);
11996	mtx_unlock(&ctl_softc->ctl_lock);
11997
11998	return (retval);
11999}
12000
12001/*
12002 * The LUN should always be set.  The I/O is optional, and is used to
12003 * distinguish between I/Os sent by this initiator, and by other
12004 * initiators.  We set unit attention for initiators other than this one.
12005 * SAM-3 is vague on this point.  It does say that a unit attention should
12006 * be established for other initiators when a LUN is reset (see section
12007 * 5.7.3), but it doesn't specifically say that the unit attention should
12008 * be established for this particular initiator when a LUN is reset.  Here
12009 * is the relevant text, from SAM-3 rev 8:
12010 *
12011 * 5.7.2 When a SCSI initiator port aborts its own tasks
12012 *
12013 * When a SCSI initiator port causes its own task(s) to be aborted, no
12014 * notification that the task(s) have been aborted shall be returned to
12015 * the SCSI initiator port other than the completion response for the
12016 * command or task management function action that caused the task(s) to
12017 * be aborted and notification(s) associated with related effects of the
12018 * action (e.g., a reset unit attention condition).
12019 *
12020 * XXX KDM for now, we're setting unit attention for all initiators.
12021 */
12022static int
12023ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
12024{
12025	union ctl_io *xio;
12026#if 0
12027	uint32_t initindex;
12028#endif
12029	int i;
12030
12031	mtx_lock(&lun->lun_lock);
12032	/*
12033	 * Run through the OOA queue and abort each I/O.
12034	 */
12035#if 0
12036	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12037#endif
12038	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12039	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12040		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
12041	}
12042
12043	/*
12044	 * This version sets unit attention for every
12045	 */
12046#if 0
12047	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12048	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12049		if (initindex == i)
12050			continue;
12051		lun->pending_ua[i] |= ua_type;
12052	}
12053#endif
12054
12055	/*
12056	 * A reset (any kind, really) clears reservations established with
12057	 * RESERVE/RELEASE.  It does not clear reservations established
12058	 * with PERSISTENT RESERVE OUT, but we don't support that at the
12059	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
12060	 * reservations made with the RESERVE/RELEASE commands, because
12061	 * those commands are obsolete in SPC-3.
12062	 */
12063	lun->flags &= ~CTL_LUN_RESERVED;
12064
12065	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12066#ifdef CTL_WITH_CA
12067		ctl_clear_mask(lun->have_ca, i);
12068#endif
12069		lun->pending_ua[i] |= ua_type;
12070	}
12071	mtx_unlock(&lun->lun_lock);
12072
12073	return (0);
12074}
12075
12076static void
12077ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
12078    int other_sc)
12079{
12080	union ctl_io *xio;
12081
12082	mtx_assert(&lun->lun_lock, MA_OWNED);
12083
12084	/*
12085	 * Run through the OOA queue and attempt to find the given I/O.
12086	 * The target port, initiator ID, tag type and tag number have to
12087	 * match the values that we got from the initiator.  If we have an
12088	 * untagged command to abort, simply abort the first untagged command
12089	 * we come to.  We only allow one untagged command at a time of course.
12090	 */
12091	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12092	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12093
12094		if ((targ_port == UINT32_MAX ||
12095		     targ_port == xio->io_hdr.nexus.targ_port) &&
12096		    (init_id == UINT32_MAX ||
12097		     init_id == xio->io_hdr.nexus.initid.id)) {
12098			if (targ_port != xio->io_hdr.nexus.targ_port ||
12099			    init_id != xio->io_hdr.nexus.initid.id)
12100				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12101			xio->io_hdr.flags |= CTL_FLAG_ABORT;
12102			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12103				union ctl_ha_msg msg_info;
12104
12105				msg_info.hdr.nexus = xio->io_hdr.nexus;
12106				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12107				msg_info.task.tag_num = xio->scsiio.tag_num;
12108				msg_info.task.tag_type = xio->scsiio.tag_type;
12109				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12110				msg_info.hdr.original_sc = NULL;
12111				msg_info.hdr.serializing_sc = NULL;
12112				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12113				    (void *)&msg_info, sizeof(msg_info), 0);
12114			}
12115		}
12116	}
12117}
12118
12119static int
12120ctl_abort_task_set(union ctl_io *io)
12121{
12122	struct ctl_softc *softc = control_softc;
12123	struct ctl_lun *lun;
12124	uint32_t targ_lun;
12125
12126	/*
12127	 * Look up the LUN.
12128	 */
12129	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12130	mtx_lock(&softc->ctl_lock);
12131	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12132		lun = softc->ctl_luns[targ_lun];
12133	else {
12134		mtx_unlock(&softc->ctl_lock);
12135		return (1);
12136	}
12137
12138	mtx_lock(&lun->lun_lock);
12139	mtx_unlock(&softc->ctl_lock);
12140	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12141		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12142		    io->io_hdr.nexus.initid.id,
12143		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12144	} else { /* CTL_TASK_CLEAR_TASK_SET */
12145		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12146		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12147	}
12148	mtx_unlock(&lun->lun_lock);
12149	return (0);
12150}
12151
12152static int
12153ctl_i_t_nexus_reset(union ctl_io *io)
12154{
12155	struct ctl_softc *softc = control_softc;
12156	struct ctl_lun *lun;
12157	uint32_t initindex, residx;
12158
12159	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12160	residx = ctl_get_resindex(&io->io_hdr.nexus);
12161	mtx_lock(&softc->ctl_lock);
12162	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12163		mtx_lock(&lun->lun_lock);
12164		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12165		    io->io_hdr.nexus.initid.id,
12166		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12167#ifdef CTL_WITH_CA
12168		ctl_clear_mask(lun->have_ca, initindex);
12169#endif
12170		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
12171			lun->flags &= ~CTL_LUN_RESERVED;
12172		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12173		mtx_unlock(&lun->lun_lock);
12174	}
12175	mtx_unlock(&softc->ctl_lock);
12176	return (0);
12177}
12178
12179static int
12180ctl_abort_task(union ctl_io *io)
12181{
12182	union ctl_io *xio;
12183	struct ctl_lun *lun;
12184	struct ctl_softc *ctl_softc;
12185#if 0
12186	struct sbuf sb;
12187	char printbuf[128];
12188#endif
12189	int found;
12190	uint32_t targ_lun;
12191
12192	ctl_softc = control_softc;
12193	found = 0;
12194
12195	/*
12196	 * Look up the LUN.
12197	 */
12198	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12199	mtx_lock(&ctl_softc->ctl_lock);
12200	if ((targ_lun < CTL_MAX_LUNS)
12201	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12202		lun = ctl_softc->ctl_luns[targ_lun];
12203	else {
12204		mtx_unlock(&ctl_softc->ctl_lock);
12205		return (1);
12206	}
12207
12208#if 0
12209	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12210	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12211#endif
12212
12213	mtx_lock(&lun->lun_lock);
12214	mtx_unlock(&ctl_softc->ctl_lock);
12215	/*
12216	 * Run through the OOA queue and attempt to find the given I/O.
12217	 * The target port, initiator ID, tag type and tag number have to
12218	 * match the values that we got from the initiator.  If we have an
12219	 * untagged command to abort, simply abort the first untagged command
12220	 * we come to.  We only allow one untagged command at a time of course.
12221	 */
12222#if 0
12223	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12224#endif
12225	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12226	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12227#if 0
12228		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12229
12230		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12231			    lun->lun, xio->scsiio.tag_num,
12232			    xio->scsiio.tag_type,
12233			    (xio->io_hdr.blocked_links.tqe_prev
12234			    == NULL) ? "" : " BLOCKED",
12235			    (xio->io_hdr.flags &
12236			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12237			    (xio->io_hdr.flags &
12238			    CTL_FLAG_ABORT) ? " ABORT" : "",
12239			    (xio->io_hdr.flags &
12240			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12241		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12242		sbuf_finish(&sb);
12243		printf("%s\n", sbuf_data(&sb));
12244#endif
12245
12246		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12247		 && (xio->io_hdr.nexus.initid.id ==
12248		     io->io_hdr.nexus.initid.id)) {
12249			/*
12250			 * If the abort says that the task is untagged, the
12251			 * task in the queue must be untagged.  Otherwise,
12252			 * we just check to see whether the tag numbers
12253			 * match.  This is because the QLogic firmware
12254			 * doesn't pass back the tag type in an abort
12255			 * request.
12256			 */
12257#if 0
12258			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12259			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12260			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12261#endif
12262			/*
12263			 * XXX KDM we've got problems with FC, because it
12264			 * doesn't send down a tag type with aborts.  So we
12265			 * can only really go by the tag number...
12266			 * This may cause problems with parallel SCSI.
12267			 * Need to figure that out!!
12268			 */
12269			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12270				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12271				found = 1;
12272				if ((io->io_hdr.flags &
12273				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12274				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12275					union ctl_ha_msg msg_info;
12276
12277					io->io_hdr.flags |=
12278					                CTL_FLAG_SENT_2OTHER_SC;
12279					msg_info.hdr.nexus = io->io_hdr.nexus;
12280					msg_info.task.task_action =
12281						CTL_TASK_ABORT_TASK;
12282					msg_info.task.tag_num =
12283						io->taskio.tag_num;
12284					msg_info.task.tag_type =
12285						io->taskio.tag_type;
12286					msg_info.hdr.msg_type =
12287						CTL_MSG_MANAGE_TASKS;
12288					msg_info.hdr.original_sc = NULL;
12289					msg_info.hdr.serializing_sc = NULL;
12290#if 0
12291					printf("Sent Abort to other side\n");
12292#endif
12293					if (CTL_HA_STATUS_SUCCESS !=
12294					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12295		    				(void *)&msg_info,
12296						sizeof(msg_info), 0)) {
12297					}
12298				}
12299#if 0
12300				printf("ctl_abort_task: found I/O to abort\n");
12301#endif
12302				break;
12303			}
12304		}
12305	}
12306	mtx_unlock(&lun->lun_lock);
12307
12308	if (found == 0) {
12309		/*
12310		 * This isn't really an error.  It's entirely possible for
12311		 * the abort and command completion to cross on the wire.
12312		 * This is more of an informative/diagnostic error.
12313		 */
12314#if 0
12315		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12316		       "%d:%d:%d:%d tag %d type %d\n",
12317		       io->io_hdr.nexus.initid.id,
12318		       io->io_hdr.nexus.targ_port,
12319		       io->io_hdr.nexus.targ_target.id,
12320		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12321		       io->taskio.tag_type);
12322#endif
12323	}
12324	return (0);
12325}
12326
12327static void
12328ctl_run_task(union ctl_io *io)
12329{
12330	struct ctl_softc *ctl_softc = control_softc;
12331	int retval = 1;
12332	const char *task_desc;
12333
12334	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12335
12336	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12337	    ("ctl_run_task: Unextected io_type %d\n",
12338	     io->io_hdr.io_type));
12339
12340	task_desc = ctl_scsi_task_string(&io->taskio);
12341	if (task_desc != NULL) {
12342#ifdef NEEDTOPORT
12343		csevent_log(CSC_CTL | CSC_SHELF_SW |
12344			    CTL_TASK_REPORT,
12345			    csevent_LogType_Trace,
12346			    csevent_Severity_Information,
12347			    csevent_AlertLevel_Green,
12348			    csevent_FRU_Firmware,
12349			    csevent_FRU_Unknown,
12350			    "CTL: received task: %s",task_desc);
12351#endif
12352	} else {
12353#ifdef NEEDTOPORT
12354		csevent_log(CSC_CTL | CSC_SHELF_SW |
12355			    CTL_TASK_REPORT,
12356			    csevent_LogType_Trace,
12357			    csevent_Severity_Information,
12358			    csevent_AlertLevel_Green,
12359			    csevent_FRU_Firmware,
12360			    csevent_FRU_Unknown,
12361			    "CTL: received unknown task "
12362			    "type: %d (%#x)",
12363			    io->taskio.task_action,
12364			    io->taskio.task_action);
12365#endif
12366	}
12367	switch (io->taskio.task_action) {
12368	case CTL_TASK_ABORT_TASK:
12369		retval = ctl_abort_task(io);
12370		break;
12371	case CTL_TASK_ABORT_TASK_SET:
12372	case CTL_TASK_CLEAR_TASK_SET:
12373		retval = ctl_abort_task_set(io);
12374		break;
12375	case CTL_TASK_CLEAR_ACA:
12376		break;
12377	case CTL_TASK_I_T_NEXUS_RESET:
12378		retval = ctl_i_t_nexus_reset(io);
12379		break;
12380	case CTL_TASK_LUN_RESET: {
12381		struct ctl_lun *lun;
12382		uint32_t targ_lun;
12383
12384		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12385		mtx_lock(&ctl_softc->ctl_lock);
12386		if ((targ_lun < CTL_MAX_LUNS)
12387		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12388			lun = ctl_softc->ctl_luns[targ_lun];
12389		else {
12390			mtx_unlock(&ctl_softc->ctl_lock);
12391			retval = 1;
12392			break;
12393		}
12394
12395		if (!(io->io_hdr.flags &
12396		    CTL_FLAG_FROM_OTHER_SC)) {
12397			union ctl_ha_msg msg_info;
12398
12399			io->io_hdr.flags |=
12400				CTL_FLAG_SENT_2OTHER_SC;
12401			msg_info.hdr.msg_type =
12402				CTL_MSG_MANAGE_TASKS;
12403			msg_info.hdr.nexus = io->io_hdr.nexus;
12404			msg_info.task.task_action =
12405				CTL_TASK_LUN_RESET;
12406			msg_info.hdr.original_sc = NULL;
12407			msg_info.hdr.serializing_sc = NULL;
12408			if (CTL_HA_STATUS_SUCCESS !=
12409			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12410			    (void *)&msg_info,
12411			    sizeof(msg_info), 0)) {
12412			}
12413		}
12414
12415		retval = ctl_lun_reset(lun, io,
12416				       CTL_UA_LUN_RESET);
12417		mtx_unlock(&ctl_softc->ctl_lock);
12418		break;
12419	}
12420	case CTL_TASK_TARGET_RESET:
12421		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12422		break;
12423	case CTL_TASK_BUS_RESET:
12424		retval = ctl_bus_reset(ctl_softc, io);
12425		break;
12426	case CTL_TASK_PORT_LOGIN:
12427		break;
12428	case CTL_TASK_PORT_LOGOUT:
12429		break;
12430	default:
12431		printf("ctl_run_task: got unknown task management event %d\n",
12432		       io->taskio.task_action);
12433		break;
12434	}
12435	if (retval == 0)
12436		io->io_hdr.status = CTL_SUCCESS;
12437	else
12438		io->io_hdr.status = CTL_ERROR;
12439	ctl_done(io);
12440}
12441
12442/*
12443 * For HA operation.  Handle commands that come in from the other
12444 * controller.
12445 */
12446static void
12447ctl_handle_isc(union ctl_io *io)
12448{
12449	int free_io;
12450	struct ctl_lun *lun;
12451	struct ctl_softc *ctl_softc;
12452	uint32_t targ_lun;
12453
12454	ctl_softc = control_softc;
12455
12456	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12457	lun = ctl_softc->ctl_luns[targ_lun];
12458
12459	switch (io->io_hdr.msg_type) {
12460	case CTL_MSG_SERIALIZE:
12461		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12462		break;
12463	case CTL_MSG_R2R: {
12464		const struct ctl_cmd_entry *entry;
12465
12466		/*
12467		 * This is only used in SER_ONLY mode.
12468		 */
12469		free_io = 0;
12470		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12471		mtx_lock(&lun->lun_lock);
12472		if (ctl_scsiio_lun_check(ctl_softc, lun,
12473		    entry, (struct ctl_scsiio *)io) != 0) {
12474			mtx_unlock(&lun->lun_lock);
12475			ctl_done(io);
12476			break;
12477		}
12478		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12479		mtx_unlock(&lun->lun_lock);
12480		ctl_enqueue_rtr(io);
12481		break;
12482	}
12483	case CTL_MSG_FINISH_IO:
12484		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12485			free_io = 0;
12486			ctl_done(io);
12487		} else {
12488			free_io = 1;
12489			mtx_lock(&lun->lun_lock);
12490			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12491				     ooa_links);
12492			ctl_check_blocked(lun);
12493			mtx_unlock(&lun->lun_lock);
12494		}
12495		break;
12496	case CTL_MSG_PERS_ACTION:
12497		ctl_hndl_per_res_out_on_other_sc(
12498			(union ctl_ha_msg *)&io->presio.pr_msg);
12499		free_io = 1;
12500		break;
12501	case CTL_MSG_BAD_JUJU:
12502		free_io = 0;
12503		ctl_done(io);
12504		break;
12505	case CTL_MSG_DATAMOVE:
12506		/* Only used in XFER mode */
12507		free_io = 0;
12508		ctl_datamove_remote(io);
12509		break;
12510	case CTL_MSG_DATAMOVE_DONE:
12511		/* Only used in XFER mode */
12512		free_io = 0;
12513		io->scsiio.be_move_done(io);
12514		break;
12515	default:
12516		free_io = 1;
12517		printf("%s: Invalid message type %d\n",
12518		       __func__, io->io_hdr.msg_type);
12519		break;
12520	}
12521	if (free_io)
12522		ctl_free_io(io);
12523
12524}
12525
12526
12527/*
12528 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12529 * there is no match.
12530 */
12531static ctl_lun_error_pattern
12532ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12533{
12534	const struct ctl_cmd_entry *entry;
12535	ctl_lun_error_pattern filtered_pattern, pattern;
12536
12537	pattern = desc->error_pattern;
12538
12539	/*
12540	 * XXX KDM we need more data passed into this function to match a
12541	 * custom pattern, and we actually need to implement custom pattern
12542	 * matching.
12543	 */
12544	if (pattern & CTL_LUN_PAT_CMD)
12545		return (CTL_LUN_PAT_CMD);
12546
12547	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12548		return (CTL_LUN_PAT_ANY);
12549
12550	entry = ctl_get_cmd_entry(ctsio, NULL);
12551
12552	filtered_pattern = entry->pattern & pattern;
12553
12554	/*
12555	 * If the user requested specific flags in the pattern (e.g.
12556	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12557	 * flags.
12558	 *
12559	 * If the user did not specify any flags, it doesn't matter whether
12560	 * or not the command supports the flags.
12561	 */
12562	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12563	     (pattern & ~CTL_LUN_PAT_MASK))
12564		return (CTL_LUN_PAT_NONE);
12565
12566	/*
12567	 * If the user asked for a range check, see if the requested LBA
12568	 * range overlaps with this command's LBA range.
12569	 */
12570	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12571		uint64_t lba1;
12572		uint64_t len1;
12573		ctl_action action;
12574		int retval;
12575
12576		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12577		if (retval != 0)
12578			return (CTL_LUN_PAT_NONE);
12579
12580		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12581					      desc->lba_range.len);
12582		/*
12583		 * A "pass" means that the LBA ranges don't overlap, so
12584		 * this doesn't match the user's range criteria.
12585		 */
12586		if (action == CTL_ACTION_PASS)
12587			return (CTL_LUN_PAT_NONE);
12588	}
12589
12590	return (filtered_pattern);
12591}
12592
12593static void
12594ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12595{
12596	struct ctl_error_desc *desc, *desc2;
12597
12598	mtx_assert(&lun->lun_lock, MA_OWNED);
12599
12600	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12601		ctl_lun_error_pattern pattern;
12602		/*
12603		 * Check to see whether this particular command matches
12604		 * the pattern in the descriptor.
12605		 */
12606		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12607		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12608			continue;
12609
12610		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12611		case CTL_LUN_INJ_ABORTED:
12612			ctl_set_aborted(&io->scsiio);
12613			break;
12614		case CTL_LUN_INJ_MEDIUM_ERR:
12615			ctl_set_medium_error(&io->scsiio);
12616			break;
12617		case CTL_LUN_INJ_UA:
12618			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12619			 * OCCURRED */
12620			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12621			break;
12622		case CTL_LUN_INJ_CUSTOM:
12623			/*
12624			 * We're assuming the user knows what he is doing.
12625			 * Just copy the sense information without doing
12626			 * checks.
12627			 */
12628			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12629			      ctl_min(sizeof(desc->custom_sense),
12630				      sizeof(io->scsiio.sense_data)));
12631			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12632			io->scsiio.sense_len = SSD_FULL_SIZE;
12633			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12634			break;
12635		case CTL_LUN_INJ_NONE:
12636		default:
12637			/*
12638			 * If this is an error injection type we don't know
12639			 * about, clear the continuous flag (if it is set)
12640			 * so it will get deleted below.
12641			 */
12642			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12643			break;
12644		}
12645		/*
12646		 * By default, each error injection action is a one-shot
12647		 */
12648		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12649			continue;
12650
12651		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12652
12653		free(desc, M_CTL);
12654	}
12655}
12656
12657#ifdef CTL_IO_DELAY
12658static void
12659ctl_datamove_timer_wakeup(void *arg)
12660{
12661	union ctl_io *io;
12662
12663	io = (union ctl_io *)arg;
12664
12665	ctl_datamove(io);
12666}
12667#endif /* CTL_IO_DELAY */
12668
12669void
12670ctl_datamove(union ctl_io *io)
12671{
12672	void (*fe_datamove)(union ctl_io *io);
12673
12674	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12675
12676	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12677
12678#ifdef CTL_TIME_IO
12679	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12680		char str[256];
12681		char path_str[64];
12682		struct sbuf sb;
12683
12684		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12685		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12686
12687		sbuf_cat(&sb, path_str);
12688		switch (io->io_hdr.io_type) {
12689		case CTL_IO_SCSI:
12690			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12691			sbuf_printf(&sb, "\n");
12692			sbuf_cat(&sb, path_str);
12693			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12694				    io->scsiio.tag_num, io->scsiio.tag_type);
12695			break;
12696		case CTL_IO_TASK:
12697			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12698				    "Tag Type: %d\n", io->taskio.task_action,
12699				    io->taskio.tag_num, io->taskio.tag_type);
12700			break;
12701		default:
12702			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12703			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12704			break;
12705		}
12706		sbuf_cat(&sb, path_str);
12707		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12708			    (intmax_t)time_uptime - io->io_hdr.start_time);
12709		sbuf_finish(&sb);
12710		printf("%s", sbuf_data(&sb));
12711	}
12712#endif /* CTL_TIME_IO */
12713
12714#ifdef CTL_IO_DELAY
12715	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12716		struct ctl_lun *lun;
12717
12718		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12719
12720		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12721	} else {
12722		struct ctl_lun *lun;
12723
12724		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12725		if ((lun != NULL)
12726		 && (lun->delay_info.datamove_delay > 0)) {
12727			struct callout *callout;
12728
12729			callout = (struct callout *)&io->io_hdr.timer_bytes;
12730			callout_init(callout, /*mpsafe*/ 1);
12731			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12732			callout_reset(callout,
12733				      lun->delay_info.datamove_delay * hz,
12734				      ctl_datamove_timer_wakeup, io);
12735			if (lun->delay_info.datamove_type ==
12736			    CTL_DELAY_TYPE_ONESHOT)
12737				lun->delay_info.datamove_delay = 0;
12738			return;
12739		}
12740	}
12741#endif
12742
12743	/*
12744	 * This command has been aborted.  Set the port status, so we fail
12745	 * the data move.
12746	 */
12747	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12748		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12749		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12750		       io->io_hdr.nexus.targ_port,
12751		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12752		       io->io_hdr.nexus.targ_lun);
12753		io->io_hdr.port_status = 31337;
12754		/*
12755		 * Note that the backend, in this case, will get the
12756		 * callback in its context.  In other cases it may get
12757		 * called in the frontend's interrupt thread context.
12758		 */
12759		io->scsiio.be_move_done(io);
12760		return;
12761	}
12762
12763	/*
12764	 * If we're in XFER mode and this I/O is from the other shelf
12765	 * controller, we need to send the DMA to the other side to
12766	 * actually transfer the data to/from the host.  In serialize only
12767	 * mode the transfer happens below CTL and ctl_datamove() is only
12768	 * called on the machine that originally received the I/O.
12769	 */
12770	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12771	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12772		union ctl_ha_msg msg;
12773		uint32_t sg_entries_sent;
12774		int do_sg_copy;
12775		int i;
12776
12777		memset(&msg, 0, sizeof(msg));
12778		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12779		msg.hdr.original_sc = io->io_hdr.original_sc;
12780		msg.hdr.serializing_sc = io;
12781		msg.hdr.nexus = io->io_hdr.nexus;
12782		msg.dt.flags = io->io_hdr.flags;
12783		/*
12784		 * We convert everything into a S/G list here.  We can't
12785		 * pass by reference, only by value between controllers.
12786		 * So we can't pass a pointer to the S/G list, only as many
12787		 * S/G entries as we can fit in here.  If it's possible for
12788		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12789		 * then we need to break this up into multiple transfers.
12790		 */
12791		if (io->scsiio.kern_sg_entries == 0) {
12792			msg.dt.kern_sg_entries = 1;
12793			/*
12794			 * If this is in cached memory, flush the cache
12795			 * before we send the DMA request to the other
12796			 * controller.  We want to do this in either the
12797			 * read or the write case.  The read case is
12798			 * straightforward.  In the write case, we want to
12799			 * make sure nothing is in the local cache that
12800			 * could overwrite the DMAed data.
12801			 */
12802			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12803				/*
12804				 * XXX KDM use bus_dmamap_sync() here.
12805				 */
12806			}
12807
12808			/*
12809			 * Convert to a physical address if this is a
12810			 * virtual address.
12811			 */
12812			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12813				msg.dt.sg_list[0].addr =
12814					io->scsiio.kern_data_ptr;
12815			} else {
12816				/*
12817				 * XXX KDM use busdma here!
12818				 */
12819#if 0
12820				msg.dt.sg_list[0].addr = (void *)
12821					vtophys(io->scsiio.kern_data_ptr);
12822#endif
12823			}
12824
12825			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12826			do_sg_copy = 0;
12827		} else {
12828			struct ctl_sg_entry *sgl;
12829
12830			do_sg_copy = 1;
12831			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12832			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12833			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12834				/*
12835				 * XXX KDM use bus_dmamap_sync() here.
12836				 */
12837			}
12838		}
12839
12840		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12841		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12842		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12843		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12844		msg.dt.sg_sequence = 0;
12845
12846		/*
12847		 * Loop until we've sent all of the S/G entries.  On the
12848		 * other end, we'll recompose these S/G entries into one
12849		 * contiguous list before passing it to the
12850		 */
12851		for (sg_entries_sent = 0; sg_entries_sent <
12852		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12853			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12854				sizeof(msg.dt.sg_list[0])),
12855				msg.dt.kern_sg_entries - sg_entries_sent);
12856
12857			if (do_sg_copy != 0) {
12858				struct ctl_sg_entry *sgl;
12859				int j;
12860
12861				sgl = (struct ctl_sg_entry *)
12862					io->scsiio.kern_data_ptr;
12863				/*
12864				 * If this is in cached memory, flush the cache
12865				 * before we send the DMA request to the other
12866				 * controller.  We want to do this in either
12867				 * the * read or the write case.  The read
12868				 * case is straightforward.  In the write
12869				 * case, we want to make sure nothing is
12870				 * in the local cache that could overwrite
12871				 * the DMAed data.
12872				 */
12873
12874				for (i = sg_entries_sent, j = 0;
12875				     i < msg.dt.cur_sg_entries; i++, j++) {
12876					if ((io->io_hdr.flags &
12877					     CTL_FLAG_NO_DATASYNC) == 0) {
12878						/*
12879						 * XXX KDM use bus_dmamap_sync()
12880						 */
12881					}
12882					if ((io->io_hdr.flags &
12883					     CTL_FLAG_BUS_ADDR) == 0) {
12884						/*
12885						 * XXX KDM use busdma.
12886						 */
12887#if 0
12888						msg.dt.sg_list[j].addr =(void *)
12889						       vtophys(sgl[i].addr);
12890#endif
12891					} else {
12892						msg.dt.sg_list[j].addr =
12893							sgl[i].addr;
12894					}
12895					msg.dt.sg_list[j].len = sgl[i].len;
12896				}
12897			}
12898
12899			sg_entries_sent += msg.dt.cur_sg_entries;
12900			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12901				msg.dt.sg_last = 1;
12902			else
12903				msg.dt.sg_last = 0;
12904
12905			/*
12906			 * XXX KDM drop and reacquire the lock here?
12907			 */
12908			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12909			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12910				/*
12911				 * XXX do something here.
12912				 */
12913			}
12914
12915			msg.dt.sent_sg_entries = sg_entries_sent;
12916		}
12917		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12918		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12919			ctl_failover_io(io, /*have_lock*/ 0);
12920
12921	} else {
12922
12923		/*
12924		 * Lookup the fe_datamove() function for this particular
12925		 * front end.
12926		 */
12927		fe_datamove =
12928		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12929
12930		fe_datamove(io);
12931	}
12932}
12933
12934static void
12935ctl_send_datamove_done(union ctl_io *io, int have_lock)
12936{
12937	union ctl_ha_msg msg;
12938	int isc_status;
12939
12940	memset(&msg, 0, sizeof(msg));
12941
12942	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12943	msg.hdr.original_sc = io;
12944	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12945	msg.hdr.nexus = io->io_hdr.nexus;
12946	msg.hdr.status = io->io_hdr.status;
12947	msg.scsi.tag_num = io->scsiio.tag_num;
12948	msg.scsi.tag_type = io->scsiio.tag_type;
12949	msg.scsi.scsi_status = io->scsiio.scsi_status;
12950	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12951	       sizeof(io->scsiio.sense_data));
12952	msg.scsi.sense_len = io->scsiio.sense_len;
12953	msg.scsi.sense_residual = io->scsiio.sense_residual;
12954	msg.scsi.fetd_status = io->io_hdr.port_status;
12955	msg.scsi.residual = io->scsiio.residual;
12956	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12957
12958	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12959		ctl_failover_io(io, /*have_lock*/ have_lock);
12960		return;
12961	}
12962
12963	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12964	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12965		/* XXX do something if this fails */
12966	}
12967
12968}
12969
12970/*
12971 * The DMA to the remote side is done, now we need to tell the other side
12972 * we're done so it can continue with its data movement.
12973 */
12974static void
12975ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12976{
12977	union ctl_io *io;
12978
12979	io = rq->context;
12980
12981	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12982		printf("%s: ISC DMA write failed with error %d", __func__,
12983		       rq->ret);
12984		ctl_set_internal_failure(&io->scsiio,
12985					 /*sks_valid*/ 1,
12986					 /*retry_count*/ rq->ret);
12987	}
12988
12989	ctl_dt_req_free(rq);
12990
12991	/*
12992	 * In this case, we had to malloc the memory locally.  Free it.
12993	 */
12994	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12995		int i;
12996		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12997			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12998	}
12999	/*
13000	 * The data is in local and remote memory, so now we need to send
13001	 * status (good or back) back to the other side.
13002	 */
13003	ctl_send_datamove_done(io, /*have_lock*/ 0);
13004}
13005
13006/*
13007 * We've moved the data from the host/controller into local memory.  Now we
13008 * need to push it over to the remote controller's memory.
13009 */
13010static int
13011ctl_datamove_remote_dm_write_cb(union ctl_io *io)
13012{
13013	int retval;
13014
13015	retval = 0;
13016
13017	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
13018					  ctl_datamove_remote_write_cb);
13019
13020	return (retval);
13021}
13022
13023static void
13024ctl_datamove_remote_write(union ctl_io *io)
13025{
13026	int retval;
13027	void (*fe_datamove)(union ctl_io *io);
13028
13029	/*
13030	 * - Get the data from the host/HBA into local memory.
13031	 * - DMA memory from the local controller to the remote controller.
13032	 * - Send status back to the remote controller.
13033	 */
13034
13035	retval = ctl_datamove_remote_sgl_setup(io);
13036	if (retval != 0)
13037		return;
13038
13039	/* Switch the pointer over so the FETD knows what to do */
13040	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13041
13042	/*
13043	 * Use a custom move done callback, since we need to send completion
13044	 * back to the other controller, not to the backend on this side.
13045	 */
13046	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
13047
13048	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13049
13050	fe_datamove(io);
13051
13052	return;
13053
13054}
13055
13056static int
13057ctl_datamove_remote_dm_read_cb(union ctl_io *io)
13058{
13059#if 0
13060	char str[256];
13061	char path_str[64];
13062	struct sbuf sb;
13063#endif
13064
13065	/*
13066	 * In this case, we had to malloc the memory locally.  Free it.
13067	 */
13068	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13069		int i;
13070		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13071			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13072	}
13073
13074#if 0
13075	scsi_path_string(io, path_str, sizeof(path_str));
13076	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13077	sbuf_cat(&sb, path_str);
13078	scsi_command_string(&io->scsiio, NULL, &sb);
13079	sbuf_printf(&sb, "\n");
13080	sbuf_cat(&sb, path_str);
13081	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13082		    io->scsiio.tag_num, io->scsiio.tag_type);
13083	sbuf_cat(&sb, path_str);
13084	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
13085		    io->io_hdr.flags, io->io_hdr.status);
13086	sbuf_finish(&sb);
13087	printk("%s", sbuf_data(&sb));
13088#endif
13089
13090
13091	/*
13092	 * The read is done, now we need to send status (good or bad) back
13093	 * to the other side.
13094	 */
13095	ctl_send_datamove_done(io, /*have_lock*/ 0);
13096
13097	return (0);
13098}
13099
13100static void
13101ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
13102{
13103	union ctl_io *io;
13104	void (*fe_datamove)(union ctl_io *io);
13105
13106	io = rq->context;
13107
13108	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13109		printf("%s: ISC DMA read failed with error %d", __func__,
13110		       rq->ret);
13111		ctl_set_internal_failure(&io->scsiio,
13112					 /*sks_valid*/ 1,
13113					 /*retry_count*/ rq->ret);
13114	}
13115
13116	ctl_dt_req_free(rq);
13117
13118	/* Switch the pointer over so the FETD knows what to do */
13119	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13120
13121	/*
13122	 * Use a custom move done callback, since we need to send completion
13123	 * back to the other controller, not to the backend on this side.
13124	 */
13125	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13126
13127	/* XXX KDM add checks like the ones in ctl_datamove? */
13128
13129	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13130
13131	fe_datamove(io);
13132}
13133
13134static int
13135ctl_datamove_remote_sgl_setup(union ctl_io *io)
13136{
13137	struct ctl_sg_entry *local_sglist, *remote_sglist;
13138	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13139	struct ctl_softc *softc;
13140	int retval;
13141	int i;
13142
13143	retval = 0;
13144	softc = control_softc;
13145
13146	local_sglist = io->io_hdr.local_sglist;
13147	local_dma_sglist = io->io_hdr.local_dma_sglist;
13148	remote_sglist = io->io_hdr.remote_sglist;
13149	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13150
13151	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13152		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13153			local_sglist[i].len = remote_sglist[i].len;
13154
13155			/*
13156			 * XXX Detect the situation where the RS-level I/O
13157			 * redirector on the other side has already read the
13158			 * data off of the AOR RS on this side, and
13159			 * transferred it to remote (mirror) memory on the
13160			 * other side.  Since we already have the data in
13161			 * memory here, we just need to use it.
13162			 *
13163			 * XXX KDM this can probably be removed once we
13164			 * get the cache device code in and take the
13165			 * current AOR implementation out.
13166			 */
13167#ifdef NEEDTOPORT
13168			if ((remote_sglist[i].addr >=
13169			     (void *)vtophys(softc->mirr->addr))
13170			 && (remote_sglist[i].addr <
13171			     ((void *)vtophys(softc->mirr->addr) +
13172			     CacheMirrorOffset))) {
13173				local_sglist[i].addr = remote_sglist[i].addr -
13174					CacheMirrorOffset;
13175				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13176				     CTL_FLAG_DATA_IN)
13177					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13178			} else {
13179				local_sglist[i].addr = remote_sglist[i].addr +
13180					CacheMirrorOffset;
13181			}
13182#endif
13183#if 0
13184			printf("%s: local %p, remote %p, len %d\n",
13185			       __func__, local_sglist[i].addr,
13186			       remote_sglist[i].addr, local_sglist[i].len);
13187#endif
13188		}
13189	} else {
13190		uint32_t len_to_go;
13191
13192		/*
13193		 * In this case, we don't have automatically allocated
13194		 * memory for this I/O on this controller.  This typically
13195		 * happens with internal CTL I/O -- e.g. inquiry, mode
13196		 * sense, etc.  Anything coming from RAIDCore will have
13197		 * a mirror area available.
13198		 */
13199		len_to_go = io->scsiio.kern_data_len;
13200
13201		/*
13202		 * Clear the no datasync flag, we have to use malloced
13203		 * buffers.
13204		 */
13205		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13206
13207		/*
13208		 * The difficult thing here is that the size of the various
13209		 * S/G segments may be different than the size from the
13210		 * remote controller.  That'll make it harder when DMAing
13211		 * the data back to the other side.
13212		 */
13213		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13214		     sizeof(io->io_hdr.remote_sglist[0])) &&
13215		     (len_to_go > 0); i++) {
13216			local_sglist[i].len = ctl_min(len_to_go, 131072);
13217			CTL_SIZE_8B(local_dma_sglist[i].len,
13218				    local_sglist[i].len);
13219			local_sglist[i].addr =
13220				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13221
13222			local_dma_sglist[i].addr = local_sglist[i].addr;
13223
13224			if (local_sglist[i].addr == NULL) {
13225				int j;
13226
13227				printf("malloc failed for %zd bytes!",
13228				       local_dma_sglist[i].len);
13229				for (j = 0; j < i; j++) {
13230					free(local_sglist[j].addr, M_CTL);
13231				}
13232				ctl_set_internal_failure(&io->scsiio,
13233							 /*sks_valid*/ 1,
13234							 /*retry_count*/ 4857);
13235				retval = 1;
13236				goto bailout_error;
13237
13238			}
13239			/* XXX KDM do we need a sync here? */
13240
13241			len_to_go -= local_sglist[i].len;
13242		}
13243		/*
13244		 * Reset the number of S/G entries accordingly.  The
13245		 * original number of S/G entries is available in
13246		 * rem_sg_entries.
13247		 */
13248		io->scsiio.kern_sg_entries = i;
13249
13250#if 0
13251		printf("%s: kern_sg_entries = %d\n", __func__,
13252		       io->scsiio.kern_sg_entries);
13253		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13254			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13255			       local_sglist[i].addr, local_sglist[i].len,
13256			       local_dma_sglist[i].len);
13257#endif
13258	}
13259
13260
13261	return (retval);
13262
13263bailout_error:
13264
13265	ctl_send_datamove_done(io, /*have_lock*/ 0);
13266
13267	return (retval);
13268}
13269
13270static int
13271ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13272			 ctl_ha_dt_cb callback)
13273{
13274	struct ctl_ha_dt_req *rq;
13275	struct ctl_sg_entry *remote_sglist, *local_sglist;
13276	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13277	uint32_t local_used, remote_used, total_used;
13278	int retval;
13279	int i, j;
13280
13281	retval = 0;
13282
13283	rq = ctl_dt_req_alloc();
13284
13285	/*
13286	 * If we failed to allocate the request, and if the DMA didn't fail
13287	 * anyway, set busy status.  This is just a resource allocation
13288	 * failure.
13289	 */
13290	if ((rq == NULL)
13291	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13292		ctl_set_busy(&io->scsiio);
13293
13294	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13295
13296		if (rq != NULL)
13297			ctl_dt_req_free(rq);
13298
13299		/*
13300		 * The data move failed.  We need to return status back
13301		 * to the other controller.  No point in trying to DMA
13302		 * data to the remote controller.
13303		 */
13304
13305		ctl_send_datamove_done(io, /*have_lock*/ 0);
13306
13307		retval = 1;
13308
13309		goto bailout;
13310	}
13311
13312	local_sglist = io->io_hdr.local_sglist;
13313	local_dma_sglist = io->io_hdr.local_dma_sglist;
13314	remote_sglist = io->io_hdr.remote_sglist;
13315	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13316	local_used = 0;
13317	remote_used = 0;
13318	total_used = 0;
13319
13320	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13321		rq->ret = CTL_HA_STATUS_SUCCESS;
13322		rq->context = io;
13323		callback(rq);
13324		goto bailout;
13325	}
13326
13327	/*
13328	 * Pull/push the data over the wire from/to the other controller.
13329	 * This takes into account the possibility that the local and
13330	 * remote sglists may not be identical in terms of the size of
13331	 * the elements and the number of elements.
13332	 *
13333	 * One fundamental assumption here is that the length allocated for
13334	 * both the local and remote sglists is identical.  Otherwise, we've
13335	 * essentially got a coding error of some sort.
13336	 */
13337	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13338		int isc_ret;
13339		uint32_t cur_len, dma_length;
13340		uint8_t *tmp_ptr;
13341
13342		rq->id = CTL_HA_DATA_CTL;
13343		rq->command = command;
13344		rq->context = io;
13345
13346		/*
13347		 * Both pointers should be aligned.  But it is possible
13348		 * that the allocation length is not.  They should both
13349		 * also have enough slack left over at the end, though,
13350		 * to round up to the next 8 byte boundary.
13351		 */
13352		cur_len = ctl_min(local_sglist[i].len - local_used,
13353				  remote_sglist[j].len - remote_used);
13354
13355		/*
13356		 * In this case, we have a size issue and need to decrease
13357		 * the size, except in the case where we actually have less
13358		 * than 8 bytes left.  In that case, we need to increase
13359		 * the DMA length to get the last bit.
13360		 */
13361		if ((cur_len & 0x7) != 0) {
13362			if (cur_len > 0x7) {
13363				cur_len = cur_len - (cur_len & 0x7);
13364				dma_length = cur_len;
13365			} else {
13366				CTL_SIZE_8B(dma_length, cur_len);
13367			}
13368
13369		} else
13370			dma_length = cur_len;
13371
13372		/*
13373		 * If we had to allocate memory for this I/O, instead of using
13374		 * the non-cached mirror memory, we'll need to flush the cache
13375		 * before trying to DMA to the other controller.
13376		 *
13377		 * We could end up doing this multiple times for the same
13378		 * segment if we have a larger local segment than remote
13379		 * segment.  That shouldn't be an issue.
13380		 */
13381		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13382			/*
13383			 * XXX KDM use bus_dmamap_sync() here.
13384			 */
13385		}
13386
13387		rq->size = dma_length;
13388
13389		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13390		tmp_ptr += local_used;
13391
13392		/* Use physical addresses when talking to ISC hardware */
13393		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13394			/* XXX KDM use busdma */
13395#if 0
13396			rq->local = vtophys(tmp_ptr);
13397#endif
13398		} else
13399			rq->local = tmp_ptr;
13400
13401		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13402		tmp_ptr += remote_used;
13403		rq->remote = tmp_ptr;
13404
13405		rq->callback = NULL;
13406
13407		local_used += cur_len;
13408		if (local_used >= local_sglist[i].len) {
13409			i++;
13410			local_used = 0;
13411		}
13412
13413		remote_used += cur_len;
13414		if (remote_used >= remote_sglist[j].len) {
13415			j++;
13416			remote_used = 0;
13417		}
13418		total_used += cur_len;
13419
13420		if (total_used >= io->scsiio.kern_data_len)
13421			rq->callback = callback;
13422
13423		if ((rq->size & 0x7) != 0) {
13424			printf("%s: warning: size %d is not on 8b boundary\n",
13425			       __func__, rq->size);
13426		}
13427		if (((uintptr_t)rq->local & 0x7) != 0) {
13428			printf("%s: warning: local %p not on 8b boundary\n",
13429			       __func__, rq->local);
13430		}
13431		if (((uintptr_t)rq->remote & 0x7) != 0) {
13432			printf("%s: warning: remote %p not on 8b boundary\n",
13433			       __func__, rq->local);
13434		}
13435#if 0
13436		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13437		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13438		       rq->local, rq->remote, rq->size);
13439#endif
13440
13441		isc_ret = ctl_dt_single(rq);
13442		if (isc_ret == CTL_HA_STATUS_WAIT)
13443			continue;
13444
13445		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13446			rq->ret = CTL_HA_STATUS_SUCCESS;
13447		} else {
13448			rq->ret = isc_ret;
13449		}
13450		callback(rq);
13451		goto bailout;
13452	}
13453
13454bailout:
13455	return (retval);
13456
13457}
13458
13459static void
13460ctl_datamove_remote_read(union ctl_io *io)
13461{
13462	int retval;
13463	int i;
13464
13465	/*
13466	 * This will send an error to the other controller in the case of a
13467	 * failure.
13468	 */
13469	retval = ctl_datamove_remote_sgl_setup(io);
13470	if (retval != 0)
13471		return;
13472
13473	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13474					  ctl_datamove_remote_read_cb);
13475	if ((retval != 0)
13476	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13477		/*
13478		 * Make sure we free memory if there was an error..  The
13479		 * ctl_datamove_remote_xfer() function will send the
13480		 * datamove done message, or call the callback with an
13481		 * error if there is a problem.
13482		 */
13483		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13484			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13485	}
13486
13487	return;
13488}
13489
13490/*
13491 * Process a datamove request from the other controller.  This is used for
13492 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13493 * first.  Once that is complete, the data gets DMAed into the remote
13494 * controller's memory.  For reads, we DMA from the remote controller's
13495 * memory into our memory first, and then move it out to the FETD.
13496 */
13497static void
13498ctl_datamove_remote(union ctl_io *io)
13499{
13500	struct ctl_softc *softc;
13501
13502	softc = control_softc;
13503
13504	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13505
13506	/*
13507	 * Note that we look for an aborted I/O here, but don't do some of
13508	 * the other checks that ctl_datamove() normally does.
13509	 * We don't need to run the datamove delay code, since that should
13510	 * have been done if need be on the other controller.
13511	 */
13512	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13513		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13514		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13515		       io->io_hdr.nexus.targ_port,
13516		       io->io_hdr.nexus.targ_target.id,
13517		       io->io_hdr.nexus.targ_lun);
13518		io->io_hdr.port_status = 31338;
13519		ctl_send_datamove_done(io, /*have_lock*/ 0);
13520		return;
13521	}
13522
13523	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13524		ctl_datamove_remote_write(io);
13525	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13526		ctl_datamove_remote_read(io);
13527	} else {
13528		union ctl_ha_msg msg;
13529		struct scsi_sense_data *sense;
13530		uint8_t sks[3];
13531		int retry_count;
13532
13533		memset(&msg, 0, sizeof(msg));
13534
13535		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13536		msg.hdr.status = CTL_SCSI_ERROR;
13537		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13538
13539		retry_count = 4243;
13540
13541		sense = &msg.scsi.sense_data;
13542		sks[0] = SSD_SCS_VALID;
13543		sks[1] = (retry_count >> 8) & 0xff;
13544		sks[2] = retry_count & 0xff;
13545
13546		/* "Internal target failure" */
13547		scsi_set_sense_data(sense,
13548				    /*sense_format*/ SSD_TYPE_NONE,
13549				    /*current_error*/ 1,
13550				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13551				    /*asc*/ 0x44,
13552				    /*ascq*/ 0x00,
13553				    /*type*/ SSD_ELEM_SKS,
13554				    /*size*/ sizeof(sks),
13555				    /*data*/ sks,
13556				    SSD_ELEM_NONE);
13557
13558		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13559		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13560			ctl_failover_io(io, /*have_lock*/ 1);
13561			return;
13562		}
13563
13564		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13565		    CTL_HA_STATUS_SUCCESS) {
13566			/* XXX KDM what to do if this fails? */
13567		}
13568		return;
13569	}
13570
13571}
13572
13573static int
13574ctl_process_done(union ctl_io *io)
13575{
13576	struct ctl_lun *lun;
13577	struct ctl_softc *ctl_softc;
13578	void (*fe_done)(union ctl_io *io);
13579	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13580
13581	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13582
13583	fe_done =
13584	    control_softc->ctl_ports[targ_port]->fe_done;
13585
13586#ifdef CTL_TIME_IO
13587	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13588		char str[256];
13589		char path_str[64];
13590		struct sbuf sb;
13591
13592		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13593		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13594
13595		sbuf_cat(&sb, path_str);
13596		switch (io->io_hdr.io_type) {
13597		case CTL_IO_SCSI:
13598			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13599			sbuf_printf(&sb, "\n");
13600			sbuf_cat(&sb, path_str);
13601			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13602				    io->scsiio.tag_num, io->scsiio.tag_type);
13603			break;
13604		case CTL_IO_TASK:
13605			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13606				    "Tag Type: %d\n", io->taskio.task_action,
13607				    io->taskio.tag_num, io->taskio.tag_type);
13608			break;
13609		default:
13610			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13611			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13612			break;
13613		}
13614		sbuf_cat(&sb, path_str);
13615		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13616			    (intmax_t)time_uptime - io->io_hdr.start_time);
13617		sbuf_finish(&sb);
13618		printf("%s", sbuf_data(&sb));
13619	}
13620#endif /* CTL_TIME_IO */
13621
13622	switch (io->io_hdr.io_type) {
13623	case CTL_IO_SCSI:
13624		break;
13625	case CTL_IO_TASK:
13626		if (bootverbose || verbose > 0)
13627			ctl_io_error_print(io, NULL);
13628		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13629			ctl_free_io(io);
13630		else
13631			fe_done(io);
13632		return (CTL_RETVAL_COMPLETE);
13633		break;
13634	default:
13635		printf("ctl_process_done: invalid io type %d\n",
13636		       io->io_hdr.io_type);
13637		panic("ctl_process_done: invalid io type %d\n",
13638		      io->io_hdr.io_type);
13639		break; /* NOTREACHED */
13640	}
13641
13642	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13643	if (lun == NULL) {
13644		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13645				 io->io_hdr.nexus.targ_mapped_lun));
13646		fe_done(io);
13647		goto bailout;
13648	}
13649	ctl_softc = lun->ctl_softc;
13650
13651	mtx_lock(&lun->lun_lock);
13652
13653	/*
13654	 * Check to see if we have any errors to inject here.  We only
13655	 * inject errors for commands that don't already have errors set.
13656	 */
13657	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13658	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13659		ctl_inject_error(lun, io);
13660
13661	/*
13662	 * XXX KDM how do we treat commands that aren't completed
13663	 * successfully?
13664	 *
13665	 * XXX KDM should we also track I/O latency?
13666	 */
13667	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13668	    io->io_hdr.io_type == CTL_IO_SCSI) {
13669#ifdef CTL_TIME_IO
13670		struct bintime cur_bt;
13671#endif
13672		int type;
13673
13674		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13675		    CTL_FLAG_DATA_IN)
13676			type = CTL_STATS_READ;
13677		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13678		    CTL_FLAG_DATA_OUT)
13679			type = CTL_STATS_WRITE;
13680		else
13681			type = CTL_STATS_NO_IO;
13682
13683		lun->stats.ports[targ_port].bytes[type] +=
13684		    io->scsiio.kern_total_len;
13685		lun->stats.ports[targ_port].operations[type]++;
13686#ifdef CTL_TIME_IO
13687		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13688		   &io->io_hdr.dma_bt);
13689		lun->stats.ports[targ_port].num_dmas[type] +=
13690		    io->io_hdr.num_dmas;
13691		getbintime(&cur_bt);
13692		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13693		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13694#endif
13695	}
13696
13697	/*
13698	 * Remove this from the OOA queue.
13699	 */
13700	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13701
13702	/*
13703	 * Run through the blocked queue on this LUN and see if anything
13704	 * has become unblocked, now that this transaction is done.
13705	 */
13706	ctl_check_blocked(lun);
13707
13708	/*
13709	 * If the LUN has been invalidated, free it if there is nothing
13710	 * left on its OOA queue.
13711	 */
13712	if ((lun->flags & CTL_LUN_INVALID)
13713	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13714		mtx_unlock(&lun->lun_lock);
13715		mtx_lock(&ctl_softc->ctl_lock);
13716		ctl_free_lun(lun);
13717		mtx_unlock(&ctl_softc->ctl_lock);
13718	} else
13719		mtx_unlock(&lun->lun_lock);
13720
13721	/*
13722	 * If this command has been aborted, make sure we set the status
13723	 * properly.  The FETD is responsible for freeing the I/O and doing
13724	 * whatever it needs to do to clean up its state.
13725	 */
13726	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13727		ctl_set_task_aborted(&io->scsiio);
13728
13729	/*
13730	 * We print out status for every task management command.  For SCSI
13731	 * commands, we filter out any unit attention errors; they happen
13732	 * on every boot, and would clutter up the log.  Note:  task
13733	 * management commands aren't printed here, they are printed above,
13734	 * since they should never even make it down here.
13735	 */
13736	switch (io->io_hdr.io_type) {
13737	case CTL_IO_SCSI: {
13738		int error_code, sense_key, asc, ascq;
13739
13740		sense_key = 0;
13741
13742		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13743		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13744			/*
13745			 * Since this is just for printing, no need to
13746			 * show errors here.
13747			 */
13748			scsi_extract_sense_len(&io->scsiio.sense_data,
13749					       io->scsiio.sense_len,
13750					       &error_code,
13751					       &sense_key,
13752					       &asc,
13753					       &ascq,
13754					       /*show_errors*/ 0);
13755		}
13756
13757		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13758		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13759		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13760		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13761
13762			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13763				ctl_softc->skipped_prints++;
13764			} else {
13765				uint32_t skipped_prints;
13766
13767				skipped_prints = ctl_softc->skipped_prints;
13768
13769				ctl_softc->skipped_prints = 0;
13770				ctl_softc->last_print_jiffies = time_uptime;
13771
13772				if (skipped_prints > 0) {
13773#ifdef NEEDTOPORT
13774					csevent_log(CSC_CTL | CSC_SHELF_SW |
13775					    CTL_ERROR_REPORT,
13776					    csevent_LogType_Trace,
13777					    csevent_Severity_Information,
13778					    csevent_AlertLevel_Green,
13779					    csevent_FRU_Firmware,
13780					    csevent_FRU_Unknown,
13781					    "High CTL error volume, %d prints "
13782					    "skipped", skipped_prints);
13783#endif
13784				}
13785				if (bootverbose || verbose > 0)
13786					ctl_io_error_print(io, NULL);
13787			}
13788		}
13789		break;
13790	}
13791	case CTL_IO_TASK:
13792		if (bootverbose || verbose > 0)
13793			ctl_io_error_print(io, NULL);
13794		break;
13795	default:
13796		break;
13797	}
13798
13799	/*
13800	 * Tell the FETD or the other shelf controller we're done with this
13801	 * command.  Note that only SCSI commands get to this point.  Task
13802	 * management commands are completed above.
13803	 *
13804	 * We only send status to the other controller if we're in XFER
13805	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13806	 * received the I/O (from CTL's perspective), and so the status is
13807	 * generated there.
13808	 *
13809	 * XXX KDM if we hold the lock here, we could cause a deadlock
13810	 * if the frontend comes back in in this context to queue
13811	 * something.
13812	 */
13813	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13814	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13815		union ctl_ha_msg msg;
13816
13817		memset(&msg, 0, sizeof(msg));
13818		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13819		msg.hdr.original_sc = io->io_hdr.original_sc;
13820		msg.hdr.nexus = io->io_hdr.nexus;
13821		msg.hdr.status = io->io_hdr.status;
13822		msg.scsi.scsi_status = io->scsiio.scsi_status;
13823		msg.scsi.tag_num = io->scsiio.tag_num;
13824		msg.scsi.tag_type = io->scsiio.tag_type;
13825		msg.scsi.sense_len = io->scsiio.sense_len;
13826		msg.scsi.sense_residual = io->scsiio.sense_residual;
13827		msg.scsi.residual = io->scsiio.residual;
13828		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13829		       sizeof(io->scsiio.sense_data));
13830		/*
13831		 * We copy this whether or not this is an I/O-related
13832		 * command.  Otherwise, we'd have to go and check to see
13833		 * whether it's a read/write command, and it really isn't
13834		 * worth it.
13835		 */
13836		memcpy(&msg.scsi.lbalen,
13837		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13838		       sizeof(msg.scsi.lbalen));
13839
13840		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13841				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13842			/* XXX do something here */
13843		}
13844
13845		ctl_free_io(io);
13846	} else
13847		fe_done(io);
13848
13849bailout:
13850
13851	return (CTL_RETVAL_COMPLETE);
13852}
13853
13854#ifdef CTL_WITH_CA
13855/*
13856 * Front end should call this if it doesn't do autosense.  When the request
13857 * sense comes back in from the initiator, we'll dequeue this and send it.
13858 */
13859int
13860ctl_queue_sense(union ctl_io *io)
13861{
13862	struct ctl_lun *lun;
13863	struct ctl_softc *ctl_softc;
13864	uint32_t initidx, targ_lun;
13865
13866	ctl_softc = control_softc;
13867
13868	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13869
13870	/*
13871	 * LUN lookup will likely move to the ctl_work_thread() once we
13872	 * have our new queueing infrastructure (that doesn't put things on
13873	 * a per-LUN queue initially).  That is so that we can handle
13874	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13875	 * can't deal with that right now.
13876	 */
13877	mtx_lock(&ctl_softc->ctl_lock);
13878
13879	/*
13880	 * If we don't have a LUN for this, just toss the sense
13881	 * information.
13882	 */
13883	targ_lun = io->io_hdr.nexus.targ_lun;
13884	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13885	if ((targ_lun < CTL_MAX_LUNS)
13886	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13887		lun = ctl_softc->ctl_luns[targ_lun];
13888	else
13889		goto bailout;
13890
13891	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13892
13893	mtx_lock(&lun->lun_lock);
13894	/*
13895	 * Already have CA set for this LUN...toss the sense information.
13896	 */
13897	if (ctl_is_set(lun->have_ca, initidx)) {
13898		mtx_unlock(&lun->lun_lock);
13899		goto bailout;
13900	}
13901
13902	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13903	       ctl_min(sizeof(lun->pending_sense[initidx]),
13904	       sizeof(io->scsiio.sense_data)));
13905	ctl_set_mask(lun->have_ca, initidx);
13906	mtx_unlock(&lun->lun_lock);
13907
13908bailout:
13909	mtx_unlock(&ctl_softc->ctl_lock);
13910
13911	ctl_free_io(io);
13912
13913	return (CTL_RETVAL_COMPLETE);
13914}
13915#endif
13916
13917/*
13918 * Primary command inlet from frontend ports.  All SCSI and task I/O
13919 * requests must go through this function.
13920 */
13921int
13922ctl_queue(union ctl_io *io)
13923{
13924	struct ctl_softc *ctl_softc;
13925
13926	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13927
13928	ctl_softc = control_softc;
13929
13930#ifdef CTL_TIME_IO
13931	io->io_hdr.start_time = time_uptime;
13932	getbintime(&io->io_hdr.start_bt);
13933#endif /* CTL_TIME_IO */
13934
13935	/* Map FE-specific LUN ID into global one. */
13936	io->io_hdr.nexus.targ_mapped_lun =
13937	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13938
13939	switch (io->io_hdr.io_type) {
13940	case CTL_IO_SCSI:
13941	case CTL_IO_TASK:
13942		ctl_enqueue_incoming(io);
13943		break;
13944	default:
13945		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13946		return (EINVAL);
13947	}
13948
13949	return (CTL_RETVAL_COMPLETE);
13950}
13951
13952#ifdef CTL_IO_DELAY
13953static void
13954ctl_done_timer_wakeup(void *arg)
13955{
13956	union ctl_io *io;
13957
13958	io = (union ctl_io *)arg;
13959	ctl_done(io);
13960}
13961#endif /* CTL_IO_DELAY */
13962
13963void
13964ctl_done(union ctl_io *io)
13965{
13966	struct ctl_softc *ctl_softc;
13967
13968	ctl_softc = control_softc;
13969
13970	/*
13971	 * Enable this to catch duplicate completion issues.
13972	 */
13973#if 0
13974	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13975		printf("%s: type %d msg %d cdb %x iptl: "
13976		       "%d:%d:%d:%d tag 0x%04x "
13977		       "flag %#x status %x\n",
13978			__func__,
13979			io->io_hdr.io_type,
13980			io->io_hdr.msg_type,
13981			io->scsiio.cdb[0],
13982			io->io_hdr.nexus.initid.id,
13983			io->io_hdr.nexus.targ_port,
13984			io->io_hdr.nexus.targ_target.id,
13985			io->io_hdr.nexus.targ_lun,
13986			(io->io_hdr.io_type ==
13987			CTL_IO_TASK) ?
13988			io->taskio.tag_num :
13989			io->scsiio.tag_num,
13990		        io->io_hdr.flags,
13991			io->io_hdr.status);
13992	} else
13993		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13994#endif
13995
13996	/*
13997	 * This is an internal copy of an I/O, and should not go through
13998	 * the normal done processing logic.
13999	 */
14000	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
14001		return;
14002
14003	/*
14004	 * We need to send a msg to the serializing shelf to finish the IO
14005	 * as well.  We don't send a finish message to the other shelf if
14006	 * this is a task management command.  Task management commands
14007	 * aren't serialized in the OOA queue, but rather just executed on
14008	 * both shelf controllers for commands that originated on that
14009	 * controller.
14010	 */
14011	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
14012	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
14013		union ctl_ha_msg msg_io;
14014
14015		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
14016		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
14017		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
14018		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
14019		}
14020		/* continue on to finish IO */
14021	}
14022#ifdef CTL_IO_DELAY
14023	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
14024		struct ctl_lun *lun;
14025
14026		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14027
14028		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
14029	} else {
14030		struct ctl_lun *lun;
14031
14032		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14033
14034		if ((lun != NULL)
14035		 && (lun->delay_info.done_delay > 0)) {
14036			struct callout *callout;
14037
14038			callout = (struct callout *)&io->io_hdr.timer_bytes;
14039			callout_init(callout, /*mpsafe*/ 1);
14040			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
14041			callout_reset(callout,
14042				      lun->delay_info.done_delay * hz,
14043				      ctl_done_timer_wakeup, io);
14044			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
14045				lun->delay_info.done_delay = 0;
14046			return;
14047		}
14048	}
14049#endif /* CTL_IO_DELAY */
14050
14051	ctl_enqueue_done(io);
14052}
14053
14054int
14055ctl_isc(struct ctl_scsiio *ctsio)
14056{
14057	struct ctl_lun *lun;
14058	int retval;
14059
14060	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14061
14062	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
14063
14064	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
14065
14066	retval = lun->backend->data_submit((union ctl_io *)ctsio);
14067
14068	return (retval);
14069}
14070
14071
14072static void
14073ctl_work_thread(void *arg)
14074{
14075	struct ctl_thread *thr = (struct ctl_thread *)arg;
14076	struct ctl_softc *softc = thr->ctl_softc;
14077	union ctl_io *io;
14078	int retval;
14079
14080	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
14081
14082	for (;;) {
14083		retval = 0;
14084
14085		/*
14086		 * We handle the queues in this order:
14087		 * - ISC
14088		 * - done queue (to free up resources, unblock other commands)
14089		 * - RtR queue
14090		 * - incoming queue
14091		 *
14092		 * If those queues are empty, we break out of the loop and
14093		 * go to sleep.
14094		 */
14095		mtx_lock(&thr->queue_lock);
14096		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
14097		if (io != NULL) {
14098			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
14099			mtx_unlock(&thr->queue_lock);
14100			ctl_handle_isc(io);
14101			continue;
14102		}
14103		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
14104		if (io != NULL) {
14105			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
14106			/* clear any blocked commands, call fe_done */
14107			mtx_unlock(&thr->queue_lock);
14108			retval = ctl_process_done(io);
14109			continue;
14110		}
14111		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
14112		if (io != NULL) {
14113			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
14114			mtx_unlock(&thr->queue_lock);
14115			if (io->io_hdr.io_type == CTL_IO_TASK)
14116				ctl_run_task(io);
14117			else
14118				ctl_scsiio_precheck(softc, &io->scsiio);
14119			continue;
14120		}
14121		if (!ctl_pause_rtr) {
14122			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
14123			if (io != NULL) {
14124				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
14125				mtx_unlock(&thr->queue_lock);
14126				retval = ctl_scsiio(&io->scsiio);
14127				if (retval != CTL_RETVAL_COMPLETE)
14128					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
14129				continue;
14130			}
14131		}
14132
14133		/* Sleep until we have something to do. */
14134		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
14135	}
14136}
14137
14138static void
14139ctl_lun_thread(void *arg)
14140{
14141	struct ctl_softc *softc = (struct ctl_softc *)arg;
14142	struct ctl_be_lun *be_lun;
14143	int retval;
14144
14145	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14146
14147	for (;;) {
14148		retval = 0;
14149		mtx_lock(&softc->ctl_lock);
14150		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14151		if (be_lun != NULL) {
14152			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14153			mtx_unlock(&softc->ctl_lock);
14154			ctl_create_lun(be_lun);
14155			continue;
14156		}
14157
14158		/* Sleep until we have something to do. */
14159		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14160		    PDROP | PRIBIO, "-", 0);
14161	}
14162}
14163
14164static void
14165ctl_enqueue_incoming(union ctl_io *io)
14166{
14167	struct ctl_softc *softc = control_softc;
14168	struct ctl_thread *thr;
14169	u_int idx;
14170
14171	idx = (io->io_hdr.nexus.targ_port * 127 +
14172	       io->io_hdr.nexus.initid.id) % worker_threads;
14173	thr = &softc->threads[idx];
14174	mtx_lock(&thr->queue_lock);
14175	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14176	mtx_unlock(&thr->queue_lock);
14177	wakeup(thr);
14178}
14179
14180static void
14181ctl_enqueue_rtr(union ctl_io *io)
14182{
14183	struct ctl_softc *softc = control_softc;
14184	struct ctl_thread *thr;
14185
14186	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14187	mtx_lock(&thr->queue_lock);
14188	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14189	mtx_unlock(&thr->queue_lock);
14190	wakeup(thr);
14191}
14192
14193static void
14194ctl_enqueue_done(union ctl_io *io)
14195{
14196	struct ctl_softc *softc = control_softc;
14197	struct ctl_thread *thr;
14198
14199	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14200	mtx_lock(&thr->queue_lock);
14201	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14202	mtx_unlock(&thr->queue_lock);
14203	wakeup(thr);
14204}
14205
14206static void
14207ctl_enqueue_isc(union ctl_io *io)
14208{
14209	struct ctl_softc *softc = control_softc;
14210	struct ctl_thread *thr;
14211
14212	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14213	mtx_lock(&thr->queue_lock);
14214	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14215	mtx_unlock(&thr->queue_lock);
14216	wakeup(thr);
14217}
14218
14219/* Initialization and failover */
14220
14221void
14222ctl_init_isc_msg(void)
14223{
14224	printf("CTL: Still calling this thing\n");
14225}
14226
14227/*
14228 * Init component
14229 * 	Initializes component into configuration defined by bootMode
14230 *	(see hasc-sv.c)
14231 *  	returns hasc_Status:
14232 * 		OK
14233 *		ERROR - fatal error
14234 */
14235static ctl_ha_comp_status
14236ctl_isc_init(struct ctl_ha_component *c)
14237{
14238	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14239
14240	c->status = ret;
14241	return ret;
14242}
14243
14244/* Start component
14245 * 	Starts component in state requested. If component starts successfully,
14246 *	it must set its own state to the requestrd state
14247 *	When requested state is HASC_STATE_HA, the component may refine it
14248 * 	by adding _SLAVE or _MASTER flags.
14249 *	Currently allowed state transitions are:
14250 *	UNKNOWN->HA		- initial startup
14251 *	UNKNOWN->SINGLE - initial startup when no parter detected
14252 *	HA->SINGLE		- failover
14253 * returns ctl_ha_comp_status:
14254 * 		OK	- component successfully started in requested state
14255 *		FAILED  - could not start the requested state, failover may
14256 * 			  be possible
14257 *		ERROR	- fatal error detected, no future startup possible
14258 */
14259static ctl_ha_comp_status
14260ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14261{
14262	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14263
14264	printf("%s: go\n", __func__);
14265
14266	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14267	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14268		ctl_is_single = 0;
14269		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14270		    != CTL_HA_STATUS_SUCCESS) {
14271			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14272			ret = CTL_HA_COMP_STATUS_ERROR;
14273		}
14274	} else if (CTL_HA_STATE_IS_HA(c->state)
14275		&& CTL_HA_STATE_IS_SINGLE(state)){
14276		// HA->SINGLE transition
14277	        ctl_failover();
14278		ctl_is_single = 1;
14279	} else {
14280		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14281		       c->state, state);
14282		ret = CTL_HA_COMP_STATUS_ERROR;
14283	}
14284	if (CTL_HA_STATE_IS_SINGLE(state))
14285		ctl_is_single = 1;
14286
14287	c->state = state;
14288	c->status = ret;
14289	return ret;
14290}
14291
14292/*
14293 * Quiesce component
14294 * The component must clear any error conditions (set status to OK) and
14295 * prepare itself to another Start call
14296 * returns ctl_ha_comp_status:
14297 * 	OK
14298 *	ERROR
14299 */
14300static ctl_ha_comp_status
14301ctl_isc_quiesce(struct ctl_ha_component *c)
14302{
14303	int ret = CTL_HA_COMP_STATUS_OK;
14304
14305	ctl_pause_rtr = 1;
14306	c->status = ret;
14307	return ret;
14308}
14309
14310struct ctl_ha_component ctl_ha_component_ctlisc =
14311{
14312	.name = "CTL ISC",
14313	.state = CTL_HA_STATE_UNKNOWN,
14314	.init = ctl_isc_init,
14315	.start = ctl_isc_start,
14316	.quiesce = ctl_isc_quiesce
14317};
14318
14319/*
14320 *  vim: ts=8
14321 */
14322