ctl.c revision 288803
1/*-
2 * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3 * Copyright (c) 2012 The FreeBSD Foundation
4 * Copyright (c) 2015 Alexander Motin <mav@FreeBSD.org>
5 * All rights reserved.
6 *
7 * Portions of this software were developed by Edward Tomasz Napierala
8 * under sponsorship from the FreeBSD Foundation.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 *    notice, this list of conditions, and the following disclaimer,
15 *    without modification.
16 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
17 *    substantially similar to the "NO WARRANTY" disclaimer below
18 *    ("Disclaimer") and any redistribution must be conditioned upon
19 *    including a substantially similar Disclaimer requirement for further
20 *    binary redistribution.
21 *
22 * NO WARRANTY
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
31 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
32 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33 * POSSIBILITY OF SUCH DAMAGES.
34 *
35 * $Id$
36 */
37/*
38 * CAM Target Layer, a SCSI device emulation subsystem.
39 *
40 * Author: Ken Merry <ken@FreeBSD.org>
41 */
42
43#define _CTL_C
44
45#include <sys/cdefs.h>
46__FBSDID("$FreeBSD: stable/10/sys/cam/ctl/ctl.c 288803 2015-10-05 11:05:56Z mav $");
47
48#include <sys/param.h>
49#include <sys/systm.h>
50#include <sys/ctype.h>
51#include <sys/kernel.h>
52#include <sys/types.h>
53#include <sys/kthread.h>
54#include <sys/bio.h>
55#include <sys/fcntl.h>
56#include <sys/lock.h>
57#include <sys/module.h>
58#include <sys/mutex.h>
59#include <sys/condvar.h>
60#include <sys/malloc.h>
61#include <sys/conf.h>
62#include <sys/ioccom.h>
63#include <sys/queue.h>
64#include <sys/sbuf.h>
65#include <sys/smp.h>
66#include <sys/endian.h>
67#include <sys/sysctl.h>
68#include <vm/uma.h>
69
70#include <cam/cam.h>
71#include <cam/scsi/scsi_all.h>
72#include <cam/scsi/scsi_da.h>
73#include <cam/ctl/ctl_io.h>
74#include <cam/ctl/ctl.h>
75#include <cam/ctl/ctl_frontend.h>
76#include <cam/ctl/ctl_util.h>
77#include <cam/ctl/ctl_backend.h>
78#include <cam/ctl/ctl_ioctl.h>
79#include <cam/ctl/ctl_ha.h>
80#include <cam/ctl/ctl_private.h>
81#include <cam/ctl/ctl_debug.h>
82#include <cam/ctl/ctl_scsi_all.h>
83#include <cam/ctl/ctl_error.h>
84
85struct ctl_softc *control_softc = NULL;
86
87/*
88 * Template mode pages.
89 */
90
91/*
92 * Note that these are default values only.  The actual values will be
93 * filled in when the user does a mode sense.
94 */
95const static struct copan_debugconf_subpage debugconf_page_default = {
96	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
97	DBGCNF_SUBPAGE_CODE,		/* subpage */
98	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
99	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
100	DBGCNF_VERSION,			/* page_version */
101	{CTL_TIME_IO_DEFAULT_SECS>>8,
102	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
103};
104
105const static struct copan_debugconf_subpage debugconf_page_changeable = {
106	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
107	DBGCNF_SUBPAGE_CODE,		/* subpage */
108	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
109	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
110	0,				/* page_version */
111	{0xff,0xff},			/* ctl_time_io_secs */
112};
113
114const static struct scsi_da_rw_recovery_page rw_er_page_default = {
115	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
116	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
117	/*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
118	/*read_retry_count*/0,
119	/*correction_span*/0,
120	/*head_offset_count*/0,
121	/*data_strobe_offset_cnt*/0,
122	/*byte8*/SMS_RWER_LBPERE,
123	/*write_retry_count*/0,
124	/*reserved2*/0,
125	/*recovery_time_limit*/{0, 0},
126};
127
128const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
129	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
130	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
131	/*byte3*/0,
132	/*read_retry_count*/0,
133	/*correction_span*/0,
134	/*head_offset_count*/0,
135	/*data_strobe_offset_cnt*/0,
136	/*byte8*/0,
137	/*write_retry_count*/0,
138	/*reserved2*/0,
139	/*recovery_time_limit*/{0, 0},
140};
141
142const static struct scsi_format_page format_page_default = {
143	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
144	/*page_length*/sizeof(struct scsi_format_page) - 2,
145	/*tracks_per_zone*/ {0, 0},
146	/*alt_sectors_per_zone*/ {0, 0},
147	/*alt_tracks_per_zone*/ {0, 0},
148	/*alt_tracks_per_lun*/ {0, 0},
149	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
150			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
151	/*bytes_per_sector*/ {0, 0},
152	/*interleave*/ {0, 0},
153	/*track_skew*/ {0, 0},
154	/*cylinder_skew*/ {0, 0},
155	/*flags*/ SFP_HSEC,
156	/*reserved*/ {0, 0, 0}
157};
158
159const static struct scsi_format_page format_page_changeable = {
160	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
161	/*page_length*/sizeof(struct scsi_format_page) - 2,
162	/*tracks_per_zone*/ {0, 0},
163	/*alt_sectors_per_zone*/ {0, 0},
164	/*alt_tracks_per_zone*/ {0, 0},
165	/*alt_tracks_per_lun*/ {0, 0},
166	/*sectors_per_track*/ {0, 0},
167	/*bytes_per_sector*/ {0, 0},
168	/*interleave*/ {0, 0},
169	/*track_skew*/ {0, 0},
170	/*cylinder_skew*/ {0, 0},
171	/*flags*/ 0,
172	/*reserved*/ {0, 0, 0}
173};
174
175const static struct scsi_rigid_disk_page rigid_disk_page_default = {
176	/*page_code*/SMS_RIGID_DISK_PAGE,
177	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
178	/*cylinders*/ {0, 0, 0},
179	/*heads*/ CTL_DEFAULT_HEADS,
180	/*start_write_precomp*/ {0, 0, 0},
181	/*start_reduced_current*/ {0, 0, 0},
182	/*step_rate*/ {0, 0},
183	/*landing_zone_cylinder*/ {0, 0, 0},
184	/*rpl*/ SRDP_RPL_DISABLED,
185	/*rotational_offset*/ 0,
186	/*reserved1*/ 0,
187	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
188			   CTL_DEFAULT_ROTATION_RATE & 0xff},
189	/*reserved2*/ {0, 0}
190};
191
192const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
193	/*page_code*/SMS_RIGID_DISK_PAGE,
194	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
195	/*cylinders*/ {0, 0, 0},
196	/*heads*/ 0,
197	/*start_write_precomp*/ {0, 0, 0},
198	/*start_reduced_current*/ {0, 0, 0},
199	/*step_rate*/ {0, 0},
200	/*landing_zone_cylinder*/ {0, 0, 0},
201	/*rpl*/ 0,
202	/*rotational_offset*/ 0,
203	/*reserved1*/ 0,
204	/*rotation_rate*/ {0, 0},
205	/*reserved2*/ {0, 0}
206};
207
208const static struct scsi_caching_page caching_page_default = {
209	/*page_code*/SMS_CACHING_PAGE,
210	/*page_length*/sizeof(struct scsi_caching_page) - 2,
211	/*flags1*/ SCP_DISC | SCP_WCE,
212	/*ret_priority*/ 0,
213	/*disable_pf_transfer_len*/ {0xff, 0xff},
214	/*min_prefetch*/ {0, 0},
215	/*max_prefetch*/ {0xff, 0xff},
216	/*max_pf_ceiling*/ {0xff, 0xff},
217	/*flags2*/ 0,
218	/*cache_segments*/ 0,
219	/*cache_seg_size*/ {0, 0},
220	/*reserved*/ 0,
221	/*non_cache_seg_size*/ {0, 0, 0}
222};
223
224const static struct scsi_caching_page caching_page_changeable = {
225	/*page_code*/SMS_CACHING_PAGE,
226	/*page_length*/sizeof(struct scsi_caching_page) - 2,
227	/*flags1*/ SCP_WCE | SCP_RCD,
228	/*ret_priority*/ 0,
229	/*disable_pf_transfer_len*/ {0, 0},
230	/*min_prefetch*/ {0, 0},
231	/*max_prefetch*/ {0, 0},
232	/*max_pf_ceiling*/ {0, 0},
233	/*flags2*/ 0,
234	/*cache_segments*/ 0,
235	/*cache_seg_size*/ {0, 0},
236	/*reserved*/ 0,
237	/*non_cache_seg_size*/ {0, 0, 0}
238};
239
240const static struct scsi_control_page control_page_default = {
241	/*page_code*/SMS_CONTROL_MODE_PAGE,
242	/*page_length*/sizeof(struct scsi_control_page) - 2,
243	/*rlec*/0,
244	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
245	/*eca_and_aen*/0,
246	/*flags4*/SCP_TAS,
247	/*aen_holdoff_period*/{0, 0},
248	/*busy_timeout_period*/{0, 0},
249	/*extended_selftest_completion_time*/{0, 0}
250};
251
252const static struct scsi_control_page control_page_changeable = {
253	/*page_code*/SMS_CONTROL_MODE_PAGE,
254	/*page_length*/sizeof(struct scsi_control_page) - 2,
255	/*rlec*/SCP_DSENSE,
256	/*queue_flags*/SCP_QUEUE_ALG_MASK,
257	/*eca_and_aen*/SCP_SWP,
258	/*flags4*/0,
259	/*aen_holdoff_period*/{0, 0},
260	/*busy_timeout_period*/{0, 0},
261	/*extended_selftest_completion_time*/{0, 0}
262};
263
264#define CTL_CEM_LEN	(sizeof(struct scsi_control_ext_page) - 4)
265
266const static struct scsi_control_ext_page control_ext_page_default = {
267	/*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
268	/*subpage_code*/0x01,
269	/*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
270	/*flags*/0,
271	/*prio*/0,
272	/*max_sense*/0
273};
274
275const static struct scsi_control_ext_page control_ext_page_changeable = {
276	/*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
277	/*subpage_code*/0x01,
278	/*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
279	/*flags*/0,
280	/*prio*/0,
281	/*max_sense*/0
282};
283
284const static struct scsi_info_exceptions_page ie_page_default = {
285	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
286	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
287	/*info_flags*/SIEP_FLAGS_DEXCPT,
288	/*mrie*/0,
289	/*interval_timer*/{0, 0, 0, 0},
290	/*report_count*/{0, 0, 0, 0}
291};
292
293const static struct scsi_info_exceptions_page ie_page_changeable = {
294	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
295	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
296	/*info_flags*/0,
297	/*mrie*/0,
298	/*interval_timer*/{0, 0, 0, 0},
299	/*report_count*/{0, 0, 0, 0}
300};
301
302#define CTL_LBPM_LEN	(sizeof(struct ctl_logical_block_provisioning_page) - 4)
303
304const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
305	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
306	/*subpage_code*/0x02,
307	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
308	/*flags*/0,
309	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
310	/*descr*/{}},
311	{{/*flags*/0,
312	  /*resource*/0x01,
313	  /*reserved*/{0, 0},
314	  /*count*/{0, 0, 0, 0}},
315	 {/*flags*/0,
316	  /*resource*/0x02,
317	  /*reserved*/{0, 0},
318	  /*count*/{0, 0, 0, 0}},
319	 {/*flags*/0,
320	  /*resource*/0xf1,
321	  /*reserved*/{0, 0},
322	  /*count*/{0, 0, 0, 0}},
323	 {/*flags*/0,
324	  /*resource*/0xf2,
325	  /*reserved*/{0, 0},
326	  /*count*/{0, 0, 0, 0}}
327	}
328};
329
330const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
331	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
332	/*subpage_code*/0x02,
333	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
334	/*flags*/0,
335	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
336	/*descr*/{}},
337	{{/*flags*/0,
338	  /*resource*/0,
339	  /*reserved*/{0, 0},
340	  /*count*/{0, 0, 0, 0}},
341	 {/*flags*/0,
342	  /*resource*/0,
343	  /*reserved*/{0, 0},
344	  /*count*/{0, 0, 0, 0}},
345	 {/*flags*/0,
346	  /*resource*/0,
347	  /*reserved*/{0, 0},
348	  /*count*/{0, 0, 0, 0}},
349	 {/*flags*/0,
350	  /*resource*/0,
351	  /*reserved*/{0, 0},
352	  /*count*/{0, 0, 0, 0}}
353	}
354};
355
356SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
357static int worker_threads = -1;
358TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
359SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
360    &worker_threads, 1, "Number of worker threads");
361static int ctl_debug = CTL_DEBUG_NONE;
362TUNABLE_INT("kern.cam.ctl.debug", &ctl_debug);
363SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
364    &ctl_debug, 0, "Enabled debug flags");
365
366/*
367 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
368 * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
369 * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
370 * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
371 */
372#define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
373
374static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
375				  int param);
376static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
377static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
378static int ctl_init(void);
379void ctl_shutdown(void);
380static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
381static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
382static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
383static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
384			      struct ctl_ooa *ooa_hdr,
385			      struct ctl_ooa_entry *kern_entries);
386static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
387		     struct thread *td);
388static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
389			 struct ctl_be_lun *be_lun);
390static int ctl_free_lun(struct ctl_lun *lun);
391static void ctl_create_lun(struct ctl_be_lun *be_lun);
392static struct ctl_port * ctl_io_port(struct ctl_io_hdr *io_hdr);
393
394static int ctl_do_mode_select(union ctl_io *io);
395static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
396			   uint64_t res_key, uint64_t sa_res_key,
397			   uint8_t type, uint32_t residx,
398			   struct ctl_scsiio *ctsio,
399			   struct scsi_per_res_out *cdb,
400			   struct scsi_per_res_out_parms* param);
401static void ctl_pro_preempt_other(struct ctl_lun *lun,
402				  union ctl_ha_msg *msg);
403static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
404static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
405static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
406static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
407static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
408static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
409static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
410					 int alloc_len);
411static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
412					 int alloc_len);
413static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
414static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
415static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
416static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
417static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
418static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
419    bool seq);
420static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
421static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
422    union ctl_io *pending_io, union ctl_io *ooa_io);
423static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
424				union ctl_io *starting_io);
425static int ctl_check_blocked(struct ctl_lun *lun);
426static int ctl_scsiio_lun_check(struct ctl_lun *lun,
427				const struct ctl_cmd_entry *entry,
428				struct ctl_scsiio *ctsio);
429static void ctl_failover_lun(union ctl_io *io);
430static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
431			       struct ctl_scsiio *ctsio);
432static int ctl_scsiio(struct ctl_scsiio *ctsio);
433
434static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
435static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
436			    ctl_ua_type ua_type);
437static int ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io,
438			 ctl_ua_type ua_type);
439static int ctl_lun_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
440static int ctl_abort_task(union ctl_io *io);
441static int ctl_abort_task_set(union ctl_io *io);
442static int ctl_query_task(union ctl_io *io, int task_set);
443static int ctl_i_t_nexus_reset(union ctl_io *io);
444static int ctl_query_async_event(union ctl_io *io);
445static void ctl_run_task(union ctl_io *io);
446#ifdef CTL_IO_DELAY
447static void ctl_datamove_timer_wakeup(void *arg);
448static void ctl_done_timer_wakeup(void *arg);
449#endif /* CTL_IO_DELAY */
450
451static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
452static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
453static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
454static void ctl_datamove_remote_write(union ctl_io *io);
455static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
456static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
457static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
458static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
459				    ctl_ha_dt_cb callback);
460static void ctl_datamove_remote_read(union ctl_io *io);
461static void ctl_datamove_remote(union ctl_io *io);
462static void ctl_process_done(union ctl_io *io);
463static void ctl_lun_thread(void *arg);
464static void ctl_thresh_thread(void *arg);
465static void ctl_work_thread(void *arg);
466static void ctl_enqueue_incoming(union ctl_io *io);
467static void ctl_enqueue_rtr(union ctl_io *io);
468static void ctl_enqueue_done(union ctl_io *io);
469static void ctl_enqueue_isc(union ctl_io *io);
470static const struct ctl_cmd_entry *
471    ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
472static const struct ctl_cmd_entry *
473    ctl_validate_command(struct ctl_scsiio *ctsio);
474static int ctl_cmd_applicable(uint8_t lun_type,
475    const struct ctl_cmd_entry *entry);
476
477static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
478static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
479static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
480static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
481
482/*
483 * Load the serialization table.  This isn't very pretty, but is probably
484 * the easiest way to do it.
485 */
486#include "ctl_ser_table.c"
487
488/*
489 * We only need to define open, close and ioctl routines for this driver.
490 */
491static struct cdevsw ctl_cdevsw = {
492	.d_version =	D_VERSION,
493	.d_flags =	0,
494	.d_open =	ctl_open,
495	.d_close =	ctl_close,
496	.d_ioctl =	ctl_ioctl,
497	.d_name =	"ctl",
498};
499
500
501MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
502
503static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
504
505static moduledata_t ctl_moduledata = {
506	"ctl",
507	ctl_module_event_handler,
508	NULL
509};
510
511DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
512MODULE_VERSION(ctl, 1);
513
514static struct ctl_frontend ha_frontend =
515{
516	.name = "ha",
517};
518
519static void
520ctl_ha_datamove(union ctl_io *io)
521{
522	struct ctl_lun *lun;
523	struct ctl_sg_entry *sgl;
524	union ctl_ha_msg msg;
525	uint32_t sg_entries_sent;
526	int do_sg_copy, i, j;
527
528	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
529	memset(&msg.dt, 0, sizeof(msg.dt));
530	msg.hdr.msg_type = CTL_MSG_DATAMOVE;
531	msg.hdr.original_sc = io->io_hdr.original_sc;
532	msg.hdr.serializing_sc = io;
533	msg.hdr.nexus = io->io_hdr.nexus;
534	msg.hdr.status = io->io_hdr.status;
535	msg.dt.flags = io->io_hdr.flags;
536
537	/*
538	 * We convert everything into a S/G list here.  We can't
539	 * pass by reference, only by value between controllers.
540	 * So we can't pass a pointer to the S/G list, only as many
541	 * S/G entries as we can fit in here.  If it's possible for
542	 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
543	 * then we need to break this up into multiple transfers.
544	 */
545	if (io->scsiio.kern_sg_entries == 0) {
546		msg.dt.kern_sg_entries = 1;
547#if 0
548		if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
549			msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
550		} else {
551			/* XXX KDM use busdma here! */
552			msg.dt.sg_list[0].addr =
553			    (void *)vtophys(io->scsiio.kern_data_ptr);
554		}
555#else
556		KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
557		    ("HA does not support BUS_ADDR"));
558		msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
559#endif
560		msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
561		do_sg_copy = 0;
562	} else {
563		msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
564		do_sg_copy = 1;
565	}
566
567	msg.dt.kern_data_len = io->scsiio.kern_data_len;
568	msg.dt.kern_total_len = io->scsiio.kern_total_len;
569	msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
570	msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
571	msg.dt.sg_sequence = 0;
572
573	/*
574	 * Loop until we've sent all of the S/G entries.  On the
575	 * other end, we'll recompose these S/G entries into one
576	 * contiguous list before processing.
577	 */
578	for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
579	    msg.dt.sg_sequence++) {
580		msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
581		    sizeof(msg.dt.sg_list[0])),
582		    msg.dt.kern_sg_entries - sg_entries_sent);
583		if (do_sg_copy != 0) {
584			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
585			for (i = sg_entries_sent, j = 0;
586			     i < msg.dt.cur_sg_entries; i++, j++) {
587#if 0
588				if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
589					msg.dt.sg_list[j].addr = sgl[i].addr;
590				} else {
591					/* XXX KDM use busdma here! */
592					msg.dt.sg_list[j].addr =
593					    (void *)vtophys(sgl[i].addr);
594				}
595#else
596				KASSERT((io->io_hdr.flags &
597				    CTL_FLAG_BUS_ADDR) == 0,
598				    ("HA does not support BUS_ADDR"));
599				msg.dt.sg_list[j].addr = sgl[i].addr;
600#endif
601				msg.dt.sg_list[j].len = sgl[i].len;
602			}
603		}
604
605		sg_entries_sent += msg.dt.cur_sg_entries;
606		msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
607		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
608		    sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
609		    sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
610		    M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
611			io->io_hdr.port_status = 31341;
612			io->scsiio.be_move_done(io);
613			return;
614		}
615		msg.dt.sent_sg_entries = sg_entries_sent;
616	}
617
618	/*
619	 * Officially handover the request from us to peer.
620	 * If failover has just happened, then we must return error.
621	 * If failover happen just after, then it is not our problem.
622	 */
623	if (lun)
624		mtx_lock(&lun->lun_lock);
625	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
626		if (lun)
627			mtx_unlock(&lun->lun_lock);
628		io->io_hdr.port_status = 31342;
629		io->scsiio.be_move_done(io);
630		return;
631	}
632	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
633	io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
634	if (lun)
635		mtx_unlock(&lun->lun_lock);
636}
637
638static void
639ctl_ha_done(union ctl_io *io)
640{
641	union ctl_ha_msg msg;
642
643	if (io->io_hdr.io_type == CTL_IO_SCSI) {
644		memset(&msg, 0, sizeof(msg));
645		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
646		msg.hdr.original_sc = io->io_hdr.original_sc;
647		msg.hdr.nexus = io->io_hdr.nexus;
648		msg.hdr.status = io->io_hdr.status;
649		msg.scsi.scsi_status = io->scsiio.scsi_status;
650		msg.scsi.tag_num = io->scsiio.tag_num;
651		msg.scsi.tag_type = io->scsiio.tag_type;
652		msg.scsi.sense_len = io->scsiio.sense_len;
653		msg.scsi.sense_residual = io->scsiio.sense_residual;
654		msg.scsi.residual = io->scsiio.residual;
655		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
656		    io->scsiio.sense_len);
657		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
658		    sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
659		    msg.scsi.sense_len, M_WAITOK);
660	}
661	ctl_free_io(io);
662}
663
664static void
665ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
666			    union ctl_ha_msg *msg_info)
667{
668	struct ctl_scsiio *ctsio;
669
670	if (msg_info->hdr.original_sc == NULL) {
671		printf("%s: original_sc == NULL!\n", __func__);
672		/* XXX KDM now what? */
673		return;
674	}
675
676	ctsio = &msg_info->hdr.original_sc->scsiio;
677	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
678	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
679	ctsio->io_hdr.status = msg_info->hdr.status;
680	ctsio->scsi_status = msg_info->scsi.scsi_status;
681	ctsio->sense_len = msg_info->scsi.sense_len;
682	ctsio->sense_residual = msg_info->scsi.sense_residual;
683	ctsio->residual = msg_info->scsi.residual;
684	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
685	       msg_info->scsi.sense_len);
686	ctl_enqueue_isc((union ctl_io *)ctsio);
687}
688
689static void
690ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
691				union ctl_ha_msg *msg_info)
692{
693	struct ctl_scsiio *ctsio;
694
695	if (msg_info->hdr.serializing_sc == NULL) {
696		printf("%s: serializing_sc == NULL!\n", __func__);
697		/* XXX KDM now what? */
698		return;
699	}
700
701	ctsio = &msg_info->hdr.serializing_sc->scsiio;
702	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
703	ctl_enqueue_isc((union ctl_io *)ctsio);
704}
705
706void
707ctl_isc_announce_lun(struct ctl_lun *lun)
708{
709	struct ctl_softc *softc = lun->ctl_softc;
710	union ctl_ha_msg *msg;
711	struct ctl_ha_msg_lun_pr_key pr_key;
712	int i, k;
713
714	if (softc->ha_link != CTL_HA_LINK_ONLINE)
715		return;
716	mtx_lock(&lun->lun_lock);
717	i = sizeof(msg->lun);
718	if (lun->lun_devid)
719		i += lun->lun_devid->len;
720	i += sizeof(pr_key) * lun->pr_key_count;
721alloc:
722	mtx_unlock(&lun->lun_lock);
723	msg = malloc(i, M_CTL, M_WAITOK);
724	mtx_lock(&lun->lun_lock);
725	k = sizeof(msg->lun);
726	if (lun->lun_devid)
727		k += lun->lun_devid->len;
728	k += sizeof(pr_key) * lun->pr_key_count;
729	if (i < k) {
730		free(msg, M_CTL);
731		i = k;
732		goto alloc;
733	}
734	bzero(&msg->lun, sizeof(msg->lun));
735	msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
736	msg->hdr.nexus.targ_lun = lun->lun;
737	msg->hdr.nexus.targ_mapped_lun = lun->lun;
738	msg->lun.flags = lun->flags;
739	msg->lun.pr_generation = lun->PRGeneration;
740	msg->lun.pr_res_idx = lun->pr_res_idx;
741	msg->lun.pr_res_type = lun->res_type;
742	msg->lun.pr_key_count = lun->pr_key_count;
743	i = 0;
744	if (lun->lun_devid) {
745		msg->lun.lun_devid_len = lun->lun_devid->len;
746		memcpy(&msg->lun.data[i], lun->lun_devid->data,
747		    msg->lun.lun_devid_len);
748		i += msg->lun.lun_devid_len;
749	}
750	for (k = 0; k < CTL_MAX_INITIATORS; k++) {
751		if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
752			continue;
753		pr_key.pr_iid = k;
754		memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
755		i += sizeof(pr_key);
756	}
757	mtx_unlock(&lun->lun_lock);
758	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
759	    M_WAITOK);
760	free(msg, M_CTL);
761
762	if (lun->flags & CTL_LUN_PRIMARY_SC) {
763		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
764			ctl_isc_announce_mode(lun, -1,
765			    lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
766			    lun->mode_pages.index[i].subpage);
767		}
768	}
769}
770
771void
772ctl_isc_announce_port(struct ctl_port *port)
773{
774	struct ctl_softc *softc = port->ctl_softc;
775	union ctl_ha_msg *msg;
776	int i;
777
778	if (port->targ_port < softc->port_min ||
779	    port->targ_port >= softc->port_max ||
780	    softc->ha_link != CTL_HA_LINK_ONLINE)
781		return;
782	i = sizeof(msg->port) + strlen(port->port_name) + 1;
783	if (port->lun_map)
784		i += sizeof(uint32_t) * CTL_MAX_LUNS;
785	if (port->port_devid)
786		i += port->port_devid->len;
787	if (port->target_devid)
788		i += port->target_devid->len;
789	if (port->init_devid)
790		i += port->init_devid->len;
791	msg = malloc(i, M_CTL, M_WAITOK);
792	bzero(&msg->port, sizeof(msg->port));
793	msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
794	msg->hdr.nexus.targ_port = port->targ_port;
795	msg->port.port_type = port->port_type;
796	msg->port.physical_port = port->physical_port;
797	msg->port.virtual_port = port->virtual_port;
798	msg->port.status = port->status;
799	i = 0;
800	msg->port.name_len = sprintf(&msg->port.data[i],
801	    "%d:%s", softc->ha_id, port->port_name) + 1;
802	i += msg->port.name_len;
803	if (port->lun_map) {
804		msg->port.lun_map_len = sizeof(uint32_t) * CTL_MAX_LUNS;
805		memcpy(&msg->port.data[i], port->lun_map,
806		    msg->port.lun_map_len);
807		i += msg->port.lun_map_len;
808	}
809	if (port->port_devid) {
810		msg->port.port_devid_len = port->port_devid->len;
811		memcpy(&msg->port.data[i], port->port_devid->data,
812		    msg->port.port_devid_len);
813		i += msg->port.port_devid_len;
814	}
815	if (port->target_devid) {
816		msg->port.target_devid_len = port->target_devid->len;
817		memcpy(&msg->port.data[i], port->target_devid->data,
818		    msg->port.target_devid_len);
819		i += msg->port.target_devid_len;
820	}
821	if (port->init_devid) {
822		msg->port.init_devid_len = port->init_devid->len;
823		memcpy(&msg->port.data[i], port->init_devid->data,
824		    msg->port.init_devid_len);
825		i += msg->port.init_devid_len;
826	}
827	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
828	    M_WAITOK);
829	free(msg, M_CTL);
830}
831
832void
833ctl_isc_announce_iid(struct ctl_port *port, int iid)
834{
835	struct ctl_softc *softc = port->ctl_softc;
836	union ctl_ha_msg *msg;
837	int i, l;
838
839	if (port->targ_port < softc->port_min ||
840	    port->targ_port >= softc->port_max ||
841	    softc->ha_link != CTL_HA_LINK_ONLINE)
842		return;
843	mtx_lock(&softc->ctl_lock);
844	i = sizeof(msg->iid);
845	l = 0;
846	if (port->wwpn_iid[iid].name)
847		l = strlen(port->wwpn_iid[iid].name) + 1;
848	i += l;
849	msg = malloc(i, M_CTL, M_NOWAIT);
850	if (msg == NULL) {
851		mtx_unlock(&softc->ctl_lock);
852		return;
853	}
854	bzero(&msg->iid, sizeof(msg->iid));
855	msg->hdr.msg_type = CTL_MSG_IID_SYNC;
856	msg->hdr.nexus.targ_port = port->targ_port;
857	msg->hdr.nexus.initid = iid;
858	msg->iid.in_use = port->wwpn_iid[iid].in_use;
859	msg->iid.name_len = l;
860	msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
861	if (port->wwpn_iid[iid].name)
862		strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
863	mtx_unlock(&softc->ctl_lock);
864	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
865	free(msg, M_CTL);
866}
867
868void
869ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
870    uint8_t page, uint8_t subpage)
871{
872	struct ctl_softc *softc = lun->ctl_softc;
873	union ctl_ha_msg msg;
874	int i;
875
876	if (softc->ha_link != CTL_HA_LINK_ONLINE)
877		return;
878	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
879		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
880		    page && lun->mode_pages.index[i].subpage == subpage)
881			break;
882	}
883	if (i == CTL_NUM_MODE_PAGES)
884		return;
885	bzero(&msg.mode, sizeof(msg.mode));
886	msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
887	msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
888	msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
889	msg.hdr.nexus.targ_lun = lun->lun;
890	msg.hdr.nexus.targ_mapped_lun = lun->lun;
891	msg.mode.page_code = page;
892	msg.mode.subpage = subpage;
893	msg.mode.page_len = lun->mode_pages.index[i].page_len;
894	memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
895	    msg.mode.page_len);
896	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
897	    M_WAITOK);
898}
899
900static void
901ctl_isc_ha_link_up(struct ctl_softc *softc)
902{
903	struct ctl_port *port;
904	struct ctl_lun *lun;
905	union ctl_ha_msg msg;
906	int i;
907
908	/* Announce this node parameters to peer for validation. */
909	msg.login.msg_type = CTL_MSG_LOGIN;
910	msg.login.version = CTL_HA_VERSION;
911	msg.login.ha_mode = softc->ha_mode;
912	msg.login.ha_id = softc->ha_id;
913	msg.login.max_luns = CTL_MAX_LUNS;
914	msg.login.max_ports = CTL_MAX_PORTS;
915	msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
916	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
917	    M_WAITOK);
918
919	STAILQ_FOREACH(port, &softc->port_list, links) {
920		ctl_isc_announce_port(port);
921		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
922			if (port->wwpn_iid[i].in_use)
923				ctl_isc_announce_iid(port, i);
924		}
925	}
926	STAILQ_FOREACH(lun, &softc->lun_list, links)
927		ctl_isc_announce_lun(lun);
928}
929
930static void
931ctl_isc_ha_link_down(struct ctl_softc *softc)
932{
933	struct ctl_port *port;
934	struct ctl_lun *lun;
935	union ctl_io *io;
936	int i;
937
938	mtx_lock(&softc->ctl_lock);
939	STAILQ_FOREACH(lun, &softc->lun_list, links) {
940		mtx_lock(&lun->lun_lock);
941		if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
942			lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
943			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
944		}
945		mtx_unlock(&lun->lun_lock);
946
947		mtx_unlock(&softc->ctl_lock);
948		io = ctl_alloc_io(softc->othersc_pool);
949		mtx_lock(&softc->ctl_lock);
950		ctl_zero_io(io);
951		io->io_hdr.msg_type = CTL_MSG_FAILOVER;
952		io->io_hdr.nexus.targ_mapped_lun = lun->lun;
953		ctl_enqueue_isc(io);
954	}
955
956	STAILQ_FOREACH(port, &softc->port_list, links) {
957		if (port->targ_port >= softc->port_min &&
958		    port->targ_port < softc->port_max)
959			continue;
960		port->status &= ~CTL_PORT_STATUS_ONLINE;
961		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
962			port->wwpn_iid[i].in_use = 0;
963			free(port->wwpn_iid[i].name, M_CTL);
964			port->wwpn_iid[i].name = NULL;
965		}
966	}
967	mtx_unlock(&softc->ctl_lock);
968}
969
970static void
971ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
972{
973	struct ctl_lun *lun;
974	uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
975
976	mtx_lock(&softc->ctl_lock);
977	if (msg->hdr.nexus.targ_lun < CTL_MAX_LUNS &&
978	    (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) != NULL) {
979		mtx_lock(&lun->lun_lock);
980		mtx_unlock(&softc->ctl_lock);
981		if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES &&
982		    msg->ua.ua_set)
983			memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
984		if (msg->ua.ua_all) {
985			if (msg->ua.ua_set)
986				ctl_est_ua_all(lun, iid, msg->ua.ua_type);
987			else
988				ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
989		} else {
990			if (msg->ua.ua_set)
991				ctl_est_ua(lun, iid, msg->ua.ua_type);
992			else
993				ctl_clr_ua(lun, iid, msg->ua.ua_type);
994		}
995		mtx_unlock(&lun->lun_lock);
996	} else
997		mtx_unlock(&softc->ctl_lock);
998}
999
1000static void
1001ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1002{
1003	struct ctl_lun *lun;
1004	struct ctl_ha_msg_lun_pr_key pr_key;
1005	int i, k;
1006	ctl_lun_flags oflags;
1007	uint32_t targ_lun;
1008
1009	targ_lun = msg->hdr.nexus.targ_mapped_lun;
1010	mtx_lock(&softc->ctl_lock);
1011	if ((targ_lun >= CTL_MAX_LUNS) ||
1012	    ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
1013		mtx_unlock(&softc->ctl_lock);
1014		return;
1015	}
1016	mtx_lock(&lun->lun_lock);
1017	mtx_unlock(&softc->ctl_lock);
1018	if (lun->flags & CTL_LUN_DISABLED) {
1019		mtx_unlock(&lun->lun_lock);
1020		return;
1021	}
1022	i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1023	if (msg->lun.lun_devid_len != i || (i > 0 &&
1024	    memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1025		mtx_unlock(&lun->lun_lock);
1026		printf("%s: Received conflicting HA LUN %d\n",
1027		    __func__, msg->hdr.nexus.targ_lun);
1028		return;
1029	} else {
1030		/* Record whether peer is primary. */
1031		oflags = lun->flags;
1032		if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1033		    (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1034			lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1035		else
1036			lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1037		if (oflags != lun->flags)
1038			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1039
1040		/* If peer is primary and we are not -- use data */
1041		if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1042		    (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1043			lun->PRGeneration = msg->lun.pr_generation;
1044			lun->pr_res_idx = msg->lun.pr_res_idx;
1045			lun->res_type = msg->lun.pr_res_type;
1046			lun->pr_key_count = msg->lun.pr_key_count;
1047			for (k = 0; k < CTL_MAX_INITIATORS; k++)
1048				ctl_clr_prkey(lun, k);
1049			for (k = 0; k < msg->lun.pr_key_count; k++) {
1050				memcpy(&pr_key, &msg->lun.data[i],
1051				    sizeof(pr_key));
1052				ctl_alloc_prkey(lun, pr_key.pr_iid);
1053				ctl_set_prkey(lun, pr_key.pr_iid,
1054				    pr_key.pr_key);
1055				i += sizeof(pr_key);
1056			}
1057		}
1058
1059		mtx_unlock(&lun->lun_lock);
1060		CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1061		    __func__, msg->hdr.nexus.targ_lun,
1062		    (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1063		    "primary" : "secondary"));
1064
1065		/* If we are primary but peer doesn't know -- notify */
1066		if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1067		    (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1068			ctl_isc_announce_lun(lun);
1069	}
1070}
1071
1072static void
1073ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1074{
1075	struct ctl_port *port;
1076	struct ctl_lun *lun;
1077	int i, new;
1078
1079	port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1080	if (port == NULL) {
1081		CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1082		    msg->hdr.nexus.targ_port));
1083		new = 1;
1084		port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1085		port->frontend = &ha_frontend;
1086		port->targ_port = msg->hdr.nexus.targ_port;
1087		port->fe_datamove = ctl_ha_datamove;
1088		port->fe_done = ctl_ha_done;
1089	} else if (port->frontend == &ha_frontend) {
1090		CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1091		    msg->hdr.nexus.targ_port));
1092		new = 0;
1093	} else {
1094		printf("%s: Received conflicting HA port %d\n",
1095		    __func__, msg->hdr.nexus.targ_port);
1096		return;
1097	}
1098	port->port_type = msg->port.port_type;
1099	port->physical_port = msg->port.physical_port;
1100	port->virtual_port = msg->port.virtual_port;
1101	port->status = msg->port.status;
1102	i = 0;
1103	free(port->port_name, M_CTL);
1104	port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1105	    M_CTL);
1106	i += msg->port.name_len;
1107	if (msg->port.lun_map_len != 0) {
1108		if (port->lun_map == NULL)
1109			port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
1110			    M_CTL, M_WAITOK);
1111		memcpy(port->lun_map, &msg->port.data[i],
1112		    sizeof(uint32_t) * CTL_MAX_LUNS);
1113		i += msg->port.lun_map_len;
1114	} else {
1115		free(port->lun_map, M_CTL);
1116		port->lun_map = NULL;
1117	}
1118	if (msg->port.port_devid_len != 0) {
1119		if (port->port_devid == NULL ||
1120		    port->port_devid->len != msg->port.port_devid_len) {
1121			free(port->port_devid, M_CTL);
1122			port->port_devid = malloc(sizeof(struct ctl_devid) +
1123			    msg->port.port_devid_len, M_CTL, M_WAITOK);
1124		}
1125		memcpy(port->port_devid->data, &msg->port.data[i],
1126		    msg->port.port_devid_len);
1127		port->port_devid->len = msg->port.port_devid_len;
1128		i += msg->port.port_devid_len;
1129	} else {
1130		free(port->port_devid, M_CTL);
1131		port->port_devid = NULL;
1132	}
1133	if (msg->port.target_devid_len != 0) {
1134		if (port->target_devid == NULL ||
1135		    port->target_devid->len != msg->port.target_devid_len) {
1136			free(port->target_devid, M_CTL);
1137			port->target_devid = malloc(sizeof(struct ctl_devid) +
1138			    msg->port.target_devid_len, M_CTL, M_WAITOK);
1139		}
1140		memcpy(port->target_devid->data, &msg->port.data[i],
1141		    msg->port.target_devid_len);
1142		port->target_devid->len = msg->port.target_devid_len;
1143		i += msg->port.target_devid_len;
1144	} else {
1145		free(port->target_devid, M_CTL);
1146		port->target_devid = NULL;
1147	}
1148	if (msg->port.init_devid_len != 0) {
1149		if (port->init_devid == NULL ||
1150		    port->init_devid->len != msg->port.init_devid_len) {
1151			free(port->init_devid, M_CTL);
1152			port->init_devid = malloc(sizeof(struct ctl_devid) +
1153			    msg->port.init_devid_len, M_CTL, M_WAITOK);
1154		}
1155		memcpy(port->init_devid->data, &msg->port.data[i],
1156		    msg->port.init_devid_len);
1157		port->init_devid->len = msg->port.init_devid_len;
1158		i += msg->port.init_devid_len;
1159	} else {
1160		free(port->init_devid, M_CTL);
1161		port->init_devid = NULL;
1162	}
1163	if (new) {
1164		if (ctl_port_register(port) != 0) {
1165			printf("%s: ctl_port_register() failed with error\n",
1166			    __func__);
1167		}
1168	}
1169	mtx_lock(&softc->ctl_lock);
1170	STAILQ_FOREACH(lun, &softc->lun_list, links) {
1171		if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
1172			continue;
1173		mtx_lock(&lun->lun_lock);
1174		ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1175		mtx_unlock(&lun->lun_lock);
1176	}
1177	mtx_unlock(&softc->ctl_lock);
1178}
1179
1180static void
1181ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1182{
1183	struct ctl_port *port;
1184	int iid;
1185
1186	port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1187	if (port == NULL) {
1188		printf("%s: Received IID for unknown port %d\n",
1189		    __func__, msg->hdr.nexus.targ_port);
1190		return;
1191	}
1192	iid = msg->hdr.nexus.initid;
1193	port->wwpn_iid[iid].in_use = msg->iid.in_use;
1194	port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1195	free(port->wwpn_iid[iid].name, M_CTL);
1196	if (msg->iid.name_len) {
1197		port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1198		    msg->iid.name_len, M_CTL);
1199	} else
1200		port->wwpn_iid[iid].name = NULL;
1201}
1202
1203static void
1204ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1205{
1206
1207	if (msg->login.version != CTL_HA_VERSION) {
1208		printf("CTL HA peers have different versions %d != %d\n",
1209		    msg->login.version, CTL_HA_VERSION);
1210		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1211		return;
1212	}
1213	if (msg->login.ha_mode != softc->ha_mode) {
1214		printf("CTL HA peers have different ha_mode %d != %d\n",
1215		    msg->login.ha_mode, softc->ha_mode);
1216		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1217		return;
1218	}
1219	if (msg->login.ha_id == softc->ha_id) {
1220		printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1221		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1222		return;
1223	}
1224	if (msg->login.max_luns != CTL_MAX_LUNS ||
1225	    msg->login.max_ports != CTL_MAX_PORTS ||
1226	    msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1227		printf("CTL HA peers have different limits\n");
1228		ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1229		return;
1230	}
1231}
1232
1233static void
1234ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1235{
1236	struct ctl_lun *lun;
1237	int i;
1238	uint32_t initidx, targ_lun;
1239
1240	targ_lun = msg->hdr.nexus.targ_mapped_lun;
1241	mtx_lock(&softc->ctl_lock);
1242	if ((targ_lun >= CTL_MAX_LUNS) ||
1243	    ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
1244		mtx_unlock(&softc->ctl_lock);
1245		return;
1246	}
1247	mtx_lock(&lun->lun_lock);
1248	mtx_unlock(&softc->ctl_lock);
1249	if (lun->flags & CTL_LUN_DISABLED) {
1250		mtx_unlock(&lun->lun_lock);
1251		return;
1252	}
1253	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1254		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1255		    msg->mode.page_code &&
1256		    lun->mode_pages.index[i].subpage == msg->mode.subpage)
1257			break;
1258	}
1259	if (i == CTL_NUM_MODE_PAGES) {
1260		mtx_unlock(&lun->lun_lock);
1261		return;
1262	}
1263	memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1264	    lun->mode_pages.index[i].page_len);
1265	initidx = ctl_get_initindex(&msg->hdr.nexus);
1266	if (initidx != -1)
1267		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1268	mtx_unlock(&lun->lun_lock);
1269}
1270
1271/*
1272 * ISC (Inter Shelf Communication) event handler.  Events from the HA
1273 * subsystem come in here.
1274 */
1275static void
1276ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1277{
1278	struct ctl_softc *softc = control_softc;
1279	union ctl_io *io;
1280	struct ctl_prio *presio;
1281	ctl_ha_status isc_status;
1282
1283	CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1284	if (event == CTL_HA_EVT_MSG_RECV) {
1285		union ctl_ha_msg *msg, msgbuf;
1286
1287		if (param > sizeof(msgbuf))
1288			msg = malloc(param, M_CTL, M_WAITOK);
1289		else
1290			msg = &msgbuf;
1291		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1292		    M_WAITOK);
1293		if (isc_status != CTL_HA_STATUS_SUCCESS) {
1294			printf("%s: Error receiving message: %d\n",
1295			    __func__, isc_status);
1296			if (msg != &msgbuf)
1297				free(msg, M_CTL);
1298			return;
1299		}
1300
1301		CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1302		switch (msg->hdr.msg_type) {
1303		case CTL_MSG_SERIALIZE:
1304			io = ctl_alloc_io(softc->othersc_pool);
1305			ctl_zero_io(io);
1306			// populate ctsio from msg
1307			io->io_hdr.io_type = CTL_IO_SCSI;
1308			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1309			io->io_hdr.original_sc = msg->hdr.original_sc;
1310			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1311					    CTL_FLAG_IO_ACTIVE;
1312			/*
1313			 * If we're in serialization-only mode, we don't
1314			 * want to go through full done processing.  Thus
1315			 * the COPY flag.
1316			 *
1317			 * XXX KDM add another flag that is more specific.
1318			 */
1319			if (softc->ha_mode != CTL_HA_MODE_XFER)
1320				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1321			io->io_hdr.nexus = msg->hdr.nexus;
1322#if 0
1323			printf("port %u, iid %u, lun %u\n",
1324			       io->io_hdr.nexus.targ_port,
1325			       io->io_hdr.nexus.initid,
1326			       io->io_hdr.nexus.targ_lun);
1327#endif
1328			io->scsiio.tag_num = msg->scsi.tag_num;
1329			io->scsiio.tag_type = msg->scsi.tag_type;
1330#ifdef CTL_TIME_IO
1331			io->io_hdr.start_time = time_uptime;
1332			getbinuptime(&io->io_hdr.start_bt);
1333#endif /* CTL_TIME_IO */
1334			io->scsiio.cdb_len = msg->scsi.cdb_len;
1335			memcpy(io->scsiio.cdb, msg->scsi.cdb,
1336			       CTL_MAX_CDBLEN);
1337			if (softc->ha_mode == CTL_HA_MODE_XFER) {
1338				const struct ctl_cmd_entry *entry;
1339
1340				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1341				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1342				io->io_hdr.flags |=
1343					entry->flags & CTL_FLAG_DATA_MASK;
1344			}
1345			ctl_enqueue_isc(io);
1346			break;
1347
1348		/* Performed on the Originating SC, XFER mode only */
1349		case CTL_MSG_DATAMOVE: {
1350			struct ctl_sg_entry *sgl;
1351			int i, j;
1352
1353			io = msg->hdr.original_sc;
1354			if (io == NULL) {
1355				printf("%s: original_sc == NULL!\n", __func__);
1356				/* XXX KDM do something here */
1357				break;
1358			}
1359			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1360			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1361			/*
1362			 * Keep track of this, we need to send it back over
1363			 * when the datamove is complete.
1364			 */
1365			io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1366			if (msg->hdr.status == CTL_SUCCESS)
1367				io->io_hdr.status = msg->hdr.status;
1368
1369			if (msg->dt.sg_sequence == 0) {
1370#ifdef CTL_TIME_IO
1371				getbinuptime(&io->io_hdr.dma_start_bt);
1372#endif
1373				i = msg->dt.kern_sg_entries +
1374				    msg->dt.kern_data_len /
1375				    CTL_HA_DATAMOVE_SEGMENT + 1;
1376				sgl = malloc(sizeof(*sgl) * i, M_CTL,
1377				    M_WAITOK | M_ZERO);
1378				io->io_hdr.remote_sglist = sgl;
1379				io->io_hdr.local_sglist =
1380				    &sgl[msg->dt.kern_sg_entries];
1381
1382				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1383
1384				io->scsiio.kern_sg_entries =
1385					msg->dt.kern_sg_entries;
1386				io->scsiio.rem_sg_entries =
1387					msg->dt.kern_sg_entries;
1388				io->scsiio.kern_data_len =
1389					msg->dt.kern_data_len;
1390				io->scsiio.kern_total_len =
1391					msg->dt.kern_total_len;
1392				io->scsiio.kern_data_resid =
1393					msg->dt.kern_data_resid;
1394				io->scsiio.kern_rel_offset =
1395					msg->dt.kern_rel_offset;
1396				io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1397				io->io_hdr.flags |= msg->dt.flags &
1398				    CTL_FLAG_BUS_ADDR;
1399			} else
1400				sgl = (struct ctl_sg_entry *)
1401					io->scsiio.kern_data_ptr;
1402
1403			for (i = msg->dt.sent_sg_entries, j = 0;
1404			     i < (msg->dt.sent_sg_entries +
1405			     msg->dt.cur_sg_entries); i++, j++) {
1406				sgl[i].addr = msg->dt.sg_list[j].addr;
1407				sgl[i].len = msg->dt.sg_list[j].len;
1408
1409#if 0
1410				printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1411				    __func__, sgl[i].addr, sgl[i].len, j, i);
1412#endif
1413			}
1414
1415			/*
1416			 * If this is the last piece of the I/O, we've got
1417			 * the full S/G list.  Queue processing in the thread.
1418			 * Otherwise wait for the next piece.
1419			 */
1420			if (msg->dt.sg_last != 0)
1421				ctl_enqueue_isc(io);
1422			break;
1423		}
1424		/* Performed on the Serializing (primary) SC, XFER mode only */
1425		case CTL_MSG_DATAMOVE_DONE: {
1426			if (msg->hdr.serializing_sc == NULL) {
1427				printf("%s: serializing_sc == NULL!\n",
1428				       __func__);
1429				/* XXX KDM now what? */
1430				break;
1431			}
1432			/*
1433			 * We grab the sense information here in case
1434			 * there was a failure, so we can return status
1435			 * back to the initiator.
1436			 */
1437			io = msg->hdr.serializing_sc;
1438			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1439			io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1440			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1441			io->io_hdr.port_status = msg->scsi.fetd_status;
1442			io->scsiio.residual = msg->scsi.residual;
1443			if (msg->hdr.status != CTL_STATUS_NONE) {
1444				io->io_hdr.status = msg->hdr.status;
1445				io->scsiio.scsi_status = msg->scsi.scsi_status;
1446				io->scsiio.sense_len = msg->scsi.sense_len;
1447				io->scsiio.sense_residual =msg->scsi.sense_residual;
1448				memcpy(&io->scsiio.sense_data,
1449				    &msg->scsi.sense_data,
1450				    msg->scsi.sense_len);
1451				if (msg->hdr.status == CTL_SUCCESS)
1452					io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1453			}
1454			ctl_enqueue_isc(io);
1455			break;
1456		}
1457
1458		/* Preformed on Originating SC, SER_ONLY mode */
1459		case CTL_MSG_R2R:
1460			io = msg->hdr.original_sc;
1461			if (io == NULL) {
1462				printf("%s: original_sc == NULL!\n",
1463				    __func__);
1464				break;
1465			}
1466			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1467			io->io_hdr.msg_type = CTL_MSG_R2R;
1468			io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1469			ctl_enqueue_isc(io);
1470			break;
1471
1472		/*
1473		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1474		 * mode.
1475		 * Performed on the Originating (i.e. secondary) SC in XFER
1476		 * mode
1477		 */
1478		case CTL_MSG_FINISH_IO:
1479			if (softc->ha_mode == CTL_HA_MODE_XFER)
1480				ctl_isc_handler_finish_xfer(softc, msg);
1481			else
1482				ctl_isc_handler_finish_ser_only(softc, msg);
1483			break;
1484
1485		/* Preformed on Originating SC */
1486		case CTL_MSG_BAD_JUJU:
1487			io = msg->hdr.original_sc;
1488			if (io == NULL) {
1489				printf("%s: Bad JUJU!, original_sc is NULL!\n",
1490				       __func__);
1491				break;
1492			}
1493			ctl_copy_sense_data(msg, io);
1494			/*
1495			 * IO should have already been cleaned up on other
1496			 * SC so clear this flag so we won't send a message
1497			 * back to finish the IO there.
1498			 */
1499			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1500			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1501
1502			/* io = msg->hdr.serializing_sc; */
1503			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1504			ctl_enqueue_isc(io);
1505			break;
1506
1507		/* Handle resets sent from the other side */
1508		case CTL_MSG_MANAGE_TASKS: {
1509			struct ctl_taskio *taskio;
1510			taskio = (struct ctl_taskio *)ctl_alloc_io(
1511			    softc->othersc_pool);
1512			ctl_zero_io((union ctl_io *)taskio);
1513			taskio->io_hdr.io_type = CTL_IO_TASK;
1514			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1515			taskio->io_hdr.nexus = msg->hdr.nexus;
1516			taskio->task_action = msg->task.task_action;
1517			taskio->tag_num = msg->task.tag_num;
1518			taskio->tag_type = msg->task.tag_type;
1519#ifdef CTL_TIME_IO
1520			taskio->io_hdr.start_time = time_uptime;
1521			getbinuptime(&taskio->io_hdr.start_bt);
1522#endif /* CTL_TIME_IO */
1523			ctl_run_task((union ctl_io *)taskio);
1524			break;
1525		}
1526		/* Persistent Reserve action which needs attention */
1527		case CTL_MSG_PERS_ACTION:
1528			presio = (struct ctl_prio *)ctl_alloc_io(
1529			    softc->othersc_pool);
1530			ctl_zero_io((union ctl_io *)presio);
1531			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1532			presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1533			presio->io_hdr.nexus = msg->hdr.nexus;
1534			presio->pr_msg = msg->pr;
1535			ctl_enqueue_isc((union ctl_io *)presio);
1536			break;
1537		case CTL_MSG_UA:
1538			ctl_isc_ua(softc, msg, param);
1539			break;
1540		case CTL_MSG_PORT_SYNC:
1541			ctl_isc_port_sync(softc, msg, param);
1542			break;
1543		case CTL_MSG_LUN_SYNC:
1544			ctl_isc_lun_sync(softc, msg, param);
1545			break;
1546		case CTL_MSG_IID_SYNC:
1547			ctl_isc_iid_sync(softc, msg, param);
1548			break;
1549		case CTL_MSG_LOGIN:
1550			ctl_isc_login(softc, msg, param);
1551			break;
1552		case CTL_MSG_MODE_SYNC:
1553			ctl_isc_mode_sync(softc, msg, param);
1554			break;
1555		default:
1556			printf("Received HA message of unknown type %d\n",
1557			    msg->hdr.msg_type);
1558			ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1559			break;
1560		}
1561		if (msg != &msgbuf)
1562			free(msg, M_CTL);
1563	} else if (event == CTL_HA_EVT_LINK_CHANGE) {
1564		printf("CTL: HA link status changed from %d to %d\n",
1565		    softc->ha_link, param);
1566		if (param == softc->ha_link)
1567			return;
1568		if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1569			softc->ha_link = param;
1570			ctl_isc_ha_link_down(softc);
1571		} else {
1572			softc->ha_link = param;
1573			if (softc->ha_link == CTL_HA_LINK_ONLINE)
1574				ctl_isc_ha_link_up(softc);
1575		}
1576		return;
1577	} else {
1578		printf("ctl_isc_event_handler: Unknown event %d\n", event);
1579		return;
1580	}
1581}
1582
1583static void
1584ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1585{
1586
1587	memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1588	    src->scsi.sense_len);
1589	dest->scsiio.scsi_status = src->scsi.scsi_status;
1590	dest->scsiio.sense_len = src->scsi.sense_len;
1591	dest->io_hdr.status = src->hdr.status;
1592}
1593
1594static void
1595ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1596{
1597
1598	memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1599	    src->scsiio.sense_len);
1600	dest->scsi.scsi_status = src->scsiio.scsi_status;
1601	dest->scsi.sense_len = src->scsiio.sense_len;
1602	dest->hdr.status = src->io_hdr.status;
1603}
1604
1605void
1606ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1607{
1608	struct ctl_softc *softc = lun->ctl_softc;
1609	ctl_ua_type *pu;
1610
1611	if (initidx < softc->init_min || initidx >= softc->init_max)
1612		return;
1613	mtx_assert(&lun->lun_lock, MA_OWNED);
1614	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1615	if (pu == NULL)
1616		return;
1617	pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1618}
1619
1620void
1621ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1622{
1623	int i;
1624
1625	mtx_assert(&lun->lun_lock, MA_OWNED);
1626	if (lun->pending_ua[port] == NULL)
1627		return;
1628	for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1629		if (port * CTL_MAX_INIT_PER_PORT + i == except)
1630			continue;
1631		lun->pending_ua[port][i] |= ua;
1632	}
1633}
1634
1635void
1636ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1637{
1638	struct ctl_softc *softc = lun->ctl_softc;
1639	int i;
1640
1641	mtx_assert(&lun->lun_lock, MA_OWNED);
1642	for (i = softc->port_min; i < softc->port_max; i++)
1643		ctl_est_ua_port(lun, i, except, ua);
1644}
1645
1646void
1647ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1648{
1649	struct ctl_softc *softc = lun->ctl_softc;
1650	ctl_ua_type *pu;
1651
1652	if (initidx < softc->init_min || initidx >= softc->init_max)
1653		return;
1654	mtx_assert(&lun->lun_lock, MA_OWNED);
1655	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1656	if (pu == NULL)
1657		return;
1658	pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1659}
1660
1661void
1662ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1663{
1664	struct ctl_softc *softc = lun->ctl_softc;
1665	int i, j;
1666
1667	mtx_assert(&lun->lun_lock, MA_OWNED);
1668	for (i = softc->port_min; i < softc->port_max; i++) {
1669		if (lun->pending_ua[i] == NULL)
1670			continue;
1671		for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1672			if (i * CTL_MAX_INIT_PER_PORT + j == except)
1673				continue;
1674			lun->pending_ua[i][j] &= ~ua;
1675		}
1676	}
1677}
1678
1679void
1680ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1681    ctl_ua_type ua_type)
1682{
1683	struct ctl_lun *lun;
1684
1685	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1686	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1687		mtx_lock(&lun->lun_lock);
1688		ctl_clr_ua(lun, initidx, ua_type);
1689		mtx_unlock(&lun->lun_lock);
1690	}
1691}
1692
1693static int
1694ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1695{
1696	struct ctl_softc *softc = (struct ctl_softc *)arg1;
1697	struct ctl_lun *lun;
1698	struct ctl_lun_req ireq;
1699	int error, value;
1700
1701	value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1702	error = sysctl_handle_int(oidp, &value, 0, req);
1703	if ((error != 0) || (req->newptr == NULL))
1704		return (error);
1705
1706	mtx_lock(&softc->ctl_lock);
1707	if (value == 0)
1708		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1709	else
1710		softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1711	STAILQ_FOREACH(lun, &softc->lun_list, links) {
1712		mtx_unlock(&softc->ctl_lock);
1713		bzero(&ireq, sizeof(ireq));
1714		ireq.reqtype = CTL_LUNREQ_MODIFY;
1715		ireq.reqdata.modify.lun_id = lun->lun;
1716		lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1717		    curthread);
1718		if (ireq.status != CTL_LUN_OK) {
1719			printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1720			    __func__, ireq.status, ireq.error_str);
1721		}
1722		mtx_lock(&softc->ctl_lock);
1723	}
1724	mtx_unlock(&softc->ctl_lock);
1725	return (0);
1726}
1727
1728static int
1729ctl_init(void)
1730{
1731	struct ctl_softc *softc;
1732	void *other_pool;
1733	int i, error;
1734
1735	softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1736			       M_WAITOK | M_ZERO);
1737
1738	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
1739			      "cam/ctl");
1740	softc->dev->si_drv1 = softc;
1741
1742	sysctl_ctx_init(&softc->sysctl_ctx);
1743	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1744		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1745		CTLFLAG_RD, 0, "CAM Target Layer");
1746
1747	if (softc->sysctl_tree == NULL) {
1748		printf("%s: unable to allocate sysctl tree\n", __func__);
1749		destroy_dev(softc->dev);
1750		free(control_softc, M_DEVBUF);
1751		control_softc = NULL;
1752		return (ENOMEM);
1753	}
1754
1755	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1756	softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1757	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1758	softc->open_count = 0;
1759
1760	/*
1761	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1762	 * the drive.
1763	 */
1764	softc->flags = CTL_FLAG_REAL_SYNC;
1765
1766	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1767	    OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1768	    "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1769
1770	/*
1771	 * In Copan's HA scheme, the "master" and "slave" roles are
1772	 * figured out through the slot the controller is in.  Although it
1773	 * is an active/active system, someone has to be in charge.
1774	 */
1775	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1776	    OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1777	    "HA head ID (0 - no HA)");
1778	if (softc->ha_id == 0 || softc->ha_id > NUM_TARGET_PORT_GROUPS) {
1779		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1780		softc->is_single = 1;
1781		softc->port_cnt = CTL_MAX_PORTS;
1782		softc->port_min = 0;
1783	} else {
1784		softc->port_cnt = CTL_MAX_PORTS / NUM_TARGET_PORT_GROUPS;
1785		softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1786	}
1787	softc->port_max = softc->port_min + softc->port_cnt;
1788	softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1789	softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1790
1791	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1792	    OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1793	    "HA link state (0 - offline, 1 - unknown, 2 - online)");
1794
1795	STAILQ_INIT(&softc->lun_list);
1796	STAILQ_INIT(&softc->pending_lun_queue);
1797	STAILQ_INIT(&softc->fe_list);
1798	STAILQ_INIT(&softc->port_list);
1799	STAILQ_INIT(&softc->be_list);
1800	ctl_tpc_init(softc);
1801
1802	if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1803	                    &other_pool) != 0)
1804	{
1805		printf("ctl: can't allocate %d entry other SC pool, "
1806		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1807		return (ENOMEM);
1808	}
1809	softc->othersc_pool = other_pool;
1810
1811	if (worker_threads <= 0)
1812		worker_threads = max(1, mp_ncpus / 4);
1813	if (worker_threads > CTL_MAX_THREADS)
1814		worker_threads = CTL_MAX_THREADS;
1815
1816	for (i = 0; i < worker_threads; i++) {
1817		struct ctl_thread *thr = &softc->threads[i];
1818
1819		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1820		thr->ctl_softc = softc;
1821		STAILQ_INIT(&thr->incoming_queue);
1822		STAILQ_INIT(&thr->rtr_queue);
1823		STAILQ_INIT(&thr->done_queue);
1824		STAILQ_INIT(&thr->isc_queue);
1825
1826		error = kproc_kthread_add(ctl_work_thread, thr,
1827		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1828		if (error != 0) {
1829			printf("error creating CTL work thread!\n");
1830			ctl_pool_free(other_pool);
1831			return (error);
1832		}
1833	}
1834	error = kproc_kthread_add(ctl_lun_thread, softc,
1835	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1836	if (error != 0) {
1837		printf("error creating CTL lun thread!\n");
1838		ctl_pool_free(other_pool);
1839		return (error);
1840	}
1841	error = kproc_kthread_add(ctl_thresh_thread, softc,
1842	    &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1843	if (error != 0) {
1844		printf("error creating CTL threshold thread!\n");
1845		ctl_pool_free(other_pool);
1846		return (error);
1847	}
1848
1849	SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1850	    OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1851	    softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1852
1853	if (softc->is_single == 0) {
1854		ctl_frontend_register(&ha_frontend);
1855		if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
1856			printf("ctl_init: ctl_ha_msg_init failed.\n");
1857			softc->is_single = 1;
1858		} else
1859		if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
1860		    != CTL_HA_STATUS_SUCCESS) {
1861			printf("ctl_init: ctl_ha_msg_register failed.\n");
1862			softc->is_single = 1;
1863		}
1864	}
1865	return (0);
1866}
1867
1868void
1869ctl_shutdown(void)
1870{
1871	struct ctl_softc *softc = control_softc;
1872	struct ctl_lun *lun, *next_lun;
1873
1874	if (softc->is_single == 0) {
1875		ctl_ha_msg_shutdown(softc);
1876		if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL)
1877		    != CTL_HA_STATUS_SUCCESS)
1878			printf("%s: ctl_ha_msg_deregister failed.\n", __func__);
1879		if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
1880			printf("%s: ctl_ha_msg_destroy failed.\n", __func__);
1881		ctl_frontend_deregister(&ha_frontend);
1882	}
1883
1884	mtx_lock(&softc->ctl_lock);
1885
1886	/*
1887	 * Free up each LUN.
1888	 */
1889	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1890		next_lun = STAILQ_NEXT(lun, links);
1891		ctl_free_lun(lun);
1892	}
1893
1894	mtx_unlock(&softc->ctl_lock);
1895
1896#if 0
1897	ctl_shutdown_thread(softc->work_thread);
1898	mtx_destroy(&softc->queue_lock);
1899#endif
1900
1901	ctl_tpc_shutdown(softc);
1902	uma_zdestroy(softc->io_zone);
1903	mtx_destroy(&softc->ctl_lock);
1904
1905	destroy_dev(softc->dev);
1906
1907	sysctl_ctx_free(&softc->sysctl_ctx);
1908
1909	free(control_softc, M_DEVBUF);
1910	control_softc = NULL;
1911}
1912
1913static int
1914ctl_module_event_handler(module_t mod, int what, void *arg)
1915{
1916
1917	switch (what) {
1918	case MOD_LOAD:
1919		return (ctl_init());
1920	case MOD_UNLOAD:
1921		return (EBUSY);
1922	default:
1923		return (EOPNOTSUPP);
1924	}
1925}
1926
1927/*
1928 * XXX KDM should we do some access checks here?  Bump a reference count to
1929 * prevent a CTL module from being unloaded while someone has it open?
1930 */
1931static int
1932ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1933{
1934	return (0);
1935}
1936
1937static int
1938ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1939{
1940	return (0);
1941}
1942
1943/*
1944 * Remove an initiator by port number and initiator ID.
1945 * Returns 0 for success, -1 for failure.
1946 */
1947int
1948ctl_remove_initiator(struct ctl_port *port, int iid)
1949{
1950	struct ctl_softc *softc = port->ctl_softc;
1951
1952	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1953
1954	if (iid > CTL_MAX_INIT_PER_PORT) {
1955		printf("%s: initiator ID %u > maximun %u!\n",
1956		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1957		return (-1);
1958	}
1959
1960	mtx_lock(&softc->ctl_lock);
1961	port->wwpn_iid[iid].in_use--;
1962	port->wwpn_iid[iid].last_use = time_uptime;
1963	mtx_unlock(&softc->ctl_lock);
1964	ctl_isc_announce_iid(port, iid);
1965
1966	return (0);
1967}
1968
1969/*
1970 * Add an initiator to the initiator map.
1971 * Returns iid for success, < 0 for failure.
1972 */
1973int
1974ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1975{
1976	struct ctl_softc *softc = port->ctl_softc;
1977	time_t best_time;
1978	int i, best;
1979
1980	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1981
1982	if (iid >= CTL_MAX_INIT_PER_PORT) {
1983		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1984		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1985		free(name, M_CTL);
1986		return (-1);
1987	}
1988
1989	mtx_lock(&softc->ctl_lock);
1990
1991	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1992		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1993			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1994				iid = i;
1995				break;
1996			}
1997			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1998			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1999				iid = i;
2000				break;
2001			}
2002		}
2003	}
2004
2005	if (iid < 0) {
2006		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2007			if (port->wwpn_iid[i].in_use == 0 &&
2008			    port->wwpn_iid[i].wwpn == 0 &&
2009			    port->wwpn_iid[i].name == NULL) {
2010				iid = i;
2011				break;
2012			}
2013		}
2014	}
2015
2016	if (iid < 0) {
2017		best = -1;
2018		best_time = INT32_MAX;
2019		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2020			if (port->wwpn_iid[i].in_use == 0) {
2021				if (port->wwpn_iid[i].last_use < best_time) {
2022					best = i;
2023					best_time = port->wwpn_iid[i].last_use;
2024				}
2025			}
2026		}
2027		iid = best;
2028	}
2029
2030	if (iid < 0) {
2031		mtx_unlock(&softc->ctl_lock);
2032		free(name, M_CTL);
2033		return (-2);
2034	}
2035
2036	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2037		/*
2038		 * This is not an error yet.
2039		 */
2040		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2041#if 0
2042			printf("%s: port %d iid %u WWPN %#jx arrived"
2043			    " again\n", __func__, port->targ_port,
2044			    iid, (uintmax_t)wwpn);
2045#endif
2046			goto take;
2047		}
2048		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2049		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
2050#if 0
2051			printf("%s: port %d iid %u name '%s' arrived"
2052			    " again\n", __func__, port->targ_port,
2053			    iid, name);
2054#endif
2055			goto take;
2056		}
2057
2058		/*
2059		 * This is an error, but what do we do about it?  The
2060		 * driver is telling us we have a new WWPN for this
2061		 * initiator ID, so we pretty much need to use it.
2062		 */
2063		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2064		    " but WWPN %#jx '%s' is still at that address\n",
2065		    __func__, port->targ_port, iid, wwpn, name,
2066		    (uintmax_t)port->wwpn_iid[iid].wwpn,
2067		    port->wwpn_iid[iid].name);
2068
2069		/*
2070		 * XXX KDM clear have_ca and ua_pending on each LUN for
2071		 * this initiator.
2072		 */
2073	}
2074take:
2075	free(port->wwpn_iid[iid].name, M_CTL);
2076	port->wwpn_iid[iid].name = name;
2077	port->wwpn_iid[iid].wwpn = wwpn;
2078	port->wwpn_iid[iid].in_use++;
2079	mtx_unlock(&softc->ctl_lock);
2080	ctl_isc_announce_iid(port, iid);
2081
2082	return (iid);
2083}
2084
2085static int
2086ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2087{
2088	int len;
2089
2090	switch (port->port_type) {
2091	case CTL_PORT_FC:
2092	{
2093		struct scsi_transportid_fcp *id =
2094		    (struct scsi_transportid_fcp *)buf;
2095		if (port->wwpn_iid[iid].wwpn == 0)
2096			return (0);
2097		memset(id, 0, sizeof(*id));
2098		id->format_protocol = SCSI_PROTO_FC;
2099		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2100		return (sizeof(*id));
2101	}
2102	case CTL_PORT_ISCSI:
2103	{
2104		struct scsi_transportid_iscsi_port *id =
2105		    (struct scsi_transportid_iscsi_port *)buf;
2106		if (port->wwpn_iid[iid].name == NULL)
2107			return (0);
2108		memset(id, 0, 256);
2109		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2110		    SCSI_PROTO_ISCSI;
2111		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2112		len = roundup2(min(len, 252), 4);
2113		scsi_ulto2b(len, id->additional_length);
2114		return (sizeof(*id) + len);
2115	}
2116	case CTL_PORT_SAS:
2117	{
2118		struct scsi_transportid_sas *id =
2119		    (struct scsi_transportid_sas *)buf;
2120		if (port->wwpn_iid[iid].wwpn == 0)
2121			return (0);
2122		memset(id, 0, sizeof(*id));
2123		id->format_protocol = SCSI_PROTO_SAS;
2124		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2125		return (sizeof(*id));
2126	}
2127	default:
2128	{
2129		struct scsi_transportid_spi *id =
2130		    (struct scsi_transportid_spi *)buf;
2131		memset(id, 0, sizeof(*id));
2132		id->format_protocol = SCSI_PROTO_SPI;
2133		scsi_ulto2b(iid, id->scsi_addr);
2134		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2135		return (sizeof(*id));
2136	}
2137	}
2138}
2139
2140/*
2141 * Serialize a command that went down the "wrong" side, and so was sent to
2142 * this controller for execution.  The logic is a little different than the
2143 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2144 * sent back to the other side, but in the success case, we execute the
2145 * command on this side (XFER mode) or tell the other side to execute it
2146 * (SER_ONLY mode).
2147 */
2148static int
2149ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2150{
2151	struct ctl_softc *softc = control_softc;
2152	union ctl_ha_msg msg_info;
2153	struct ctl_port *port;
2154	struct ctl_lun *lun;
2155	const struct ctl_cmd_entry *entry;
2156	int retval = 0;
2157	uint32_t targ_lun;
2158
2159	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2160	mtx_lock(&softc->ctl_lock);
2161
2162	/* Make sure that we know about this port. */
2163	port = ctl_io_port(&ctsio->io_hdr);
2164	if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2165		ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2166					 /*retry_count*/ 1);
2167		goto badjuju;
2168	}
2169
2170	/* Make sure that we know about this LUN. */
2171	if ((targ_lun < CTL_MAX_LUNS) &&
2172	    ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
2173		mtx_lock(&lun->lun_lock);
2174		mtx_unlock(&softc->ctl_lock);
2175		/*
2176		 * If the LUN is invalid, pretend that it doesn't exist.
2177		 * It will go away as soon as all pending I/O has been
2178		 * completed.
2179		 */
2180		if (lun->flags & CTL_LUN_DISABLED) {
2181			mtx_unlock(&lun->lun_lock);
2182			lun = NULL;
2183		}
2184	} else {
2185		mtx_unlock(&softc->ctl_lock);
2186		lun = NULL;
2187	}
2188	if (lun == NULL) {
2189		/*
2190		 * The other node would not send this request to us unless
2191		 * received announce that we are primary node for this LUN.
2192		 * If this LUN does not exist now, it is probably result of
2193		 * a race, so respond to initiator in the most opaque way.
2194		 */
2195		ctl_set_busy(ctsio);
2196		goto badjuju;
2197	}
2198
2199	entry = ctl_get_cmd_entry(ctsio, NULL);
2200	if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2201		mtx_unlock(&lun->lun_lock);
2202		goto badjuju;
2203	}
2204
2205	ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
2206	ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = lun->be_lun;
2207
2208	/*
2209	 * Every I/O goes into the OOA queue for a
2210	 * particular LUN, and stays there until completion.
2211	 */
2212#ifdef CTL_TIME_IO
2213	if (TAILQ_EMPTY(&lun->ooa_queue))
2214		lun->idle_time += getsbinuptime() - lun->last_busy;
2215#endif
2216	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2217
2218	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2219		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2220		 ooa_links))) {
2221	case CTL_ACTION_BLOCK:
2222		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2223		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2224				  blocked_links);
2225		mtx_unlock(&lun->lun_lock);
2226		break;
2227	case CTL_ACTION_PASS:
2228	case CTL_ACTION_SKIP:
2229		if (softc->ha_mode == CTL_HA_MODE_XFER) {
2230			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2231			ctl_enqueue_rtr((union ctl_io *)ctsio);
2232			mtx_unlock(&lun->lun_lock);
2233		} else {
2234			ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2235			mtx_unlock(&lun->lun_lock);
2236
2237			/* send msg back to other side */
2238			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2239			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2240			msg_info.hdr.msg_type = CTL_MSG_R2R;
2241			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2242			    sizeof(msg_info.hdr), M_WAITOK);
2243		}
2244		break;
2245	case CTL_ACTION_OVERLAP:
2246		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2247		mtx_unlock(&lun->lun_lock);
2248		ctl_set_overlapped_cmd(ctsio);
2249		goto badjuju;
2250	case CTL_ACTION_OVERLAP_TAG:
2251		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2252		mtx_unlock(&lun->lun_lock);
2253		ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2254		goto badjuju;
2255	case CTL_ACTION_ERROR:
2256	default:
2257		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2258		mtx_unlock(&lun->lun_lock);
2259
2260		ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2261					 /*retry_count*/ 0);
2262badjuju:
2263		ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2264		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2265		msg_info.hdr.serializing_sc = NULL;
2266		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2267		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2268		    sizeof(msg_info.scsi), M_WAITOK);
2269		retval = 1;
2270		break;
2271	}
2272	return (retval);
2273}
2274
2275/*
2276 * Returns 0 for success, errno for failure.
2277 */
2278static void
2279ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2280		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2281{
2282	union ctl_io *io;
2283
2284	mtx_lock(&lun->lun_lock);
2285	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2286	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2287	     ooa_links)) {
2288		struct ctl_ooa_entry *entry;
2289
2290		/*
2291		 * If we've got more than we can fit, just count the
2292		 * remaining entries.
2293		 */
2294		if (*cur_fill_num >= ooa_hdr->alloc_num)
2295			continue;
2296
2297		entry = &kern_entries[*cur_fill_num];
2298
2299		entry->tag_num = io->scsiio.tag_num;
2300		entry->lun_num = lun->lun;
2301#ifdef CTL_TIME_IO
2302		entry->start_bt = io->io_hdr.start_bt;
2303#endif
2304		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2305		entry->cdb_len = io->scsiio.cdb_len;
2306		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2307			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2308
2309		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2310			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2311
2312		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2313			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2314
2315		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2316			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2317
2318		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2319			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2320	}
2321	mtx_unlock(&lun->lun_lock);
2322}
2323
2324static void *
2325ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2326		 size_t error_str_len)
2327{
2328	void *kptr;
2329
2330	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2331
2332	if (copyin(user_addr, kptr, len) != 0) {
2333		snprintf(error_str, error_str_len, "Error copying %d bytes "
2334			 "from user address %p to kernel address %p", len,
2335			 user_addr, kptr);
2336		free(kptr, M_CTL);
2337		return (NULL);
2338	}
2339
2340	return (kptr);
2341}
2342
2343static void
2344ctl_free_args(int num_args, struct ctl_be_arg *args)
2345{
2346	int i;
2347
2348	if (args == NULL)
2349		return;
2350
2351	for (i = 0; i < num_args; i++) {
2352		free(args[i].kname, M_CTL);
2353		free(args[i].kvalue, M_CTL);
2354	}
2355
2356	free(args, M_CTL);
2357}
2358
2359static struct ctl_be_arg *
2360ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2361		char *error_str, size_t error_str_len)
2362{
2363	struct ctl_be_arg *args;
2364	int i;
2365
2366	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2367				error_str, error_str_len);
2368
2369	if (args == NULL)
2370		goto bailout;
2371
2372	for (i = 0; i < num_args; i++) {
2373		args[i].kname = NULL;
2374		args[i].kvalue = NULL;
2375	}
2376
2377	for (i = 0; i < num_args; i++) {
2378		uint8_t *tmpptr;
2379
2380		args[i].kname = ctl_copyin_alloc(args[i].name,
2381			args[i].namelen, error_str, error_str_len);
2382		if (args[i].kname == NULL)
2383			goto bailout;
2384
2385		if (args[i].kname[args[i].namelen - 1] != '\0') {
2386			snprintf(error_str, error_str_len, "Argument %d "
2387				 "name is not NUL-terminated", i);
2388			goto bailout;
2389		}
2390
2391		if (args[i].flags & CTL_BEARG_RD) {
2392			tmpptr = ctl_copyin_alloc(args[i].value,
2393				args[i].vallen, error_str, error_str_len);
2394			if (tmpptr == NULL)
2395				goto bailout;
2396			if ((args[i].flags & CTL_BEARG_ASCII)
2397			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2398				snprintf(error_str, error_str_len, "Argument "
2399				    "%d value is not NUL-terminated", i);
2400				goto bailout;
2401			}
2402			args[i].kvalue = tmpptr;
2403		} else {
2404			args[i].kvalue = malloc(args[i].vallen,
2405			    M_CTL, M_WAITOK | M_ZERO);
2406		}
2407	}
2408
2409	return (args);
2410bailout:
2411
2412	ctl_free_args(num_args, args);
2413
2414	return (NULL);
2415}
2416
2417static void
2418ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2419{
2420	int i;
2421
2422	for (i = 0; i < num_args; i++) {
2423		if (args[i].flags & CTL_BEARG_WR)
2424			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2425	}
2426}
2427
2428/*
2429 * Escape characters that are illegal or not recommended in XML.
2430 */
2431int
2432ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2433{
2434	char *end = str + size;
2435	int retval;
2436
2437	retval = 0;
2438
2439	for (; *str && str < end; str++) {
2440		switch (*str) {
2441		case '&':
2442			retval = sbuf_printf(sb, "&amp;");
2443			break;
2444		case '>':
2445			retval = sbuf_printf(sb, "&gt;");
2446			break;
2447		case '<':
2448			retval = sbuf_printf(sb, "&lt;");
2449			break;
2450		default:
2451			retval = sbuf_putc(sb, *str);
2452			break;
2453		}
2454
2455		if (retval != 0)
2456			break;
2457
2458	}
2459
2460	return (retval);
2461}
2462
2463static void
2464ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2465{
2466	struct scsi_vpd_id_descriptor *desc;
2467	int i;
2468
2469	if (id == NULL || id->len < 4)
2470		return;
2471	desc = (struct scsi_vpd_id_descriptor *)id->data;
2472	switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2473	case SVPD_ID_TYPE_T10:
2474		sbuf_printf(sb, "t10.");
2475		break;
2476	case SVPD_ID_TYPE_EUI64:
2477		sbuf_printf(sb, "eui.");
2478		break;
2479	case SVPD_ID_TYPE_NAA:
2480		sbuf_printf(sb, "naa.");
2481		break;
2482	case SVPD_ID_TYPE_SCSI_NAME:
2483		break;
2484	}
2485	switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2486	case SVPD_ID_CODESET_BINARY:
2487		for (i = 0; i < desc->length; i++)
2488			sbuf_printf(sb, "%02x", desc->identifier[i]);
2489		break;
2490	case SVPD_ID_CODESET_ASCII:
2491		sbuf_printf(sb, "%.*s", (int)desc->length,
2492		    (char *)desc->identifier);
2493		break;
2494	case SVPD_ID_CODESET_UTF8:
2495		sbuf_printf(sb, "%s", (char *)desc->identifier);
2496		break;
2497	}
2498}
2499
2500static int
2501ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2502	  struct thread *td)
2503{
2504	struct ctl_softc *softc = dev->si_drv1;
2505	struct ctl_lun *lun;
2506	int retval;
2507
2508	retval = 0;
2509
2510	switch (cmd) {
2511	case CTL_IO:
2512		retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2513		break;
2514	case CTL_ENABLE_PORT:
2515	case CTL_DISABLE_PORT:
2516	case CTL_SET_PORT_WWNS: {
2517		struct ctl_port *port;
2518		struct ctl_port_entry *entry;
2519
2520		entry = (struct ctl_port_entry *)addr;
2521
2522		mtx_lock(&softc->ctl_lock);
2523		STAILQ_FOREACH(port, &softc->port_list, links) {
2524			int action, done;
2525
2526			if (port->targ_port < softc->port_min ||
2527			    port->targ_port >= softc->port_max)
2528				continue;
2529
2530			action = 0;
2531			done = 0;
2532			if ((entry->port_type == CTL_PORT_NONE)
2533			 && (entry->targ_port == port->targ_port)) {
2534				/*
2535				 * If the user only wants to enable or
2536				 * disable or set WWNs on a specific port,
2537				 * do the operation and we're done.
2538				 */
2539				action = 1;
2540				done = 1;
2541			} else if (entry->port_type & port->port_type) {
2542				/*
2543				 * Compare the user's type mask with the
2544				 * particular frontend type to see if we
2545				 * have a match.
2546				 */
2547				action = 1;
2548				done = 0;
2549
2550				/*
2551				 * Make sure the user isn't trying to set
2552				 * WWNs on multiple ports at the same time.
2553				 */
2554				if (cmd == CTL_SET_PORT_WWNS) {
2555					printf("%s: Can't set WWNs on "
2556					       "multiple ports\n", __func__);
2557					retval = EINVAL;
2558					break;
2559				}
2560			}
2561			if (action == 0)
2562				continue;
2563
2564			/*
2565			 * XXX KDM we have to drop the lock here, because
2566			 * the online/offline operations can potentially
2567			 * block.  We need to reference count the frontends
2568			 * so they can't go away,
2569			 */
2570			if (cmd == CTL_ENABLE_PORT) {
2571				mtx_unlock(&softc->ctl_lock);
2572				ctl_port_online(port);
2573				mtx_lock(&softc->ctl_lock);
2574			} else if (cmd == CTL_DISABLE_PORT) {
2575				mtx_unlock(&softc->ctl_lock);
2576				ctl_port_offline(port);
2577				mtx_lock(&softc->ctl_lock);
2578			} else if (cmd == CTL_SET_PORT_WWNS) {
2579				ctl_port_set_wwns(port,
2580				    (entry->flags & CTL_PORT_WWNN_VALID) ?
2581				    1 : 0, entry->wwnn,
2582				    (entry->flags & CTL_PORT_WWPN_VALID) ?
2583				    1 : 0, entry->wwpn);
2584			}
2585			if (done != 0)
2586				break;
2587		}
2588		mtx_unlock(&softc->ctl_lock);
2589		break;
2590	}
2591	case CTL_GET_PORT_LIST: {
2592		struct ctl_port *port;
2593		struct ctl_port_list *list;
2594		int i;
2595
2596		list = (struct ctl_port_list *)addr;
2597
2598		if (list->alloc_len != (list->alloc_num *
2599		    sizeof(struct ctl_port_entry))) {
2600			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2601			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2602			       "%zu\n", __func__, list->alloc_len,
2603			       list->alloc_num, sizeof(struct ctl_port_entry));
2604			retval = EINVAL;
2605			break;
2606		}
2607		list->fill_len = 0;
2608		list->fill_num = 0;
2609		list->dropped_num = 0;
2610		i = 0;
2611		mtx_lock(&softc->ctl_lock);
2612		STAILQ_FOREACH(port, &softc->port_list, links) {
2613			struct ctl_port_entry entry, *list_entry;
2614
2615			if (list->fill_num >= list->alloc_num) {
2616				list->dropped_num++;
2617				continue;
2618			}
2619
2620			entry.port_type = port->port_type;
2621			strlcpy(entry.port_name, port->port_name,
2622				sizeof(entry.port_name));
2623			entry.targ_port = port->targ_port;
2624			entry.physical_port = port->physical_port;
2625			entry.virtual_port = port->virtual_port;
2626			entry.wwnn = port->wwnn;
2627			entry.wwpn = port->wwpn;
2628			if (port->status & CTL_PORT_STATUS_ONLINE)
2629				entry.online = 1;
2630			else
2631				entry.online = 0;
2632
2633			list_entry = &list->entries[i];
2634
2635			retval = copyout(&entry, list_entry, sizeof(entry));
2636			if (retval != 0) {
2637				printf("%s: CTL_GET_PORT_LIST: copyout "
2638				       "returned %d\n", __func__, retval);
2639				break;
2640			}
2641			i++;
2642			list->fill_num++;
2643			list->fill_len += sizeof(entry);
2644		}
2645		mtx_unlock(&softc->ctl_lock);
2646
2647		/*
2648		 * If this is non-zero, we had a copyout fault, so there's
2649		 * probably no point in attempting to set the status inside
2650		 * the structure.
2651		 */
2652		if (retval != 0)
2653			break;
2654
2655		if (list->dropped_num > 0)
2656			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2657		else
2658			list->status = CTL_PORT_LIST_OK;
2659		break;
2660	}
2661	case CTL_DUMP_OOA: {
2662		union ctl_io *io;
2663		char printbuf[128];
2664		struct sbuf sb;
2665
2666		mtx_lock(&softc->ctl_lock);
2667		printf("Dumping OOA queues:\n");
2668		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2669			mtx_lock(&lun->lun_lock);
2670			for (io = (union ctl_io *)TAILQ_FIRST(
2671			     &lun->ooa_queue); io != NULL;
2672			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2673			     ooa_links)) {
2674				sbuf_new(&sb, printbuf, sizeof(printbuf),
2675					 SBUF_FIXEDLEN);
2676				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2677					    (intmax_t)lun->lun,
2678					    io->scsiio.tag_num,
2679					    (io->io_hdr.flags &
2680					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2681					    (io->io_hdr.flags &
2682					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2683					    (io->io_hdr.flags &
2684					    CTL_FLAG_ABORT) ? " ABORT" : "",
2685			                    (io->io_hdr.flags &
2686		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2687				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2688				sbuf_finish(&sb);
2689				printf("%s\n", sbuf_data(&sb));
2690			}
2691			mtx_unlock(&lun->lun_lock);
2692		}
2693		printf("OOA queues dump done\n");
2694		mtx_unlock(&softc->ctl_lock);
2695		break;
2696	}
2697	case CTL_GET_OOA: {
2698		struct ctl_ooa *ooa_hdr;
2699		struct ctl_ooa_entry *entries;
2700		uint32_t cur_fill_num;
2701
2702		ooa_hdr = (struct ctl_ooa *)addr;
2703
2704		if ((ooa_hdr->alloc_len == 0)
2705		 || (ooa_hdr->alloc_num == 0)) {
2706			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2707			       "must be non-zero\n", __func__,
2708			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2709			retval = EINVAL;
2710			break;
2711		}
2712
2713		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2714		    sizeof(struct ctl_ooa_entry))) {
2715			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2716			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2717			       __func__, ooa_hdr->alloc_len,
2718			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2719			retval = EINVAL;
2720			break;
2721		}
2722
2723		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2724		if (entries == NULL) {
2725			printf("%s: could not allocate %d bytes for OOA "
2726			       "dump\n", __func__, ooa_hdr->alloc_len);
2727			retval = ENOMEM;
2728			break;
2729		}
2730
2731		mtx_lock(&softc->ctl_lock);
2732		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2733		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2734		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2735			mtx_unlock(&softc->ctl_lock);
2736			free(entries, M_CTL);
2737			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2738			       __func__, (uintmax_t)ooa_hdr->lun_num);
2739			retval = EINVAL;
2740			break;
2741		}
2742
2743		cur_fill_num = 0;
2744
2745		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2746			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2747				ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2748				    ooa_hdr, entries);
2749			}
2750		} else {
2751			lun = softc->ctl_luns[ooa_hdr->lun_num];
2752			ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2753			    entries);
2754		}
2755		mtx_unlock(&softc->ctl_lock);
2756
2757		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2758		ooa_hdr->fill_len = ooa_hdr->fill_num *
2759			sizeof(struct ctl_ooa_entry);
2760		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2761		if (retval != 0) {
2762			printf("%s: error copying out %d bytes for OOA dump\n",
2763			       __func__, ooa_hdr->fill_len);
2764		}
2765
2766		getbinuptime(&ooa_hdr->cur_bt);
2767
2768		if (cur_fill_num > ooa_hdr->alloc_num) {
2769			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2770			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2771		} else {
2772			ooa_hdr->dropped_num = 0;
2773			ooa_hdr->status = CTL_OOA_OK;
2774		}
2775
2776		free(entries, M_CTL);
2777		break;
2778	}
2779	case CTL_CHECK_OOA: {
2780		union ctl_io *io;
2781		struct ctl_ooa_info *ooa_info;
2782
2783
2784		ooa_info = (struct ctl_ooa_info *)addr;
2785
2786		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2787			ooa_info->status = CTL_OOA_INVALID_LUN;
2788			break;
2789		}
2790		mtx_lock(&softc->ctl_lock);
2791		lun = softc->ctl_luns[ooa_info->lun_id];
2792		if (lun == NULL) {
2793			mtx_unlock(&softc->ctl_lock);
2794			ooa_info->status = CTL_OOA_INVALID_LUN;
2795			break;
2796		}
2797		mtx_lock(&lun->lun_lock);
2798		mtx_unlock(&softc->ctl_lock);
2799		ooa_info->num_entries = 0;
2800		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2801		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2802		     &io->io_hdr, ooa_links)) {
2803			ooa_info->num_entries++;
2804		}
2805		mtx_unlock(&lun->lun_lock);
2806
2807		ooa_info->status = CTL_OOA_SUCCESS;
2808
2809		break;
2810	}
2811	case CTL_DELAY_IO: {
2812		struct ctl_io_delay_info *delay_info;
2813
2814		delay_info = (struct ctl_io_delay_info *)addr;
2815
2816#ifdef CTL_IO_DELAY
2817		mtx_lock(&softc->ctl_lock);
2818
2819		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2820		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2821			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2822		} else {
2823			lun = softc->ctl_luns[delay_info->lun_id];
2824			mtx_lock(&lun->lun_lock);
2825
2826			delay_info->status = CTL_DELAY_STATUS_OK;
2827
2828			switch (delay_info->delay_type) {
2829			case CTL_DELAY_TYPE_CONT:
2830				break;
2831			case CTL_DELAY_TYPE_ONESHOT:
2832				break;
2833			default:
2834				delay_info->status =
2835					CTL_DELAY_STATUS_INVALID_TYPE;
2836				break;
2837			}
2838
2839			switch (delay_info->delay_loc) {
2840			case CTL_DELAY_LOC_DATAMOVE:
2841				lun->delay_info.datamove_type =
2842					delay_info->delay_type;
2843				lun->delay_info.datamove_delay =
2844					delay_info->delay_secs;
2845				break;
2846			case CTL_DELAY_LOC_DONE:
2847				lun->delay_info.done_type =
2848					delay_info->delay_type;
2849				lun->delay_info.done_delay =
2850					delay_info->delay_secs;
2851				break;
2852			default:
2853				delay_info->status =
2854					CTL_DELAY_STATUS_INVALID_LOC;
2855				break;
2856			}
2857			mtx_unlock(&lun->lun_lock);
2858		}
2859
2860		mtx_unlock(&softc->ctl_lock);
2861#else
2862		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2863#endif /* CTL_IO_DELAY */
2864		break;
2865	}
2866	case CTL_REALSYNC_SET: {
2867		int *syncstate;
2868
2869		syncstate = (int *)addr;
2870
2871		mtx_lock(&softc->ctl_lock);
2872		switch (*syncstate) {
2873		case 0:
2874			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2875			break;
2876		case 1:
2877			softc->flags |= CTL_FLAG_REAL_SYNC;
2878			break;
2879		default:
2880			retval = EINVAL;
2881			break;
2882		}
2883		mtx_unlock(&softc->ctl_lock);
2884		break;
2885	}
2886	case CTL_REALSYNC_GET: {
2887		int *syncstate;
2888
2889		syncstate = (int*)addr;
2890
2891		mtx_lock(&softc->ctl_lock);
2892		if (softc->flags & CTL_FLAG_REAL_SYNC)
2893			*syncstate = 1;
2894		else
2895			*syncstate = 0;
2896		mtx_unlock(&softc->ctl_lock);
2897
2898		break;
2899	}
2900	case CTL_SETSYNC:
2901	case CTL_GETSYNC: {
2902		struct ctl_sync_info *sync_info;
2903
2904		sync_info = (struct ctl_sync_info *)addr;
2905
2906		mtx_lock(&softc->ctl_lock);
2907		lun = softc->ctl_luns[sync_info->lun_id];
2908		if (lun == NULL) {
2909			mtx_unlock(&softc->ctl_lock);
2910			sync_info->status = CTL_GS_SYNC_NO_LUN;
2911			break;
2912		}
2913		/*
2914		 * Get or set the sync interval.  We're not bounds checking
2915		 * in the set case, hopefully the user won't do something
2916		 * silly.
2917		 */
2918		mtx_lock(&lun->lun_lock);
2919		mtx_unlock(&softc->ctl_lock);
2920		if (cmd == CTL_GETSYNC)
2921			sync_info->sync_interval = lun->sync_interval;
2922		else
2923			lun->sync_interval = sync_info->sync_interval;
2924		mtx_unlock(&lun->lun_lock);
2925
2926		sync_info->status = CTL_GS_SYNC_OK;
2927
2928		break;
2929	}
2930	case CTL_GETSTATS: {
2931		struct ctl_stats *stats;
2932		int i;
2933
2934		stats = (struct ctl_stats *)addr;
2935
2936		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2937		     stats->alloc_len) {
2938			stats->status = CTL_SS_NEED_MORE_SPACE;
2939			stats->num_luns = softc->num_luns;
2940			break;
2941		}
2942		/*
2943		 * XXX KDM no locking here.  If the LUN list changes,
2944		 * things can blow up.
2945		 */
2946		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2947		     i++, lun = STAILQ_NEXT(lun, links)) {
2948			retval = copyout(&lun->stats, &stats->lun_stats[i],
2949					 sizeof(lun->stats));
2950			if (retval != 0)
2951				break;
2952		}
2953		stats->num_luns = softc->num_luns;
2954		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2955				 softc->num_luns;
2956		stats->status = CTL_SS_OK;
2957#ifdef CTL_TIME_IO
2958		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2959#else
2960		stats->flags = CTL_STATS_FLAG_NONE;
2961#endif
2962		getnanouptime(&stats->timestamp);
2963		break;
2964	}
2965	case CTL_ERROR_INJECT: {
2966		struct ctl_error_desc *err_desc, *new_err_desc;
2967
2968		err_desc = (struct ctl_error_desc *)addr;
2969
2970		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2971				      M_WAITOK | M_ZERO);
2972		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2973
2974		mtx_lock(&softc->ctl_lock);
2975		lun = softc->ctl_luns[err_desc->lun_id];
2976		if (lun == NULL) {
2977			mtx_unlock(&softc->ctl_lock);
2978			free(new_err_desc, M_CTL);
2979			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2980			       __func__, (uintmax_t)err_desc->lun_id);
2981			retval = EINVAL;
2982			break;
2983		}
2984		mtx_lock(&lun->lun_lock);
2985		mtx_unlock(&softc->ctl_lock);
2986
2987		/*
2988		 * We could do some checking here to verify the validity
2989		 * of the request, but given the complexity of error
2990		 * injection requests, the checking logic would be fairly
2991		 * complex.
2992		 *
2993		 * For now, if the request is invalid, it just won't get
2994		 * executed and might get deleted.
2995		 */
2996		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2997
2998		/*
2999		 * XXX KDM check to make sure the serial number is unique,
3000		 * in case we somehow manage to wrap.  That shouldn't
3001		 * happen for a very long time, but it's the right thing to
3002		 * do.
3003		 */
3004		new_err_desc->serial = lun->error_serial;
3005		err_desc->serial = lun->error_serial;
3006		lun->error_serial++;
3007
3008		mtx_unlock(&lun->lun_lock);
3009		break;
3010	}
3011	case CTL_ERROR_INJECT_DELETE: {
3012		struct ctl_error_desc *delete_desc, *desc, *desc2;
3013		int delete_done;
3014
3015		delete_desc = (struct ctl_error_desc *)addr;
3016		delete_done = 0;
3017
3018		mtx_lock(&softc->ctl_lock);
3019		lun = softc->ctl_luns[delete_desc->lun_id];
3020		if (lun == NULL) {
3021			mtx_unlock(&softc->ctl_lock);
3022			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
3023			       __func__, (uintmax_t)delete_desc->lun_id);
3024			retval = EINVAL;
3025			break;
3026		}
3027		mtx_lock(&lun->lun_lock);
3028		mtx_unlock(&softc->ctl_lock);
3029		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
3030			if (desc->serial != delete_desc->serial)
3031				continue;
3032
3033			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
3034				      links);
3035			free(desc, M_CTL);
3036			delete_done = 1;
3037		}
3038		mtx_unlock(&lun->lun_lock);
3039		if (delete_done == 0) {
3040			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
3041			       "error serial %ju on LUN %u\n", __func__,
3042			       delete_desc->serial, delete_desc->lun_id);
3043			retval = EINVAL;
3044			break;
3045		}
3046		break;
3047	}
3048	case CTL_DUMP_STRUCTS: {
3049		int i, j, k;
3050		struct ctl_port *port;
3051		struct ctl_frontend *fe;
3052
3053		mtx_lock(&softc->ctl_lock);
3054		printf("CTL Persistent Reservation information start:\n");
3055		for (i = 0; i < CTL_MAX_LUNS; i++) {
3056			lun = softc->ctl_luns[i];
3057
3058			if ((lun == NULL)
3059			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
3060				continue;
3061
3062			for (j = 0; j < CTL_MAX_PORTS; j++) {
3063				if (lun->pr_keys[j] == NULL)
3064					continue;
3065				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
3066					if (lun->pr_keys[j][k] == 0)
3067						continue;
3068					printf("  LUN %d port %d iid %d key "
3069					       "%#jx\n", i, j, k,
3070					       (uintmax_t)lun->pr_keys[j][k]);
3071				}
3072			}
3073		}
3074		printf("CTL Persistent Reservation information end\n");
3075		printf("CTL Ports:\n");
3076		STAILQ_FOREACH(port, &softc->port_list, links) {
3077			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3078			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3079			       port->frontend->name, port->port_type,
3080			       port->physical_port, port->virtual_port,
3081			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3082			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3083				if (port->wwpn_iid[j].in_use == 0 &&
3084				    port->wwpn_iid[j].wwpn == 0 &&
3085				    port->wwpn_iid[j].name == NULL)
3086					continue;
3087
3088				printf("    iid %u use %d WWPN %#jx '%s'\n",
3089				    j, port->wwpn_iid[j].in_use,
3090				    (uintmax_t)port->wwpn_iid[j].wwpn,
3091				    port->wwpn_iid[j].name);
3092			}
3093		}
3094		printf("CTL Port information end\n");
3095		mtx_unlock(&softc->ctl_lock);
3096		/*
3097		 * XXX KDM calling this without a lock.  We'd likely want
3098		 * to drop the lock before calling the frontend's dump
3099		 * routine anyway.
3100		 */
3101		printf("CTL Frontends:\n");
3102		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3103			printf("  Frontend '%s'\n", fe->name);
3104			if (fe->fe_dump != NULL)
3105				fe->fe_dump();
3106		}
3107		printf("CTL Frontend information end\n");
3108		break;
3109	}
3110	case CTL_LUN_REQ: {
3111		struct ctl_lun_req *lun_req;
3112		struct ctl_backend_driver *backend;
3113
3114		lun_req = (struct ctl_lun_req *)addr;
3115
3116		backend = ctl_backend_find(lun_req->backend);
3117		if (backend == NULL) {
3118			lun_req->status = CTL_LUN_ERROR;
3119			snprintf(lun_req->error_str,
3120				 sizeof(lun_req->error_str),
3121				 "Backend \"%s\" not found.",
3122				 lun_req->backend);
3123			break;
3124		}
3125		if (lun_req->num_be_args > 0) {
3126			lun_req->kern_be_args = ctl_copyin_args(
3127				lun_req->num_be_args,
3128				lun_req->be_args,
3129				lun_req->error_str,
3130				sizeof(lun_req->error_str));
3131			if (lun_req->kern_be_args == NULL) {
3132				lun_req->status = CTL_LUN_ERROR;
3133				break;
3134			}
3135		}
3136
3137		retval = backend->ioctl(dev, cmd, addr, flag, td);
3138
3139		if (lun_req->num_be_args > 0) {
3140			ctl_copyout_args(lun_req->num_be_args,
3141				      lun_req->kern_be_args);
3142			ctl_free_args(lun_req->num_be_args,
3143				      lun_req->kern_be_args);
3144		}
3145		break;
3146	}
3147	case CTL_LUN_LIST: {
3148		struct sbuf *sb;
3149		struct ctl_lun_list *list;
3150		struct ctl_option *opt;
3151
3152		list = (struct ctl_lun_list *)addr;
3153
3154		/*
3155		 * Allocate a fixed length sbuf here, based on the length
3156		 * of the user's buffer.  We could allocate an auto-extending
3157		 * buffer, and then tell the user how much larger our
3158		 * amount of data is than his buffer, but that presents
3159		 * some problems:
3160		 *
3161		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3162		 *     we can't hold a lock while calling them with an
3163		 *     auto-extending buffer.
3164 		 *
3165		 * 2.  There is not currently a LUN reference counting
3166		 *     mechanism, outside of outstanding transactions on
3167		 *     the LUN's OOA queue.  So a LUN could go away on us
3168		 *     while we're getting the LUN number, backend-specific
3169		 *     information, etc.  Thus, given the way things
3170		 *     currently work, we need to hold the CTL lock while
3171		 *     grabbing LUN information.
3172		 *
3173		 * So, from the user's standpoint, the best thing to do is
3174		 * allocate what he thinks is a reasonable buffer length,
3175		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3176		 * double the buffer length and try again.  (And repeat
3177		 * that until he succeeds.)
3178		 */
3179		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3180		if (sb == NULL) {
3181			list->status = CTL_LUN_LIST_ERROR;
3182			snprintf(list->error_str, sizeof(list->error_str),
3183				 "Unable to allocate %d bytes for LUN list",
3184				 list->alloc_len);
3185			break;
3186		}
3187
3188		sbuf_printf(sb, "<ctllunlist>\n");
3189
3190		mtx_lock(&softc->ctl_lock);
3191		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3192			mtx_lock(&lun->lun_lock);
3193			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3194					     (uintmax_t)lun->lun);
3195
3196			/*
3197			 * Bail out as soon as we see that we've overfilled
3198			 * the buffer.
3199			 */
3200			if (retval != 0)
3201				break;
3202
3203			retval = sbuf_printf(sb, "\t<backend_type>%s"
3204					     "</backend_type>\n",
3205					     (lun->backend == NULL) ?  "none" :
3206					     lun->backend->name);
3207
3208			if (retval != 0)
3209				break;
3210
3211			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3212					     lun->be_lun->lun_type);
3213
3214			if (retval != 0)
3215				break;
3216
3217			if (lun->backend == NULL) {
3218				retval = sbuf_printf(sb, "</lun>\n");
3219				if (retval != 0)
3220					break;
3221				continue;
3222			}
3223
3224			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3225					     (lun->be_lun->maxlba > 0) ?
3226					     lun->be_lun->maxlba + 1 : 0);
3227
3228			if (retval != 0)
3229				break;
3230
3231			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3232					     lun->be_lun->blocksize);
3233
3234			if (retval != 0)
3235				break;
3236
3237			retval = sbuf_printf(sb, "\t<serial_number>");
3238
3239			if (retval != 0)
3240				break;
3241
3242			retval = ctl_sbuf_printf_esc(sb,
3243			    lun->be_lun->serial_num,
3244			    sizeof(lun->be_lun->serial_num));
3245
3246			if (retval != 0)
3247				break;
3248
3249			retval = sbuf_printf(sb, "</serial_number>\n");
3250
3251			if (retval != 0)
3252				break;
3253
3254			retval = sbuf_printf(sb, "\t<device_id>");
3255
3256			if (retval != 0)
3257				break;
3258
3259			retval = ctl_sbuf_printf_esc(sb,
3260			    lun->be_lun->device_id,
3261			    sizeof(lun->be_lun->device_id));
3262
3263			if (retval != 0)
3264				break;
3265
3266			retval = sbuf_printf(sb, "</device_id>\n");
3267
3268			if (retval != 0)
3269				break;
3270
3271			if (lun->backend->lun_info != NULL) {
3272				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3273				if (retval != 0)
3274					break;
3275			}
3276			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3277				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3278				    opt->name, opt->value, opt->name);
3279				if (retval != 0)
3280					break;
3281			}
3282
3283			retval = sbuf_printf(sb, "</lun>\n");
3284
3285			if (retval != 0)
3286				break;
3287			mtx_unlock(&lun->lun_lock);
3288		}
3289		if (lun != NULL)
3290			mtx_unlock(&lun->lun_lock);
3291		mtx_unlock(&softc->ctl_lock);
3292
3293		if ((retval != 0)
3294		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3295			retval = 0;
3296			sbuf_delete(sb);
3297			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3298			snprintf(list->error_str, sizeof(list->error_str),
3299				 "Out of space, %d bytes is too small",
3300				 list->alloc_len);
3301			break;
3302		}
3303
3304		sbuf_finish(sb);
3305
3306		retval = copyout(sbuf_data(sb), list->lun_xml,
3307				 sbuf_len(sb) + 1);
3308
3309		list->fill_len = sbuf_len(sb) + 1;
3310		list->status = CTL_LUN_LIST_OK;
3311		sbuf_delete(sb);
3312		break;
3313	}
3314	case CTL_ISCSI: {
3315		struct ctl_iscsi *ci;
3316		struct ctl_frontend *fe;
3317
3318		ci = (struct ctl_iscsi *)addr;
3319
3320		fe = ctl_frontend_find("iscsi");
3321		if (fe == NULL) {
3322			ci->status = CTL_ISCSI_ERROR;
3323			snprintf(ci->error_str, sizeof(ci->error_str),
3324			    "Frontend \"iscsi\" not found.");
3325			break;
3326		}
3327
3328		retval = fe->ioctl(dev, cmd, addr, flag, td);
3329		break;
3330	}
3331	case CTL_PORT_REQ: {
3332		struct ctl_req *req;
3333		struct ctl_frontend *fe;
3334
3335		req = (struct ctl_req *)addr;
3336
3337		fe = ctl_frontend_find(req->driver);
3338		if (fe == NULL) {
3339			req->status = CTL_LUN_ERROR;
3340			snprintf(req->error_str, sizeof(req->error_str),
3341			    "Frontend \"%s\" not found.", req->driver);
3342			break;
3343		}
3344		if (req->num_args > 0) {
3345			req->kern_args = ctl_copyin_args(req->num_args,
3346			    req->args, req->error_str, sizeof(req->error_str));
3347			if (req->kern_args == NULL) {
3348				req->status = CTL_LUN_ERROR;
3349				break;
3350			}
3351		}
3352
3353		if (fe->ioctl)
3354			retval = fe->ioctl(dev, cmd, addr, flag, td);
3355		else
3356			retval = ENODEV;
3357
3358		if (req->num_args > 0) {
3359			ctl_copyout_args(req->num_args, req->kern_args);
3360			ctl_free_args(req->num_args, req->kern_args);
3361		}
3362		break;
3363	}
3364	case CTL_PORT_LIST: {
3365		struct sbuf *sb;
3366		struct ctl_port *port;
3367		struct ctl_lun_list *list;
3368		struct ctl_option *opt;
3369		int j;
3370		uint32_t plun;
3371
3372		list = (struct ctl_lun_list *)addr;
3373
3374		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3375		if (sb == NULL) {
3376			list->status = CTL_LUN_LIST_ERROR;
3377			snprintf(list->error_str, sizeof(list->error_str),
3378				 "Unable to allocate %d bytes for LUN list",
3379				 list->alloc_len);
3380			break;
3381		}
3382
3383		sbuf_printf(sb, "<ctlportlist>\n");
3384
3385		mtx_lock(&softc->ctl_lock);
3386		STAILQ_FOREACH(port, &softc->port_list, links) {
3387			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3388					     (uintmax_t)port->targ_port);
3389
3390			/*
3391			 * Bail out as soon as we see that we've overfilled
3392			 * the buffer.
3393			 */
3394			if (retval != 0)
3395				break;
3396
3397			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3398			    "</frontend_type>\n", port->frontend->name);
3399			if (retval != 0)
3400				break;
3401
3402			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3403					     port->port_type);
3404			if (retval != 0)
3405				break;
3406
3407			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3408			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3409			if (retval != 0)
3410				break;
3411
3412			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3413			    port->port_name);
3414			if (retval != 0)
3415				break;
3416
3417			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3418			    port->physical_port);
3419			if (retval != 0)
3420				break;
3421
3422			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3423			    port->virtual_port);
3424			if (retval != 0)
3425				break;
3426
3427			if (port->target_devid != NULL) {
3428				sbuf_printf(sb, "\t<target>");
3429				ctl_id_sbuf(port->target_devid, sb);
3430				sbuf_printf(sb, "</target>\n");
3431			}
3432
3433			if (port->port_devid != NULL) {
3434				sbuf_printf(sb, "\t<port>");
3435				ctl_id_sbuf(port->port_devid, sb);
3436				sbuf_printf(sb, "</port>\n");
3437			}
3438
3439			if (port->port_info != NULL) {
3440				retval = port->port_info(port->onoff_arg, sb);
3441				if (retval != 0)
3442					break;
3443			}
3444			STAILQ_FOREACH(opt, &port->options, links) {
3445				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3446				    opt->name, opt->value, opt->name);
3447				if (retval != 0)
3448					break;
3449			}
3450
3451			if (port->lun_map != NULL) {
3452				sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3453				for (j = 0; j < CTL_MAX_LUNS; j++) {
3454					plun = ctl_lun_map_from_port(port, j);
3455					if (plun >= CTL_MAX_LUNS)
3456						continue;
3457					sbuf_printf(sb,
3458					    "\t<lun id=\"%u\">%u</lun>\n",
3459					    j, plun);
3460				}
3461			}
3462
3463			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3464				if (port->wwpn_iid[j].in_use == 0 ||
3465				    (port->wwpn_iid[j].wwpn == 0 &&
3466				     port->wwpn_iid[j].name == NULL))
3467					continue;
3468
3469				if (port->wwpn_iid[j].name != NULL)
3470					retval = sbuf_printf(sb,
3471					    "\t<initiator id=\"%u\">%s</initiator>\n",
3472					    j, port->wwpn_iid[j].name);
3473				else
3474					retval = sbuf_printf(sb,
3475					    "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3476					    j, port->wwpn_iid[j].wwpn);
3477				if (retval != 0)
3478					break;
3479			}
3480			if (retval != 0)
3481				break;
3482
3483			retval = sbuf_printf(sb, "</targ_port>\n");
3484			if (retval != 0)
3485				break;
3486		}
3487		mtx_unlock(&softc->ctl_lock);
3488
3489		if ((retval != 0)
3490		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3491			retval = 0;
3492			sbuf_delete(sb);
3493			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3494			snprintf(list->error_str, sizeof(list->error_str),
3495				 "Out of space, %d bytes is too small",
3496				 list->alloc_len);
3497			break;
3498		}
3499
3500		sbuf_finish(sb);
3501
3502		retval = copyout(sbuf_data(sb), list->lun_xml,
3503				 sbuf_len(sb) + 1);
3504
3505		list->fill_len = sbuf_len(sb) + 1;
3506		list->status = CTL_LUN_LIST_OK;
3507		sbuf_delete(sb);
3508		break;
3509	}
3510	case CTL_LUN_MAP: {
3511		struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3512		struct ctl_port *port;
3513
3514		mtx_lock(&softc->ctl_lock);
3515		if (lm->port < softc->port_min ||
3516		    lm->port >= softc->port_max ||
3517		    (port = softc->ctl_ports[lm->port]) == NULL) {
3518			mtx_unlock(&softc->ctl_lock);
3519			return (ENXIO);
3520		}
3521		if (port->status & CTL_PORT_STATUS_ONLINE) {
3522			STAILQ_FOREACH(lun, &softc->lun_list, links) {
3523				if (ctl_lun_map_to_port(port, lun->lun) >=
3524				    CTL_MAX_LUNS)
3525					continue;
3526				mtx_lock(&lun->lun_lock);
3527				ctl_est_ua_port(lun, lm->port, -1,
3528				    CTL_UA_LUN_CHANGE);
3529				mtx_unlock(&lun->lun_lock);
3530			}
3531		}
3532		mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3533		if (lm->plun < CTL_MAX_LUNS) {
3534			if (lm->lun == UINT32_MAX)
3535				retval = ctl_lun_map_unset(port, lm->plun);
3536			else if (lm->lun < CTL_MAX_LUNS &&
3537			    softc->ctl_luns[lm->lun] != NULL)
3538				retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3539			else
3540				return (ENXIO);
3541		} else if (lm->plun == UINT32_MAX) {
3542			if (lm->lun == UINT32_MAX)
3543				retval = ctl_lun_map_deinit(port);
3544			else
3545				retval = ctl_lun_map_init(port);
3546		} else
3547			return (ENXIO);
3548		if (port->status & CTL_PORT_STATUS_ONLINE)
3549			ctl_isc_announce_port(port);
3550		break;
3551	}
3552	default: {
3553		/* XXX KDM should we fix this? */
3554#if 0
3555		struct ctl_backend_driver *backend;
3556		unsigned int type;
3557		int found;
3558
3559		found = 0;
3560
3561		/*
3562		 * We encode the backend type as the ioctl type for backend
3563		 * ioctls.  So parse it out here, and then search for a
3564		 * backend of this type.
3565		 */
3566		type = _IOC_TYPE(cmd);
3567
3568		STAILQ_FOREACH(backend, &softc->be_list, links) {
3569			if (backend->type == type) {
3570				found = 1;
3571				break;
3572			}
3573		}
3574		if (found == 0) {
3575			printf("ctl: unknown ioctl command %#lx or backend "
3576			       "%d\n", cmd, type);
3577			retval = EINVAL;
3578			break;
3579		}
3580		retval = backend->ioctl(dev, cmd, addr, flag, td);
3581#endif
3582		retval = ENOTTY;
3583		break;
3584	}
3585	}
3586	return (retval);
3587}
3588
3589uint32_t
3590ctl_get_initindex(struct ctl_nexus *nexus)
3591{
3592	return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3593}
3594
3595int
3596ctl_lun_map_init(struct ctl_port *port)
3597{
3598	struct ctl_softc *softc = port->ctl_softc;
3599	struct ctl_lun *lun;
3600	uint32_t i;
3601
3602	if (port->lun_map == NULL)
3603		port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
3604		    M_CTL, M_NOWAIT);
3605	if (port->lun_map == NULL)
3606		return (ENOMEM);
3607	for (i = 0; i < CTL_MAX_LUNS; i++)
3608		port->lun_map[i] = UINT32_MAX;
3609	if (port->status & CTL_PORT_STATUS_ONLINE) {
3610		if (port->lun_disable != NULL) {
3611			STAILQ_FOREACH(lun, &softc->lun_list, links)
3612				port->lun_disable(port->targ_lun_arg, lun->lun);
3613		}
3614		ctl_isc_announce_port(port);
3615	}
3616	return (0);
3617}
3618
3619int
3620ctl_lun_map_deinit(struct ctl_port *port)
3621{
3622	struct ctl_softc *softc = port->ctl_softc;
3623	struct ctl_lun *lun;
3624
3625	if (port->lun_map == NULL)
3626		return (0);
3627	free(port->lun_map, M_CTL);
3628	port->lun_map = NULL;
3629	if (port->status & CTL_PORT_STATUS_ONLINE) {
3630		if (port->lun_enable != NULL) {
3631			STAILQ_FOREACH(lun, &softc->lun_list, links)
3632				port->lun_enable(port->targ_lun_arg, lun->lun);
3633		}
3634		ctl_isc_announce_port(port);
3635	}
3636	return (0);
3637}
3638
3639int
3640ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3641{
3642	int status;
3643	uint32_t old;
3644
3645	if (port->lun_map == NULL) {
3646		status = ctl_lun_map_init(port);
3647		if (status != 0)
3648			return (status);
3649	}
3650	old = port->lun_map[plun];
3651	port->lun_map[plun] = glun;
3652	if ((port->status & CTL_PORT_STATUS_ONLINE) && old >= CTL_MAX_LUNS) {
3653		if (port->lun_enable != NULL)
3654			port->lun_enable(port->targ_lun_arg, plun);
3655		ctl_isc_announce_port(port);
3656	}
3657	return (0);
3658}
3659
3660int
3661ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3662{
3663	uint32_t old;
3664
3665	if (port->lun_map == NULL)
3666		return (0);
3667	old = port->lun_map[plun];
3668	port->lun_map[plun] = UINT32_MAX;
3669	if ((port->status & CTL_PORT_STATUS_ONLINE) && old < CTL_MAX_LUNS) {
3670		if (port->lun_disable != NULL)
3671			port->lun_disable(port->targ_lun_arg, plun);
3672		ctl_isc_announce_port(port);
3673	}
3674	return (0);
3675}
3676
3677uint32_t
3678ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3679{
3680
3681	if (port == NULL)
3682		return (UINT32_MAX);
3683	if (port->lun_map == NULL || lun_id >= CTL_MAX_LUNS)
3684		return (lun_id);
3685	return (port->lun_map[lun_id]);
3686}
3687
3688uint32_t
3689ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3690{
3691	uint32_t i;
3692
3693	if (port == NULL)
3694		return (UINT32_MAX);
3695	if (port->lun_map == NULL)
3696		return (lun_id);
3697	for (i = 0; i < CTL_MAX_LUNS; i++) {
3698		if (port->lun_map[i] == lun_id)
3699			return (i);
3700	}
3701	return (UINT32_MAX);
3702}
3703
3704static struct ctl_port *
3705ctl_io_port(struct ctl_io_hdr *io_hdr)
3706{
3707
3708	return (control_softc->ctl_ports[io_hdr->nexus.targ_port]);
3709}
3710
3711int
3712ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3713{
3714	int i;
3715
3716	for (i = first; i < last; i++) {
3717		if ((mask[i / 32] & (1 << (i % 32))) == 0)
3718			return (i);
3719	}
3720	return (-1);
3721}
3722
3723int
3724ctl_set_mask(uint32_t *mask, uint32_t bit)
3725{
3726	uint32_t chunk, piece;
3727
3728	chunk = bit >> 5;
3729	piece = bit % (sizeof(uint32_t) * 8);
3730
3731	if ((mask[chunk] & (1 << piece)) != 0)
3732		return (-1);
3733	else
3734		mask[chunk] |= (1 << piece);
3735
3736	return (0);
3737}
3738
3739int
3740ctl_clear_mask(uint32_t *mask, uint32_t bit)
3741{
3742	uint32_t chunk, piece;
3743
3744	chunk = bit >> 5;
3745	piece = bit % (sizeof(uint32_t) * 8);
3746
3747	if ((mask[chunk] & (1 << piece)) == 0)
3748		return (-1);
3749	else
3750		mask[chunk] &= ~(1 << piece);
3751
3752	return (0);
3753}
3754
3755int
3756ctl_is_set(uint32_t *mask, uint32_t bit)
3757{
3758	uint32_t chunk, piece;
3759
3760	chunk = bit >> 5;
3761	piece = bit % (sizeof(uint32_t) * 8);
3762
3763	if ((mask[chunk] & (1 << piece)) == 0)
3764		return (0);
3765	else
3766		return (1);
3767}
3768
3769static uint64_t
3770ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3771{
3772	uint64_t *t;
3773
3774	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3775	if (t == NULL)
3776		return (0);
3777	return (t[residx % CTL_MAX_INIT_PER_PORT]);
3778}
3779
3780static void
3781ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3782{
3783	uint64_t *t;
3784
3785	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3786	if (t == NULL)
3787		return;
3788	t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3789}
3790
3791static void
3792ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3793{
3794	uint64_t *p;
3795	u_int i;
3796
3797	i = residx/CTL_MAX_INIT_PER_PORT;
3798	if (lun->pr_keys[i] != NULL)
3799		return;
3800	mtx_unlock(&lun->lun_lock);
3801	p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3802	    M_WAITOK | M_ZERO);
3803	mtx_lock(&lun->lun_lock);
3804	if (lun->pr_keys[i] == NULL)
3805		lun->pr_keys[i] = p;
3806	else
3807		free(p, M_CTL);
3808}
3809
3810static void
3811ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3812{
3813	uint64_t *t;
3814
3815	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3816	KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3817	t[residx % CTL_MAX_INIT_PER_PORT] = key;
3818}
3819
3820/*
3821 * ctl_softc, pool_name, total_ctl_io are passed in.
3822 * npool is passed out.
3823 */
3824int
3825ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3826		uint32_t total_ctl_io, void **npool)
3827{
3828#ifdef IO_POOLS
3829	struct ctl_io_pool *pool;
3830
3831	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3832					    M_NOWAIT | M_ZERO);
3833	if (pool == NULL)
3834		return (ENOMEM);
3835
3836	snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3837	pool->ctl_softc = ctl_softc;
3838	pool->zone = uma_zsecond_create(pool->name, NULL,
3839	    NULL, NULL, NULL, ctl_softc->io_zone);
3840	/* uma_prealloc(pool->zone, total_ctl_io); */
3841
3842	*npool = pool;
3843#else
3844	*npool = ctl_softc->io_zone;
3845#endif
3846	return (0);
3847}
3848
3849void
3850ctl_pool_free(struct ctl_io_pool *pool)
3851{
3852
3853	if (pool == NULL)
3854		return;
3855
3856#ifdef IO_POOLS
3857	uma_zdestroy(pool->zone);
3858	free(pool, M_CTL);
3859#endif
3860}
3861
3862union ctl_io *
3863ctl_alloc_io(void *pool_ref)
3864{
3865	union ctl_io *io;
3866#ifdef IO_POOLS
3867	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3868
3869	io = uma_zalloc(pool->zone, M_WAITOK);
3870#else
3871	io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3872#endif
3873	if (io != NULL)
3874		io->io_hdr.pool = pool_ref;
3875	return (io);
3876}
3877
3878union ctl_io *
3879ctl_alloc_io_nowait(void *pool_ref)
3880{
3881	union ctl_io *io;
3882#ifdef IO_POOLS
3883	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3884
3885	io = uma_zalloc(pool->zone, M_NOWAIT);
3886#else
3887	io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3888#endif
3889	if (io != NULL)
3890		io->io_hdr.pool = pool_ref;
3891	return (io);
3892}
3893
3894void
3895ctl_free_io(union ctl_io *io)
3896{
3897#ifdef IO_POOLS
3898	struct ctl_io_pool *pool;
3899#endif
3900
3901	if (io == NULL)
3902		return;
3903
3904#ifdef IO_POOLS
3905	pool = (struct ctl_io_pool *)io->io_hdr.pool;
3906	uma_zfree(pool->zone, io);
3907#else
3908	uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3909#endif
3910}
3911
3912void
3913ctl_zero_io(union ctl_io *io)
3914{
3915	void *pool_ref;
3916
3917	if (io == NULL)
3918		return;
3919
3920	/*
3921	 * May need to preserve linked list pointers at some point too.
3922	 */
3923	pool_ref = io->io_hdr.pool;
3924	memset(io, 0, sizeof(*io));
3925	io->io_hdr.pool = pool_ref;
3926}
3927
3928/*
3929 * This routine is currently used for internal copies of ctl_ios that need
3930 * to persist for some reason after we've already returned status to the
3931 * FETD.  (Thus the flag set.)
3932 *
3933 * XXX XXX
3934 * Note that this makes a blind copy of all fields in the ctl_io, except
3935 * for the pool reference.  This includes any memory that has been
3936 * allocated!  That memory will no longer be valid after done has been
3937 * called, so this would be VERY DANGEROUS for command that actually does
3938 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3939 * start and stop commands, which don't transfer any data, so this is not a
3940 * problem.  If it is used for anything else, the caller would also need to
3941 * allocate data buffer space and this routine would need to be modified to
3942 * copy the data buffer(s) as well.
3943 */
3944void
3945ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3946{
3947	void *pool_ref;
3948
3949	if ((src == NULL)
3950	 || (dest == NULL))
3951		return;
3952
3953	/*
3954	 * May need to preserve linked list pointers at some point too.
3955	 */
3956	pool_ref = dest->io_hdr.pool;
3957
3958	memcpy(dest, src, MIN(sizeof(*src), sizeof(*dest)));
3959
3960	dest->io_hdr.pool = pool_ref;
3961	/*
3962	 * We need to know that this is an internal copy, and doesn't need
3963	 * to get passed back to the FETD that allocated it.
3964	 */
3965	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3966}
3967
3968int
3969ctl_expand_number(const char *buf, uint64_t *num)
3970{
3971	char *endptr;
3972	uint64_t number;
3973	unsigned shift;
3974
3975	number = strtoq(buf, &endptr, 0);
3976
3977	switch (tolower((unsigned char)*endptr)) {
3978	case 'e':
3979		shift = 60;
3980		break;
3981	case 'p':
3982		shift = 50;
3983		break;
3984	case 't':
3985		shift = 40;
3986		break;
3987	case 'g':
3988		shift = 30;
3989		break;
3990	case 'm':
3991		shift = 20;
3992		break;
3993	case 'k':
3994		shift = 10;
3995		break;
3996	case 'b':
3997	case '\0': /* No unit. */
3998		*num = number;
3999		return (0);
4000	default:
4001		/* Unrecognized unit. */
4002		return (-1);
4003	}
4004
4005	if ((number << shift) >> shift != number) {
4006		/* Overflow */
4007		return (-1);
4008	}
4009	*num = number << shift;
4010	return (0);
4011}
4012
4013
4014/*
4015 * This routine could be used in the future to load default and/or saved
4016 * mode page parameters for a particuar lun.
4017 */
4018static int
4019ctl_init_page_index(struct ctl_lun *lun)
4020{
4021	int i;
4022	struct ctl_page_index *page_index;
4023	const char *value;
4024	uint64_t ival;
4025
4026	memcpy(&lun->mode_pages.index, page_index_template,
4027	       sizeof(page_index_template));
4028
4029	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4030
4031		page_index = &lun->mode_pages.index[i];
4032		/*
4033		 * If this is a disk-only mode page, there's no point in
4034		 * setting it up.  For some pages, we have to have some
4035		 * basic information about the disk in order to calculate the
4036		 * mode page data.
4037		 */
4038		if ((lun->be_lun->lun_type != T_DIRECT)
4039		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4040			continue;
4041
4042		switch (page_index->page_code & SMPH_PC_MASK) {
4043		case SMS_RW_ERROR_RECOVERY_PAGE: {
4044			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4045				panic("subpage is incorrect!");
4046			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4047			       &rw_er_page_default,
4048			       sizeof(rw_er_page_default));
4049			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4050			       &rw_er_page_changeable,
4051			       sizeof(rw_er_page_changeable));
4052			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4053			       &rw_er_page_default,
4054			       sizeof(rw_er_page_default));
4055			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4056			       &rw_er_page_default,
4057			       sizeof(rw_er_page_default));
4058			page_index->page_data =
4059				(uint8_t *)lun->mode_pages.rw_er_page;
4060			break;
4061		}
4062		case SMS_FORMAT_DEVICE_PAGE: {
4063			struct scsi_format_page *format_page;
4064
4065			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4066				panic("subpage is incorrect!");
4067
4068			/*
4069			 * Sectors per track are set above.  Bytes per
4070			 * sector need to be set here on a per-LUN basis.
4071			 */
4072			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4073			       &format_page_default,
4074			       sizeof(format_page_default));
4075			memcpy(&lun->mode_pages.format_page[
4076			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4077			       sizeof(format_page_changeable));
4078			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4079			       &format_page_default,
4080			       sizeof(format_page_default));
4081			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4082			       &format_page_default,
4083			       sizeof(format_page_default));
4084
4085			format_page = &lun->mode_pages.format_page[
4086				CTL_PAGE_CURRENT];
4087			scsi_ulto2b(lun->be_lun->blocksize,
4088				    format_page->bytes_per_sector);
4089
4090			format_page = &lun->mode_pages.format_page[
4091				CTL_PAGE_DEFAULT];
4092			scsi_ulto2b(lun->be_lun->blocksize,
4093				    format_page->bytes_per_sector);
4094
4095			format_page = &lun->mode_pages.format_page[
4096				CTL_PAGE_SAVED];
4097			scsi_ulto2b(lun->be_lun->blocksize,
4098				    format_page->bytes_per_sector);
4099
4100			page_index->page_data =
4101				(uint8_t *)lun->mode_pages.format_page;
4102			break;
4103		}
4104		case SMS_RIGID_DISK_PAGE: {
4105			struct scsi_rigid_disk_page *rigid_disk_page;
4106			uint32_t sectors_per_cylinder;
4107			uint64_t cylinders;
4108#ifndef	__XSCALE__
4109			int shift;
4110#endif /* !__XSCALE__ */
4111
4112			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4113				panic("invalid subpage value %d",
4114				      page_index->subpage);
4115
4116			/*
4117			 * Rotation rate and sectors per track are set
4118			 * above.  We calculate the cylinders here based on
4119			 * capacity.  Due to the number of heads and
4120			 * sectors per track we're using, smaller arrays
4121			 * may turn out to have 0 cylinders.  Linux and
4122			 * FreeBSD don't pay attention to these mode pages
4123			 * to figure out capacity, but Solaris does.  It
4124			 * seems to deal with 0 cylinders just fine, and
4125			 * works out a fake geometry based on the capacity.
4126			 */
4127			memcpy(&lun->mode_pages.rigid_disk_page[
4128			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4129			       sizeof(rigid_disk_page_default));
4130			memcpy(&lun->mode_pages.rigid_disk_page[
4131			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4132			       sizeof(rigid_disk_page_changeable));
4133
4134			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4135				CTL_DEFAULT_HEADS;
4136
4137			/*
4138			 * The divide method here will be more accurate,
4139			 * probably, but results in floating point being
4140			 * used in the kernel on i386 (__udivdi3()).  On the
4141			 * XScale, though, __udivdi3() is implemented in
4142			 * software.
4143			 *
4144			 * The shift method for cylinder calculation is
4145			 * accurate if sectors_per_cylinder is a power of
4146			 * 2.  Otherwise it might be slightly off -- you
4147			 * might have a bit of a truncation problem.
4148			 */
4149#ifdef	__XSCALE__
4150			cylinders = (lun->be_lun->maxlba + 1) /
4151				sectors_per_cylinder;
4152#else
4153			for (shift = 31; shift > 0; shift--) {
4154				if (sectors_per_cylinder & (1 << shift))
4155					break;
4156			}
4157			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4158#endif
4159
4160			/*
4161			 * We've basically got 3 bytes, or 24 bits for the
4162			 * cylinder size in the mode page.  If we're over,
4163			 * just round down to 2^24.
4164			 */
4165			if (cylinders > 0xffffff)
4166				cylinders = 0xffffff;
4167
4168			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4169				CTL_PAGE_DEFAULT];
4170			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4171
4172			if ((value = ctl_get_opt(&lun->be_lun->options,
4173			    "rpm")) != NULL) {
4174				scsi_ulto2b(strtol(value, NULL, 0),
4175				     rigid_disk_page->rotation_rate);
4176			}
4177
4178			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4179			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4180			       sizeof(rigid_disk_page_default));
4181			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4182			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4183			       sizeof(rigid_disk_page_default));
4184
4185			page_index->page_data =
4186				(uint8_t *)lun->mode_pages.rigid_disk_page;
4187			break;
4188		}
4189		case SMS_CACHING_PAGE: {
4190			struct scsi_caching_page *caching_page;
4191
4192			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4193				panic("invalid subpage value %d",
4194				      page_index->subpage);
4195			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4196			       &caching_page_default,
4197			       sizeof(caching_page_default));
4198			memcpy(&lun->mode_pages.caching_page[
4199			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4200			       sizeof(caching_page_changeable));
4201			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4202			       &caching_page_default,
4203			       sizeof(caching_page_default));
4204			caching_page = &lun->mode_pages.caching_page[
4205			    CTL_PAGE_SAVED];
4206			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4207			if (value != NULL && strcmp(value, "off") == 0)
4208				caching_page->flags1 &= ~SCP_WCE;
4209			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4210			if (value != NULL && strcmp(value, "off") == 0)
4211				caching_page->flags1 |= SCP_RCD;
4212			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4213			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4214			       sizeof(caching_page_default));
4215			page_index->page_data =
4216				(uint8_t *)lun->mode_pages.caching_page;
4217			break;
4218		}
4219		case SMS_CONTROL_MODE_PAGE: {
4220			switch (page_index->subpage) {
4221			case SMS_SUBPAGE_PAGE_0: {
4222				struct scsi_control_page *control_page;
4223
4224				memcpy(&lun->mode_pages.control_page[
4225				    CTL_PAGE_DEFAULT],
4226				       &control_page_default,
4227				       sizeof(control_page_default));
4228				memcpy(&lun->mode_pages.control_page[
4229				    CTL_PAGE_CHANGEABLE],
4230				       &control_page_changeable,
4231				       sizeof(control_page_changeable));
4232				memcpy(&lun->mode_pages.control_page[
4233				    CTL_PAGE_SAVED],
4234				       &control_page_default,
4235				       sizeof(control_page_default));
4236				control_page = &lun->mode_pages.control_page[
4237				    CTL_PAGE_SAVED];
4238				value = ctl_get_opt(&lun->be_lun->options,
4239				    "reordering");
4240				if (value != NULL &&
4241				    strcmp(value, "unrestricted") == 0) {
4242					control_page->queue_flags &=
4243					    ~SCP_QUEUE_ALG_MASK;
4244					control_page->queue_flags |=
4245					    SCP_QUEUE_ALG_UNRESTRICTED;
4246				}
4247				memcpy(&lun->mode_pages.control_page[
4248				    CTL_PAGE_CURRENT],
4249				       &lun->mode_pages.control_page[
4250				    CTL_PAGE_SAVED],
4251				       sizeof(control_page_default));
4252				page_index->page_data =
4253				    (uint8_t *)lun->mode_pages.control_page;
4254				break;
4255			}
4256			case 0x01:
4257				memcpy(&lun->mode_pages.control_ext_page[
4258				    CTL_PAGE_DEFAULT],
4259				       &control_ext_page_default,
4260				       sizeof(control_ext_page_default));
4261				memcpy(&lun->mode_pages.control_ext_page[
4262				    CTL_PAGE_CHANGEABLE],
4263				       &control_ext_page_changeable,
4264				       sizeof(control_ext_page_changeable));
4265				memcpy(&lun->mode_pages.control_ext_page[
4266				    CTL_PAGE_SAVED],
4267				       &control_ext_page_default,
4268				       sizeof(control_ext_page_default));
4269				memcpy(&lun->mode_pages.control_ext_page[
4270				    CTL_PAGE_CURRENT],
4271				       &lun->mode_pages.control_ext_page[
4272				    CTL_PAGE_SAVED],
4273				       sizeof(control_ext_page_default));
4274				page_index->page_data =
4275				    (uint8_t *)lun->mode_pages.control_ext_page;
4276				break;
4277			}
4278			break;
4279		}
4280		case SMS_INFO_EXCEPTIONS_PAGE: {
4281			switch (page_index->subpage) {
4282			case SMS_SUBPAGE_PAGE_0:
4283				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4284				       &ie_page_default,
4285				       sizeof(ie_page_default));
4286				memcpy(&lun->mode_pages.ie_page[
4287				       CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4288				       sizeof(ie_page_changeable));
4289				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4290				       &ie_page_default,
4291				       sizeof(ie_page_default));
4292				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4293				       &ie_page_default,
4294				       sizeof(ie_page_default));
4295				page_index->page_data =
4296					(uint8_t *)lun->mode_pages.ie_page;
4297				break;
4298			case 0x02: {
4299				struct ctl_logical_block_provisioning_page *page;
4300
4301				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4302				       &lbp_page_default,
4303				       sizeof(lbp_page_default));
4304				memcpy(&lun->mode_pages.lbp_page[
4305				       CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4306				       sizeof(lbp_page_changeable));
4307				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4308				       &lbp_page_default,
4309				       sizeof(lbp_page_default));
4310				page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4311				value = ctl_get_opt(&lun->be_lun->options,
4312				    "avail-threshold");
4313				if (value != NULL &&
4314				    ctl_expand_number(value, &ival) == 0) {
4315					page->descr[0].flags |= SLBPPD_ENABLED |
4316					    SLBPPD_ARMING_DEC;
4317					if (lun->be_lun->blocksize)
4318						ival /= lun->be_lun->blocksize;
4319					else
4320						ival /= 512;
4321					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4322					    page->descr[0].count);
4323				}
4324				value = ctl_get_opt(&lun->be_lun->options,
4325				    "used-threshold");
4326				if (value != NULL &&
4327				    ctl_expand_number(value, &ival) == 0) {
4328					page->descr[1].flags |= SLBPPD_ENABLED |
4329					    SLBPPD_ARMING_INC;
4330					if (lun->be_lun->blocksize)
4331						ival /= lun->be_lun->blocksize;
4332					else
4333						ival /= 512;
4334					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4335					    page->descr[1].count);
4336				}
4337				value = ctl_get_opt(&lun->be_lun->options,
4338				    "pool-avail-threshold");
4339				if (value != NULL &&
4340				    ctl_expand_number(value, &ival) == 0) {
4341					page->descr[2].flags |= SLBPPD_ENABLED |
4342					    SLBPPD_ARMING_DEC;
4343					if (lun->be_lun->blocksize)
4344						ival /= lun->be_lun->blocksize;
4345					else
4346						ival /= 512;
4347					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4348					    page->descr[2].count);
4349				}
4350				value = ctl_get_opt(&lun->be_lun->options,
4351				    "pool-used-threshold");
4352				if (value != NULL &&
4353				    ctl_expand_number(value, &ival) == 0) {
4354					page->descr[3].flags |= SLBPPD_ENABLED |
4355					    SLBPPD_ARMING_INC;
4356					if (lun->be_lun->blocksize)
4357						ival /= lun->be_lun->blocksize;
4358					else
4359						ival /= 512;
4360					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4361					    page->descr[3].count);
4362				}
4363				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4364				       &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4365				       sizeof(lbp_page_default));
4366				page_index->page_data =
4367					(uint8_t *)lun->mode_pages.lbp_page;
4368			}}
4369			break;
4370		}
4371		case SMS_VENDOR_SPECIFIC_PAGE:{
4372			switch (page_index->subpage) {
4373			case DBGCNF_SUBPAGE_CODE: {
4374				memcpy(&lun->mode_pages.debugconf_subpage[
4375				       CTL_PAGE_CURRENT],
4376				       &debugconf_page_default,
4377				       sizeof(debugconf_page_default));
4378				memcpy(&lun->mode_pages.debugconf_subpage[
4379				       CTL_PAGE_CHANGEABLE],
4380				       &debugconf_page_changeable,
4381				       sizeof(debugconf_page_changeable));
4382				memcpy(&lun->mode_pages.debugconf_subpage[
4383				       CTL_PAGE_DEFAULT],
4384				       &debugconf_page_default,
4385				       sizeof(debugconf_page_default));
4386				memcpy(&lun->mode_pages.debugconf_subpage[
4387				       CTL_PAGE_SAVED],
4388				       &debugconf_page_default,
4389				       sizeof(debugconf_page_default));
4390				page_index->page_data =
4391				    (uint8_t *)lun->mode_pages.debugconf_subpage;
4392				break;
4393			}
4394			default:
4395				panic("invalid subpage value %d",
4396				      page_index->subpage);
4397				break;
4398			}
4399   			break;
4400		}
4401		default:
4402			panic("invalid page value %d",
4403			      page_index->page_code & SMPH_PC_MASK);
4404			break;
4405    	}
4406	}
4407
4408	return (CTL_RETVAL_COMPLETE);
4409}
4410
4411static int
4412ctl_init_log_page_index(struct ctl_lun *lun)
4413{
4414	struct ctl_page_index *page_index;
4415	int i, j, k, prev;
4416
4417	memcpy(&lun->log_pages.index, log_page_index_template,
4418	       sizeof(log_page_index_template));
4419
4420	prev = -1;
4421	for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4422
4423		page_index = &lun->log_pages.index[i];
4424		/*
4425		 * If this is a disk-only mode page, there's no point in
4426		 * setting it up.  For some pages, we have to have some
4427		 * basic information about the disk in order to calculate the
4428		 * mode page data.
4429		 */
4430		if ((lun->be_lun->lun_type != T_DIRECT)
4431		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4432			continue;
4433
4434		if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4435		     lun->backend->lun_attr == NULL)
4436			continue;
4437
4438		if (page_index->page_code != prev) {
4439			lun->log_pages.pages_page[j] = page_index->page_code;
4440			prev = page_index->page_code;
4441			j++;
4442		}
4443		lun->log_pages.subpages_page[k*2] = page_index->page_code;
4444		lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4445		k++;
4446	}
4447	lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4448	lun->log_pages.index[0].page_len = j;
4449	lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4450	lun->log_pages.index[1].page_len = k * 2;
4451	lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4452	lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4453	lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4454	lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4455
4456	return (CTL_RETVAL_COMPLETE);
4457}
4458
4459static int
4460hex2bin(const char *str, uint8_t *buf, int buf_size)
4461{
4462	int i;
4463	u_char c;
4464
4465	memset(buf, 0, buf_size);
4466	while (isspace(str[0]))
4467		str++;
4468	if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4469		str += 2;
4470	buf_size *= 2;
4471	for (i = 0; str[i] != 0 && i < buf_size; i++) {
4472		c = str[i];
4473		if (isdigit(c))
4474			c -= '0';
4475		else if (isalpha(c))
4476			c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4477		else
4478			break;
4479		if (c >= 16)
4480			break;
4481		if ((i & 1) == 0)
4482			buf[i / 2] |= (c << 4);
4483		else
4484			buf[i / 2] |= c;
4485	}
4486	return ((i + 1) / 2);
4487}
4488
4489/*
4490 * LUN allocation.
4491 *
4492 * Requirements:
4493 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4494 *   wants us to allocate the LUN and he can block.
4495 * - ctl_softc is always set
4496 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4497 *
4498 * Returns 0 for success, non-zero (errno) for failure.
4499 */
4500static int
4501ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4502	      struct ctl_be_lun *const be_lun)
4503{
4504	struct ctl_lun *nlun, *lun;
4505	struct scsi_vpd_id_descriptor *desc;
4506	struct scsi_vpd_id_t10 *t10id;
4507	const char *eui, *naa, *scsiname, *vendor;
4508	int lun_number, i, lun_malloced;
4509	int devidlen, idlen1, idlen2 = 0, len;
4510
4511	if (be_lun == NULL)
4512		return (EINVAL);
4513
4514	/*
4515	 * We currently only support Direct Access or Processor LUN types.
4516	 */
4517	switch (be_lun->lun_type) {
4518	case T_DIRECT:
4519		break;
4520	case T_PROCESSOR:
4521		break;
4522	case T_SEQUENTIAL:
4523	case T_CHANGER:
4524	default:
4525		be_lun->lun_config_status(be_lun->be_lun,
4526					  CTL_LUN_CONFIG_FAILURE);
4527		break;
4528	}
4529	if (ctl_lun == NULL) {
4530		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4531		lun_malloced = 1;
4532	} else {
4533		lun_malloced = 0;
4534		lun = ctl_lun;
4535	}
4536
4537	memset(lun, 0, sizeof(*lun));
4538	if (lun_malloced)
4539		lun->flags = CTL_LUN_MALLOCED;
4540
4541	/* Generate LUN ID. */
4542	devidlen = max(CTL_DEVID_MIN_LEN,
4543	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4544	idlen1 = sizeof(*t10id) + devidlen;
4545	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4546	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4547	if (scsiname != NULL) {
4548		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4549		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4550	}
4551	eui = ctl_get_opt(&be_lun->options, "eui");
4552	if (eui != NULL) {
4553		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4554	}
4555	naa = ctl_get_opt(&be_lun->options, "naa");
4556	if (naa != NULL) {
4557		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4558	}
4559	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4560	    M_CTL, M_WAITOK | M_ZERO);
4561	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4562	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4563	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4564	desc->length = idlen1;
4565	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4566	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4567	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4568		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4569	} else {
4570		strncpy(t10id->vendor, vendor,
4571		    min(sizeof(t10id->vendor), strlen(vendor)));
4572	}
4573	strncpy((char *)t10id->vendor_spec_id,
4574	    (char *)be_lun->device_id, devidlen);
4575	if (scsiname != NULL) {
4576		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4577		    desc->length);
4578		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4579		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4580		    SVPD_ID_TYPE_SCSI_NAME;
4581		desc->length = idlen2;
4582		strlcpy(desc->identifier, scsiname, idlen2);
4583	}
4584	if (eui != NULL) {
4585		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4586		    desc->length);
4587		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4588		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4589		    SVPD_ID_TYPE_EUI64;
4590		desc->length = hex2bin(eui, desc->identifier, 16);
4591		desc->length = desc->length > 12 ? 16 :
4592		    (desc->length > 8 ? 12 : 8);
4593		len -= 16 - desc->length;
4594	}
4595	if (naa != NULL) {
4596		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4597		    desc->length);
4598		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4599		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4600		    SVPD_ID_TYPE_NAA;
4601		desc->length = hex2bin(naa, desc->identifier, 16);
4602		desc->length = desc->length > 8 ? 16 : 8;
4603		len -= 16 - desc->length;
4604	}
4605	lun->lun_devid->len = len;
4606
4607	mtx_lock(&ctl_softc->ctl_lock);
4608	/*
4609	 * See if the caller requested a particular LUN number.  If so, see
4610	 * if it is available.  Otherwise, allocate the first available LUN.
4611	 */
4612	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4613		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4614		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4615			mtx_unlock(&ctl_softc->ctl_lock);
4616			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4617				printf("ctl: requested LUN ID %d is higher "
4618				       "than CTL_MAX_LUNS - 1 (%d)\n",
4619				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4620			} else {
4621				/*
4622				 * XXX KDM return an error, or just assign
4623				 * another LUN ID in this case??
4624				 */
4625				printf("ctl: requested LUN ID %d is already "
4626				       "in use\n", be_lun->req_lun_id);
4627			}
4628			if (lun->flags & CTL_LUN_MALLOCED)
4629				free(lun, M_CTL);
4630			be_lun->lun_config_status(be_lun->be_lun,
4631						  CTL_LUN_CONFIG_FAILURE);
4632			return (ENOSPC);
4633		}
4634		lun_number = be_lun->req_lun_id;
4635	} else {
4636		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4637		if (lun_number == -1) {
4638			mtx_unlock(&ctl_softc->ctl_lock);
4639			printf("ctl: can't allocate LUN, out of LUNs\n");
4640			if (lun->flags & CTL_LUN_MALLOCED)
4641				free(lun, M_CTL);
4642			be_lun->lun_config_status(be_lun->be_lun,
4643						  CTL_LUN_CONFIG_FAILURE);
4644			return (ENOSPC);
4645		}
4646	}
4647	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4648
4649	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4650	lun->lun = lun_number;
4651	lun->be_lun = be_lun;
4652	/*
4653	 * The processor LUN is always enabled.  Disk LUNs come on line
4654	 * disabled, and must be enabled by the backend.
4655	 */
4656	lun->flags |= CTL_LUN_DISABLED;
4657	lun->backend = be_lun->be;
4658	be_lun->ctl_lun = lun;
4659	be_lun->lun_id = lun_number;
4660	atomic_add_int(&be_lun->be->num_luns, 1);
4661	if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4662		lun->flags |= CTL_LUN_OFFLINE;
4663
4664	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4665		lun->flags |= CTL_LUN_STOPPED;
4666
4667	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4668		lun->flags |= CTL_LUN_INOPERABLE;
4669
4670	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4671		lun->flags |= CTL_LUN_PRIMARY_SC;
4672
4673	lun->ctl_softc = ctl_softc;
4674#ifdef CTL_TIME_IO
4675	lun->last_busy = getsbinuptime();
4676#endif
4677	TAILQ_INIT(&lun->ooa_queue);
4678	TAILQ_INIT(&lun->blocked_queue);
4679	STAILQ_INIT(&lun->error_list);
4680	ctl_tpc_lun_init(lun);
4681
4682	/*
4683	 * Initialize the mode and log page index.
4684	 */
4685	ctl_init_page_index(lun);
4686	ctl_init_log_page_index(lun);
4687
4688	/*
4689	 * Now, before we insert this lun on the lun list, set the lun
4690	 * inventory changed UA for all other luns.
4691	 */
4692	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4693		mtx_lock(&nlun->lun_lock);
4694		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4695		mtx_unlock(&nlun->lun_lock);
4696	}
4697
4698	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4699
4700	ctl_softc->ctl_luns[lun_number] = lun;
4701
4702	ctl_softc->num_luns++;
4703
4704	/* Setup statistics gathering */
4705	lun->stats.device_type = be_lun->lun_type;
4706	lun->stats.lun_number = lun_number;
4707	if (lun->stats.device_type == T_DIRECT)
4708		lun->stats.blocksize = be_lun->blocksize;
4709	else
4710		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4711	for (i = 0;i < CTL_MAX_PORTS;i++)
4712		lun->stats.ports[i].targ_port = i;
4713
4714	mtx_unlock(&ctl_softc->ctl_lock);
4715
4716	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4717	return (0);
4718}
4719
4720/*
4721 * Delete a LUN.
4722 * Assumptions:
4723 * - LUN has already been marked invalid and any pending I/O has been taken
4724 *   care of.
4725 */
4726static int
4727ctl_free_lun(struct ctl_lun *lun)
4728{
4729	struct ctl_softc *softc;
4730	struct ctl_lun *nlun;
4731	int i;
4732
4733	softc = lun->ctl_softc;
4734
4735	mtx_assert(&softc->ctl_lock, MA_OWNED);
4736
4737	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4738
4739	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4740
4741	softc->ctl_luns[lun->lun] = NULL;
4742
4743	if (!TAILQ_EMPTY(&lun->ooa_queue))
4744		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4745
4746	softc->num_luns--;
4747
4748	/*
4749	 * Tell the backend to free resources, if this LUN has a backend.
4750	 */
4751	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4752	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4753
4754	ctl_tpc_lun_shutdown(lun);
4755	mtx_destroy(&lun->lun_lock);
4756	free(lun->lun_devid, M_CTL);
4757	for (i = 0; i < CTL_MAX_PORTS; i++)
4758		free(lun->pending_ua[i], M_CTL);
4759	for (i = 0; i < CTL_MAX_PORTS; i++)
4760		free(lun->pr_keys[i], M_CTL);
4761	free(lun->write_buffer, M_CTL);
4762	if (lun->flags & CTL_LUN_MALLOCED)
4763		free(lun, M_CTL);
4764
4765	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4766		mtx_lock(&nlun->lun_lock);
4767		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4768		mtx_unlock(&nlun->lun_lock);
4769	}
4770
4771	return (0);
4772}
4773
4774static void
4775ctl_create_lun(struct ctl_be_lun *be_lun)
4776{
4777
4778	/*
4779	 * ctl_alloc_lun() should handle all potential failure cases.
4780	 */
4781	ctl_alloc_lun(control_softc, NULL, be_lun);
4782}
4783
4784int
4785ctl_add_lun(struct ctl_be_lun *be_lun)
4786{
4787	struct ctl_softc *softc = control_softc;
4788
4789	mtx_lock(&softc->ctl_lock);
4790	STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4791	mtx_unlock(&softc->ctl_lock);
4792	wakeup(&softc->pending_lun_queue);
4793
4794	return (0);
4795}
4796
4797int
4798ctl_enable_lun(struct ctl_be_lun *be_lun)
4799{
4800	struct ctl_softc *softc;
4801	struct ctl_port *port, *nport;
4802	struct ctl_lun *lun;
4803	int retval;
4804
4805	lun = (struct ctl_lun *)be_lun->ctl_lun;
4806	softc = lun->ctl_softc;
4807
4808	mtx_lock(&softc->ctl_lock);
4809	mtx_lock(&lun->lun_lock);
4810	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4811		/*
4812		 * eh?  Why did we get called if the LUN is already
4813		 * enabled?
4814		 */
4815		mtx_unlock(&lun->lun_lock);
4816		mtx_unlock(&softc->ctl_lock);
4817		return (0);
4818	}
4819	lun->flags &= ~CTL_LUN_DISABLED;
4820	mtx_unlock(&lun->lun_lock);
4821
4822	for (port = STAILQ_FIRST(&softc->port_list); port != NULL; port = nport) {
4823		nport = STAILQ_NEXT(port, links);
4824		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4825		    port->lun_map != NULL || port->lun_enable == NULL)
4826			continue;
4827
4828		/*
4829		 * Drop the lock while we call the FETD's enable routine.
4830		 * This can lead to a callback into CTL (at least in the
4831		 * case of the internal initiator frontend.
4832		 */
4833		mtx_unlock(&softc->ctl_lock);
4834		retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4835		mtx_lock(&softc->ctl_lock);
4836		if (retval != 0) {
4837			printf("%s: FETD %s port %d returned error "
4838			       "%d for lun_enable on lun %jd\n",
4839			       __func__, port->port_name, port->targ_port,
4840			       retval, (intmax_t)lun->lun);
4841		}
4842	}
4843
4844	mtx_unlock(&softc->ctl_lock);
4845	ctl_isc_announce_lun(lun);
4846
4847	return (0);
4848}
4849
4850int
4851ctl_disable_lun(struct ctl_be_lun *be_lun)
4852{
4853	struct ctl_softc *softc;
4854	struct ctl_port *port;
4855	struct ctl_lun *lun;
4856	int retval;
4857
4858	lun = (struct ctl_lun *)be_lun->ctl_lun;
4859	softc = lun->ctl_softc;
4860
4861	mtx_lock(&softc->ctl_lock);
4862	mtx_lock(&lun->lun_lock);
4863	if (lun->flags & CTL_LUN_DISABLED) {
4864		mtx_unlock(&lun->lun_lock);
4865		mtx_unlock(&softc->ctl_lock);
4866		return (0);
4867	}
4868	lun->flags |= CTL_LUN_DISABLED;
4869	mtx_unlock(&lun->lun_lock);
4870
4871	STAILQ_FOREACH(port, &softc->port_list, links) {
4872		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4873		    port->lun_map != NULL || port->lun_disable == NULL)
4874			continue;
4875
4876		/*
4877		 * Drop the lock before we call the frontend's disable
4878		 * routine, to avoid lock order reversals.
4879		 *
4880		 * XXX KDM what happens if the frontend list changes while
4881		 * we're traversing it?  It's unlikely, but should be handled.
4882		 */
4883		mtx_unlock(&softc->ctl_lock);
4884		retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4885		mtx_lock(&softc->ctl_lock);
4886		if (retval != 0) {
4887			printf("%s: FETD %s port %d returned error "
4888			       "%d for lun_disable on lun %jd\n",
4889			       __func__, port->port_name, port->targ_port,
4890			       retval, (intmax_t)lun->lun);
4891		}
4892	}
4893
4894	mtx_unlock(&softc->ctl_lock);
4895	ctl_isc_announce_lun(lun);
4896
4897	return (0);
4898}
4899
4900int
4901ctl_start_lun(struct ctl_be_lun *be_lun)
4902{
4903	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4904
4905	mtx_lock(&lun->lun_lock);
4906	lun->flags &= ~CTL_LUN_STOPPED;
4907	mtx_unlock(&lun->lun_lock);
4908	return (0);
4909}
4910
4911int
4912ctl_stop_lun(struct ctl_be_lun *be_lun)
4913{
4914	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4915
4916	mtx_lock(&lun->lun_lock);
4917	lun->flags |= CTL_LUN_STOPPED;
4918	mtx_unlock(&lun->lun_lock);
4919	return (0);
4920}
4921
4922int
4923ctl_lun_offline(struct ctl_be_lun *be_lun)
4924{
4925	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4926
4927	mtx_lock(&lun->lun_lock);
4928	lun->flags |= CTL_LUN_OFFLINE;
4929	mtx_unlock(&lun->lun_lock);
4930	return (0);
4931}
4932
4933int
4934ctl_lun_online(struct ctl_be_lun *be_lun)
4935{
4936	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4937
4938	mtx_lock(&lun->lun_lock);
4939	lun->flags &= ~CTL_LUN_OFFLINE;
4940	mtx_unlock(&lun->lun_lock);
4941	return (0);
4942}
4943
4944int
4945ctl_lun_primary(struct ctl_be_lun *be_lun)
4946{
4947	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4948
4949	mtx_lock(&lun->lun_lock);
4950	lun->flags |= CTL_LUN_PRIMARY_SC;
4951	ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4952	mtx_unlock(&lun->lun_lock);
4953	ctl_isc_announce_lun(lun);
4954	return (0);
4955}
4956
4957int
4958ctl_lun_secondary(struct ctl_be_lun *be_lun)
4959{
4960	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4961
4962	mtx_lock(&lun->lun_lock);
4963	lun->flags &= ~CTL_LUN_PRIMARY_SC;
4964	ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4965	mtx_unlock(&lun->lun_lock);
4966	ctl_isc_announce_lun(lun);
4967	return (0);
4968}
4969
4970int
4971ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4972{
4973	struct ctl_softc *softc;
4974	struct ctl_lun *lun;
4975
4976	lun = (struct ctl_lun *)be_lun->ctl_lun;
4977	softc = lun->ctl_softc;
4978
4979	mtx_lock(&lun->lun_lock);
4980
4981	/*
4982	 * The LUN needs to be disabled before it can be marked invalid.
4983	 */
4984	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4985		mtx_unlock(&lun->lun_lock);
4986		return (-1);
4987	}
4988	/*
4989	 * Mark the LUN invalid.
4990	 */
4991	lun->flags |= CTL_LUN_INVALID;
4992
4993	/*
4994	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4995	 * If we have something in the OOA queue, we'll free it when the
4996	 * last I/O completes.
4997	 */
4998	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4999		mtx_unlock(&lun->lun_lock);
5000		mtx_lock(&softc->ctl_lock);
5001		ctl_free_lun(lun);
5002		mtx_unlock(&softc->ctl_lock);
5003	} else
5004		mtx_unlock(&lun->lun_lock);
5005
5006	return (0);
5007}
5008
5009int
5010ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5011{
5012	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5013
5014	mtx_lock(&lun->lun_lock);
5015	lun->flags |= CTL_LUN_INOPERABLE;
5016	mtx_unlock(&lun->lun_lock);
5017	return (0);
5018}
5019
5020int
5021ctl_lun_operable(struct ctl_be_lun *be_lun)
5022{
5023	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5024
5025	mtx_lock(&lun->lun_lock);
5026	lun->flags &= ~CTL_LUN_INOPERABLE;
5027	mtx_unlock(&lun->lun_lock);
5028	return (0);
5029}
5030
5031void
5032ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5033{
5034	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5035	union ctl_ha_msg msg;
5036
5037	mtx_lock(&lun->lun_lock);
5038	ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGED);
5039	mtx_unlock(&lun->lun_lock);
5040	if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5041		/* Send msg to other side. */
5042		bzero(&msg.ua, sizeof(msg.ua));
5043		msg.hdr.msg_type = CTL_MSG_UA;
5044		msg.hdr.nexus.initid = -1;
5045		msg.hdr.nexus.targ_port = -1;
5046		msg.hdr.nexus.targ_lun = lun->lun;
5047		msg.hdr.nexus.targ_mapped_lun = lun->lun;
5048		msg.ua.ua_all = 1;
5049		msg.ua.ua_set = 1;
5050		msg.ua.ua_type = CTL_UA_CAPACITY_CHANGED;
5051		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5052		    M_WAITOK);
5053	}
5054}
5055
5056/*
5057 * Backend "memory move is complete" callback for requests that never
5058 * make it down to say RAIDCore's configuration code.
5059 */
5060int
5061ctl_config_move_done(union ctl_io *io)
5062{
5063	int retval;
5064
5065	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5066	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5067	    ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5068
5069	if ((io->io_hdr.port_status != 0) &&
5070	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5071	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5072		/*
5073		 * For hardware error sense keys, the sense key
5074		 * specific value is defined to be a retry count,
5075		 * but we use it to pass back an internal FETD
5076		 * error code.  XXX KDM  Hopefully the FETD is only
5077		 * using 16 bits for an error code, since that's
5078		 * all the space we have in the sks field.
5079		 */
5080		ctl_set_internal_failure(&io->scsiio,
5081					 /*sks_valid*/ 1,
5082					 /*retry_count*/
5083					 io->io_hdr.port_status);
5084	}
5085
5086	if (ctl_debug & CTL_DEBUG_CDB_DATA)
5087		ctl_data_print(io);
5088	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5089	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5090	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5091	    ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5092		/*
5093		 * XXX KDM just assuming a single pointer here, and not a
5094		 * S/G list.  If we start using S/G lists for config data,
5095		 * we'll need to know how to clean them up here as well.
5096		 */
5097		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5098			free(io->scsiio.kern_data_ptr, M_CTL);
5099		ctl_done(io);
5100		retval = CTL_RETVAL_COMPLETE;
5101	} else {
5102		/*
5103		 * XXX KDM now we need to continue data movement.  Some
5104		 * options:
5105		 * - call ctl_scsiio() again?  We don't do this for data
5106		 *   writes, because for those at least we know ahead of
5107		 *   time where the write will go and how long it is.  For
5108		 *   config writes, though, that information is largely
5109		 *   contained within the write itself, thus we need to
5110		 *   parse out the data again.
5111		 *
5112		 * - Call some other function once the data is in?
5113		 */
5114
5115		/*
5116		 * XXX KDM call ctl_scsiio() again for now, and check flag
5117		 * bits to see whether we're allocated or not.
5118		 */
5119		retval = ctl_scsiio(&io->scsiio);
5120	}
5121	return (retval);
5122}
5123
5124/*
5125 * This gets called by a backend driver when it is done with a
5126 * data_submit method.
5127 */
5128void
5129ctl_data_submit_done(union ctl_io *io)
5130{
5131	/*
5132	 * If the IO_CONT flag is set, we need to call the supplied
5133	 * function to continue processing the I/O, instead of completing
5134	 * the I/O just yet.
5135	 *
5136	 * If there is an error, though, we don't want to keep processing.
5137	 * Instead, just send status back to the initiator.
5138	 */
5139	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5140	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5141	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5142	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5143		io->scsiio.io_cont(io);
5144		return;
5145	}
5146	ctl_done(io);
5147}
5148
5149/*
5150 * This gets called by a backend driver when it is done with a
5151 * configuration write.
5152 */
5153void
5154ctl_config_write_done(union ctl_io *io)
5155{
5156	uint8_t *buf;
5157
5158	/*
5159	 * If the IO_CONT flag is set, we need to call the supplied
5160	 * function to continue processing the I/O, instead of completing
5161	 * the I/O just yet.
5162	 *
5163	 * If there is an error, though, we don't want to keep processing.
5164	 * Instead, just send status back to the initiator.
5165	 */
5166	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5167	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5168	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5169	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5170		io->scsiio.io_cont(io);
5171		return;
5172	}
5173	/*
5174	 * Since a configuration write can be done for commands that actually
5175	 * have data allocated, like write buffer, and commands that have
5176	 * no data, like start/stop unit, we need to check here.
5177	 */
5178	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5179		buf = io->scsiio.kern_data_ptr;
5180	else
5181		buf = NULL;
5182	ctl_done(io);
5183	if (buf)
5184		free(buf, M_CTL);
5185}
5186
5187void
5188ctl_config_read_done(union ctl_io *io)
5189{
5190	uint8_t *buf;
5191
5192	/*
5193	 * If there is some error -- we are done, skip data transfer.
5194	 */
5195	if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5196	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5197	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5198		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5199			buf = io->scsiio.kern_data_ptr;
5200		else
5201			buf = NULL;
5202		ctl_done(io);
5203		if (buf)
5204			free(buf, M_CTL);
5205		return;
5206	}
5207
5208	/*
5209	 * If the IO_CONT flag is set, we need to call the supplied
5210	 * function to continue processing the I/O, instead of completing
5211	 * the I/O just yet.
5212	 */
5213	if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5214		io->scsiio.io_cont(io);
5215		return;
5216	}
5217
5218	ctl_datamove(io);
5219}
5220
5221/*
5222 * SCSI release command.
5223 */
5224int
5225ctl_scsi_release(struct ctl_scsiio *ctsio)
5226{
5227	struct ctl_lun *lun;
5228	uint32_t residx;
5229
5230	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5231
5232	residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5233	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5234
5235	/*
5236	 * XXX KDM right now, we only support LUN reservation.  We don't
5237	 * support 3rd party reservations, or extent reservations, which
5238	 * might actually need the parameter list.  If we've gotten this
5239	 * far, we've got a LUN reservation.  Anything else got kicked out
5240	 * above.  So, according to SPC, ignore the length.
5241	 */
5242
5243	mtx_lock(&lun->lun_lock);
5244
5245	/*
5246	 * According to SPC, it is not an error for an intiator to attempt
5247	 * to release a reservation on a LUN that isn't reserved, or that
5248	 * is reserved by another initiator.  The reservation can only be
5249	 * released, though, by the initiator who made it or by one of
5250	 * several reset type events.
5251	 */
5252	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5253			lun->flags &= ~CTL_LUN_RESERVED;
5254
5255	mtx_unlock(&lun->lun_lock);
5256
5257	ctl_set_success(ctsio);
5258	ctl_done((union ctl_io *)ctsio);
5259	return (CTL_RETVAL_COMPLETE);
5260}
5261
5262int
5263ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5264{
5265	struct ctl_lun *lun;
5266	uint32_t residx;
5267
5268	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5269
5270	residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5271	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5272
5273	/*
5274	 * XXX KDM right now, we only support LUN reservation.  We don't
5275	 * support 3rd party reservations, or extent reservations, which
5276	 * might actually need the parameter list.  If we've gotten this
5277	 * far, we've got a LUN reservation.  Anything else got kicked out
5278	 * above.  So, according to SPC, ignore the length.
5279	 */
5280
5281	mtx_lock(&lun->lun_lock);
5282	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5283		ctl_set_reservation_conflict(ctsio);
5284		goto bailout;
5285	}
5286	lun->flags |= CTL_LUN_RESERVED;
5287	lun->res_idx = residx;
5288	ctl_set_success(ctsio);
5289
5290bailout:
5291	mtx_unlock(&lun->lun_lock);
5292	ctl_done((union ctl_io *)ctsio);
5293	return (CTL_RETVAL_COMPLETE);
5294}
5295
5296int
5297ctl_start_stop(struct ctl_scsiio *ctsio)
5298{
5299	struct scsi_start_stop_unit *cdb;
5300	struct ctl_lun *lun;
5301	int retval;
5302
5303	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5304
5305	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5306	retval = 0;
5307
5308	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5309
5310	/*
5311	 * XXX KDM
5312	 * We don't support the immediate bit on a stop unit.  In order to
5313	 * do that, we would need to code up a way to know that a stop is
5314	 * pending, and hold off any new commands until it completes, one
5315	 * way or another.  Then we could accept or reject those commands
5316	 * depending on its status.  We would almost need to do the reverse
5317	 * of what we do below for an immediate start -- return the copy of
5318	 * the ctl_io to the FETD with status to send to the host (and to
5319	 * free the copy!) and then free the original I/O once the stop
5320	 * actually completes.  That way, the OOA queue mechanism can work
5321	 * to block commands that shouldn't proceed.  Another alternative
5322	 * would be to put the copy in the queue in place of the original,
5323	 * and return the original back to the caller.  That could be
5324	 * slightly safer..
5325	 */
5326	if ((cdb->byte2 & SSS_IMMED)
5327	 && ((cdb->how & SSS_START) == 0)) {
5328		ctl_set_invalid_field(ctsio,
5329				      /*sks_valid*/ 1,
5330				      /*command*/ 1,
5331				      /*field*/ 1,
5332				      /*bit_valid*/ 1,
5333				      /*bit*/ 0);
5334		ctl_done((union ctl_io *)ctsio);
5335		return (CTL_RETVAL_COMPLETE);
5336	}
5337
5338	if ((lun->flags & CTL_LUN_PR_RESERVED)
5339	 && ((cdb->how & SSS_START)==0)) {
5340		uint32_t residx;
5341
5342		residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5343		if (ctl_get_prkey(lun, residx) == 0
5344		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5345
5346			ctl_set_reservation_conflict(ctsio);
5347			ctl_done((union ctl_io *)ctsio);
5348			return (CTL_RETVAL_COMPLETE);
5349		}
5350	}
5351
5352	/*
5353	 * If there is no backend on this device, we can't start or stop
5354	 * it.  In theory we shouldn't get any start/stop commands in the
5355	 * first place at this level if the LUN doesn't have a backend.
5356	 * That should get stopped by the command decode code.
5357	 */
5358	if (lun->backend == NULL) {
5359		ctl_set_invalid_opcode(ctsio);
5360		ctl_done((union ctl_io *)ctsio);
5361		return (CTL_RETVAL_COMPLETE);
5362	}
5363
5364	/*
5365	 * XXX KDM Copan-specific offline behavior.
5366	 * Figure out a reasonable way to port this?
5367	 */
5368#ifdef NEEDTOPORT
5369	mtx_lock(&lun->lun_lock);
5370
5371	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5372	 && (lun->flags & CTL_LUN_OFFLINE)) {
5373		/*
5374		 * If the LUN is offline, and the on/offline bit isn't set,
5375		 * reject the start or stop.  Otherwise, let it through.
5376		 */
5377		mtx_unlock(&lun->lun_lock);
5378		ctl_set_lun_not_ready(ctsio);
5379		ctl_done((union ctl_io *)ctsio);
5380	} else {
5381		mtx_unlock(&lun->lun_lock);
5382#endif /* NEEDTOPORT */
5383		/*
5384		 * This could be a start or a stop when we're online,
5385		 * or a stop/offline or start/online.  A start or stop when
5386		 * we're offline is covered in the case above.
5387		 */
5388		/*
5389		 * In the non-immediate case, we send the request to
5390		 * the backend and return status to the user when
5391		 * it is done.
5392		 *
5393		 * In the immediate case, we allocate a new ctl_io
5394		 * to hold a copy of the request, and send that to
5395		 * the backend.  We then set good status on the
5396		 * user's request and return it immediately.
5397		 */
5398		if (cdb->byte2 & SSS_IMMED) {
5399			union ctl_io *new_io;
5400
5401			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5402			ctl_copy_io((union ctl_io *)ctsio, new_io);
5403			retval = lun->backend->config_write(new_io);
5404			ctl_set_success(ctsio);
5405			ctl_done((union ctl_io *)ctsio);
5406		} else {
5407			retval = lun->backend->config_write(
5408				(union ctl_io *)ctsio);
5409		}
5410#ifdef NEEDTOPORT
5411	}
5412#endif
5413	return (retval);
5414}
5415
5416/*
5417 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5418 * we don't really do anything with the LBA and length fields if the user
5419 * passes them in.  Instead we'll just flush out the cache for the entire
5420 * LUN.
5421 */
5422int
5423ctl_sync_cache(struct ctl_scsiio *ctsio)
5424{
5425	struct ctl_lun *lun;
5426	struct ctl_softc *softc;
5427	struct ctl_lba_len_flags *lbalen;
5428	uint64_t starting_lba;
5429	uint32_t block_count;
5430	int retval;
5431	uint8_t byte2;
5432
5433	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5434
5435	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5436	softc = lun->ctl_softc;
5437	retval = 0;
5438
5439	switch (ctsio->cdb[0]) {
5440	case SYNCHRONIZE_CACHE: {
5441		struct scsi_sync_cache *cdb;
5442		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5443
5444		starting_lba = scsi_4btoul(cdb->begin_lba);
5445		block_count = scsi_2btoul(cdb->lb_count);
5446		byte2 = cdb->byte2;
5447		break;
5448	}
5449	case SYNCHRONIZE_CACHE_16: {
5450		struct scsi_sync_cache_16 *cdb;
5451		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5452
5453		starting_lba = scsi_8btou64(cdb->begin_lba);
5454		block_count = scsi_4btoul(cdb->lb_count);
5455		byte2 = cdb->byte2;
5456		break;
5457	}
5458	default:
5459		ctl_set_invalid_opcode(ctsio);
5460		ctl_done((union ctl_io *)ctsio);
5461		goto bailout;
5462		break; /* NOTREACHED */
5463	}
5464
5465	/*
5466	 * We check the LBA and length, but don't do anything with them.
5467	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5468	 * get flushed.  This check will just help satisfy anyone who wants
5469	 * to see an error for an out of range LBA.
5470	 */
5471	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5472		ctl_set_lba_out_of_range(ctsio);
5473		ctl_done((union ctl_io *)ctsio);
5474		goto bailout;
5475	}
5476
5477	/*
5478	 * If this LUN has no backend, we can't flush the cache anyway.
5479	 */
5480	if (lun->backend == NULL) {
5481		ctl_set_invalid_opcode(ctsio);
5482		ctl_done((union ctl_io *)ctsio);
5483		goto bailout;
5484	}
5485
5486	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5487	lbalen->lba = starting_lba;
5488	lbalen->len = block_count;
5489	lbalen->flags = byte2;
5490
5491	/*
5492	 * Check to see whether we're configured to send the SYNCHRONIZE
5493	 * CACHE command directly to the back end.
5494	 */
5495	mtx_lock(&lun->lun_lock);
5496	if ((softc->flags & CTL_FLAG_REAL_SYNC)
5497	 && (++(lun->sync_count) >= lun->sync_interval)) {
5498		lun->sync_count = 0;
5499		mtx_unlock(&lun->lun_lock);
5500		retval = lun->backend->config_write((union ctl_io *)ctsio);
5501	} else {
5502		mtx_unlock(&lun->lun_lock);
5503		ctl_set_success(ctsio);
5504		ctl_done((union ctl_io *)ctsio);
5505	}
5506
5507bailout:
5508
5509	return (retval);
5510}
5511
5512int
5513ctl_format(struct ctl_scsiio *ctsio)
5514{
5515	struct scsi_format *cdb;
5516	struct ctl_lun *lun;
5517	int length, defect_list_len;
5518
5519	CTL_DEBUG_PRINT(("ctl_format\n"));
5520
5521	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5522
5523	cdb = (struct scsi_format *)ctsio->cdb;
5524
5525	length = 0;
5526	if (cdb->byte2 & SF_FMTDATA) {
5527		if (cdb->byte2 & SF_LONGLIST)
5528			length = sizeof(struct scsi_format_header_long);
5529		else
5530			length = sizeof(struct scsi_format_header_short);
5531	}
5532
5533	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5534	 && (length > 0)) {
5535		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5536		ctsio->kern_data_len = length;
5537		ctsio->kern_total_len = length;
5538		ctsio->kern_data_resid = 0;
5539		ctsio->kern_rel_offset = 0;
5540		ctsio->kern_sg_entries = 0;
5541		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5542		ctsio->be_move_done = ctl_config_move_done;
5543		ctl_datamove((union ctl_io *)ctsio);
5544
5545		return (CTL_RETVAL_COMPLETE);
5546	}
5547
5548	defect_list_len = 0;
5549
5550	if (cdb->byte2 & SF_FMTDATA) {
5551		if (cdb->byte2 & SF_LONGLIST) {
5552			struct scsi_format_header_long *header;
5553
5554			header = (struct scsi_format_header_long *)
5555				ctsio->kern_data_ptr;
5556
5557			defect_list_len = scsi_4btoul(header->defect_list_len);
5558			if (defect_list_len != 0) {
5559				ctl_set_invalid_field(ctsio,
5560						      /*sks_valid*/ 1,
5561						      /*command*/ 0,
5562						      /*field*/ 2,
5563						      /*bit_valid*/ 0,
5564						      /*bit*/ 0);
5565				goto bailout;
5566			}
5567		} else {
5568			struct scsi_format_header_short *header;
5569
5570			header = (struct scsi_format_header_short *)
5571				ctsio->kern_data_ptr;
5572
5573			defect_list_len = scsi_2btoul(header->defect_list_len);
5574			if (defect_list_len != 0) {
5575				ctl_set_invalid_field(ctsio,
5576						      /*sks_valid*/ 1,
5577						      /*command*/ 0,
5578						      /*field*/ 2,
5579						      /*bit_valid*/ 0,
5580						      /*bit*/ 0);
5581				goto bailout;
5582			}
5583		}
5584	}
5585
5586	/*
5587	 * The format command will clear out the "Medium format corrupted"
5588	 * status if set by the configuration code.  That status is really
5589	 * just a way to notify the host that we have lost the media, and
5590	 * get them to issue a command that will basically make them think
5591	 * they're blowing away the media.
5592	 */
5593	mtx_lock(&lun->lun_lock);
5594	lun->flags &= ~CTL_LUN_INOPERABLE;
5595	mtx_unlock(&lun->lun_lock);
5596
5597	ctl_set_success(ctsio);
5598bailout:
5599
5600	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5601		free(ctsio->kern_data_ptr, M_CTL);
5602		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5603	}
5604
5605	ctl_done((union ctl_io *)ctsio);
5606	return (CTL_RETVAL_COMPLETE);
5607}
5608
5609int
5610ctl_read_buffer(struct ctl_scsiio *ctsio)
5611{
5612	struct ctl_lun *lun;
5613	uint64_t buffer_offset;
5614	uint32_t len;
5615	uint8_t byte2;
5616	static uint8_t descr[4];
5617	static uint8_t echo_descr[4] = { 0 };
5618
5619	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5620	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5621	switch (ctsio->cdb[0]) {
5622	case READ_BUFFER: {
5623		struct scsi_read_buffer *cdb;
5624
5625		cdb = (struct scsi_read_buffer *)ctsio->cdb;
5626		buffer_offset = scsi_3btoul(cdb->offset);
5627		len = scsi_3btoul(cdb->length);
5628		byte2 = cdb->byte2;
5629		break;
5630	}
5631	case READ_BUFFER_16: {
5632		struct scsi_read_buffer_16 *cdb;
5633
5634		cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5635		buffer_offset = scsi_8btou64(cdb->offset);
5636		len = scsi_4btoul(cdb->length);
5637		byte2 = cdb->byte2;
5638		break;
5639	}
5640	default: /* This shouldn't happen. */
5641		ctl_set_invalid_opcode(ctsio);
5642		ctl_done((union ctl_io *)ctsio);
5643		return (CTL_RETVAL_COMPLETE);
5644	}
5645
5646	if ((byte2 & RWB_MODE) != RWB_MODE_DATA &&
5647	    (byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5648	    (byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5649		ctl_set_invalid_field(ctsio,
5650				      /*sks_valid*/ 1,
5651				      /*command*/ 1,
5652				      /*field*/ 1,
5653				      /*bit_valid*/ 1,
5654				      /*bit*/ 4);
5655		ctl_done((union ctl_io *)ctsio);
5656		return (CTL_RETVAL_COMPLETE);
5657	}
5658
5659	if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5660	    buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5661		ctl_set_invalid_field(ctsio,
5662				      /*sks_valid*/ 1,
5663				      /*command*/ 1,
5664				      /*field*/ 6,
5665				      /*bit_valid*/ 0,
5666				      /*bit*/ 0);
5667		ctl_done((union ctl_io *)ctsio);
5668		return (CTL_RETVAL_COMPLETE);
5669	}
5670
5671	if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5672		descr[0] = 0;
5673		scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5674		ctsio->kern_data_ptr = descr;
5675		len = min(len, sizeof(descr));
5676	} else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5677		ctsio->kern_data_ptr = echo_descr;
5678		len = min(len, sizeof(echo_descr));
5679	} else {
5680		if (lun->write_buffer == NULL) {
5681			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5682			    M_CTL, M_WAITOK);
5683		}
5684		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5685	}
5686	ctsio->kern_data_len = len;
5687	ctsio->kern_total_len = len;
5688	ctsio->kern_data_resid = 0;
5689	ctsio->kern_rel_offset = 0;
5690	ctsio->kern_sg_entries = 0;
5691	ctl_set_success(ctsio);
5692	ctsio->be_move_done = ctl_config_move_done;
5693	ctl_datamove((union ctl_io *)ctsio);
5694	return (CTL_RETVAL_COMPLETE);
5695}
5696
5697int
5698ctl_write_buffer(struct ctl_scsiio *ctsio)
5699{
5700	struct scsi_write_buffer *cdb;
5701	struct ctl_lun *lun;
5702	int buffer_offset, len;
5703
5704	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5705
5706	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5707	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5708
5709	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5710		ctl_set_invalid_field(ctsio,
5711				      /*sks_valid*/ 1,
5712				      /*command*/ 1,
5713				      /*field*/ 1,
5714				      /*bit_valid*/ 1,
5715				      /*bit*/ 4);
5716		ctl_done((union ctl_io *)ctsio);
5717		return (CTL_RETVAL_COMPLETE);
5718	}
5719
5720	len = scsi_3btoul(cdb->length);
5721	buffer_offset = scsi_3btoul(cdb->offset);
5722
5723	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5724		ctl_set_invalid_field(ctsio,
5725				      /*sks_valid*/ 1,
5726				      /*command*/ 1,
5727				      /*field*/ 6,
5728				      /*bit_valid*/ 0,
5729				      /*bit*/ 0);
5730		ctl_done((union ctl_io *)ctsio);
5731		return (CTL_RETVAL_COMPLETE);
5732	}
5733
5734	/*
5735	 * If we've got a kernel request that hasn't been malloced yet,
5736	 * malloc it and tell the caller the data buffer is here.
5737	 */
5738	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5739		if (lun->write_buffer == NULL) {
5740			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5741			    M_CTL, M_WAITOK);
5742		}
5743		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5744		ctsio->kern_data_len = len;
5745		ctsio->kern_total_len = len;
5746		ctsio->kern_data_resid = 0;
5747		ctsio->kern_rel_offset = 0;
5748		ctsio->kern_sg_entries = 0;
5749		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5750		ctsio->be_move_done = ctl_config_move_done;
5751		ctl_datamove((union ctl_io *)ctsio);
5752
5753		return (CTL_RETVAL_COMPLETE);
5754	}
5755
5756	ctl_set_success(ctsio);
5757	ctl_done((union ctl_io *)ctsio);
5758	return (CTL_RETVAL_COMPLETE);
5759}
5760
5761int
5762ctl_write_same(struct ctl_scsiio *ctsio)
5763{
5764	struct ctl_lun *lun;
5765	struct ctl_lba_len_flags *lbalen;
5766	uint64_t lba;
5767	uint32_t num_blocks;
5768	int len, retval;
5769	uint8_t byte2;
5770
5771	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5772
5773	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5774
5775	switch (ctsio->cdb[0]) {
5776	case WRITE_SAME_10: {
5777		struct scsi_write_same_10 *cdb;
5778
5779		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5780
5781		lba = scsi_4btoul(cdb->addr);
5782		num_blocks = scsi_2btoul(cdb->length);
5783		byte2 = cdb->byte2;
5784		break;
5785	}
5786	case WRITE_SAME_16: {
5787		struct scsi_write_same_16 *cdb;
5788
5789		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5790
5791		lba = scsi_8btou64(cdb->addr);
5792		num_blocks = scsi_4btoul(cdb->length);
5793		byte2 = cdb->byte2;
5794		break;
5795	}
5796	default:
5797		/*
5798		 * We got a command we don't support.  This shouldn't
5799		 * happen, commands should be filtered out above us.
5800		 */
5801		ctl_set_invalid_opcode(ctsio);
5802		ctl_done((union ctl_io *)ctsio);
5803
5804		return (CTL_RETVAL_COMPLETE);
5805		break; /* NOTREACHED */
5806	}
5807
5808	/* ANCHOR flag can be used only together with UNMAP */
5809	if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5810		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5811		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5812		ctl_done((union ctl_io *)ctsio);
5813		return (CTL_RETVAL_COMPLETE);
5814	}
5815
5816	/*
5817	 * The first check is to make sure we're in bounds, the second
5818	 * check is to catch wrap-around problems.  If the lba + num blocks
5819	 * is less than the lba, then we've wrapped around and the block
5820	 * range is invalid anyway.
5821	 */
5822	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5823	 || ((lba + num_blocks) < lba)) {
5824		ctl_set_lba_out_of_range(ctsio);
5825		ctl_done((union ctl_io *)ctsio);
5826		return (CTL_RETVAL_COMPLETE);
5827	}
5828
5829	/* Zero number of blocks means "to the last logical block" */
5830	if (num_blocks == 0) {
5831		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5832			ctl_set_invalid_field(ctsio,
5833					      /*sks_valid*/ 0,
5834					      /*command*/ 1,
5835					      /*field*/ 0,
5836					      /*bit_valid*/ 0,
5837					      /*bit*/ 0);
5838			ctl_done((union ctl_io *)ctsio);
5839			return (CTL_RETVAL_COMPLETE);
5840		}
5841		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5842	}
5843
5844	len = lun->be_lun->blocksize;
5845
5846	/*
5847	 * If we've got a kernel request that hasn't been malloced yet,
5848	 * malloc it and tell the caller the data buffer is here.
5849	 */
5850	if ((byte2 & SWS_NDOB) == 0 &&
5851	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5852		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5853		ctsio->kern_data_len = len;
5854		ctsio->kern_total_len = len;
5855		ctsio->kern_data_resid = 0;
5856		ctsio->kern_rel_offset = 0;
5857		ctsio->kern_sg_entries = 0;
5858		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5859		ctsio->be_move_done = ctl_config_move_done;
5860		ctl_datamove((union ctl_io *)ctsio);
5861
5862		return (CTL_RETVAL_COMPLETE);
5863	}
5864
5865	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5866	lbalen->lba = lba;
5867	lbalen->len = num_blocks;
5868	lbalen->flags = byte2;
5869	retval = lun->backend->config_write((union ctl_io *)ctsio);
5870
5871	return (retval);
5872}
5873
5874int
5875ctl_unmap(struct ctl_scsiio *ctsio)
5876{
5877	struct ctl_lun *lun;
5878	struct scsi_unmap *cdb;
5879	struct ctl_ptr_len_flags *ptrlen;
5880	struct scsi_unmap_header *hdr;
5881	struct scsi_unmap_desc *buf, *end, *endnz, *range;
5882	uint64_t lba;
5883	uint32_t num_blocks;
5884	int len, retval;
5885	uint8_t byte2;
5886
5887	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5888
5889	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5890	cdb = (struct scsi_unmap *)ctsio->cdb;
5891
5892	len = scsi_2btoul(cdb->length);
5893	byte2 = cdb->byte2;
5894
5895	/*
5896	 * If we've got a kernel request that hasn't been malloced yet,
5897	 * malloc it and tell the caller the data buffer is here.
5898	 */
5899	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5900		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5901		ctsio->kern_data_len = len;
5902		ctsio->kern_total_len = len;
5903		ctsio->kern_data_resid = 0;
5904		ctsio->kern_rel_offset = 0;
5905		ctsio->kern_sg_entries = 0;
5906		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5907		ctsio->be_move_done = ctl_config_move_done;
5908		ctl_datamove((union ctl_io *)ctsio);
5909
5910		return (CTL_RETVAL_COMPLETE);
5911	}
5912
5913	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5914	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5915	if (len < sizeof (*hdr) ||
5916	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5917	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5918	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5919		ctl_set_invalid_field(ctsio,
5920				      /*sks_valid*/ 0,
5921				      /*command*/ 0,
5922				      /*field*/ 0,
5923				      /*bit_valid*/ 0,
5924				      /*bit*/ 0);
5925		goto done;
5926	}
5927	len = scsi_2btoul(hdr->desc_length);
5928	buf = (struct scsi_unmap_desc *)(hdr + 1);
5929	end = buf + len / sizeof(*buf);
5930
5931	endnz = buf;
5932	for (range = buf; range < end; range++) {
5933		lba = scsi_8btou64(range->lba);
5934		num_blocks = scsi_4btoul(range->length);
5935		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5936		 || ((lba + num_blocks) < lba)) {
5937			ctl_set_lba_out_of_range(ctsio);
5938			ctl_done((union ctl_io *)ctsio);
5939			return (CTL_RETVAL_COMPLETE);
5940		}
5941		if (num_blocks != 0)
5942			endnz = range + 1;
5943	}
5944
5945	/*
5946	 * Block backend can not handle zero last range.
5947	 * Filter it out and return if there is nothing left.
5948	 */
5949	len = (uint8_t *)endnz - (uint8_t *)buf;
5950	if (len == 0) {
5951		ctl_set_success(ctsio);
5952		goto done;
5953	}
5954
5955	mtx_lock(&lun->lun_lock);
5956	ptrlen = (struct ctl_ptr_len_flags *)
5957	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5958	ptrlen->ptr = (void *)buf;
5959	ptrlen->len = len;
5960	ptrlen->flags = byte2;
5961	ctl_check_blocked(lun);
5962	mtx_unlock(&lun->lun_lock);
5963
5964	retval = lun->backend->config_write((union ctl_io *)ctsio);
5965	return (retval);
5966
5967done:
5968	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5969		free(ctsio->kern_data_ptr, M_CTL);
5970		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5971	}
5972	ctl_done((union ctl_io *)ctsio);
5973	return (CTL_RETVAL_COMPLETE);
5974}
5975
5976/*
5977 * Note that this function currently doesn't actually do anything inside
5978 * CTL to enforce things if the DQue bit is turned on.
5979 *
5980 * Also note that this function can't be used in the default case, because
5981 * the DQue bit isn't set in the changeable mask for the control mode page
5982 * anyway.  This is just here as an example for how to implement a page
5983 * handler, and a placeholder in case we want to allow the user to turn
5984 * tagged queueing on and off.
5985 *
5986 * The D_SENSE bit handling is functional, however, and will turn
5987 * descriptor sense on and off for a given LUN.
5988 */
5989int
5990ctl_control_page_handler(struct ctl_scsiio *ctsio,
5991			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5992{
5993	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5994	struct ctl_lun *lun;
5995	int set_ua;
5996	uint32_t initidx;
5997
5998	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5999	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6000	set_ua = 0;
6001
6002	user_cp = (struct scsi_control_page *)page_ptr;
6003	current_cp = (struct scsi_control_page *)
6004		(page_index->page_data + (page_index->page_len *
6005		CTL_PAGE_CURRENT));
6006	saved_cp = (struct scsi_control_page *)
6007		(page_index->page_data + (page_index->page_len *
6008		CTL_PAGE_SAVED));
6009
6010	mtx_lock(&lun->lun_lock);
6011	if (((current_cp->rlec & SCP_DSENSE) == 0)
6012	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6013		/*
6014		 * Descriptor sense is currently turned off and the user
6015		 * wants to turn it on.
6016		 */
6017		current_cp->rlec |= SCP_DSENSE;
6018		saved_cp->rlec |= SCP_DSENSE;
6019		lun->flags |= CTL_LUN_SENSE_DESC;
6020		set_ua = 1;
6021	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6022		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6023		/*
6024		 * Descriptor sense is currently turned on, and the user
6025		 * wants to turn it off.
6026		 */
6027		current_cp->rlec &= ~SCP_DSENSE;
6028		saved_cp->rlec &= ~SCP_DSENSE;
6029		lun->flags &= ~CTL_LUN_SENSE_DESC;
6030		set_ua = 1;
6031	}
6032	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6033	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6034		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6035		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6036		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6037		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6038		set_ua = 1;
6039	}
6040	if ((current_cp->eca_and_aen & SCP_SWP) !=
6041	    (user_cp->eca_and_aen & SCP_SWP)) {
6042		current_cp->eca_and_aen &= ~SCP_SWP;
6043		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6044		saved_cp->eca_and_aen &= ~SCP_SWP;
6045		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6046		set_ua = 1;
6047	}
6048	if (set_ua != 0)
6049		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6050	mtx_unlock(&lun->lun_lock);
6051	if (set_ua) {
6052		ctl_isc_announce_mode(lun,
6053		    ctl_get_initindex(&ctsio->io_hdr.nexus),
6054		    page_index->page_code, page_index->subpage);
6055	}
6056	return (0);
6057}
6058
6059int
6060ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6061		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6062{
6063	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6064	struct ctl_lun *lun;
6065	int set_ua;
6066	uint32_t initidx;
6067
6068	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6069	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6070	set_ua = 0;
6071
6072	user_cp = (struct scsi_caching_page *)page_ptr;
6073	current_cp = (struct scsi_caching_page *)
6074		(page_index->page_data + (page_index->page_len *
6075		CTL_PAGE_CURRENT));
6076	saved_cp = (struct scsi_caching_page *)
6077		(page_index->page_data + (page_index->page_len *
6078		CTL_PAGE_SAVED));
6079
6080	mtx_lock(&lun->lun_lock);
6081	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6082	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6083		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6084		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6085		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6086		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6087		set_ua = 1;
6088	}
6089	if (set_ua != 0)
6090		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6091	mtx_unlock(&lun->lun_lock);
6092	if (set_ua) {
6093		ctl_isc_announce_mode(lun,
6094		    ctl_get_initindex(&ctsio->io_hdr.nexus),
6095		    page_index->page_code, page_index->subpage);
6096	}
6097	return (0);
6098}
6099
6100int
6101ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6102				struct ctl_page_index *page_index,
6103				uint8_t *page_ptr)
6104{
6105	uint8_t *c;
6106	int i;
6107
6108	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6109	ctl_time_io_secs =
6110		(c[0] << 8) |
6111		(c[1] << 0) |
6112		0;
6113	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6114	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6115	printf("page data:");
6116	for (i=0; i<8; i++)
6117		printf(" %.2x",page_ptr[i]);
6118	printf("\n");
6119	return (0);
6120}
6121
6122int
6123ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6124			       struct ctl_page_index *page_index,
6125			       int pc)
6126{
6127	struct copan_debugconf_subpage *page;
6128
6129	page = (struct copan_debugconf_subpage *)page_index->page_data +
6130		(page_index->page_len * pc);
6131
6132	switch (pc) {
6133	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6134	case SMS_PAGE_CTRL_DEFAULT >> 6:
6135	case SMS_PAGE_CTRL_SAVED >> 6:
6136		/*
6137		 * We don't update the changable or default bits for this page.
6138		 */
6139		break;
6140	case SMS_PAGE_CTRL_CURRENT >> 6:
6141		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6142		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6143		break;
6144	default:
6145#ifdef NEEDTOPORT
6146		EPRINT(0, "Invalid PC %d!!", pc);
6147#endif /* NEEDTOPORT */
6148		break;
6149	}
6150	return (0);
6151}
6152
6153
6154static int
6155ctl_do_mode_select(union ctl_io *io)
6156{
6157	struct scsi_mode_page_header *page_header;
6158	struct ctl_page_index *page_index;
6159	struct ctl_scsiio *ctsio;
6160	int control_dev, page_len;
6161	int page_len_offset, page_len_size;
6162	union ctl_modepage_info *modepage_info;
6163	struct ctl_lun *lun;
6164	int *len_left, *len_used;
6165	int retval, i;
6166
6167	ctsio = &io->scsiio;
6168	page_index = NULL;
6169	page_len = 0;
6170	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6171
6172	if (lun->be_lun->lun_type != T_DIRECT)
6173		control_dev = 1;
6174	else
6175		control_dev = 0;
6176
6177	modepage_info = (union ctl_modepage_info *)
6178		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6179	len_left = &modepage_info->header.len_left;
6180	len_used = &modepage_info->header.len_used;
6181
6182do_next_page:
6183
6184	page_header = (struct scsi_mode_page_header *)
6185		(ctsio->kern_data_ptr + *len_used);
6186
6187	if (*len_left == 0) {
6188		free(ctsio->kern_data_ptr, M_CTL);
6189		ctl_set_success(ctsio);
6190		ctl_done((union ctl_io *)ctsio);
6191		return (CTL_RETVAL_COMPLETE);
6192	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6193
6194		free(ctsio->kern_data_ptr, M_CTL);
6195		ctl_set_param_len_error(ctsio);
6196		ctl_done((union ctl_io *)ctsio);
6197		return (CTL_RETVAL_COMPLETE);
6198
6199	} else if ((page_header->page_code & SMPH_SPF)
6200		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6201
6202		free(ctsio->kern_data_ptr, M_CTL);
6203		ctl_set_param_len_error(ctsio);
6204		ctl_done((union ctl_io *)ctsio);
6205		return (CTL_RETVAL_COMPLETE);
6206	}
6207
6208
6209	/*
6210	 * XXX KDM should we do something with the block descriptor?
6211	 */
6212	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6213
6214		if ((control_dev != 0)
6215		 && (lun->mode_pages.index[i].page_flags &
6216		     CTL_PAGE_FLAG_DISK_ONLY))
6217			continue;
6218
6219		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6220		    (page_header->page_code & SMPH_PC_MASK))
6221			continue;
6222
6223		/*
6224		 * If neither page has a subpage code, then we've got a
6225		 * match.
6226		 */
6227		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6228		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6229			page_index = &lun->mode_pages.index[i];
6230			page_len = page_header->page_length;
6231			break;
6232		}
6233
6234		/*
6235		 * If both pages have subpages, then the subpage numbers
6236		 * have to match.
6237		 */
6238		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6239		  && (page_header->page_code & SMPH_SPF)) {
6240			struct scsi_mode_page_header_sp *sph;
6241
6242			sph = (struct scsi_mode_page_header_sp *)page_header;
6243
6244			if (lun->mode_pages.index[i].subpage ==
6245			    sph->subpage) {
6246				page_index = &lun->mode_pages.index[i];
6247				page_len = scsi_2btoul(sph->page_length);
6248				break;
6249			}
6250		}
6251	}
6252
6253	/*
6254	 * If we couldn't find the page, or if we don't have a mode select
6255	 * handler for it, send back an error to the user.
6256	 */
6257	if ((page_index == NULL)
6258	 || (page_index->select_handler == NULL)) {
6259		ctl_set_invalid_field(ctsio,
6260				      /*sks_valid*/ 1,
6261				      /*command*/ 0,
6262				      /*field*/ *len_used,
6263				      /*bit_valid*/ 0,
6264				      /*bit*/ 0);
6265		free(ctsio->kern_data_ptr, M_CTL);
6266		ctl_done((union ctl_io *)ctsio);
6267		return (CTL_RETVAL_COMPLETE);
6268	}
6269
6270	if (page_index->page_code & SMPH_SPF) {
6271		page_len_offset = 2;
6272		page_len_size = 2;
6273	} else {
6274		page_len_size = 1;
6275		page_len_offset = 1;
6276	}
6277
6278	/*
6279	 * If the length the initiator gives us isn't the one we specify in
6280	 * the mode page header, or if they didn't specify enough data in
6281	 * the CDB to avoid truncating this page, kick out the request.
6282	 */
6283	if ((page_len != (page_index->page_len - page_len_offset -
6284			  page_len_size))
6285	 || (*len_left < page_index->page_len)) {
6286
6287
6288		ctl_set_invalid_field(ctsio,
6289				      /*sks_valid*/ 1,
6290				      /*command*/ 0,
6291				      /*field*/ *len_used + page_len_offset,
6292				      /*bit_valid*/ 0,
6293				      /*bit*/ 0);
6294		free(ctsio->kern_data_ptr, M_CTL);
6295		ctl_done((union ctl_io *)ctsio);
6296		return (CTL_RETVAL_COMPLETE);
6297	}
6298
6299	/*
6300	 * Run through the mode page, checking to make sure that the bits
6301	 * the user changed are actually legal for him to change.
6302	 */
6303	for (i = 0; i < page_index->page_len; i++) {
6304		uint8_t *user_byte, *change_mask, *current_byte;
6305		int bad_bit;
6306		int j;
6307
6308		user_byte = (uint8_t *)page_header + i;
6309		change_mask = page_index->page_data +
6310			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6311		current_byte = page_index->page_data +
6312			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6313
6314		/*
6315		 * Check to see whether the user set any bits in this byte
6316		 * that he is not allowed to set.
6317		 */
6318		if ((*user_byte & ~(*change_mask)) ==
6319		    (*current_byte & ~(*change_mask)))
6320			continue;
6321
6322		/*
6323		 * Go through bit by bit to determine which one is illegal.
6324		 */
6325		bad_bit = 0;
6326		for (j = 7; j >= 0; j--) {
6327			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6328			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6329				bad_bit = i;
6330				break;
6331			}
6332		}
6333		ctl_set_invalid_field(ctsio,
6334				      /*sks_valid*/ 1,
6335				      /*command*/ 0,
6336				      /*field*/ *len_used + i,
6337				      /*bit_valid*/ 1,
6338				      /*bit*/ bad_bit);
6339		free(ctsio->kern_data_ptr, M_CTL);
6340		ctl_done((union ctl_io *)ctsio);
6341		return (CTL_RETVAL_COMPLETE);
6342	}
6343
6344	/*
6345	 * Decrement these before we call the page handler, since we may
6346	 * end up getting called back one way or another before the handler
6347	 * returns to this context.
6348	 */
6349	*len_left -= page_index->page_len;
6350	*len_used += page_index->page_len;
6351
6352	retval = page_index->select_handler(ctsio, page_index,
6353					    (uint8_t *)page_header);
6354
6355	/*
6356	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6357	 * wait until this queued command completes to finish processing
6358	 * the mode page.  If it returns anything other than
6359	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6360	 * already set the sense information, freed the data pointer, and
6361	 * completed the io for us.
6362	 */
6363	if (retval != CTL_RETVAL_COMPLETE)
6364		goto bailout_no_done;
6365
6366	/*
6367	 * If the initiator sent us more than one page, parse the next one.
6368	 */
6369	if (*len_left > 0)
6370		goto do_next_page;
6371
6372	ctl_set_success(ctsio);
6373	free(ctsio->kern_data_ptr, M_CTL);
6374	ctl_done((union ctl_io *)ctsio);
6375
6376bailout_no_done:
6377
6378	return (CTL_RETVAL_COMPLETE);
6379
6380}
6381
6382int
6383ctl_mode_select(struct ctl_scsiio *ctsio)
6384{
6385	int param_len, pf, sp;
6386	int header_size, bd_len;
6387	union ctl_modepage_info *modepage_info;
6388
6389	switch (ctsio->cdb[0]) {
6390	case MODE_SELECT_6: {
6391		struct scsi_mode_select_6 *cdb;
6392
6393		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6394
6395		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6396		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6397		param_len = cdb->length;
6398		header_size = sizeof(struct scsi_mode_header_6);
6399		break;
6400	}
6401	case MODE_SELECT_10: {
6402		struct scsi_mode_select_10 *cdb;
6403
6404		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6405
6406		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6407		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6408		param_len = scsi_2btoul(cdb->length);
6409		header_size = sizeof(struct scsi_mode_header_10);
6410		break;
6411	}
6412	default:
6413		ctl_set_invalid_opcode(ctsio);
6414		ctl_done((union ctl_io *)ctsio);
6415		return (CTL_RETVAL_COMPLETE);
6416	}
6417
6418	/*
6419	 * From SPC-3:
6420	 * "A parameter list length of zero indicates that the Data-Out Buffer
6421	 * shall be empty. This condition shall not be considered as an error."
6422	 */
6423	if (param_len == 0) {
6424		ctl_set_success(ctsio);
6425		ctl_done((union ctl_io *)ctsio);
6426		return (CTL_RETVAL_COMPLETE);
6427	}
6428
6429	/*
6430	 * Since we'll hit this the first time through, prior to
6431	 * allocation, we don't need to free a data buffer here.
6432	 */
6433	if (param_len < header_size) {
6434		ctl_set_param_len_error(ctsio);
6435		ctl_done((union ctl_io *)ctsio);
6436		return (CTL_RETVAL_COMPLETE);
6437	}
6438
6439	/*
6440	 * Allocate the data buffer and grab the user's data.  In theory,
6441	 * we shouldn't have to sanity check the parameter list length here
6442	 * because the maximum size is 64K.  We should be able to malloc
6443	 * that much without too many problems.
6444	 */
6445	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6446		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6447		ctsio->kern_data_len = param_len;
6448		ctsio->kern_total_len = param_len;
6449		ctsio->kern_data_resid = 0;
6450		ctsio->kern_rel_offset = 0;
6451		ctsio->kern_sg_entries = 0;
6452		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6453		ctsio->be_move_done = ctl_config_move_done;
6454		ctl_datamove((union ctl_io *)ctsio);
6455
6456		return (CTL_RETVAL_COMPLETE);
6457	}
6458
6459	switch (ctsio->cdb[0]) {
6460	case MODE_SELECT_6: {
6461		struct scsi_mode_header_6 *mh6;
6462
6463		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6464		bd_len = mh6->blk_desc_len;
6465		break;
6466	}
6467	case MODE_SELECT_10: {
6468		struct scsi_mode_header_10 *mh10;
6469
6470		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6471		bd_len = scsi_2btoul(mh10->blk_desc_len);
6472		break;
6473	}
6474	default:
6475		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6476		break;
6477	}
6478
6479	if (param_len < (header_size + bd_len)) {
6480		free(ctsio->kern_data_ptr, M_CTL);
6481		ctl_set_param_len_error(ctsio);
6482		ctl_done((union ctl_io *)ctsio);
6483		return (CTL_RETVAL_COMPLETE);
6484	}
6485
6486	/*
6487	 * Set the IO_CONT flag, so that if this I/O gets passed to
6488	 * ctl_config_write_done(), it'll get passed back to
6489	 * ctl_do_mode_select() for further processing, or completion if
6490	 * we're all done.
6491	 */
6492	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6493	ctsio->io_cont = ctl_do_mode_select;
6494
6495	modepage_info = (union ctl_modepage_info *)
6496		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6497	memset(modepage_info, 0, sizeof(*modepage_info));
6498	modepage_info->header.len_left = param_len - header_size - bd_len;
6499	modepage_info->header.len_used = header_size + bd_len;
6500
6501	return (ctl_do_mode_select((union ctl_io *)ctsio));
6502}
6503
6504int
6505ctl_mode_sense(struct ctl_scsiio *ctsio)
6506{
6507	struct ctl_lun *lun;
6508	int pc, page_code, dbd, llba, subpage;
6509	int alloc_len, page_len, header_len, total_len;
6510	struct scsi_mode_block_descr *block_desc;
6511	struct ctl_page_index *page_index;
6512	int control_dev;
6513
6514	dbd = 0;
6515	llba = 0;
6516	block_desc = NULL;
6517	page_index = NULL;
6518
6519	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6520
6521	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6522
6523	if (lun->be_lun->lun_type != T_DIRECT)
6524		control_dev = 1;
6525	else
6526		control_dev = 0;
6527
6528	switch (ctsio->cdb[0]) {
6529	case MODE_SENSE_6: {
6530		struct scsi_mode_sense_6 *cdb;
6531
6532		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6533
6534		header_len = sizeof(struct scsi_mode_hdr_6);
6535		if (cdb->byte2 & SMS_DBD)
6536			dbd = 1;
6537		else
6538			header_len += sizeof(struct scsi_mode_block_descr);
6539
6540		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6541		page_code = cdb->page & SMS_PAGE_CODE;
6542		subpage = cdb->subpage;
6543		alloc_len = cdb->length;
6544		break;
6545	}
6546	case MODE_SENSE_10: {
6547		struct scsi_mode_sense_10 *cdb;
6548
6549		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6550
6551		header_len = sizeof(struct scsi_mode_hdr_10);
6552
6553		if (cdb->byte2 & SMS_DBD)
6554			dbd = 1;
6555		else
6556			header_len += sizeof(struct scsi_mode_block_descr);
6557		if (cdb->byte2 & SMS10_LLBAA)
6558			llba = 1;
6559		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6560		page_code = cdb->page & SMS_PAGE_CODE;
6561		subpage = cdb->subpage;
6562		alloc_len = scsi_2btoul(cdb->length);
6563		break;
6564	}
6565	default:
6566		ctl_set_invalid_opcode(ctsio);
6567		ctl_done((union ctl_io *)ctsio);
6568		return (CTL_RETVAL_COMPLETE);
6569		break; /* NOTREACHED */
6570	}
6571
6572	/*
6573	 * We have to make a first pass through to calculate the size of
6574	 * the pages that match the user's query.  Then we allocate enough
6575	 * memory to hold it, and actually copy the data into the buffer.
6576	 */
6577	switch (page_code) {
6578	case SMS_ALL_PAGES_PAGE: {
6579		int i;
6580
6581		page_len = 0;
6582
6583		/*
6584		 * At the moment, values other than 0 and 0xff here are
6585		 * reserved according to SPC-3.
6586		 */
6587		if ((subpage != SMS_SUBPAGE_PAGE_0)
6588		 && (subpage != SMS_SUBPAGE_ALL)) {
6589			ctl_set_invalid_field(ctsio,
6590					      /*sks_valid*/ 1,
6591					      /*command*/ 1,
6592					      /*field*/ 3,
6593					      /*bit_valid*/ 0,
6594					      /*bit*/ 0);
6595			ctl_done((union ctl_io *)ctsio);
6596			return (CTL_RETVAL_COMPLETE);
6597		}
6598
6599		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6600			if ((control_dev != 0)
6601			 && (lun->mode_pages.index[i].page_flags &
6602			     CTL_PAGE_FLAG_DISK_ONLY))
6603				continue;
6604
6605			/*
6606			 * We don't use this subpage if the user didn't
6607			 * request all subpages.
6608			 */
6609			if ((lun->mode_pages.index[i].subpage != 0)
6610			 && (subpage == SMS_SUBPAGE_PAGE_0))
6611				continue;
6612
6613#if 0
6614			printf("found page %#x len %d\n",
6615			       lun->mode_pages.index[i].page_code &
6616			       SMPH_PC_MASK,
6617			       lun->mode_pages.index[i].page_len);
6618#endif
6619			page_len += lun->mode_pages.index[i].page_len;
6620		}
6621		break;
6622	}
6623	default: {
6624		int i;
6625
6626		page_len = 0;
6627
6628		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6629			/* Look for the right page code */
6630			if ((lun->mode_pages.index[i].page_code &
6631			     SMPH_PC_MASK) != page_code)
6632				continue;
6633
6634			/* Look for the right subpage or the subpage wildcard*/
6635			if ((lun->mode_pages.index[i].subpage != subpage)
6636			 && (subpage != SMS_SUBPAGE_ALL))
6637				continue;
6638
6639			/* Make sure the page is supported for this dev type */
6640			if ((control_dev != 0)
6641			 && (lun->mode_pages.index[i].page_flags &
6642			     CTL_PAGE_FLAG_DISK_ONLY))
6643				continue;
6644
6645#if 0
6646			printf("found page %#x len %d\n",
6647			       lun->mode_pages.index[i].page_code &
6648			       SMPH_PC_MASK,
6649			       lun->mode_pages.index[i].page_len);
6650#endif
6651
6652			page_len += lun->mode_pages.index[i].page_len;
6653		}
6654
6655		if (page_len == 0) {
6656			ctl_set_invalid_field(ctsio,
6657					      /*sks_valid*/ 1,
6658					      /*command*/ 1,
6659					      /*field*/ 2,
6660					      /*bit_valid*/ 1,
6661					      /*bit*/ 5);
6662			ctl_done((union ctl_io *)ctsio);
6663			return (CTL_RETVAL_COMPLETE);
6664		}
6665		break;
6666	}
6667	}
6668
6669	total_len = header_len + page_len;
6670#if 0
6671	printf("header_len = %d, page_len = %d, total_len = %d\n",
6672	       header_len, page_len, total_len);
6673#endif
6674
6675	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6676	ctsio->kern_sg_entries = 0;
6677	ctsio->kern_data_resid = 0;
6678	ctsio->kern_rel_offset = 0;
6679	if (total_len < alloc_len) {
6680		ctsio->residual = alloc_len - total_len;
6681		ctsio->kern_data_len = total_len;
6682		ctsio->kern_total_len = total_len;
6683	} else {
6684		ctsio->residual = 0;
6685		ctsio->kern_data_len = alloc_len;
6686		ctsio->kern_total_len = alloc_len;
6687	}
6688
6689	switch (ctsio->cdb[0]) {
6690	case MODE_SENSE_6: {
6691		struct scsi_mode_hdr_6 *header;
6692
6693		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6694
6695		header->datalen = MIN(total_len - 1, 254);
6696		if (control_dev == 0) {
6697			header->dev_specific = 0x10; /* DPOFUA */
6698			if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6699			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6700			    .eca_and_aen & SCP_SWP) != 0)
6701				    header->dev_specific |= 0x80; /* WP */
6702		}
6703		if (dbd)
6704			header->block_descr_len = 0;
6705		else
6706			header->block_descr_len =
6707				sizeof(struct scsi_mode_block_descr);
6708		block_desc = (struct scsi_mode_block_descr *)&header[1];
6709		break;
6710	}
6711	case MODE_SENSE_10: {
6712		struct scsi_mode_hdr_10 *header;
6713		int datalen;
6714
6715		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6716
6717		datalen = MIN(total_len - 2, 65533);
6718		scsi_ulto2b(datalen, header->datalen);
6719		if (control_dev == 0) {
6720			header->dev_specific = 0x10; /* DPOFUA */
6721			if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6722			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6723			    .eca_and_aen & SCP_SWP) != 0)
6724				    header->dev_specific |= 0x80; /* WP */
6725		}
6726		if (dbd)
6727			scsi_ulto2b(0, header->block_descr_len);
6728		else
6729			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6730				    header->block_descr_len);
6731		block_desc = (struct scsi_mode_block_descr *)&header[1];
6732		break;
6733	}
6734	default:
6735		panic("invalid CDB type %#x", ctsio->cdb[0]);
6736		break; /* NOTREACHED */
6737	}
6738
6739	/*
6740	 * If we've got a disk, use its blocksize in the block
6741	 * descriptor.  Otherwise, just set it to 0.
6742	 */
6743	if (dbd == 0) {
6744		if (control_dev == 0)
6745			scsi_ulto3b(lun->be_lun->blocksize,
6746				    block_desc->block_len);
6747		else
6748			scsi_ulto3b(0, block_desc->block_len);
6749	}
6750
6751	switch (page_code) {
6752	case SMS_ALL_PAGES_PAGE: {
6753		int i, data_used;
6754
6755		data_used = header_len;
6756		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6757			struct ctl_page_index *page_index;
6758
6759			page_index = &lun->mode_pages.index[i];
6760
6761			if ((control_dev != 0)
6762			 && (page_index->page_flags &
6763			    CTL_PAGE_FLAG_DISK_ONLY))
6764				continue;
6765
6766			/*
6767			 * We don't use this subpage if the user didn't
6768			 * request all subpages.  We already checked (above)
6769			 * to make sure the user only specified a subpage
6770			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6771			 */
6772			if ((page_index->subpage != 0)
6773			 && (subpage == SMS_SUBPAGE_PAGE_0))
6774				continue;
6775
6776			/*
6777			 * Call the handler, if it exists, to update the
6778			 * page to the latest values.
6779			 */
6780			if (page_index->sense_handler != NULL)
6781				page_index->sense_handler(ctsio, page_index,pc);
6782
6783			memcpy(ctsio->kern_data_ptr + data_used,
6784			       page_index->page_data +
6785			       (page_index->page_len * pc),
6786			       page_index->page_len);
6787			data_used += page_index->page_len;
6788		}
6789		break;
6790	}
6791	default: {
6792		int i, data_used;
6793
6794		data_used = header_len;
6795
6796		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6797			struct ctl_page_index *page_index;
6798
6799			page_index = &lun->mode_pages.index[i];
6800
6801			/* Look for the right page code */
6802			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6803				continue;
6804
6805			/* Look for the right subpage or the subpage wildcard*/
6806			if ((page_index->subpage != subpage)
6807			 && (subpage != SMS_SUBPAGE_ALL))
6808				continue;
6809
6810			/* Make sure the page is supported for this dev type */
6811			if ((control_dev != 0)
6812			 && (page_index->page_flags &
6813			     CTL_PAGE_FLAG_DISK_ONLY))
6814				continue;
6815
6816			/*
6817			 * Call the handler, if it exists, to update the
6818			 * page to the latest values.
6819			 */
6820			if (page_index->sense_handler != NULL)
6821				page_index->sense_handler(ctsio, page_index,pc);
6822
6823			memcpy(ctsio->kern_data_ptr + data_used,
6824			       page_index->page_data +
6825			       (page_index->page_len * pc),
6826			       page_index->page_len);
6827			data_used += page_index->page_len;
6828		}
6829		break;
6830	}
6831	}
6832
6833	ctl_set_success(ctsio);
6834	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6835	ctsio->be_move_done = ctl_config_move_done;
6836	ctl_datamove((union ctl_io *)ctsio);
6837	return (CTL_RETVAL_COMPLETE);
6838}
6839
6840int
6841ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6842			       struct ctl_page_index *page_index,
6843			       int pc)
6844{
6845	struct ctl_lun *lun;
6846	struct scsi_log_param_header *phdr;
6847	uint8_t *data;
6848	uint64_t val;
6849
6850	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6851	data = page_index->page_data;
6852
6853	if (lun->backend->lun_attr != NULL &&
6854	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6855	     != UINT64_MAX) {
6856		phdr = (struct scsi_log_param_header *)data;
6857		scsi_ulto2b(0x0001, phdr->param_code);
6858		phdr->param_control = SLP_LBIN | SLP_LP;
6859		phdr->param_len = 8;
6860		data = (uint8_t *)(phdr + 1);
6861		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6862		data[4] = 0x02; /* per-pool */
6863		data += phdr->param_len;
6864	}
6865
6866	if (lun->backend->lun_attr != NULL &&
6867	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6868	     != UINT64_MAX) {
6869		phdr = (struct scsi_log_param_header *)data;
6870		scsi_ulto2b(0x0002, phdr->param_code);
6871		phdr->param_control = SLP_LBIN | SLP_LP;
6872		phdr->param_len = 8;
6873		data = (uint8_t *)(phdr + 1);
6874		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6875		data[4] = 0x01; /* per-LUN */
6876		data += phdr->param_len;
6877	}
6878
6879	if (lun->backend->lun_attr != NULL &&
6880	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6881	     != UINT64_MAX) {
6882		phdr = (struct scsi_log_param_header *)data;
6883		scsi_ulto2b(0x00f1, phdr->param_code);
6884		phdr->param_control = SLP_LBIN | SLP_LP;
6885		phdr->param_len = 8;
6886		data = (uint8_t *)(phdr + 1);
6887		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6888		data[4] = 0x02; /* per-pool */
6889		data += phdr->param_len;
6890	}
6891
6892	if (lun->backend->lun_attr != NULL &&
6893	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6894	     != UINT64_MAX) {
6895		phdr = (struct scsi_log_param_header *)data;
6896		scsi_ulto2b(0x00f2, phdr->param_code);
6897		phdr->param_control = SLP_LBIN | SLP_LP;
6898		phdr->param_len = 8;
6899		data = (uint8_t *)(phdr + 1);
6900		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6901		data[4] = 0x02; /* per-pool */
6902		data += phdr->param_len;
6903	}
6904
6905	page_index->page_len = data - page_index->page_data;
6906	return (0);
6907}
6908
6909int
6910ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6911			       struct ctl_page_index *page_index,
6912			       int pc)
6913{
6914	struct ctl_lun *lun;
6915	struct stat_page *data;
6916	uint64_t rn, wn, rb, wb;
6917	struct bintime rt, wt;
6918	int i;
6919
6920	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6921	data = (struct stat_page *)page_index->page_data;
6922
6923	scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6924	data->sap.hdr.param_control = SLP_LBIN;
6925	data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6926	    sizeof(struct scsi_log_param_header);
6927	rn = wn = rb = wb = 0;
6928	bintime_clear(&rt);
6929	bintime_clear(&wt);
6930	for (i = 0; i < CTL_MAX_PORTS; i++) {
6931		rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6932		wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6933		rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6934		wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6935		bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6936		bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6937	}
6938	scsi_u64to8b(rn, data->sap.read_num);
6939	scsi_u64to8b(wn, data->sap.write_num);
6940	if (lun->stats.blocksize > 0) {
6941		scsi_u64to8b(wb / lun->stats.blocksize,
6942		    data->sap.recvieved_lba);
6943		scsi_u64to8b(rb / lun->stats.blocksize,
6944		    data->sap.transmitted_lba);
6945	}
6946	scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6947	    data->sap.read_int);
6948	scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6949	    data->sap.write_int);
6950	scsi_u64to8b(0, data->sap.weighted_num);
6951	scsi_u64to8b(0, data->sap.weighted_int);
6952	scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6953	data->it.hdr.param_control = SLP_LBIN;
6954	data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6955	    sizeof(struct scsi_log_param_header);
6956#ifdef CTL_TIME_IO
6957	scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6958#endif
6959	scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6960	data->it.hdr.param_control = SLP_LBIN;
6961	data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6962	    sizeof(struct scsi_log_param_header);
6963	scsi_ulto4b(3, data->ti.exponent);
6964	scsi_ulto4b(1, data->ti.integer);
6965
6966	page_index->page_len = sizeof(*data);
6967	return (0);
6968}
6969
6970int
6971ctl_log_sense(struct ctl_scsiio *ctsio)
6972{
6973	struct ctl_lun *lun;
6974	int i, pc, page_code, subpage;
6975	int alloc_len, total_len;
6976	struct ctl_page_index *page_index;
6977	struct scsi_log_sense *cdb;
6978	struct scsi_log_header *header;
6979
6980	CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6981
6982	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6983	cdb = (struct scsi_log_sense *)ctsio->cdb;
6984	pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6985	page_code = cdb->page & SLS_PAGE_CODE;
6986	subpage = cdb->subpage;
6987	alloc_len = scsi_2btoul(cdb->length);
6988
6989	page_index = NULL;
6990	for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6991		page_index = &lun->log_pages.index[i];
6992
6993		/* Look for the right page code */
6994		if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6995			continue;
6996
6997		/* Look for the right subpage or the subpage wildcard*/
6998		if (page_index->subpage != subpage)
6999			continue;
7000
7001		break;
7002	}
7003	if (i >= CTL_NUM_LOG_PAGES) {
7004		ctl_set_invalid_field(ctsio,
7005				      /*sks_valid*/ 1,
7006				      /*command*/ 1,
7007				      /*field*/ 2,
7008				      /*bit_valid*/ 0,
7009				      /*bit*/ 0);
7010		ctl_done((union ctl_io *)ctsio);
7011		return (CTL_RETVAL_COMPLETE);
7012	}
7013
7014	total_len = sizeof(struct scsi_log_header) + page_index->page_len;
7015
7016	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7017	ctsio->kern_sg_entries = 0;
7018	ctsio->kern_data_resid = 0;
7019	ctsio->kern_rel_offset = 0;
7020	if (total_len < alloc_len) {
7021		ctsio->residual = alloc_len - total_len;
7022		ctsio->kern_data_len = total_len;
7023		ctsio->kern_total_len = total_len;
7024	} else {
7025		ctsio->residual = 0;
7026		ctsio->kern_data_len = alloc_len;
7027		ctsio->kern_total_len = alloc_len;
7028	}
7029
7030	header = (struct scsi_log_header *)ctsio->kern_data_ptr;
7031	header->page = page_index->page_code;
7032	if (page_index->subpage) {
7033		header->page |= SL_SPF;
7034		header->subpage = page_index->subpage;
7035	}
7036	scsi_ulto2b(page_index->page_len, header->datalen);
7037
7038	/*
7039	 * Call the handler, if it exists, to update the
7040	 * page to the latest values.
7041	 */
7042	if (page_index->sense_handler != NULL)
7043		page_index->sense_handler(ctsio, page_index, pc);
7044
7045	memcpy(header + 1, page_index->page_data, page_index->page_len);
7046
7047	ctl_set_success(ctsio);
7048	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7049	ctsio->be_move_done = ctl_config_move_done;
7050	ctl_datamove((union ctl_io *)ctsio);
7051	return (CTL_RETVAL_COMPLETE);
7052}
7053
7054int
7055ctl_read_capacity(struct ctl_scsiio *ctsio)
7056{
7057	struct scsi_read_capacity *cdb;
7058	struct scsi_read_capacity_data *data;
7059	struct ctl_lun *lun;
7060	uint32_t lba;
7061
7062	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7063
7064	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7065
7066	lba = scsi_4btoul(cdb->addr);
7067	if (((cdb->pmi & SRC_PMI) == 0)
7068	 && (lba != 0)) {
7069		ctl_set_invalid_field(/*ctsio*/ ctsio,
7070				      /*sks_valid*/ 1,
7071				      /*command*/ 1,
7072				      /*field*/ 2,
7073				      /*bit_valid*/ 0,
7074				      /*bit*/ 0);
7075		ctl_done((union ctl_io *)ctsio);
7076		return (CTL_RETVAL_COMPLETE);
7077	}
7078
7079	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7080
7081	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7082	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7083	ctsio->residual = 0;
7084	ctsio->kern_data_len = sizeof(*data);
7085	ctsio->kern_total_len = sizeof(*data);
7086	ctsio->kern_data_resid = 0;
7087	ctsio->kern_rel_offset = 0;
7088	ctsio->kern_sg_entries = 0;
7089
7090	/*
7091	 * If the maximum LBA is greater than 0xfffffffe, the user must
7092	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7093	 * serivce action set.
7094	 */
7095	if (lun->be_lun->maxlba > 0xfffffffe)
7096		scsi_ulto4b(0xffffffff, data->addr);
7097	else
7098		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7099
7100	/*
7101	 * XXX KDM this may not be 512 bytes...
7102	 */
7103	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7104
7105	ctl_set_success(ctsio);
7106	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7107	ctsio->be_move_done = ctl_config_move_done;
7108	ctl_datamove((union ctl_io *)ctsio);
7109	return (CTL_RETVAL_COMPLETE);
7110}
7111
7112int
7113ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7114{
7115	struct scsi_read_capacity_16 *cdb;
7116	struct scsi_read_capacity_data_long *data;
7117	struct ctl_lun *lun;
7118	uint64_t lba;
7119	uint32_t alloc_len;
7120
7121	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7122
7123	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7124
7125	alloc_len = scsi_4btoul(cdb->alloc_len);
7126	lba = scsi_8btou64(cdb->addr);
7127
7128	if ((cdb->reladr & SRC16_PMI)
7129	 && (lba != 0)) {
7130		ctl_set_invalid_field(/*ctsio*/ ctsio,
7131				      /*sks_valid*/ 1,
7132				      /*command*/ 1,
7133				      /*field*/ 2,
7134				      /*bit_valid*/ 0,
7135				      /*bit*/ 0);
7136		ctl_done((union ctl_io *)ctsio);
7137		return (CTL_RETVAL_COMPLETE);
7138	}
7139
7140	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7141
7142	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7143	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7144
7145	if (sizeof(*data) < alloc_len) {
7146		ctsio->residual = alloc_len - sizeof(*data);
7147		ctsio->kern_data_len = sizeof(*data);
7148		ctsio->kern_total_len = sizeof(*data);
7149	} else {
7150		ctsio->residual = 0;
7151		ctsio->kern_data_len = alloc_len;
7152		ctsio->kern_total_len = alloc_len;
7153	}
7154	ctsio->kern_data_resid = 0;
7155	ctsio->kern_rel_offset = 0;
7156	ctsio->kern_sg_entries = 0;
7157
7158	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7159	/* XXX KDM this may not be 512 bytes... */
7160	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7161	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7162	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7163	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7164		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7165
7166	ctl_set_success(ctsio);
7167	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7168	ctsio->be_move_done = ctl_config_move_done;
7169	ctl_datamove((union ctl_io *)ctsio);
7170	return (CTL_RETVAL_COMPLETE);
7171}
7172
7173int
7174ctl_get_lba_status(struct ctl_scsiio *ctsio)
7175{
7176	struct scsi_get_lba_status *cdb;
7177	struct scsi_get_lba_status_data *data;
7178	struct ctl_lun *lun;
7179	struct ctl_lba_len_flags *lbalen;
7180	uint64_t lba;
7181	uint32_t alloc_len, total_len;
7182	int retval;
7183
7184	CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7185
7186	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7187	cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7188	lba = scsi_8btou64(cdb->addr);
7189	alloc_len = scsi_4btoul(cdb->alloc_len);
7190
7191	if (lba > lun->be_lun->maxlba) {
7192		ctl_set_lba_out_of_range(ctsio);
7193		ctl_done((union ctl_io *)ctsio);
7194		return (CTL_RETVAL_COMPLETE);
7195	}
7196
7197	total_len = sizeof(*data) + sizeof(data->descr[0]);
7198	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7199	data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7200
7201	if (total_len < alloc_len) {
7202		ctsio->residual = alloc_len - total_len;
7203		ctsio->kern_data_len = total_len;
7204		ctsio->kern_total_len = total_len;
7205	} else {
7206		ctsio->residual = 0;
7207		ctsio->kern_data_len = alloc_len;
7208		ctsio->kern_total_len = alloc_len;
7209	}
7210	ctsio->kern_data_resid = 0;
7211	ctsio->kern_rel_offset = 0;
7212	ctsio->kern_sg_entries = 0;
7213
7214	/* Fill dummy data in case backend can't tell anything. */
7215	scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7216	scsi_u64to8b(lba, data->descr[0].addr);
7217	scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7218	    data->descr[0].length);
7219	data->descr[0].status = 0; /* Mapped or unknown. */
7220
7221	ctl_set_success(ctsio);
7222	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7223	ctsio->be_move_done = ctl_config_move_done;
7224
7225	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7226	lbalen->lba = lba;
7227	lbalen->len = total_len;
7228	lbalen->flags = 0;
7229	retval = lun->backend->config_read((union ctl_io *)ctsio);
7230	return (CTL_RETVAL_COMPLETE);
7231}
7232
7233int
7234ctl_read_defect(struct ctl_scsiio *ctsio)
7235{
7236	struct scsi_read_defect_data_10 *ccb10;
7237	struct scsi_read_defect_data_12 *ccb12;
7238	struct scsi_read_defect_data_hdr_10 *data10;
7239	struct scsi_read_defect_data_hdr_12 *data12;
7240	uint32_t alloc_len, data_len;
7241	uint8_t format;
7242
7243	CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7244
7245	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7246		ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7247		format = ccb10->format;
7248		alloc_len = scsi_2btoul(ccb10->alloc_length);
7249		data_len = sizeof(*data10);
7250	} else {
7251		ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7252		format = ccb12->format;
7253		alloc_len = scsi_4btoul(ccb12->alloc_length);
7254		data_len = sizeof(*data12);
7255	}
7256	if (alloc_len == 0) {
7257		ctl_set_success(ctsio);
7258		ctl_done((union ctl_io *)ctsio);
7259		return (CTL_RETVAL_COMPLETE);
7260	}
7261
7262	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7263	if (data_len < alloc_len) {
7264		ctsio->residual = alloc_len - data_len;
7265		ctsio->kern_data_len = data_len;
7266		ctsio->kern_total_len = data_len;
7267	} else {
7268		ctsio->residual = 0;
7269		ctsio->kern_data_len = alloc_len;
7270		ctsio->kern_total_len = alloc_len;
7271	}
7272	ctsio->kern_data_resid = 0;
7273	ctsio->kern_rel_offset = 0;
7274	ctsio->kern_sg_entries = 0;
7275
7276	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7277		data10 = (struct scsi_read_defect_data_hdr_10 *)
7278		    ctsio->kern_data_ptr;
7279		data10->format = format;
7280		scsi_ulto2b(0, data10->length);
7281	} else {
7282		data12 = (struct scsi_read_defect_data_hdr_12 *)
7283		    ctsio->kern_data_ptr;
7284		data12->format = format;
7285		scsi_ulto2b(0, data12->generation);
7286		scsi_ulto4b(0, data12->length);
7287	}
7288
7289	ctl_set_success(ctsio);
7290	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7291	ctsio->be_move_done = ctl_config_move_done;
7292	ctl_datamove((union ctl_io *)ctsio);
7293	return (CTL_RETVAL_COMPLETE);
7294}
7295
7296int
7297ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7298{
7299	struct scsi_maintenance_in *cdb;
7300	int retval;
7301	int alloc_len, ext, total_len = 0, g, pc, pg, gs, os;
7302	int num_target_port_groups, num_target_ports;
7303	struct ctl_lun *lun;
7304	struct ctl_softc *softc;
7305	struct ctl_port *port;
7306	struct scsi_target_group_data *rtg_ptr;
7307	struct scsi_target_group_data_extended *rtg_ext_ptr;
7308	struct scsi_target_port_group_descriptor *tpg_desc;
7309
7310	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7311
7312	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7313	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7314	softc = lun->ctl_softc;
7315
7316	retval = CTL_RETVAL_COMPLETE;
7317
7318	switch (cdb->byte2 & STG_PDF_MASK) {
7319	case STG_PDF_LENGTH:
7320		ext = 0;
7321		break;
7322	case STG_PDF_EXTENDED:
7323		ext = 1;
7324		break;
7325	default:
7326		ctl_set_invalid_field(/*ctsio*/ ctsio,
7327				      /*sks_valid*/ 1,
7328				      /*command*/ 1,
7329				      /*field*/ 2,
7330				      /*bit_valid*/ 1,
7331				      /*bit*/ 5);
7332		ctl_done((union ctl_io *)ctsio);
7333		return(retval);
7334	}
7335
7336	if (softc->is_single)
7337		num_target_port_groups = 1;
7338	else
7339		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7340	num_target_ports = 0;
7341	mtx_lock(&softc->ctl_lock);
7342	STAILQ_FOREACH(port, &softc->port_list, links) {
7343		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7344			continue;
7345		if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7346			continue;
7347		num_target_ports++;
7348	}
7349	mtx_unlock(&softc->ctl_lock);
7350
7351	if (ext)
7352		total_len = sizeof(struct scsi_target_group_data_extended);
7353	else
7354		total_len = sizeof(struct scsi_target_group_data);
7355	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7356		num_target_port_groups +
7357	    sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7358
7359	alloc_len = scsi_4btoul(cdb->length);
7360
7361	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7362
7363	ctsio->kern_sg_entries = 0;
7364
7365	if (total_len < alloc_len) {
7366		ctsio->residual = alloc_len - total_len;
7367		ctsio->kern_data_len = total_len;
7368		ctsio->kern_total_len = total_len;
7369	} else {
7370		ctsio->residual = 0;
7371		ctsio->kern_data_len = alloc_len;
7372		ctsio->kern_total_len = alloc_len;
7373	}
7374	ctsio->kern_data_resid = 0;
7375	ctsio->kern_rel_offset = 0;
7376
7377	if (ext) {
7378		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7379		    ctsio->kern_data_ptr;
7380		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7381		rtg_ext_ptr->format_type = 0x10;
7382		rtg_ext_ptr->implicit_transition_time = 0;
7383		tpg_desc = &rtg_ext_ptr->groups[0];
7384	} else {
7385		rtg_ptr = (struct scsi_target_group_data *)
7386		    ctsio->kern_data_ptr;
7387		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7388		tpg_desc = &rtg_ptr->groups[0];
7389	}
7390
7391	mtx_lock(&softc->ctl_lock);
7392	pg = softc->port_min / softc->port_cnt;
7393	if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7394		gs = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7395	else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7396		gs = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7397	else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7398		gs = TPG_ASYMMETRIC_ACCESS_STANDBY;
7399	else
7400		gs = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7401	if (lun->flags & CTL_LUN_PRIMARY_SC) {
7402		os = gs;
7403		gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7404	} else
7405		os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7406	for (g = 0; g < num_target_port_groups; g++) {
7407		tpg_desc->pref_state = (g == pg) ? gs : os;
7408		tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7409		    TPG_U_SUP | TPG_T_SUP;
7410		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7411		tpg_desc->status = TPG_IMPLICIT;
7412		pc = 0;
7413		STAILQ_FOREACH(port, &softc->port_list, links) {
7414			if (port->targ_port < g * softc->port_cnt ||
7415			    port->targ_port >= (g + 1) * softc->port_cnt)
7416				continue;
7417			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7418				continue;
7419			if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7420				continue;
7421			scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7422			    relative_target_port_identifier);
7423			pc++;
7424		}
7425		tpg_desc->target_port_count = pc;
7426		tpg_desc = (struct scsi_target_port_group_descriptor *)
7427		    &tpg_desc->descriptors[pc];
7428	}
7429	mtx_unlock(&softc->ctl_lock);
7430
7431	ctl_set_success(ctsio);
7432	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7433	ctsio->be_move_done = ctl_config_move_done;
7434	ctl_datamove((union ctl_io *)ctsio);
7435	return(retval);
7436}
7437
7438int
7439ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7440{
7441	struct ctl_lun *lun;
7442	struct scsi_report_supported_opcodes *cdb;
7443	const struct ctl_cmd_entry *entry, *sentry;
7444	struct scsi_report_supported_opcodes_all *all;
7445	struct scsi_report_supported_opcodes_descr *descr;
7446	struct scsi_report_supported_opcodes_one *one;
7447	int retval;
7448	int alloc_len, total_len;
7449	int opcode, service_action, i, j, num;
7450
7451	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7452
7453	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7454	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7455
7456	retval = CTL_RETVAL_COMPLETE;
7457
7458	opcode = cdb->requested_opcode;
7459	service_action = scsi_2btoul(cdb->requested_service_action);
7460	switch (cdb->options & RSO_OPTIONS_MASK) {
7461	case RSO_OPTIONS_ALL:
7462		num = 0;
7463		for (i = 0; i < 256; i++) {
7464			entry = &ctl_cmd_table[i];
7465			if (entry->flags & CTL_CMD_FLAG_SA5) {
7466				for (j = 0; j < 32; j++) {
7467					sentry = &((const struct ctl_cmd_entry *)
7468					    entry->execute)[j];
7469					if (ctl_cmd_applicable(
7470					    lun->be_lun->lun_type, sentry))
7471						num++;
7472				}
7473			} else {
7474				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7475				    entry))
7476					num++;
7477			}
7478		}
7479		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7480		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7481		break;
7482	case RSO_OPTIONS_OC:
7483		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7484			ctl_set_invalid_field(/*ctsio*/ ctsio,
7485					      /*sks_valid*/ 1,
7486					      /*command*/ 1,
7487					      /*field*/ 2,
7488					      /*bit_valid*/ 1,
7489					      /*bit*/ 2);
7490			ctl_done((union ctl_io *)ctsio);
7491			return (CTL_RETVAL_COMPLETE);
7492		}
7493		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7494		break;
7495	case RSO_OPTIONS_OC_SA:
7496		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7497		    service_action >= 32) {
7498			ctl_set_invalid_field(/*ctsio*/ ctsio,
7499					      /*sks_valid*/ 1,
7500					      /*command*/ 1,
7501					      /*field*/ 2,
7502					      /*bit_valid*/ 1,
7503					      /*bit*/ 2);
7504			ctl_done((union ctl_io *)ctsio);
7505			return (CTL_RETVAL_COMPLETE);
7506		}
7507		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7508		break;
7509	default:
7510		ctl_set_invalid_field(/*ctsio*/ ctsio,
7511				      /*sks_valid*/ 1,
7512				      /*command*/ 1,
7513				      /*field*/ 2,
7514				      /*bit_valid*/ 1,
7515				      /*bit*/ 2);
7516		ctl_done((union ctl_io *)ctsio);
7517		return (CTL_RETVAL_COMPLETE);
7518	}
7519
7520	alloc_len = scsi_4btoul(cdb->length);
7521
7522	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7523
7524	ctsio->kern_sg_entries = 0;
7525
7526	if (total_len < alloc_len) {
7527		ctsio->residual = alloc_len - total_len;
7528		ctsio->kern_data_len = total_len;
7529		ctsio->kern_total_len = total_len;
7530	} else {
7531		ctsio->residual = 0;
7532		ctsio->kern_data_len = alloc_len;
7533		ctsio->kern_total_len = alloc_len;
7534	}
7535	ctsio->kern_data_resid = 0;
7536	ctsio->kern_rel_offset = 0;
7537
7538	switch (cdb->options & RSO_OPTIONS_MASK) {
7539	case RSO_OPTIONS_ALL:
7540		all = (struct scsi_report_supported_opcodes_all *)
7541		    ctsio->kern_data_ptr;
7542		num = 0;
7543		for (i = 0; i < 256; i++) {
7544			entry = &ctl_cmd_table[i];
7545			if (entry->flags & CTL_CMD_FLAG_SA5) {
7546				for (j = 0; j < 32; j++) {
7547					sentry = &((const struct ctl_cmd_entry *)
7548					    entry->execute)[j];
7549					if (!ctl_cmd_applicable(
7550					    lun->be_lun->lun_type, sentry))
7551						continue;
7552					descr = &all->descr[num++];
7553					descr->opcode = i;
7554					scsi_ulto2b(j, descr->service_action);
7555					descr->flags = RSO_SERVACTV;
7556					scsi_ulto2b(sentry->length,
7557					    descr->cdb_length);
7558				}
7559			} else {
7560				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7561				    entry))
7562					continue;
7563				descr = &all->descr[num++];
7564				descr->opcode = i;
7565				scsi_ulto2b(0, descr->service_action);
7566				descr->flags = 0;
7567				scsi_ulto2b(entry->length, descr->cdb_length);
7568			}
7569		}
7570		scsi_ulto4b(
7571		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7572		    all->length);
7573		break;
7574	case RSO_OPTIONS_OC:
7575		one = (struct scsi_report_supported_opcodes_one *)
7576		    ctsio->kern_data_ptr;
7577		entry = &ctl_cmd_table[opcode];
7578		goto fill_one;
7579	case RSO_OPTIONS_OC_SA:
7580		one = (struct scsi_report_supported_opcodes_one *)
7581		    ctsio->kern_data_ptr;
7582		entry = &ctl_cmd_table[opcode];
7583		entry = &((const struct ctl_cmd_entry *)
7584		    entry->execute)[service_action];
7585fill_one:
7586		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7587			one->support = 3;
7588			scsi_ulto2b(entry->length, one->cdb_length);
7589			one->cdb_usage[0] = opcode;
7590			memcpy(&one->cdb_usage[1], entry->usage,
7591			    entry->length - 1);
7592		} else
7593			one->support = 1;
7594		break;
7595	}
7596
7597	ctl_set_success(ctsio);
7598	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7599	ctsio->be_move_done = ctl_config_move_done;
7600	ctl_datamove((union ctl_io *)ctsio);
7601	return(retval);
7602}
7603
7604int
7605ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7606{
7607	struct scsi_report_supported_tmf *cdb;
7608	struct scsi_report_supported_tmf_data *data;
7609	int retval;
7610	int alloc_len, total_len;
7611
7612	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7613
7614	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7615
7616	retval = CTL_RETVAL_COMPLETE;
7617
7618	total_len = sizeof(struct scsi_report_supported_tmf_data);
7619	alloc_len = scsi_4btoul(cdb->length);
7620
7621	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7622
7623	ctsio->kern_sg_entries = 0;
7624
7625	if (total_len < alloc_len) {
7626		ctsio->residual = alloc_len - total_len;
7627		ctsio->kern_data_len = total_len;
7628		ctsio->kern_total_len = total_len;
7629	} else {
7630		ctsio->residual = 0;
7631		ctsio->kern_data_len = alloc_len;
7632		ctsio->kern_total_len = alloc_len;
7633	}
7634	ctsio->kern_data_resid = 0;
7635	ctsio->kern_rel_offset = 0;
7636
7637	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7638	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7639	    RST_TRS;
7640	data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7641
7642	ctl_set_success(ctsio);
7643	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7644	ctsio->be_move_done = ctl_config_move_done;
7645	ctl_datamove((union ctl_io *)ctsio);
7646	return (retval);
7647}
7648
7649int
7650ctl_report_timestamp(struct ctl_scsiio *ctsio)
7651{
7652	struct scsi_report_timestamp *cdb;
7653	struct scsi_report_timestamp_data *data;
7654	struct timeval tv;
7655	int64_t timestamp;
7656	int retval;
7657	int alloc_len, total_len;
7658
7659	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7660
7661	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7662
7663	retval = CTL_RETVAL_COMPLETE;
7664
7665	total_len = sizeof(struct scsi_report_timestamp_data);
7666	alloc_len = scsi_4btoul(cdb->length);
7667
7668	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7669
7670	ctsio->kern_sg_entries = 0;
7671
7672	if (total_len < alloc_len) {
7673		ctsio->residual = alloc_len - total_len;
7674		ctsio->kern_data_len = total_len;
7675		ctsio->kern_total_len = total_len;
7676	} else {
7677		ctsio->residual = 0;
7678		ctsio->kern_data_len = alloc_len;
7679		ctsio->kern_total_len = alloc_len;
7680	}
7681	ctsio->kern_data_resid = 0;
7682	ctsio->kern_rel_offset = 0;
7683
7684	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7685	scsi_ulto2b(sizeof(*data) - 2, data->length);
7686	data->origin = RTS_ORIG_OUTSIDE;
7687	getmicrotime(&tv);
7688	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7689	scsi_ulto4b(timestamp >> 16, data->timestamp);
7690	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7691
7692	ctl_set_success(ctsio);
7693	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7694	ctsio->be_move_done = ctl_config_move_done;
7695	ctl_datamove((union ctl_io *)ctsio);
7696	return (retval);
7697}
7698
7699int
7700ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7701{
7702	struct scsi_per_res_in *cdb;
7703	int alloc_len, total_len = 0;
7704	/* struct scsi_per_res_in_rsrv in_data; */
7705	struct ctl_lun *lun;
7706	struct ctl_softc *softc;
7707	uint64_t key;
7708
7709	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7710
7711	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7712
7713	alloc_len = scsi_2btoul(cdb->length);
7714
7715	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7716	softc = lun->ctl_softc;
7717
7718retry:
7719	mtx_lock(&lun->lun_lock);
7720	switch (cdb->action) {
7721	case SPRI_RK: /* read keys */
7722		total_len = sizeof(struct scsi_per_res_in_keys) +
7723			lun->pr_key_count *
7724			sizeof(struct scsi_per_res_key);
7725		break;
7726	case SPRI_RR: /* read reservation */
7727		if (lun->flags & CTL_LUN_PR_RESERVED)
7728			total_len = sizeof(struct scsi_per_res_in_rsrv);
7729		else
7730			total_len = sizeof(struct scsi_per_res_in_header);
7731		break;
7732	case SPRI_RC: /* report capabilities */
7733		total_len = sizeof(struct scsi_per_res_cap);
7734		break;
7735	case SPRI_RS: /* read full status */
7736		total_len = sizeof(struct scsi_per_res_in_header) +
7737		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7738		    lun->pr_key_count;
7739		break;
7740	default:
7741		panic("Invalid PR type %x", cdb->action);
7742	}
7743	mtx_unlock(&lun->lun_lock);
7744
7745	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7746
7747	if (total_len < alloc_len) {
7748		ctsio->residual = alloc_len - total_len;
7749		ctsio->kern_data_len = total_len;
7750		ctsio->kern_total_len = total_len;
7751	} else {
7752		ctsio->residual = 0;
7753		ctsio->kern_data_len = alloc_len;
7754		ctsio->kern_total_len = alloc_len;
7755	}
7756
7757	ctsio->kern_data_resid = 0;
7758	ctsio->kern_rel_offset = 0;
7759	ctsio->kern_sg_entries = 0;
7760
7761	mtx_lock(&lun->lun_lock);
7762	switch (cdb->action) {
7763	case SPRI_RK: { // read keys
7764        struct scsi_per_res_in_keys *res_keys;
7765		int i, key_count;
7766
7767		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7768
7769		/*
7770		 * We had to drop the lock to allocate our buffer, which
7771		 * leaves time for someone to come in with another
7772		 * persistent reservation.  (That is unlikely, though,
7773		 * since this should be the only persistent reservation
7774		 * command active right now.)
7775		 */
7776		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7777		    (lun->pr_key_count *
7778		     sizeof(struct scsi_per_res_key)))){
7779			mtx_unlock(&lun->lun_lock);
7780			free(ctsio->kern_data_ptr, M_CTL);
7781			printf("%s: reservation length changed, retrying\n",
7782			       __func__);
7783			goto retry;
7784		}
7785
7786		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7787
7788		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7789			     lun->pr_key_count, res_keys->header.length);
7790
7791		for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7792			if ((key = ctl_get_prkey(lun, i)) == 0)
7793				continue;
7794
7795			/*
7796			 * We used lun->pr_key_count to calculate the
7797			 * size to allocate.  If it turns out the number of
7798			 * initiators with the registered flag set is
7799			 * larger than that (i.e. they haven't been kept in
7800			 * sync), we've got a problem.
7801			 */
7802			if (key_count >= lun->pr_key_count) {
7803#ifdef NEEDTOPORT
7804				csevent_log(CSC_CTL | CSC_SHELF_SW |
7805					    CTL_PR_ERROR,
7806					    csevent_LogType_Fault,
7807					    csevent_AlertLevel_Yellow,
7808					    csevent_FRU_ShelfController,
7809					    csevent_FRU_Firmware,
7810				        csevent_FRU_Unknown,
7811					    "registered keys %d >= key "
7812					    "count %d", key_count,
7813					    lun->pr_key_count);
7814#endif
7815				key_count++;
7816				continue;
7817			}
7818			scsi_u64to8b(key, res_keys->keys[key_count].key);
7819			key_count++;
7820		}
7821		break;
7822	}
7823	case SPRI_RR: { // read reservation
7824		struct scsi_per_res_in_rsrv *res;
7825		int tmp_len, header_only;
7826
7827		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7828
7829		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7830
7831		if (lun->flags & CTL_LUN_PR_RESERVED)
7832		{
7833			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7834			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7835				    res->header.length);
7836			header_only = 0;
7837		} else {
7838			tmp_len = sizeof(struct scsi_per_res_in_header);
7839			scsi_ulto4b(0, res->header.length);
7840			header_only = 1;
7841		}
7842
7843		/*
7844		 * We had to drop the lock to allocate our buffer, which
7845		 * leaves time for someone to come in with another
7846		 * persistent reservation.  (That is unlikely, though,
7847		 * since this should be the only persistent reservation
7848		 * command active right now.)
7849		 */
7850		if (tmp_len != total_len) {
7851			mtx_unlock(&lun->lun_lock);
7852			free(ctsio->kern_data_ptr, M_CTL);
7853			printf("%s: reservation status changed, retrying\n",
7854			       __func__);
7855			goto retry;
7856		}
7857
7858		/*
7859		 * No reservation held, so we're done.
7860		 */
7861		if (header_only != 0)
7862			break;
7863
7864		/*
7865		 * If the registration is an All Registrants type, the key
7866		 * is 0, since it doesn't really matter.
7867		 */
7868		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7869			scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7870			    res->data.reservation);
7871		}
7872		res->data.scopetype = lun->res_type;
7873		break;
7874	}
7875	case SPRI_RC:     //report capabilities
7876	{
7877		struct scsi_per_res_cap *res_cap;
7878		uint16_t type_mask;
7879
7880		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7881		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7882		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7883		type_mask = SPRI_TM_WR_EX_AR |
7884			    SPRI_TM_EX_AC_RO |
7885			    SPRI_TM_WR_EX_RO |
7886			    SPRI_TM_EX_AC |
7887			    SPRI_TM_WR_EX |
7888			    SPRI_TM_EX_AC_AR;
7889		scsi_ulto2b(type_mask, res_cap->type_mask);
7890		break;
7891	}
7892	case SPRI_RS: { // read full status
7893		struct scsi_per_res_in_full *res_status;
7894		struct scsi_per_res_in_full_desc *res_desc;
7895		struct ctl_port *port;
7896		int i, len;
7897
7898		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7899
7900		/*
7901		 * We had to drop the lock to allocate our buffer, which
7902		 * leaves time for someone to come in with another
7903		 * persistent reservation.  (That is unlikely, though,
7904		 * since this should be the only persistent reservation
7905		 * command active right now.)
7906		 */
7907		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7908		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7909		     lun->pr_key_count)){
7910			mtx_unlock(&lun->lun_lock);
7911			free(ctsio->kern_data_ptr, M_CTL);
7912			printf("%s: reservation length changed, retrying\n",
7913			       __func__);
7914			goto retry;
7915		}
7916
7917		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7918
7919		res_desc = &res_status->desc[0];
7920		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7921			if ((key = ctl_get_prkey(lun, i)) == 0)
7922				continue;
7923
7924			scsi_u64to8b(key, res_desc->res_key.key);
7925			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7926			    (lun->pr_res_idx == i ||
7927			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7928				res_desc->flags = SPRI_FULL_R_HOLDER;
7929				res_desc->scopetype = lun->res_type;
7930			}
7931			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7932			    res_desc->rel_trgt_port_id);
7933			len = 0;
7934			port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7935			if (port != NULL)
7936				len = ctl_create_iid(port,
7937				    i % CTL_MAX_INIT_PER_PORT,
7938				    res_desc->transport_id);
7939			scsi_ulto4b(len, res_desc->additional_length);
7940			res_desc = (struct scsi_per_res_in_full_desc *)
7941			    &res_desc->transport_id[len];
7942		}
7943		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7944		    res_status->header.length);
7945		break;
7946	}
7947	default:
7948		/*
7949		 * This is a bug, because we just checked for this above,
7950		 * and should have returned an error.
7951		 */
7952		panic("Invalid PR type %x", cdb->action);
7953		break; /* NOTREACHED */
7954	}
7955	mtx_unlock(&lun->lun_lock);
7956
7957	ctl_set_success(ctsio);
7958	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7959	ctsio->be_move_done = ctl_config_move_done;
7960	ctl_datamove((union ctl_io *)ctsio);
7961	return (CTL_RETVAL_COMPLETE);
7962}
7963
7964/*
7965 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7966 * it should return.
7967 */
7968static int
7969ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7970		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7971		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7972		struct scsi_per_res_out_parms* param)
7973{
7974	union ctl_ha_msg persis_io;
7975	int i;
7976
7977	mtx_lock(&lun->lun_lock);
7978	if (sa_res_key == 0) {
7979		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7980			/* validate scope and type */
7981			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7982			     SPR_LU_SCOPE) {
7983				mtx_unlock(&lun->lun_lock);
7984				ctl_set_invalid_field(/*ctsio*/ ctsio,
7985						      /*sks_valid*/ 1,
7986						      /*command*/ 1,
7987						      /*field*/ 2,
7988						      /*bit_valid*/ 1,
7989						      /*bit*/ 4);
7990				ctl_done((union ctl_io *)ctsio);
7991				return (1);
7992			}
7993
7994		        if (type>8 || type==2 || type==4 || type==0) {
7995				mtx_unlock(&lun->lun_lock);
7996				ctl_set_invalid_field(/*ctsio*/ ctsio,
7997       	           				      /*sks_valid*/ 1,
7998						      /*command*/ 1,
7999						      /*field*/ 2,
8000						      /*bit_valid*/ 1,
8001						      /*bit*/ 0);
8002				ctl_done((union ctl_io *)ctsio);
8003				return (1);
8004		        }
8005
8006			/*
8007			 * Unregister everybody else and build UA for
8008			 * them
8009			 */
8010			for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8011				if (i == residx || ctl_get_prkey(lun, i) == 0)
8012					continue;
8013
8014				ctl_clr_prkey(lun, i);
8015				ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8016			}
8017			lun->pr_key_count = 1;
8018			lun->res_type = type;
8019			if (lun->res_type != SPR_TYPE_WR_EX_AR
8020			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8021				lun->pr_res_idx = residx;
8022			lun->PRGeneration++;
8023			mtx_unlock(&lun->lun_lock);
8024
8025			/* send msg to other side */
8026			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8027			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8028			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8029			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8030			persis_io.pr.pr_info.res_type = type;
8031			memcpy(persis_io.pr.pr_info.sa_res_key,
8032			       param->serv_act_res_key,
8033			       sizeof(param->serv_act_res_key));
8034			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8035			    sizeof(persis_io.pr), M_WAITOK);
8036		} else {
8037			/* not all registrants */
8038			mtx_unlock(&lun->lun_lock);
8039			free(ctsio->kern_data_ptr, M_CTL);
8040			ctl_set_invalid_field(ctsio,
8041					      /*sks_valid*/ 1,
8042					      /*command*/ 0,
8043					      /*field*/ 8,
8044					      /*bit_valid*/ 0,
8045					      /*bit*/ 0);
8046			ctl_done((union ctl_io *)ctsio);
8047			return (1);
8048		}
8049	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8050		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8051		int found = 0;
8052
8053		if (res_key == sa_res_key) {
8054			/* special case */
8055			/*
8056			 * The spec implies this is not good but doesn't
8057			 * say what to do. There are two choices either
8058			 * generate a res conflict or check condition
8059			 * with illegal field in parameter data. Since
8060			 * that is what is done when the sa_res_key is
8061			 * zero I'll take that approach since this has
8062			 * to do with the sa_res_key.
8063			 */
8064			mtx_unlock(&lun->lun_lock);
8065			free(ctsio->kern_data_ptr, M_CTL);
8066			ctl_set_invalid_field(ctsio,
8067					      /*sks_valid*/ 1,
8068					      /*command*/ 0,
8069					      /*field*/ 8,
8070					      /*bit_valid*/ 0,
8071					      /*bit*/ 0);
8072			ctl_done((union ctl_io *)ctsio);
8073			return (1);
8074		}
8075
8076		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8077			if (ctl_get_prkey(lun, i) != sa_res_key)
8078				continue;
8079
8080			found = 1;
8081			ctl_clr_prkey(lun, i);
8082			lun->pr_key_count--;
8083			ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8084		}
8085		if (!found) {
8086			mtx_unlock(&lun->lun_lock);
8087			free(ctsio->kern_data_ptr, M_CTL);
8088			ctl_set_reservation_conflict(ctsio);
8089			ctl_done((union ctl_io *)ctsio);
8090			return (CTL_RETVAL_COMPLETE);
8091		}
8092		lun->PRGeneration++;
8093		mtx_unlock(&lun->lun_lock);
8094
8095		/* send msg to other side */
8096		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8097		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8098		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8099		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8100		persis_io.pr.pr_info.res_type = type;
8101		memcpy(persis_io.pr.pr_info.sa_res_key,
8102		       param->serv_act_res_key,
8103		       sizeof(param->serv_act_res_key));
8104		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8105		    sizeof(persis_io.pr), M_WAITOK);
8106	} else {
8107		/* Reserved but not all registrants */
8108		/* sa_res_key is res holder */
8109		if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
8110			/* validate scope and type */
8111			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8112			     SPR_LU_SCOPE) {
8113				mtx_unlock(&lun->lun_lock);
8114				ctl_set_invalid_field(/*ctsio*/ ctsio,
8115						      /*sks_valid*/ 1,
8116						      /*command*/ 1,
8117						      /*field*/ 2,
8118						      /*bit_valid*/ 1,
8119						      /*bit*/ 4);
8120				ctl_done((union ctl_io *)ctsio);
8121				return (1);
8122			}
8123
8124			if (type>8 || type==2 || type==4 || type==0) {
8125				mtx_unlock(&lun->lun_lock);
8126				ctl_set_invalid_field(/*ctsio*/ ctsio,
8127						      /*sks_valid*/ 1,
8128						      /*command*/ 1,
8129						      /*field*/ 2,
8130						      /*bit_valid*/ 1,
8131						      /*bit*/ 0);
8132				ctl_done((union ctl_io *)ctsio);
8133				return (1);
8134			}
8135
8136			/*
8137			 * Do the following:
8138			 * if sa_res_key != res_key remove all
8139			 * registrants w/sa_res_key and generate UA
8140			 * for these registrants(Registrations
8141			 * Preempted) if it wasn't an exclusive
8142			 * reservation generate UA(Reservations
8143			 * Preempted) for all other registered nexuses
8144			 * if the type has changed. Establish the new
8145			 * reservation and holder. If res_key and
8146			 * sa_res_key are the same do the above
8147			 * except don't unregister the res holder.
8148			 */
8149
8150			for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8151				if (i == residx || ctl_get_prkey(lun, i) == 0)
8152					continue;
8153
8154				if (sa_res_key == ctl_get_prkey(lun, i)) {
8155					ctl_clr_prkey(lun, i);
8156					lun->pr_key_count--;
8157					ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8158				} else if (type != lun->res_type
8159					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8160					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8161					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8162				}
8163			}
8164			lun->res_type = type;
8165			if (lun->res_type != SPR_TYPE_WR_EX_AR
8166			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8167				lun->pr_res_idx = residx;
8168			else
8169				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8170			lun->PRGeneration++;
8171			mtx_unlock(&lun->lun_lock);
8172
8173			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8174			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8175			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8176			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8177			persis_io.pr.pr_info.res_type = type;
8178			memcpy(persis_io.pr.pr_info.sa_res_key,
8179			       param->serv_act_res_key,
8180			       sizeof(param->serv_act_res_key));
8181			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8182			    sizeof(persis_io.pr), M_WAITOK);
8183		} else {
8184			/*
8185			 * sa_res_key is not the res holder just
8186			 * remove registrants
8187			 */
8188			int found=0;
8189
8190			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8191				if (sa_res_key != ctl_get_prkey(lun, i))
8192					continue;
8193
8194				found = 1;
8195				ctl_clr_prkey(lun, i);
8196				lun->pr_key_count--;
8197				ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8198			}
8199
8200			if (!found) {
8201				mtx_unlock(&lun->lun_lock);
8202				free(ctsio->kern_data_ptr, M_CTL);
8203				ctl_set_reservation_conflict(ctsio);
8204				ctl_done((union ctl_io *)ctsio);
8205		        	return (1);
8206			}
8207			lun->PRGeneration++;
8208			mtx_unlock(&lun->lun_lock);
8209
8210			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8211			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8212			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8213			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8214			persis_io.pr.pr_info.res_type = type;
8215			memcpy(persis_io.pr.pr_info.sa_res_key,
8216			       param->serv_act_res_key,
8217			       sizeof(param->serv_act_res_key));
8218			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8219			    sizeof(persis_io.pr), M_WAITOK);
8220		}
8221	}
8222	return (0);
8223}
8224
8225static void
8226ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8227{
8228	uint64_t sa_res_key;
8229	int i;
8230
8231	sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8232
8233	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8234	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8235	 || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8236		if (sa_res_key == 0) {
8237			/*
8238			 * Unregister everybody else and build UA for
8239			 * them
8240			 */
8241			for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8242				if (i == msg->pr.pr_info.residx ||
8243				    ctl_get_prkey(lun, i) == 0)
8244					continue;
8245
8246				ctl_clr_prkey(lun, i);
8247				ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8248			}
8249
8250			lun->pr_key_count = 1;
8251			lun->res_type = msg->pr.pr_info.res_type;
8252			if (lun->res_type != SPR_TYPE_WR_EX_AR
8253			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8254				lun->pr_res_idx = msg->pr.pr_info.residx;
8255		} else {
8256		        for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8257				if (sa_res_key == ctl_get_prkey(lun, i))
8258					continue;
8259
8260				ctl_clr_prkey(lun, i);
8261				lun->pr_key_count--;
8262				ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8263			}
8264		}
8265	} else {
8266		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8267			if (i == msg->pr.pr_info.residx ||
8268			    ctl_get_prkey(lun, i) == 0)
8269				continue;
8270
8271			if (sa_res_key == ctl_get_prkey(lun, i)) {
8272				ctl_clr_prkey(lun, i);
8273				lun->pr_key_count--;
8274				ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8275			} else if (msg->pr.pr_info.res_type != lun->res_type
8276				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8277				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8278				ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8279			}
8280		}
8281		lun->res_type = msg->pr.pr_info.res_type;
8282		if (lun->res_type != SPR_TYPE_WR_EX_AR
8283		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8284			lun->pr_res_idx = msg->pr.pr_info.residx;
8285		else
8286			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8287	}
8288	lun->PRGeneration++;
8289
8290}
8291
8292
8293int
8294ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8295{
8296	int retval;
8297	u_int32_t param_len;
8298	struct scsi_per_res_out *cdb;
8299	struct ctl_lun *lun;
8300	struct scsi_per_res_out_parms* param;
8301	struct ctl_softc *softc;
8302	uint32_t residx;
8303	uint64_t res_key, sa_res_key, key;
8304	uint8_t type;
8305	union ctl_ha_msg persis_io;
8306	int    i;
8307
8308	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8309
8310	retval = CTL_RETVAL_COMPLETE;
8311
8312	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8313	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8314	softc = lun->ctl_softc;
8315
8316	/*
8317	 * We only support whole-LUN scope.  The scope & type are ignored for
8318	 * register, register and ignore existing key and clear.
8319	 * We sometimes ignore scope and type on preempts too!!
8320	 * Verify reservation type here as well.
8321	 */
8322	type = cdb->scope_type & SPR_TYPE_MASK;
8323	if ((cdb->action == SPRO_RESERVE)
8324	 || (cdb->action == SPRO_RELEASE)) {
8325		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8326			ctl_set_invalid_field(/*ctsio*/ ctsio,
8327					      /*sks_valid*/ 1,
8328					      /*command*/ 1,
8329					      /*field*/ 2,
8330					      /*bit_valid*/ 1,
8331					      /*bit*/ 4);
8332			ctl_done((union ctl_io *)ctsio);
8333			return (CTL_RETVAL_COMPLETE);
8334		}
8335
8336		if (type>8 || type==2 || type==4 || type==0) {
8337			ctl_set_invalid_field(/*ctsio*/ ctsio,
8338					      /*sks_valid*/ 1,
8339					      /*command*/ 1,
8340					      /*field*/ 2,
8341					      /*bit_valid*/ 1,
8342					      /*bit*/ 0);
8343			ctl_done((union ctl_io *)ctsio);
8344			return (CTL_RETVAL_COMPLETE);
8345		}
8346	}
8347
8348	param_len = scsi_4btoul(cdb->length);
8349
8350	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8351		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8352		ctsio->kern_data_len = param_len;
8353		ctsio->kern_total_len = param_len;
8354		ctsio->kern_data_resid = 0;
8355		ctsio->kern_rel_offset = 0;
8356		ctsio->kern_sg_entries = 0;
8357		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8358		ctsio->be_move_done = ctl_config_move_done;
8359		ctl_datamove((union ctl_io *)ctsio);
8360
8361		return (CTL_RETVAL_COMPLETE);
8362	}
8363
8364	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8365
8366	residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8367	res_key = scsi_8btou64(param->res_key.key);
8368	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8369
8370	/*
8371	 * Validate the reservation key here except for SPRO_REG_IGNO
8372	 * This must be done for all other service actions
8373	 */
8374	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8375		mtx_lock(&lun->lun_lock);
8376		if ((key = ctl_get_prkey(lun, residx)) != 0) {
8377			if (res_key != key) {
8378				/*
8379				 * The current key passed in doesn't match
8380				 * the one the initiator previously
8381				 * registered.
8382				 */
8383				mtx_unlock(&lun->lun_lock);
8384				free(ctsio->kern_data_ptr, M_CTL);
8385				ctl_set_reservation_conflict(ctsio);
8386				ctl_done((union ctl_io *)ctsio);
8387				return (CTL_RETVAL_COMPLETE);
8388			}
8389		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8390			/*
8391			 * We are not registered
8392			 */
8393			mtx_unlock(&lun->lun_lock);
8394			free(ctsio->kern_data_ptr, M_CTL);
8395			ctl_set_reservation_conflict(ctsio);
8396			ctl_done((union ctl_io *)ctsio);
8397			return (CTL_RETVAL_COMPLETE);
8398		} else if (res_key != 0) {
8399			/*
8400			 * We are not registered and trying to register but
8401			 * the register key isn't zero.
8402			 */
8403			mtx_unlock(&lun->lun_lock);
8404			free(ctsio->kern_data_ptr, M_CTL);
8405			ctl_set_reservation_conflict(ctsio);
8406			ctl_done((union ctl_io *)ctsio);
8407			return (CTL_RETVAL_COMPLETE);
8408		}
8409		mtx_unlock(&lun->lun_lock);
8410	}
8411
8412	switch (cdb->action & SPRO_ACTION_MASK) {
8413	case SPRO_REGISTER:
8414	case SPRO_REG_IGNO: {
8415
8416#if 0
8417		printf("Registration received\n");
8418#endif
8419
8420		/*
8421		 * We don't support any of these options, as we report in
8422		 * the read capabilities request (see
8423		 * ctl_persistent_reserve_in(), above).
8424		 */
8425		if ((param->flags & SPR_SPEC_I_PT)
8426		 || (param->flags & SPR_ALL_TG_PT)
8427		 || (param->flags & SPR_APTPL)) {
8428			int bit_ptr;
8429
8430			if (param->flags & SPR_APTPL)
8431				bit_ptr = 0;
8432			else if (param->flags & SPR_ALL_TG_PT)
8433				bit_ptr = 2;
8434			else /* SPR_SPEC_I_PT */
8435				bit_ptr = 3;
8436
8437			free(ctsio->kern_data_ptr, M_CTL);
8438			ctl_set_invalid_field(ctsio,
8439					      /*sks_valid*/ 1,
8440					      /*command*/ 0,
8441					      /*field*/ 20,
8442					      /*bit_valid*/ 1,
8443					      /*bit*/ bit_ptr);
8444			ctl_done((union ctl_io *)ctsio);
8445			return (CTL_RETVAL_COMPLETE);
8446		}
8447
8448		mtx_lock(&lun->lun_lock);
8449
8450		/*
8451		 * The initiator wants to clear the
8452		 * key/unregister.
8453		 */
8454		if (sa_res_key == 0) {
8455			if ((res_key == 0
8456			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8457			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8458			  && ctl_get_prkey(lun, residx) == 0)) {
8459				mtx_unlock(&lun->lun_lock);
8460				goto done;
8461			}
8462
8463			ctl_clr_prkey(lun, residx);
8464			lun->pr_key_count--;
8465
8466			if (residx == lun->pr_res_idx) {
8467				lun->flags &= ~CTL_LUN_PR_RESERVED;
8468				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8469
8470				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8471				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8472				 && lun->pr_key_count) {
8473					/*
8474					 * If the reservation is a registrants
8475					 * only type we need to generate a UA
8476					 * for other registered inits.  The
8477					 * sense code should be RESERVATIONS
8478					 * RELEASED
8479					 */
8480
8481					for (i = softc->init_min; i < softc->init_max; i++){
8482						if (ctl_get_prkey(lun, i) == 0)
8483							continue;
8484						ctl_est_ua(lun, i,
8485						    CTL_UA_RES_RELEASE);
8486					}
8487				}
8488				lun->res_type = 0;
8489			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8490				if (lun->pr_key_count==0) {
8491					lun->flags &= ~CTL_LUN_PR_RESERVED;
8492					lun->res_type = 0;
8493					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8494				}
8495			}
8496			lun->PRGeneration++;
8497			mtx_unlock(&lun->lun_lock);
8498
8499			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8500			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8501			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8502			persis_io.pr.pr_info.residx = residx;
8503			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8504			    sizeof(persis_io.pr), M_WAITOK);
8505		} else /* sa_res_key != 0 */ {
8506
8507			/*
8508			 * If we aren't registered currently then increment
8509			 * the key count and set the registered flag.
8510			 */
8511			ctl_alloc_prkey(lun, residx);
8512			if (ctl_get_prkey(lun, residx) == 0)
8513				lun->pr_key_count++;
8514			ctl_set_prkey(lun, residx, sa_res_key);
8515			lun->PRGeneration++;
8516			mtx_unlock(&lun->lun_lock);
8517
8518			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8519			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8520			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8521			persis_io.pr.pr_info.residx = residx;
8522			memcpy(persis_io.pr.pr_info.sa_res_key,
8523			       param->serv_act_res_key,
8524			       sizeof(param->serv_act_res_key));
8525			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8526			    sizeof(persis_io.pr), M_WAITOK);
8527		}
8528
8529		break;
8530	}
8531	case SPRO_RESERVE:
8532#if 0
8533                printf("Reserve executed type %d\n", type);
8534#endif
8535		mtx_lock(&lun->lun_lock);
8536		if (lun->flags & CTL_LUN_PR_RESERVED) {
8537			/*
8538			 * if this isn't the reservation holder and it's
8539			 * not a "all registrants" type or if the type is
8540			 * different then we have a conflict
8541			 */
8542			if ((lun->pr_res_idx != residx
8543			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8544			 || lun->res_type != type) {
8545				mtx_unlock(&lun->lun_lock);
8546				free(ctsio->kern_data_ptr, M_CTL);
8547				ctl_set_reservation_conflict(ctsio);
8548				ctl_done((union ctl_io *)ctsio);
8549				return (CTL_RETVAL_COMPLETE);
8550			}
8551			mtx_unlock(&lun->lun_lock);
8552		} else /* create a reservation */ {
8553			/*
8554			 * If it's not an "all registrants" type record
8555			 * reservation holder
8556			 */
8557			if (type != SPR_TYPE_WR_EX_AR
8558			 && type != SPR_TYPE_EX_AC_AR)
8559				lun->pr_res_idx = residx; /* Res holder */
8560			else
8561				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8562
8563			lun->flags |= CTL_LUN_PR_RESERVED;
8564			lun->res_type = type;
8565
8566			mtx_unlock(&lun->lun_lock);
8567
8568			/* send msg to other side */
8569			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8570			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8571			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8572			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8573			persis_io.pr.pr_info.res_type = type;
8574			ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8575			    sizeof(persis_io.pr), M_WAITOK);
8576		}
8577		break;
8578
8579	case SPRO_RELEASE:
8580		mtx_lock(&lun->lun_lock);
8581		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8582			/* No reservation exists return good status */
8583			mtx_unlock(&lun->lun_lock);
8584			goto done;
8585		}
8586		/*
8587		 * Is this nexus a reservation holder?
8588		 */
8589		if (lun->pr_res_idx != residx
8590		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8591			/*
8592			 * not a res holder return good status but
8593			 * do nothing
8594			 */
8595			mtx_unlock(&lun->lun_lock);
8596			goto done;
8597		}
8598
8599		if (lun->res_type != type) {
8600			mtx_unlock(&lun->lun_lock);
8601			free(ctsio->kern_data_ptr, M_CTL);
8602			ctl_set_illegal_pr_release(ctsio);
8603			ctl_done((union ctl_io *)ctsio);
8604			return (CTL_RETVAL_COMPLETE);
8605		}
8606
8607		/* okay to release */
8608		lun->flags &= ~CTL_LUN_PR_RESERVED;
8609		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8610		lun->res_type = 0;
8611
8612		/*
8613		 * if this isn't an exclusive access
8614		 * res generate UA for all other
8615		 * registrants.
8616		 */
8617		if (type != SPR_TYPE_EX_AC
8618		 && type != SPR_TYPE_WR_EX) {
8619			for (i = softc->init_min; i < softc->init_max; i++) {
8620				if (i == residx || ctl_get_prkey(lun, i) == 0)
8621					continue;
8622				ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8623			}
8624		}
8625		mtx_unlock(&lun->lun_lock);
8626
8627		/* Send msg to other side */
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_RELEASE;
8631		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8632		     sizeof(persis_io.pr), M_WAITOK);
8633		break;
8634
8635	case SPRO_CLEAR:
8636		/* send msg to other side */
8637
8638		mtx_lock(&lun->lun_lock);
8639		lun->flags &= ~CTL_LUN_PR_RESERVED;
8640		lun->res_type = 0;
8641		lun->pr_key_count = 0;
8642		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8643
8644		ctl_clr_prkey(lun, residx);
8645		for (i = 0; i < CTL_MAX_INITIATORS; i++)
8646			if (ctl_get_prkey(lun, i) != 0) {
8647				ctl_clr_prkey(lun, i);
8648				ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8649			}
8650		lun->PRGeneration++;
8651		mtx_unlock(&lun->lun_lock);
8652
8653		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8654		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8655		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8656		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8657		     sizeof(persis_io.pr), M_WAITOK);
8658		break;
8659
8660	case SPRO_PREEMPT:
8661	case SPRO_PRE_ABO: {
8662		int nretval;
8663
8664		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8665					  residx, ctsio, cdb, param);
8666		if (nretval != 0)
8667			return (CTL_RETVAL_COMPLETE);
8668		break;
8669	}
8670	default:
8671		panic("Invalid PR type %x", cdb->action);
8672	}
8673
8674done:
8675	free(ctsio->kern_data_ptr, M_CTL);
8676	ctl_set_success(ctsio);
8677	ctl_done((union ctl_io *)ctsio);
8678
8679	return (retval);
8680}
8681
8682/*
8683 * This routine is for handling a message from the other SC pertaining to
8684 * persistent reserve out. All the error checking will have been done
8685 * so only perorming the action need be done here to keep the two
8686 * in sync.
8687 */
8688static void
8689ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8690{
8691	struct ctl_softc *softc = control_softc;
8692	struct ctl_lun *lun;
8693	int i;
8694	uint32_t residx, targ_lun;
8695
8696	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8697	mtx_lock(&softc->ctl_lock);
8698	if ((targ_lun >= CTL_MAX_LUNS) ||
8699	    ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
8700		mtx_unlock(&softc->ctl_lock);
8701		return;
8702	}
8703	mtx_lock(&lun->lun_lock);
8704	mtx_unlock(&softc->ctl_lock);
8705	if (lun->flags & CTL_LUN_DISABLED) {
8706		mtx_unlock(&lun->lun_lock);
8707		return;
8708	}
8709	residx = ctl_get_initindex(&msg->hdr.nexus);
8710	switch(msg->pr.pr_info.action) {
8711	case CTL_PR_REG_KEY:
8712		ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8713		if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8714			lun->pr_key_count++;
8715		ctl_set_prkey(lun, msg->pr.pr_info.residx,
8716		    scsi_8btou64(msg->pr.pr_info.sa_res_key));
8717		lun->PRGeneration++;
8718		break;
8719
8720	case CTL_PR_UNREG_KEY:
8721		ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8722		lun->pr_key_count--;
8723
8724		/* XXX Need to see if the reservation has been released */
8725		/* if so do we need to generate UA? */
8726		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8727			lun->flags &= ~CTL_LUN_PR_RESERVED;
8728			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8729
8730			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8731			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8732			 && lun->pr_key_count) {
8733				/*
8734				 * If the reservation is a registrants
8735				 * only type we need to generate a UA
8736				 * for other registered inits.  The
8737				 * sense code should be RESERVATIONS
8738				 * RELEASED
8739				 */
8740
8741				for (i = softc->init_min; i < softc->init_max; i++) {
8742					if (ctl_get_prkey(lun, i) == 0)
8743						continue;
8744
8745					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8746				}
8747			}
8748			lun->res_type = 0;
8749		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8750			if (lun->pr_key_count==0) {
8751				lun->flags &= ~CTL_LUN_PR_RESERVED;
8752				lun->res_type = 0;
8753				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8754			}
8755		}
8756		lun->PRGeneration++;
8757		break;
8758
8759	case CTL_PR_RESERVE:
8760		lun->flags |= CTL_LUN_PR_RESERVED;
8761		lun->res_type = msg->pr.pr_info.res_type;
8762		lun->pr_res_idx = msg->pr.pr_info.residx;
8763
8764		break;
8765
8766	case CTL_PR_RELEASE:
8767		/*
8768		 * if this isn't an exclusive access res generate UA for all
8769		 * other registrants.
8770		 */
8771		if (lun->res_type != SPR_TYPE_EX_AC
8772		 && lun->res_type != SPR_TYPE_WR_EX) {
8773			for (i = softc->init_min; i < softc->init_max; i++)
8774				if (i == residx || ctl_get_prkey(lun, i) == 0)
8775					continue;
8776				ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8777		}
8778
8779		lun->flags &= ~CTL_LUN_PR_RESERVED;
8780		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8781		lun->res_type = 0;
8782		break;
8783
8784	case CTL_PR_PREEMPT:
8785		ctl_pro_preempt_other(lun, msg);
8786		break;
8787	case CTL_PR_CLEAR:
8788		lun->flags &= ~CTL_LUN_PR_RESERVED;
8789		lun->res_type = 0;
8790		lun->pr_key_count = 0;
8791		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8792
8793		for (i=0; i < CTL_MAX_INITIATORS; i++) {
8794			if (ctl_get_prkey(lun, i) == 0)
8795				continue;
8796			ctl_clr_prkey(lun, i);
8797			ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8798		}
8799		lun->PRGeneration++;
8800		break;
8801	}
8802
8803	mtx_unlock(&lun->lun_lock);
8804}
8805
8806int
8807ctl_read_write(struct ctl_scsiio *ctsio)
8808{
8809	struct ctl_lun *lun;
8810	struct ctl_lba_len_flags *lbalen;
8811	uint64_t lba;
8812	uint32_t num_blocks;
8813	int flags, retval;
8814	int isread;
8815
8816	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8817
8818	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8819
8820	flags = 0;
8821	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8822	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8823	switch (ctsio->cdb[0]) {
8824	case READ_6:
8825	case WRITE_6: {
8826		struct scsi_rw_6 *cdb;
8827
8828		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8829
8830		lba = scsi_3btoul(cdb->addr);
8831		/* only 5 bits are valid in the most significant address byte */
8832		lba &= 0x1fffff;
8833		num_blocks = cdb->length;
8834		/*
8835		 * This is correct according to SBC-2.
8836		 */
8837		if (num_blocks == 0)
8838			num_blocks = 256;
8839		break;
8840	}
8841	case READ_10:
8842	case WRITE_10: {
8843		struct scsi_rw_10 *cdb;
8844
8845		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8846		if (cdb->byte2 & SRW10_FUA)
8847			flags |= CTL_LLF_FUA;
8848		if (cdb->byte2 & SRW10_DPO)
8849			flags |= CTL_LLF_DPO;
8850		lba = scsi_4btoul(cdb->addr);
8851		num_blocks = scsi_2btoul(cdb->length);
8852		break;
8853	}
8854	case WRITE_VERIFY_10: {
8855		struct scsi_write_verify_10 *cdb;
8856
8857		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8858		flags |= CTL_LLF_FUA;
8859		if (cdb->byte2 & SWV_DPO)
8860			flags |= CTL_LLF_DPO;
8861		lba = scsi_4btoul(cdb->addr);
8862		num_blocks = scsi_2btoul(cdb->length);
8863		break;
8864	}
8865	case READ_12:
8866	case WRITE_12: {
8867		struct scsi_rw_12 *cdb;
8868
8869		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8870		if (cdb->byte2 & SRW12_FUA)
8871			flags |= CTL_LLF_FUA;
8872		if (cdb->byte2 & SRW12_DPO)
8873			flags |= CTL_LLF_DPO;
8874		lba = scsi_4btoul(cdb->addr);
8875		num_blocks = scsi_4btoul(cdb->length);
8876		break;
8877	}
8878	case WRITE_VERIFY_12: {
8879		struct scsi_write_verify_12 *cdb;
8880
8881		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8882		flags |= CTL_LLF_FUA;
8883		if (cdb->byte2 & SWV_DPO)
8884			flags |= CTL_LLF_DPO;
8885		lba = scsi_4btoul(cdb->addr);
8886		num_blocks = scsi_4btoul(cdb->length);
8887		break;
8888	}
8889	case READ_16:
8890	case WRITE_16: {
8891		struct scsi_rw_16 *cdb;
8892
8893		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8894		if (cdb->byte2 & SRW12_FUA)
8895			flags |= CTL_LLF_FUA;
8896		if (cdb->byte2 & SRW12_DPO)
8897			flags |= CTL_LLF_DPO;
8898		lba = scsi_8btou64(cdb->addr);
8899		num_blocks = scsi_4btoul(cdb->length);
8900		break;
8901	}
8902	case WRITE_ATOMIC_16: {
8903		struct scsi_write_atomic_16 *cdb;
8904
8905		if (lun->be_lun->atomicblock == 0) {
8906			ctl_set_invalid_opcode(ctsio);
8907			ctl_done((union ctl_io *)ctsio);
8908			return (CTL_RETVAL_COMPLETE);
8909		}
8910
8911		cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8912		if (cdb->byte2 & SRW12_FUA)
8913			flags |= CTL_LLF_FUA;
8914		if (cdb->byte2 & SRW12_DPO)
8915			flags |= CTL_LLF_DPO;
8916		lba = scsi_8btou64(cdb->addr);
8917		num_blocks = scsi_2btoul(cdb->length);
8918		if (num_blocks > lun->be_lun->atomicblock) {
8919			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8920			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8921			    /*bit*/ 0);
8922			ctl_done((union ctl_io *)ctsio);
8923			return (CTL_RETVAL_COMPLETE);
8924		}
8925		break;
8926	}
8927	case WRITE_VERIFY_16: {
8928		struct scsi_write_verify_16 *cdb;
8929
8930		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8931		flags |= CTL_LLF_FUA;
8932		if (cdb->byte2 & SWV_DPO)
8933			flags |= CTL_LLF_DPO;
8934		lba = scsi_8btou64(cdb->addr);
8935		num_blocks = scsi_4btoul(cdb->length);
8936		break;
8937	}
8938	default:
8939		/*
8940		 * We got a command we don't support.  This shouldn't
8941		 * happen, commands should be filtered out above us.
8942		 */
8943		ctl_set_invalid_opcode(ctsio);
8944		ctl_done((union ctl_io *)ctsio);
8945
8946		return (CTL_RETVAL_COMPLETE);
8947		break; /* NOTREACHED */
8948	}
8949
8950	/*
8951	 * The first check is to make sure we're in bounds, the second
8952	 * check is to catch wrap-around problems.  If the lba + num blocks
8953	 * is less than the lba, then we've wrapped around and the block
8954	 * range is invalid anyway.
8955	 */
8956	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8957	 || ((lba + num_blocks) < lba)) {
8958		ctl_set_lba_out_of_range(ctsio);
8959		ctl_done((union ctl_io *)ctsio);
8960		return (CTL_RETVAL_COMPLETE);
8961	}
8962
8963	/*
8964	 * According to SBC-3, a transfer length of 0 is not an error.
8965	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8966	 * translates to 256 blocks for those commands.
8967	 */
8968	if (num_blocks == 0) {
8969		ctl_set_success(ctsio);
8970		ctl_done((union ctl_io *)ctsio);
8971		return (CTL_RETVAL_COMPLETE);
8972	}
8973
8974	/* Set FUA and/or DPO if caches are disabled. */
8975	if (isread) {
8976		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8977		    SCP_RCD) != 0)
8978			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8979	} else {
8980		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8981		    SCP_WCE) == 0)
8982			flags |= CTL_LLF_FUA;
8983	}
8984
8985	lbalen = (struct ctl_lba_len_flags *)
8986	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8987	lbalen->lba = lba;
8988	lbalen->len = num_blocks;
8989	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8990
8991	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8992	ctsio->kern_rel_offset = 0;
8993
8994	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8995
8996	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8997	return (retval);
8998}
8999
9000static int
9001ctl_cnw_cont(union ctl_io *io)
9002{
9003	struct ctl_scsiio *ctsio;
9004	struct ctl_lun *lun;
9005	struct ctl_lba_len_flags *lbalen;
9006	int retval;
9007
9008	ctsio = &io->scsiio;
9009	ctsio->io_hdr.status = CTL_STATUS_NONE;
9010	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9011	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9012	lbalen = (struct ctl_lba_len_flags *)
9013	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9014	lbalen->flags &= ~CTL_LLF_COMPARE;
9015	lbalen->flags |= CTL_LLF_WRITE;
9016
9017	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9018	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9019	return (retval);
9020}
9021
9022int
9023ctl_cnw(struct ctl_scsiio *ctsio)
9024{
9025	struct ctl_lun *lun;
9026	struct ctl_lba_len_flags *lbalen;
9027	uint64_t lba;
9028	uint32_t num_blocks;
9029	int flags, retval;
9030
9031	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9032
9033	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9034
9035	flags = 0;
9036	switch (ctsio->cdb[0]) {
9037	case COMPARE_AND_WRITE: {
9038		struct scsi_compare_and_write *cdb;
9039
9040		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9041		if (cdb->byte2 & SRW10_FUA)
9042			flags |= CTL_LLF_FUA;
9043		if (cdb->byte2 & SRW10_DPO)
9044			flags |= CTL_LLF_DPO;
9045		lba = scsi_8btou64(cdb->addr);
9046		num_blocks = cdb->length;
9047		break;
9048	}
9049	default:
9050		/*
9051		 * We got a command we don't support.  This shouldn't
9052		 * happen, commands should be filtered out above us.
9053		 */
9054		ctl_set_invalid_opcode(ctsio);
9055		ctl_done((union ctl_io *)ctsio);
9056
9057		return (CTL_RETVAL_COMPLETE);
9058		break; /* NOTREACHED */
9059	}
9060
9061	/*
9062	 * The first check is to make sure we're in bounds, the second
9063	 * check is to catch wrap-around problems.  If the lba + num blocks
9064	 * is less than the lba, then we've wrapped around and the block
9065	 * range is invalid anyway.
9066	 */
9067	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9068	 || ((lba + num_blocks) < lba)) {
9069		ctl_set_lba_out_of_range(ctsio);
9070		ctl_done((union ctl_io *)ctsio);
9071		return (CTL_RETVAL_COMPLETE);
9072	}
9073
9074	/*
9075	 * According to SBC-3, a transfer length of 0 is not an error.
9076	 */
9077	if (num_blocks == 0) {
9078		ctl_set_success(ctsio);
9079		ctl_done((union ctl_io *)ctsio);
9080		return (CTL_RETVAL_COMPLETE);
9081	}
9082
9083	/* Set FUA if write cache is disabled. */
9084	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9085	    SCP_WCE) == 0)
9086		flags |= CTL_LLF_FUA;
9087
9088	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9089	ctsio->kern_rel_offset = 0;
9090
9091	/*
9092	 * Set the IO_CONT flag, so that if this I/O gets passed to
9093	 * ctl_data_submit_done(), it'll get passed back to
9094	 * ctl_ctl_cnw_cont() for further processing.
9095	 */
9096	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9097	ctsio->io_cont = ctl_cnw_cont;
9098
9099	lbalen = (struct ctl_lba_len_flags *)
9100	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9101	lbalen->lba = lba;
9102	lbalen->len = num_blocks;
9103	lbalen->flags = CTL_LLF_COMPARE | flags;
9104
9105	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9106	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9107	return (retval);
9108}
9109
9110int
9111ctl_verify(struct ctl_scsiio *ctsio)
9112{
9113	struct ctl_lun *lun;
9114	struct ctl_lba_len_flags *lbalen;
9115	uint64_t lba;
9116	uint32_t num_blocks;
9117	int bytchk, flags;
9118	int retval;
9119
9120	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9121
9122	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9123
9124	bytchk = 0;
9125	flags = CTL_LLF_FUA;
9126	switch (ctsio->cdb[0]) {
9127	case VERIFY_10: {
9128		struct scsi_verify_10 *cdb;
9129
9130		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9131		if (cdb->byte2 & SVFY_BYTCHK)
9132			bytchk = 1;
9133		if (cdb->byte2 & SVFY_DPO)
9134			flags |= CTL_LLF_DPO;
9135		lba = scsi_4btoul(cdb->addr);
9136		num_blocks = scsi_2btoul(cdb->length);
9137		break;
9138	}
9139	case VERIFY_12: {
9140		struct scsi_verify_12 *cdb;
9141
9142		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9143		if (cdb->byte2 & SVFY_BYTCHK)
9144			bytchk = 1;
9145		if (cdb->byte2 & SVFY_DPO)
9146			flags |= CTL_LLF_DPO;
9147		lba = scsi_4btoul(cdb->addr);
9148		num_blocks = scsi_4btoul(cdb->length);
9149		break;
9150	}
9151	case VERIFY_16: {
9152		struct scsi_rw_16 *cdb;
9153
9154		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9155		if (cdb->byte2 & SVFY_BYTCHK)
9156			bytchk = 1;
9157		if (cdb->byte2 & SVFY_DPO)
9158			flags |= CTL_LLF_DPO;
9159		lba = scsi_8btou64(cdb->addr);
9160		num_blocks = scsi_4btoul(cdb->length);
9161		break;
9162	}
9163	default:
9164		/*
9165		 * We got a command we don't support.  This shouldn't
9166		 * happen, commands should be filtered out above us.
9167		 */
9168		ctl_set_invalid_opcode(ctsio);
9169		ctl_done((union ctl_io *)ctsio);
9170		return (CTL_RETVAL_COMPLETE);
9171	}
9172
9173	/*
9174	 * The first check is to make sure we're in bounds, the second
9175	 * check is to catch wrap-around problems.  If the lba + num blocks
9176	 * is less than the lba, then we've wrapped around and the block
9177	 * range is invalid anyway.
9178	 */
9179	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9180	 || ((lba + num_blocks) < lba)) {
9181		ctl_set_lba_out_of_range(ctsio);
9182		ctl_done((union ctl_io *)ctsio);
9183		return (CTL_RETVAL_COMPLETE);
9184	}
9185
9186	/*
9187	 * According to SBC-3, a transfer length of 0 is not an error.
9188	 */
9189	if (num_blocks == 0) {
9190		ctl_set_success(ctsio);
9191		ctl_done((union ctl_io *)ctsio);
9192		return (CTL_RETVAL_COMPLETE);
9193	}
9194
9195	lbalen = (struct ctl_lba_len_flags *)
9196	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9197	lbalen->lba = lba;
9198	lbalen->len = num_blocks;
9199	if (bytchk) {
9200		lbalen->flags = CTL_LLF_COMPARE | flags;
9201		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9202	} else {
9203		lbalen->flags = CTL_LLF_VERIFY | flags;
9204		ctsio->kern_total_len = 0;
9205	}
9206	ctsio->kern_rel_offset = 0;
9207
9208	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9209	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9210	return (retval);
9211}
9212
9213int
9214ctl_report_luns(struct ctl_scsiio *ctsio)
9215{
9216	struct ctl_softc *softc;
9217	struct scsi_report_luns *cdb;
9218	struct scsi_report_luns_data *lun_data;
9219	struct ctl_lun *lun, *request_lun;
9220	struct ctl_port *port;
9221	int num_luns, retval;
9222	uint32_t alloc_len, lun_datalen;
9223	int num_filled;
9224	uint32_t initidx, targ_lun_id, lun_id;
9225
9226	retval = CTL_RETVAL_COMPLETE;
9227	cdb = (struct scsi_report_luns *)ctsio->cdb;
9228	port = ctl_io_port(&ctsio->io_hdr);
9229	softc = port->ctl_softc;
9230
9231	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9232
9233	mtx_lock(&softc->ctl_lock);
9234	num_luns = 0;
9235	for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
9236		if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
9237			num_luns++;
9238	}
9239	mtx_unlock(&softc->ctl_lock);
9240
9241	switch (cdb->select_report) {
9242	case RPL_REPORT_DEFAULT:
9243	case RPL_REPORT_ALL:
9244	case RPL_REPORT_NONSUBSID:
9245		break;
9246	case RPL_REPORT_WELLKNOWN:
9247	case RPL_REPORT_ADMIN:
9248	case RPL_REPORT_CONGLOM:
9249		num_luns = 0;
9250		break;
9251	default:
9252		ctl_set_invalid_field(ctsio,
9253				      /*sks_valid*/ 1,
9254				      /*command*/ 1,
9255				      /*field*/ 2,
9256				      /*bit_valid*/ 0,
9257				      /*bit*/ 0);
9258		ctl_done((union ctl_io *)ctsio);
9259		return (retval);
9260		break; /* NOTREACHED */
9261	}
9262
9263	alloc_len = scsi_4btoul(cdb->length);
9264	/*
9265	 * The initiator has to allocate at least 16 bytes for this request,
9266	 * so he can at least get the header and the first LUN.  Otherwise
9267	 * we reject the request (per SPC-3 rev 14, section 6.21).
9268	 */
9269	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9270	    sizeof(struct scsi_report_luns_lundata))) {
9271		ctl_set_invalid_field(ctsio,
9272				      /*sks_valid*/ 1,
9273				      /*command*/ 1,
9274				      /*field*/ 6,
9275				      /*bit_valid*/ 0,
9276				      /*bit*/ 0);
9277		ctl_done((union ctl_io *)ctsio);
9278		return (retval);
9279	}
9280
9281	request_lun = (struct ctl_lun *)
9282		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9283
9284	lun_datalen = sizeof(*lun_data) +
9285		(num_luns * sizeof(struct scsi_report_luns_lundata));
9286
9287	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9288	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9289	ctsio->kern_sg_entries = 0;
9290
9291	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9292
9293	mtx_lock(&softc->ctl_lock);
9294	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9295		lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9296		if (lun_id >= CTL_MAX_LUNS)
9297			continue;
9298		lun = softc->ctl_luns[lun_id];
9299		if (lun == NULL)
9300			continue;
9301
9302		if (targ_lun_id <= 0xff) {
9303			/*
9304			 * Peripheral addressing method, bus number 0.
9305			 */
9306			lun_data->luns[num_filled].lundata[0] =
9307				RPL_LUNDATA_ATYP_PERIPH;
9308			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9309			num_filled++;
9310		} else if (targ_lun_id <= 0x3fff) {
9311			/*
9312			 * Flat addressing method.
9313			 */
9314			lun_data->luns[num_filled].lundata[0] =
9315				RPL_LUNDATA_ATYP_FLAT | (targ_lun_id >> 8);
9316			lun_data->luns[num_filled].lundata[1] =
9317				(targ_lun_id & 0xff);
9318			num_filled++;
9319		} else if (targ_lun_id <= 0xffffff) {
9320			/*
9321			 * Extended flat addressing method.
9322			 */
9323			lun_data->luns[num_filled].lundata[0] =
9324			    RPL_LUNDATA_ATYP_EXTLUN | 0x12;
9325			scsi_ulto3b(targ_lun_id,
9326			    &lun_data->luns[num_filled].lundata[1]);
9327			num_filled++;
9328		} else {
9329			printf("ctl_report_luns: bogus LUN number %jd, "
9330			       "skipping\n", (intmax_t)targ_lun_id);
9331		}
9332		/*
9333		 * According to SPC-3, rev 14 section 6.21:
9334		 *
9335		 * "The execution of a REPORT LUNS command to any valid and
9336		 * installed logical unit shall clear the REPORTED LUNS DATA
9337		 * HAS CHANGED unit attention condition for all logical
9338		 * units of that target with respect to the requesting
9339		 * initiator. A valid and installed logical unit is one
9340		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9341		 * INQUIRY data (see 6.4.2)."
9342		 *
9343		 * If request_lun is NULL, the LUN this report luns command
9344		 * was issued to is either disabled or doesn't exist. In that
9345		 * case, we shouldn't clear any pending lun change unit
9346		 * attention.
9347		 */
9348		if (request_lun != NULL) {
9349			mtx_lock(&lun->lun_lock);
9350			ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9351			mtx_unlock(&lun->lun_lock);
9352		}
9353	}
9354	mtx_unlock(&softc->ctl_lock);
9355
9356	/*
9357	 * It's quite possible that we've returned fewer LUNs than we allocated
9358	 * space for.  Trim it.
9359	 */
9360	lun_datalen = sizeof(*lun_data) +
9361		(num_filled * sizeof(struct scsi_report_luns_lundata));
9362
9363	if (lun_datalen < alloc_len) {
9364		ctsio->residual = alloc_len - lun_datalen;
9365		ctsio->kern_data_len = lun_datalen;
9366		ctsio->kern_total_len = lun_datalen;
9367	} else {
9368		ctsio->residual = 0;
9369		ctsio->kern_data_len = alloc_len;
9370		ctsio->kern_total_len = alloc_len;
9371	}
9372	ctsio->kern_data_resid = 0;
9373	ctsio->kern_rel_offset = 0;
9374	ctsio->kern_sg_entries = 0;
9375
9376	/*
9377	 * We set this to the actual data length, regardless of how much
9378	 * space we actually have to return results.  If the user looks at
9379	 * this value, he'll know whether or not he allocated enough space
9380	 * and reissue the command if necessary.  We don't support well
9381	 * known logical units, so if the user asks for that, return none.
9382	 */
9383	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9384
9385	/*
9386	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9387	 * this request.
9388	 */
9389	ctl_set_success(ctsio);
9390	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9391	ctsio->be_move_done = ctl_config_move_done;
9392	ctl_datamove((union ctl_io *)ctsio);
9393	return (retval);
9394}
9395
9396int
9397ctl_request_sense(struct ctl_scsiio *ctsio)
9398{
9399	struct scsi_request_sense *cdb;
9400	struct scsi_sense_data *sense_ptr;
9401	struct ctl_softc *ctl_softc;
9402	struct ctl_lun *lun;
9403	uint32_t initidx;
9404	int have_error;
9405	scsi_sense_data_type sense_format;
9406	ctl_ua_type ua_type;
9407
9408	cdb = (struct scsi_request_sense *)ctsio->cdb;
9409
9410	ctl_softc = control_softc;
9411	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9412
9413	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9414
9415	/*
9416	 * Determine which sense format the user wants.
9417	 */
9418	if (cdb->byte2 & SRS_DESC)
9419		sense_format = SSD_TYPE_DESC;
9420	else
9421		sense_format = SSD_TYPE_FIXED;
9422
9423	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9424	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9425	ctsio->kern_sg_entries = 0;
9426
9427	/*
9428	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9429	 * larger than the largest allowed value for the length field in the
9430	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9431	 */
9432	ctsio->residual = 0;
9433	ctsio->kern_data_len = cdb->length;
9434	ctsio->kern_total_len = cdb->length;
9435
9436	ctsio->kern_data_resid = 0;
9437	ctsio->kern_rel_offset = 0;
9438	ctsio->kern_sg_entries = 0;
9439
9440	/*
9441	 * If we don't have a LUN, we don't have any pending sense.
9442	 */
9443	if (lun == NULL)
9444		goto no_sense;
9445
9446	have_error = 0;
9447	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9448	/*
9449	 * Check for pending sense, and then for pending unit attentions.
9450	 * Pending sense gets returned first, then pending unit attentions.
9451	 */
9452	mtx_lock(&lun->lun_lock);
9453#ifdef CTL_WITH_CA
9454	if (ctl_is_set(lun->have_ca, initidx)) {
9455		scsi_sense_data_type stored_format;
9456
9457		/*
9458		 * Check to see which sense format was used for the stored
9459		 * sense data.
9460		 */
9461		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9462
9463		/*
9464		 * If the user requested a different sense format than the
9465		 * one we stored, then we need to convert it to the other
9466		 * format.  If we're going from descriptor to fixed format
9467		 * sense data, we may lose things in translation, depending
9468		 * on what options were used.
9469		 *
9470		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9471		 * for some reason we'll just copy it out as-is.
9472		 */
9473		if ((stored_format == SSD_TYPE_FIXED)
9474		 && (sense_format == SSD_TYPE_DESC))
9475			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9476			    &lun->pending_sense[initidx],
9477			    (struct scsi_sense_data_desc *)sense_ptr);
9478		else if ((stored_format == SSD_TYPE_DESC)
9479		      && (sense_format == SSD_TYPE_FIXED))
9480			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9481			    &lun->pending_sense[initidx],
9482			    (struct scsi_sense_data_fixed *)sense_ptr);
9483		else
9484			memcpy(sense_ptr, &lun->pending_sense[initidx],
9485			       MIN(sizeof(*sense_ptr),
9486			       sizeof(lun->pending_sense[initidx])));
9487
9488		ctl_clear_mask(lun->have_ca, initidx);
9489		have_error = 1;
9490	} else
9491#endif
9492	{
9493		ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9494		if (ua_type != CTL_UA_NONE)
9495			have_error = 1;
9496		if (ua_type == CTL_UA_LUN_CHANGE) {
9497			mtx_unlock(&lun->lun_lock);
9498			mtx_lock(&ctl_softc->ctl_lock);
9499			ctl_clr_ua_allluns(ctl_softc, initidx, ua_type);
9500			mtx_unlock(&ctl_softc->ctl_lock);
9501			mtx_lock(&lun->lun_lock);
9502		}
9503
9504	}
9505	mtx_unlock(&lun->lun_lock);
9506
9507	/*
9508	 * We already have a pending error, return it.
9509	 */
9510	if (have_error != 0) {
9511		/*
9512		 * We report the SCSI status as OK, since the status of the
9513		 * request sense command itself is OK.
9514		 * We report 0 for the sense length, because we aren't doing
9515		 * autosense in this case.  We're reporting sense as
9516		 * parameter data.
9517		 */
9518		ctl_set_success(ctsio);
9519		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9520		ctsio->be_move_done = ctl_config_move_done;
9521		ctl_datamove((union ctl_io *)ctsio);
9522		return (CTL_RETVAL_COMPLETE);
9523	}
9524
9525no_sense:
9526
9527	/*
9528	 * No sense information to report, so we report that everything is
9529	 * okay.
9530	 */
9531	ctl_set_sense_data(sense_ptr,
9532			   lun,
9533			   sense_format,
9534			   /*current_error*/ 1,
9535			   /*sense_key*/ SSD_KEY_NO_SENSE,
9536			   /*asc*/ 0x00,
9537			   /*ascq*/ 0x00,
9538			   SSD_ELEM_NONE);
9539
9540	/*
9541	 * We report 0 for the sense length, because we aren't doing
9542	 * autosense in this case.  We're reporting sense as parameter data.
9543	 */
9544	ctl_set_success(ctsio);
9545	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9546	ctsio->be_move_done = ctl_config_move_done;
9547	ctl_datamove((union ctl_io *)ctsio);
9548	return (CTL_RETVAL_COMPLETE);
9549}
9550
9551int
9552ctl_tur(struct ctl_scsiio *ctsio)
9553{
9554
9555	CTL_DEBUG_PRINT(("ctl_tur\n"));
9556
9557	ctl_set_success(ctsio);
9558	ctl_done((union ctl_io *)ctsio);
9559
9560	return (CTL_RETVAL_COMPLETE);
9561}
9562
9563/*
9564 * SCSI VPD page 0x00, the Supported VPD Pages page.
9565 */
9566static int
9567ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9568{
9569	struct scsi_vpd_supported_pages *pages;
9570	int sup_page_size;
9571	struct ctl_lun *lun;
9572	int p;
9573
9574	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9575
9576	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9577	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9578	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9579	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9580	ctsio->kern_sg_entries = 0;
9581
9582	if (sup_page_size < alloc_len) {
9583		ctsio->residual = alloc_len - sup_page_size;
9584		ctsio->kern_data_len = sup_page_size;
9585		ctsio->kern_total_len = sup_page_size;
9586	} else {
9587		ctsio->residual = 0;
9588		ctsio->kern_data_len = alloc_len;
9589		ctsio->kern_total_len = alloc_len;
9590	}
9591	ctsio->kern_data_resid = 0;
9592	ctsio->kern_rel_offset = 0;
9593	ctsio->kern_sg_entries = 0;
9594
9595	/*
9596	 * The control device is always connected.  The disk device, on the
9597	 * other hand, may not be online all the time.  Need to change this
9598	 * to figure out whether the disk device is actually online or not.
9599	 */
9600	if (lun != NULL)
9601		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9602				lun->be_lun->lun_type;
9603	else
9604		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9605
9606	p = 0;
9607	/* Supported VPD pages */
9608	pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9609	/* Serial Number */
9610	pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9611	/* Device Identification */
9612	pages->page_list[p++] = SVPD_DEVICE_ID;
9613	/* Extended INQUIRY Data */
9614	pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9615	/* Mode Page Policy */
9616	pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9617	/* SCSI Ports */
9618	pages->page_list[p++] = SVPD_SCSI_PORTS;
9619	/* Third-party Copy */
9620	pages->page_list[p++] = SVPD_SCSI_TPC;
9621	if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9622		/* Block limits */
9623		pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9624		/* Block Device Characteristics */
9625		pages->page_list[p++] = SVPD_BDC;
9626		/* Logical Block Provisioning */
9627		pages->page_list[p++] = SVPD_LBP;
9628	}
9629	pages->length = p;
9630
9631	ctl_set_success(ctsio);
9632	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9633	ctsio->be_move_done = ctl_config_move_done;
9634	ctl_datamove((union ctl_io *)ctsio);
9635	return (CTL_RETVAL_COMPLETE);
9636}
9637
9638/*
9639 * SCSI VPD page 0x80, the Unit Serial Number page.
9640 */
9641static int
9642ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9643{
9644	struct scsi_vpd_unit_serial_number *sn_ptr;
9645	struct ctl_lun *lun;
9646	int data_len;
9647
9648	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9649
9650	data_len = 4 + CTL_SN_LEN;
9651	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9652	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9653	if (data_len < alloc_len) {
9654		ctsio->residual = alloc_len - data_len;
9655		ctsio->kern_data_len = data_len;
9656		ctsio->kern_total_len = data_len;
9657	} else {
9658		ctsio->residual = 0;
9659		ctsio->kern_data_len = alloc_len;
9660		ctsio->kern_total_len = alloc_len;
9661	}
9662	ctsio->kern_data_resid = 0;
9663	ctsio->kern_rel_offset = 0;
9664	ctsio->kern_sg_entries = 0;
9665
9666	/*
9667	 * The control device is always connected.  The disk device, on the
9668	 * other hand, may not be online all the time.  Need to change this
9669	 * to figure out whether the disk device is actually online or not.
9670	 */
9671	if (lun != NULL)
9672		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9673				  lun->be_lun->lun_type;
9674	else
9675		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9676
9677	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9678	sn_ptr->length = CTL_SN_LEN;
9679	/*
9680	 * If we don't have a LUN, we just leave the serial number as
9681	 * all spaces.
9682	 */
9683	if (lun != NULL) {
9684		strncpy((char *)sn_ptr->serial_num,
9685			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9686	} else
9687		memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9688
9689	ctl_set_success(ctsio);
9690	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9691	ctsio->be_move_done = ctl_config_move_done;
9692	ctl_datamove((union ctl_io *)ctsio);
9693	return (CTL_RETVAL_COMPLETE);
9694}
9695
9696
9697/*
9698 * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9699 */
9700static int
9701ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9702{
9703	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9704	struct ctl_lun *lun;
9705	int data_len;
9706
9707	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9708
9709	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9710	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9711	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9712	ctsio->kern_sg_entries = 0;
9713
9714	if (data_len < alloc_len) {
9715		ctsio->residual = alloc_len - data_len;
9716		ctsio->kern_data_len = data_len;
9717		ctsio->kern_total_len = data_len;
9718	} else {
9719		ctsio->residual = 0;
9720		ctsio->kern_data_len = alloc_len;
9721		ctsio->kern_total_len = alloc_len;
9722	}
9723	ctsio->kern_data_resid = 0;
9724	ctsio->kern_rel_offset = 0;
9725	ctsio->kern_sg_entries = 0;
9726
9727	/*
9728	 * The control device is always connected.  The disk device, on the
9729	 * other hand, may not be online all the time.
9730	 */
9731	if (lun != NULL)
9732		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9733				     lun->be_lun->lun_type;
9734	else
9735		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9736	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9737	scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9738	/*
9739	 * We support head of queue, ordered and simple tags.
9740	 */
9741	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9742	/*
9743	 * Volatile cache supported.
9744	 */
9745	eid_ptr->flags3 = SVPD_EID_V_SUP;
9746
9747	/*
9748	 * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9749	 * attention for a particular IT nexus on all LUNs once we report
9750	 * it to that nexus once.  This bit is required as of SPC-4.
9751	 */
9752	eid_ptr->flags4 = SVPD_EID_LUICLT;
9753
9754	/*
9755	 * XXX KDM in order to correctly answer this, we would need
9756	 * information from the SIM to determine how much sense data it
9757	 * can send.  So this would really be a path inquiry field, most
9758	 * likely.  This can be set to a maximum of 252 according to SPC-4,
9759	 * but the hardware may or may not be able to support that much.
9760	 * 0 just means that the maximum sense data length is not reported.
9761	 */
9762	eid_ptr->max_sense_length = 0;
9763
9764	ctl_set_success(ctsio);
9765	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9766	ctsio->be_move_done = ctl_config_move_done;
9767	ctl_datamove((union ctl_io *)ctsio);
9768	return (CTL_RETVAL_COMPLETE);
9769}
9770
9771static int
9772ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9773{
9774	struct scsi_vpd_mode_page_policy *mpp_ptr;
9775	struct ctl_lun *lun;
9776	int data_len;
9777
9778	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9779
9780	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9781	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9782
9783	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9784	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9785	ctsio->kern_sg_entries = 0;
9786
9787	if (data_len < alloc_len) {
9788		ctsio->residual = alloc_len - data_len;
9789		ctsio->kern_data_len = data_len;
9790		ctsio->kern_total_len = data_len;
9791	} else {
9792		ctsio->residual = 0;
9793		ctsio->kern_data_len = alloc_len;
9794		ctsio->kern_total_len = alloc_len;
9795	}
9796	ctsio->kern_data_resid = 0;
9797	ctsio->kern_rel_offset = 0;
9798	ctsio->kern_sg_entries = 0;
9799
9800	/*
9801	 * The control device is always connected.  The disk device, on the
9802	 * other hand, may not be online all the time.
9803	 */
9804	if (lun != NULL)
9805		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9806				     lun->be_lun->lun_type;
9807	else
9808		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9809	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9810	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9811	mpp_ptr->descr[0].page_code = 0x3f;
9812	mpp_ptr->descr[0].subpage_code = 0xff;
9813	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9814
9815	ctl_set_success(ctsio);
9816	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9817	ctsio->be_move_done = ctl_config_move_done;
9818	ctl_datamove((union ctl_io *)ctsio);
9819	return (CTL_RETVAL_COMPLETE);
9820}
9821
9822/*
9823 * SCSI VPD page 0x83, the Device Identification page.
9824 */
9825static int
9826ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9827{
9828	struct scsi_vpd_device_id *devid_ptr;
9829	struct scsi_vpd_id_descriptor *desc;
9830	struct ctl_softc *softc;
9831	struct ctl_lun *lun;
9832	struct ctl_port *port;
9833	int data_len;
9834	uint8_t proto;
9835
9836	softc = control_softc;
9837
9838	port = ctl_io_port(&ctsio->io_hdr);
9839	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9840
9841	data_len = sizeof(struct scsi_vpd_device_id) +
9842	    sizeof(struct scsi_vpd_id_descriptor) +
9843		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9844	    sizeof(struct scsi_vpd_id_descriptor) +
9845		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9846	if (lun && lun->lun_devid)
9847		data_len += lun->lun_devid->len;
9848	if (port && port->port_devid)
9849		data_len += port->port_devid->len;
9850	if (port && port->target_devid)
9851		data_len += port->target_devid->len;
9852
9853	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9854	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9855	ctsio->kern_sg_entries = 0;
9856
9857	if (data_len < alloc_len) {
9858		ctsio->residual = alloc_len - data_len;
9859		ctsio->kern_data_len = data_len;
9860		ctsio->kern_total_len = data_len;
9861	} else {
9862		ctsio->residual = 0;
9863		ctsio->kern_data_len = alloc_len;
9864		ctsio->kern_total_len = alloc_len;
9865	}
9866	ctsio->kern_data_resid = 0;
9867	ctsio->kern_rel_offset = 0;
9868	ctsio->kern_sg_entries = 0;
9869
9870	/*
9871	 * The control device is always connected.  The disk device, on the
9872	 * other hand, may not be online all the time.
9873	 */
9874	if (lun != NULL)
9875		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9876				     lun->be_lun->lun_type;
9877	else
9878		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9879	devid_ptr->page_code = SVPD_DEVICE_ID;
9880	scsi_ulto2b(data_len - 4, devid_ptr->length);
9881
9882	if (port && port->port_type == CTL_PORT_FC)
9883		proto = SCSI_PROTO_FC << 4;
9884	else if (port && port->port_type == CTL_PORT_ISCSI)
9885		proto = SCSI_PROTO_ISCSI << 4;
9886	else
9887		proto = SCSI_PROTO_SPI << 4;
9888	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9889
9890	/*
9891	 * We're using a LUN association here.  i.e., this device ID is a
9892	 * per-LUN identifier.
9893	 */
9894	if (lun && lun->lun_devid) {
9895		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9896		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9897		    lun->lun_devid->len);
9898	}
9899
9900	/*
9901	 * This is for the WWPN which is a port association.
9902	 */
9903	if (port && port->port_devid) {
9904		memcpy(desc, port->port_devid->data, port->port_devid->len);
9905		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9906		    port->port_devid->len);
9907	}
9908
9909	/*
9910	 * This is for the Relative Target Port(type 4h) identifier
9911	 */
9912	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9913	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9914	    SVPD_ID_TYPE_RELTARG;
9915	desc->length = 4;
9916	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9917	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9918	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9919
9920	/*
9921	 * This is for the Target Port Group(type 5h) identifier
9922	 */
9923	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9924	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9925	    SVPD_ID_TYPE_TPORTGRP;
9926	desc->length = 4;
9927	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / softc->port_cnt + 1,
9928	    &desc->identifier[2]);
9929	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9930	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9931
9932	/*
9933	 * This is for the Target identifier
9934	 */
9935	if (port && port->target_devid) {
9936		memcpy(desc, port->target_devid->data, port->target_devid->len);
9937	}
9938
9939	ctl_set_success(ctsio);
9940	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9941	ctsio->be_move_done = ctl_config_move_done;
9942	ctl_datamove((union ctl_io *)ctsio);
9943	return (CTL_RETVAL_COMPLETE);
9944}
9945
9946static int
9947ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9948{
9949	struct ctl_softc *softc = control_softc;
9950	struct scsi_vpd_scsi_ports *sp;
9951	struct scsi_vpd_port_designation *pd;
9952	struct scsi_vpd_port_designation_cont *pdc;
9953	struct ctl_lun *lun;
9954	struct ctl_port *port;
9955	int data_len, num_target_ports, iid_len, id_len;
9956
9957	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9958
9959	num_target_ports = 0;
9960	iid_len = 0;
9961	id_len = 0;
9962	mtx_lock(&softc->ctl_lock);
9963	STAILQ_FOREACH(port, &softc->port_list, links) {
9964		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9965			continue;
9966		if (lun != NULL &&
9967		    ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9968			continue;
9969		num_target_ports++;
9970		if (port->init_devid)
9971			iid_len += port->init_devid->len;
9972		if (port->port_devid)
9973			id_len += port->port_devid->len;
9974	}
9975	mtx_unlock(&softc->ctl_lock);
9976
9977	data_len = sizeof(struct scsi_vpd_scsi_ports) +
9978	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9979	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9980	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9981	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9982	ctsio->kern_sg_entries = 0;
9983
9984	if (data_len < alloc_len) {
9985		ctsio->residual = alloc_len - data_len;
9986		ctsio->kern_data_len = data_len;
9987		ctsio->kern_total_len = data_len;
9988	} else {
9989		ctsio->residual = 0;
9990		ctsio->kern_data_len = alloc_len;
9991		ctsio->kern_total_len = alloc_len;
9992	}
9993	ctsio->kern_data_resid = 0;
9994	ctsio->kern_rel_offset = 0;
9995	ctsio->kern_sg_entries = 0;
9996
9997	/*
9998	 * The control device is always connected.  The disk device, on the
9999	 * other hand, may not be online all the time.  Need to change this
10000	 * to figure out whether the disk device is actually online or not.
10001	 */
10002	if (lun != NULL)
10003		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10004				  lun->be_lun->lun_type;
10005	else
10006		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10007
10008	sp->page_code = SVPD_SCSI_PORTS;
10009	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10010	    sp->page_length);
10011	pd = &sp->design[0];
10012
10013	mtx_lock(&softc->ctl_lock);
10014	STAILQ_FOREACH(port, &softc->port_list, links) {
10015		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10016			continue;
10017		if (lun != NULL &&
10018		    ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
10019			continue;
10020		scsi_ulto2b(port->targ_port, pd->relative_port_id);
10021		if (port->init_devid) {
10022			iid_len = port->init_devid->len;
10023			memcpy(pd->initiator_transportid,
10024			    port->init_devid->data, port->init_devid->len);
10025		} else
10026			iid_len = 0;
10027		scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10028		pdc = (struct scsi_vpd_port_designation_cont *)
10029		    (&pd->initiator_transportid[iid_len]);
10030		if (port->port_devid) {
10031			id_len = port->port_devid->len;
10032			memcpy(pdc->target_port_descriptors,
10033			    port->port_devid->data, port->port_devid->len);
10034		} else
10035			id_len = 0;
10036		scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10037		pd = (struct scsi_vpd_port_designation *)
10038		    ((uint8_t *)pdc->target_port_descriptors + id_len);
10039	}
10040	mtx_unlock(&softc->ctl_lock);
10041
10042	ctl_set_success(ctsio);
10043	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10044	ctsio->be_move_done = ctl_config_move_done;
10045	ctl_datamove((union ctl_io *)ctsio);
10046	return (CTL_RETVAL_COMPLETE);
10047}
10048
10049static int
10050ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10051{
10052	struct scsi_vpd_block_limits *bl_ptr;
10053	struct ctl_lun *lun;
10054
10055	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10056
10057	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10058	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10059	ctsio->kern_sg_entries = 0;
10060
10061	if (sizeof(*bl_ptr) < alloc_len) {
10062		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10063		ctsio->kern_data_len = sizeof(*bl_ptr);
10064		ctsio->kern_total_len = sizeof(*bl_ptr);
10065	} else {
10066		ctsio->residual = 0;
10067		ctsio->kern_data_len = alloc_len;
10068		ctsio->kern_total_len = alloc_len;
10069	}
10070	ctsio->kern_data_resid = 0;
10071	ctsio->kern_rel_offset = 0;
10072	ctsio->kern_sg_entries = 0;
10073
10074	/*
10075	 * The control device is always connected.  The disk device, on the
10076	 * other hand, may not be online all the time.  Need to change this
10077	 * to figure out whether the disk device is actually online or not.
10078	 */
10079	if (lun != NULL)
10080		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10081				  lun->be_lun->lun_type;
10082	else
10083		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10084
10085	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10086	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
10087	bl_ptr->max_cmp_write_len = 0xff;
10088	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10089	if (lun != NULL) {
10090		scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
10091		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10092			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10093			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10094			if (lun->be_lun->ublockexp != 0) {
10095				scsi_ulto4b((1 << lun->be_lun->ublockexp),
10096				    bl_ptr->opt_unmap_grain);
10097				scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
10098				    bl_ptr->unmap_grain_align);
10099			}
10100		}
10101		scsi_ulto4b(lun->be_lun->atomicblock,
10102		    bl_ptr->max_atomic_transfer_length);
10103		scsi_ulto4b(0, bl_ptr->atomic_alignment);
10104		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10105		scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
10106		scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
10107	}
10108	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10109
10110	ctl_set_success(ctsio);
10111	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10112	ctsio->be_move_done = ctl_config_move_done;
10113	ctl_datamove((union ctl_io *)ctsio);
10114	return (CTL_RETVAL_COMPLETE);
10115}
10116
10117static int
10118ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10119{
10120	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10121	struct ctl_lun *lun;
10122	const char *value;
10123	u_int i;
10124
10125	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10126
10127	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10128	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10129	ctsio->kern_sg_entries = 0;
10130
10131	if (sizeof(*bdc_ptr) < alloc_len) {
10132		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10133		ctsio->kern_data_len = sizeof(*bdc_ptr);
10134		ctsio->kern_total_len = sizeof(*bdc_ptr);
10135	} else {
10136		ctsio->residual = 0;
10137		ctsio->kern_data_len = alloc_len;
10138		ctsio->kern_total_len = alloc_len;
10139	}
10140	ctsio->kern_data_resid = 0;
10141	ctsio->kern_rel_offset = 0;
10142	ctsio->kern_sg_entries = 0;
10143
10144	/*
10145	 * The control device is always connected.  The disk device, on the
10146	 * other hand, may not be online all the time.  Need to change this
10147	 * to figure out whether the disk device is actually online or not.
10148	 */
10149	if (lun != NULL)
10150		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10151				  lun->be_lun->lun_type;
10152	else
10153		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10154	bdc_ptr->page_code = SVPD_BDC;
10155	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10156	if (lun != NULL &&
10157	    (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
10158		i = strtol(value, NULL, 0);
10159	else
10160		i = CTL_DEFAULT_ROTATION_RATE;
10161	scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
10162	if (lun != NULL &&
10163	    (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
10164		i = strtol(value, NULL, 0);
10165	else
10166		i = 0;
10167	bdc_ptr->wab_wac_ff = (i & 0x0f);
10168	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10169
10170	ctl_set_success(ctsio);
10171	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10172	ctsio->be_move_done = ctl_config_move_done;
10173	ctl_datamove((union ctl_io *)ctsio);
10174	return (CTL_RETVAL_COMPLETE);
10175}
10176
10177static int
10178ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10179{
10180	struct scsi_vpd_logical_block_prov *lbp_ptr;
10181	struct ctl_lun *lun;
10182
10183	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10184
10185	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10186	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10187	ctsio->kern_sg_entries = 0;
10188
10189	if (sizeof(*lbp_ptr) < alloc_len) {
10190		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10191		ctsio->kern_data_len = sizeof(*lbp_ptr);
10192		ctsio->kern_total_len = sizeof(*lbp_ptr);
10193	} else {
10194		ctsio->residual = 0;
10195		ctsio->kern_data_len = alloc_len;
10196		ctsio->kern_total_len = alloc_len;
10197	}
10198	ctsio->kern_data_resid = 0;
10199	ctsio->kern_rel_offset = 0;
10200	ctsio->kern_sg_entries = 0;
10201
10202	/*
10203	 * The control device is always connected.  The disk device, on the
10204	 * other hand, may not be online all the time.  Need to change this
10205	 * to figure out whether the disk device is actually online or not.
10206	 */
10207	if (lun != NULL)
10208		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10209				  lun->be_lun->lun_type;
10210	else
10211		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10212
10213	lbp_ptr->page_code = SVPD_LBP;
10214	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10215	lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10216	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10217		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10218		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10219		lbp_ptr->prov_type = SVPD_LBP_THIN;
10220	}
10221
10222	ctl_set_success(ctsio);
10223	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10224	ctsio->be_move_done = ctl_config_move_done;
10225	ctl_datamove((union ctl_io *)ctsio);
10226	return (CTL_RETVAL_COMPLETE);
10227}
10228
10229/*
10230 * INQUIRY with the EVPD bit set.
10231 */
10232static int
10233ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10234{
10235	struct ctl_lun *lun;
10236	struct scsi_inquiry *cdb;
10237	int alloc_len, retval;
10238
10239	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10240	cdb = (struct scsi_inquiry *)ctsio->cdb;
10241	alloc_len = scsi_2btoul(cdb->length);
10242
10243	switch (cdb->page_code) {
10244	case SVPD_SUPPORTED_PAGES:
10245		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10246		break;
10247	case SVPD_UNIT_SERIAL_NUMBER:
10248		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10249		break;
10250	case SVPD_DEVICE_ID:
10251		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10252		break;
10253	case SVPD_EXTENDED_INQUIRY_DATA:
10254		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10255		break;
10256	case SVPD_MODE_PAGE_POLICY:
10257		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10258		break;
10259	case SVPD_SCSI_PORTS:
10260		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10261		break;
10262	case SVPD_SCSI_TPC:
10263		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10264		break;
10265	case SVPD_BLOCK_LIMITS:
10266		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10267			goto err;
10268		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10269		break;
10270	case SVPD_BDC:
10271		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10272			goto err;
10273		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10274		break;
10275	case SVPD_LBP:
10276		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10277			goto err;
10278		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10279		break;
10280	default:
10281err:
10282		ctl_set_invalid_field(ctsio,
10283				      /*sks_valid*/ 1,
10284				      /*command*/ 1,
10285				      /*field*/ 2,
10286				      /*bit_valid*/ 0,
10287				      /*bit*/ 0);
10288		ctl_done((union ctl_io *)ctsio);
10289		retval = CTL_RETVAL_COMPLETE;
10290		break;
10291	}
10292
10293	return (retval);
10294}
10295
10296/*
10297 * Standard INQUIRY data.
10298 */
10299static int
10300ctl_inquiry_std(struct ctl_scsiio *ctsio)
10301{
10302	struct scsi_inquiry_data *inq_ptr;
10303	struct scsi_inquiry *cdb;
10304	struct ctl_softc *softc = control_softc;
10305	struct ctl_port *port;
10306	struct ctl_lun *lun;
10307	char *val;
10308	uint32_t alloc_len, data_len;
10309	ctl_port_type port_type;
10310
10311	port = ctl_io_port(&ctsio->io_hdr);
10312	port_type = port->port_type;
10313	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10314		port_type = CTL_PORT_SCSI;
10315
10316	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10317	cdb = (struct scsi_inquiry *)ctsio->cdb;
10318	alloc_len = scsi_2btoul(cdb->length);
10319
10320	/*
10321	 * We malloc the full inquiry data size here and fill it
10322	 * in.  If the user only asks for less, we'll give him
10323	 * that much.
10324	 */
10325	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10326	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10327	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10328	ctsio->kern_sg_entries = 0;
10329	ctsio->kern_data_resid = 0;
10330	ctsio->kern_rel_offset = 0;
10331
10332	if (data_len < alloc_len) {
10333		ctsio->residual = alloc_len - data_len;
10334		ctsio->kern_data_len = data_len;
10335		ctsio->kern_total_len = data_len;
10336	} else {
10337		ctsio->residual = 0;
10338		ctsio->kern_data_len = alloc_len;
10339		ctsio->kern_total_len = alloc_len;
10340	}
10341
10342	if (lun != NULL) {
10343		if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10344		    softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10345			inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10346			    lun->be_lun->lun_type;
10347		} else {
10348			inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10349			    lun->be_lun->lun_type;
10350		}
10351	} else
10352		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10353
10354	/* RMB in byte 2 is 0 */
10355	inq_ptr->version = SCSI_REV_SPC4;
10356
10357	/*
10358	 * According to SAM-3, even if a device only supports a single
10359	 * level of LUN addressing, it should still set the HISUP bit:
10360	 *
10361	 * 4.9.1 Logical unit numbers overview
10362	 *
10363	 * All logical unit number formats described in this standard are
10364	 * hierarchical in structure even when only a single level in that
10365	 * hierarchy is used. The HISUP bit shall be set to one in the
10366	 * standard INQUIRY data (see SPC-2) when any logical unit number
10367	 * format described in this standard is used.  Non-hierarchical
10368	 * formats are outside the scope of this standard.
10369	 *
10370	 * Therefore we set the HiSup bit here.
10371	 *
10372	 * The reponse format is 2, per SPC-3.
10373	 */
10374	inq_ptr->response_format = SID_HiSup | 2;
10375
10376	inq_ptr->additional_length = data_len -
10377	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10378	CTL_DEBUG_PRINT(("additional_length = %d\n",
10379			 inq_ptr->additional_length));
10380
10381	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10382	/* 16 bit addressing */
10383	if (port_type == CTL_PORT_SCSI)
10384		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10385	/* XXX set the SID_MultiP bit here if we're actually going to
10386	   respond on multiple ports */
10387	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10388
10389	/* 16 bit data bus, synchronous transfers */
10390	if (port_type == CTL_PORT_SCSI)
10391		inq_ptr->flags = SID_WBus16 | SID_Sync;
10392	/*
10393	 * XXX KDM do we want to support tagged queueing on the control
10394	 * device at all?
10395	 */
10396	if ((lun == NULL)
10397	 || (lun->be_lun->lun_type != T_PROCESSOR))
10398		inq_ptr->flags |= SID_CmdQue;
10399	/*
10400	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10401	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10402	 * name and 4 bytes for the revision.
10403	 */
10404	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10405	    "vendor")) == NULL) {
10406		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10407	} else {
10408		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10409		strncpy(inq_ptr->vendor, val,
10410		    min(sizeof(inq_ptr->vendor), strlen(val)));
10411	}
10412	if (lun == NULL) {
10413		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10414		    sizeof(inq_ptr->product));
10415	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10416		switch (lun->be_lun->lun_type) {
10417		case T_DIRECT:
10418			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10419			    sizeof(inq_ptr->product));
10420			break;
10421		case T_PROCESSOR:
10422			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10423			    sizeof(inq_ptr->product));
10424			break;
10425		default:
10426			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10427			    sizeof(inq_ptr->product));
10428			break;
10429		}
10430	} else {
10431		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10432		strncpy(inq_ptr->product, val,
10433		    min(sizeof(inq_ptr->product), strlen(val)));
10434	}
10435
10436	/*
10437	 * XXX make this a macro somewhere so it automatically gets
10438	 * incremented when we make changes.
10439	 */
10440	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10441	    "revision")) == NULL) {
10442		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10443	} else {
10444		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10445		strncpy(inq_ptr->revision, val,
10446		    min(sizeof(inq_ptr->revision), strlen(val)));
10447	}
10448
10449	/*
10450	 * For parallel SCSI, we support double transition and single
10451	 * transition clocking.  We also support QAS (Quick Arbitration
10452	 * and Selection) and Information Unit transfers on both the
10453	 * control and array devices.
10454	 */
10455	if (port_type == CTL_PORT_SCSI)
10456		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10457				    SID_SPI_IUS;
10458
10459	/* SAM-5 (no version claimed) */
10460	scsi_ulto2b(0x00A0, inq_ptr->version1);
10461	/* SPC-4 (no version claimed) */
10462	scsi_ulto2b(0x0460, inq_ptr->version2);
10463	if (port_type == CTL_PORT_FC) {
10464		/* FCP-2 ANSI INCITS.350:2003 */
10465		scsi_ulto2b(0x0917, inq_ptr->version3);
10466	} else if (port_type == CTL_PORT_SCSI) {
10467		/* SPI-4 ANSI INCITS.362:200x */
10468		scsi_ulto2b(0x0B56, inq_ptr->version3);
10469	} else if (port_type == CTL_PORT_ISCSI) {
10470		/* iSCSI (no version claimed) */
10471		scsi_ulto2b(0x0960, inq_ptr->version3);
10472	} else if (port_type == CTL_PORT_SAS) {
10473		/* SAS (no version claimed) */
10474		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10475	}
10476
10477	if (lun == NULL) {
10478		/* SBC-4 (no version claimed) */
10479		scsi_ulto2b(0x0600, inq_ptr->version4);
10480	} else {
10481		switch (lun->be_lun->lun_type) {
10482		case T_DIRECT:
10483			/* SBC-4 (no version claimed) */
10484			scsi_ulto2b(0x0600, inq_ptr->version4);
10485			break;
10486		case T_PROCESSOR:
10487		default:
10488			break;
10489		}
10490	}
10491
10492	ctl_set_success(ctsio);
10493	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10494	ctsio->be_move_done = ctl_config_move_done;
10495	ctl_datamove((union ctl_io *)ctsio);
10496	return (CTL_RETVAL_COMPLETE);
10497}
10498
10499int
10500ctl_inquiry(struct ctl_scsiio *ctsio)
10501{
10502	struct scsi_inquiry *cdb;
10503	int retval;
10504
10505	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10506
10507	cdb = (struct scsi_inquiry *)ctsio->cdb;
10508	if (cdb->byte2 & SI_EVPD)
10509		retval = ctl_inquiry_evpd(ctsio);
10510	else if (cdb->page_code == 0)
10511		retval = ctl_inquiry_std(ctsio);
10512	else {
10513		ctl_set_invalid_field(ctsio,
10514				      /*sks_valid*/ 1,
10515				      /*command*/ 1,
10516				      /*field*/ 2,
10517				      /*bit_valid*/ 0,
10518				      /*bit*/ 0);
10519		ctl_done((union ctl_io *)ctsio);
10520		return (CTL_RETVAL_COMPLETE);
10521	}
10522
10523	return (retval);
10524}
10525
10526/*
10527 * For known CDB types, parse the LBA and length.
10528 */
10529static int
10530ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10531{
10532	if (io->io_hdr.io_type != CTL_IO_SCSI)
10533		return (1);
10534
10535	switch (io->scsiio.cdb[0]) {
10536	case COMPARE_AND_WRITE: {
10537		struct scsi_compare_and_write *cdb;
10538
10539		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10540
10541		*lba = scsi_8btou64(cdb->addr);
10542		*len = cdb->length;
10543		break;
10544	}
10545	case READ_6:
10546	case WRITE_6: {
10547		struct scsi_rw_6 *cdb;
10548
10549		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10550
10551		*lba = scsi_3btoul(cdb->addr);
10552		/* only 5 bits are valid in the most significant address byte */
10553		*lba &= 0x1fffff;
10554		*len = cdb->length;
10555		break;
10556	}
10557	case READ_10:
10558	case WRITE_10: {
10559		struct scsi_rw_10 *cdb;
10560
10561		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10562
10563		*lba = scsi_4btoul(cdb->addr);
10564		*len = scsi_2btoul(cdb->length);
10565		break;
10566	}
10567	case WRITE_VERIFY_10: {
10568		struct scsi_write_verify_10 *cdb;
10569
10570		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10571
10572		*lba = scsi_4btoul(cdb->addr);
10573		*len = scsi_2btoul(cdb->length);
10574		break;
10575	}
10576	case READ_12:
10577	case WRITE_12: {
10578		struct scsi_rw_12 *cdb;
10579
10580		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10581
10582		*lba = scsi_4btoul(cdb->addr);
10583		*len = scsi_4btoul(cdb->length);
10584		break;
10585	}
10586	case WRITE_VERIFY_12: {
10587		struct scsi_write_verify_12 *cdb;
10588
10589		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10590
10591		*lba = scsi_4btoul(cdb->addr);
10592		*len = scsi_4btoul(cdb->length);
10593		break;
10594	}
10595	case READ_16:
10596	case WRITE_16: {
10597		struct scsi_rw_16 *cdb;
10598
10599		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10600
10601		*lba = scsi_8btou64(cdb->addr);
10602		*len = scsi_4btoul(cdb->length);
10603		break;
10604	}
10605	case WRITE_ATOMIC_16: {
10606		struct scsi_write_atomic_16 *cdb;
10607
10608		cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10609
10610		*lba = scsi_8btou64(cdb->addr);
10611		*len = scsi_2btoul(cdb->length);
10612		break;
10613	}
10614	case WRITE_VERIFY_16: {
10615		struct scsi_write_verify_16 *cdb;
10616
10617		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10618
10619		*lba = scsi_8btou64(cdb->addr);
10620		*len = scsi_4btoul(cdb->length);
10621		break;
10622	}
10623	case WRITE_SAME_10: {
10624		struct scsi_write_same_10 *cdb;
10625
10626		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10627
10628		*lba = scsi_4btoul(cdb->addr);
10629		*len = scsi_2btoul(cdb->length);
10630		break;
10631	}
10632	case WRITE_SAME_16: {
10633		struct scsi_write_same_16 *cdb;
10634
10635		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10636
10637		*lba = scsi_8btou64(cdb->addr);
10638		*len = scsi_4btoul(cdb->length);
10639		break;
10640	}
10641	case VERIFY_10: {
10642		struct scsi_verify_10 *cdb;
10643
10644		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10645
10646		*lba = scsi_4btoul(cdb->addr);
10647		*len = scsi_2btoul(cdb->length);
10648		break;
10649	}
10650	case VERIFY_12: {
10651		struct scsi_verify_12 *cdb;
10652
10653		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10654
10655		*lba = scsi_4btoul(cdb->addr);
10656		*len = scsi_4btoul(cdb->length);
10657		break;
10658	}
10659	case VERIFY_16: {
10660		struct scsi_verify_16 *cdb;
10661
10662		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10663
10664		*lba = scsi_8btou64(cdb->addr);
10665		*len = scsi_4btoul(cdb->length);
10666		break;
10667	}
10668	case UNMAP: {
10669		*lba = 0;
10670		*len = UINT64_MAX;
10671		break;
10672	}
10673	case SERVICE_ACTION_IN: {	/* GET LBA STATUS */
10674		struct scsi_get_lba_status *cdb;
10675
10676		cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10677		*lba = scsi_8btou64(cdb->addr);
10678		*len = UINT32_MAX;
10679		break;
10680	}
10681	default:
10682		return (1);
10683		break; /* NOTREACHED */
10684	}
10685
10686	return (0);
10687}
10688
10689static ctl_action
10690ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10691    bool seq)
10692{
10693	uint64_t endlba1, endlba2;
10694
10695	endlba1 = lba1 + len1 - (seq ? 0 : 1);
10696	endlba2 = lba2 + len2 - 1;
10697
10698	if ((endlba1 < lba2) || (endlba2 < lba1))
10699		return (CTL_ACTION_PASS);
10700	else
10701		return (CTL_ACTION_BLOCK);
10702}
10703
10704static int
10705ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10706{
10707	struct ctl_ptr_len_flags *ptrlen;
10708	struct scsi_unmap_desc *buf, *end, *range;
10709	uint64_t lba;
10710	uint32_t len;
10711
10712	/* If not UNMAP -- go other way. */
10713	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10714	    io->scsiio.cdb[0] != UNMAP)
10715		return (CTL_ACTION_ERROR);
10716
10717	/* If UNMAP without data -- block and wait for data. */
10718	ptrlen = (struct ctl_ptr_len_flags *)
10719	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10720	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10721	    ptrlen->ptr == NULL)
10722		return (CTL_ACTION_BLOCK);
10723
10724	/* UNMAP with data -- check for collision. */
10725	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10726	end = buf + ptrlen->len / sizeof(*buf);
10727	for (range = buf; range < end; range++) {
10728		lba = scsi_8btou64(range->lba);
10729		len = scsi_4btoul(range->length);
10730		if ((lba < lba2 + len2) && (lba + len > lba2))
10731			return (CTL_ACTION_BLOCK);
10732	}
10733	return (CTL_ACTION_PASS);
10734}
10735
10736static ctl_action
10737ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10738{
10739	uint64_t lba1, lba2;
10740	uint64_t len1, len2;
10741	int retval;
10742
10743	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10744		return (CTL_ACTION_ERROR);
10745
10746	retval = ctl_extent_check_unmap(io1, lba2, len2);
10747	if (retval != CTL_ACTION_ERROR)
10748		return (retval);
10749
10750	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10751		return (CTL_ACTION_ERROR);
10752
10753	if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10754		seq = FALSE;
10755	return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10756}
10757
10758static ctl_action
10759ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10760{
10761	uint64_t lba1, lba2;
10762	uint64_t len1, len2;
10763
10764	if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10765		return (CTL_ACTION_PASS);
10766	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10767		return (CTL_ACTION_ERROR);
10768	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10769		return (CTL_ACTION_ERROR);
10770
10771	if (lba1 + len1 == lba2)
10772		return (CTL_ACTION_BLOCK);
10773	return (CTL_ACTION_PASS);
10774}
10775
10776static ctl_action
10777ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10778    union ctl_io *ooa_io)
10779{
10780	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10781	const ctl_serialize_action *serialize_row;
10782
10783	/*
10784	 * The initiator attempted multiple untagged commands at the same
10785	 * time.  Can't do that.
10786	 */
10787	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10788	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10789	 && ((pending_io->io_hdr.nexus.targ_port ==
10790	      ooa_io->io_hdr.nexus.targ_port)
10791	  && (pending_io->io_hdr.nexus.initid ==
10792	      ooa_io->io_hdr.nexus.initid))
10793	 && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10794	      CTL_FLAG_STATUS_SENT)) == 0))
10795		return (CTL_ACTION_OVERLAP);
10796
10797	/*
10798	 * The initiator attempted to send multiple tagged commands with
10799	 * the same ID.  (It's fine if different initiators have the same
10800	 * tag ID.)
10801	 *
10802	 * Even if all of those conditions are true, we don't kill the I/O
10803	 * if the command ahead of us has been aborted.  We won't end up
10804	 * sending it to the FETD, and it's perfectly legal to resend a
10805	 * command with the same tag number as long as the previous
10806	 * instance of this tag number has been aborted somehow.
10807	 */
10808	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10809	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10810	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10811	 && ((pending_io->io_hdr.nexus.targ_port ==
10812	      ooa_io->io_hdr.nexus.targ_port)
10813	  && (pending_io->io_hdr.nexus.initid ==
10814	      ooa_io->io_hdr.nexus.initid))
10815	 && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10816	      CTL_FLAG_STATUS_SENT)) == 0))
10817		return (CTL_ACTION_OVERLAP_TAG);
10818
10819	/*
10820	 * If we get a head of queue tag, SAM-3 says that we should
10821	 * immediately execute it.
10822	 *
10823	 * What happens if this command would normally block for some other
10824	 * reason?  e.g. a request sense with a head of queue tag
10825	 * immediately after a write.  Normally that would block, but this
10826	 * will result in its getting executed immediately...
10827	 *
10828	 * We currently return "pass" instead of "skip", so we'll end up
10829	 * going through the rest of the queue to check for overlapped tags.
10830	 *
10831	 * XXX KDM check for other types of blockage first??
10832	 */
10833	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10834		return (CTL_ACTION_PASS);
10835
10836	/*
10837	 * Ordered tags have to block until all items ahead of them
10838	 * have completed.  If we get called with an ordered tag, we always
10839	 * block, if something else is ahead of us in the queue.
10840	 */
10841	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10842		return (CTL_ACTION_BLOCK);
10843
10844	/*
10845	 * Simple tags get blocked until all head of queue and ordered tags
10846	 * ahead of them have completed.  I'm lumping untagged commands in
10847	 * with simple tags here.  XXX KDM is that the right thing to do?
10848	 */
10849	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10850	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10851	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10852	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10853		return (CTL_ACTION_BLOCK);
10854
10855	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10856	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10857
10858	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10859
10860	switch (serialize_row[pending_entry->seridx]) {
10861	case CTL_SER_BLOCK:
10862		return (CTL_ACTION_BLOCK);
10863	case CTL_SER_EXTENT:
10864		return (ctl_extent_check(ooa_io, pending_io,
10865		    (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10866	case CTL_SER_EXTENTOPT:
10867		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10868		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10869			return (ctl_extent_check(ooa_io, pending_io,
10870			    (lun->be_lun &&
10871			     lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10872		return (CTL_ACTION_PASS);
10873	case CTL_SER_EXTENTSEQ:
10874		if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10875			return (ctl_extent_check_seq(ooa_io, pending_io));
10876		return (CTL_ACTION_PASS);
10877	case CTL_SER_PASS:
10878		return (CTL_ACTION_PASS);
10879	case CTL_SER_BLOCKOPT:
10880		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10881		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10882			return (CTL_ACTION_BLOCK);
10883		return (CTL_ACTION_PASS);
10884	case CTL_SER_SKIP:
10885		return (CTL_ACTION_SKIP);
10886	default:
10887		panic("invalid serialization value %d",
10888		      serialize_row[pending_entry->seridx]);
10889	}
10890
10891	return (CTL_ACTION_ERROR);
10892}
10893
10894/*
10895 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10896 * Assumptions:
10897 * - pending_io is generally either incoming, or on the blocked queue
10898 * - starting I/O is the I/O we want to start the check with.
10899 */
10900static ctl_action
10901ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10902	      union ctl_io *starting_io)
10903{
10904	union ctl_io *ooa_io;
10905	ctl_action action;
10906
10907	mtx_assert(&lun->lun_lock, MA_OWNED);
10908
10909	/*
10910	 * Run back along the OOA queue, starting with the current
10911	 * blocked I/O and going through every I/O before it on the
10912	 * queue.  If starting_io is NULL, we'll just end up returning
10913	 * CTL_ACTION_PASS.
10914	 */
10915	for (ooa_io = starting_io; ooa_io != NULL;
10916	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10917	     ooa_links)){
10918
10919		/*
10920		 * This routine just checks to see whether
10921		 * cur_blocked is blocked by ooa_io, which is ahead
10922		 * of it in the queue.  It doesn't queue/dequeue
10923		 * cur_blocked.
10924		 */
10925		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10926		switch (action) {
10927		case CTL_ACTION_BLOCK:
10928		case CTL_ACTION_OVERLAP:
10929		case CTL_ACTION_OVERLAP_TAG:
10930		case CTL_ACTION_SKIP:
10931		case CTL_ACTION_ERROR:
10932			return (action);
10933			break; /* NOTREACHED */
10934		case CTL_ACTION_PASS:
10935			break;
10936		default:
10937			panic("invalid action %d", action);
10938			break;  /* NOTREACHED */
10939		}
10940	}
10941
10942	return (CTL_ACTION_PASS);
10943}
10944
10945/*
10946 * Assumptions:
10947 * - An I/O has just completed, and has been removed from the per-LUN OOA
10948 *   queue, so some items on the blocked queue may now be unblocked.
10949 */
10950static int
10951ctl_check_blocked(struct ctl_lun *lun)
10952{
10953	struct ctl_softc *softc = lun->ctl_softc;
10954	union ctl_io *cur_blocked, *next_blocked;
10955
10956	mtx_assert(&lun->lun_lock, MA_OWNED);
10957
10958	/*
10959	 * Run forward from the head of the blocked queue, checking each
10960	 * entry against the I/Os prior to it on the OOA queue to see if
10961	 * there is still any blockage.
10962	 *
10963	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10964	 * with our removing a variable on it while it is traversing the
10965	 * list.
10966	 */
10967	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10968	     cur_blocked != NULL; cur_blocked = next_blocked) {
10969		union ctl_io *prev_ooa;
10970		ctl_action action;
10971
10972		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10973							  blocked_links);
10974
10975		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10976						      ctl_ooaq, ooa_links);
10977
10978		/*
10979		 * If cur_blocked happens to be the first item in the OOA
10980		 * queue now, prev_ooa will be NULL, and the action
10981		 * returned will just be CTL_ACTION_PASS.
10982		 */
10983		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10984
10985		switch (action) {
10986		case CTL_ACTION_BLOCK:
10987			/* Nothing to do here, still blocked */
10988			break;
10989		case CTL_ACTION_OVERLAP:
10990		case CTL_ACTION_OVERLAP_TAG:
10991			/*
10992			 * This shouldn't happen!  In theory we've already
10993			 * checked this command for overlap...
10994			 */
10995			break;
10996		case CTL_ACTION_PASS:
10997		case CTL_ACTION_SKIP: {
10998			const struct ctl_cmd_entry *entry;
10999
11000			/*
11001			 * The skip case shouldn't happen, this transaction
11002			 * should have never made it onto the blocked queue.
11003			 */
11004			/*
11005			 * This I/O is no longer blocked, we can remove it
11006			 * from the blocked queue.  Since this is a TAILQ
11007			 * (doubly linked list), we can do O(1) removals
11008			 * from any place on the list.
11009			 */
11010			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11011				     blocked_links);
11012			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11013
11014			if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11015			    (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11016				/*
11017				 * Need to send IO back to original side to
11018				 * run
11019				 */
11020				union ctl_ha_msg msg_info;
11021
11022				cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11023				msg_info.hdr.original_sc =
11024					cur_blocked->io_hdr.original_sc;
11025				msg_info.hdr.serializing_sc = cur_blocked;
11026				msg_info.hdr.msg_type = CTL_MSG_R2R;
11027				ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11028				    sizeof(msg_info.hdr), M_NOWAIT);
11029				break;
11030			}
11031			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11032
11033			/*
11034			 * Check this I/O for LUN state changes that may
11035			 * have happened while this command was blocked.
11036			 * The LUN state may have been changed by a command
11037			 * ahead of us in the queue, so we need to re-check
11038			 * for any states that can be caused by SCSI
11039			 * commands.
11040			 */
11041			if (ctl_scsiio_lun_check(lun, entry,
11042						 &cur_blocked->scsiio) == 0) {
11043				cur_blocked->io_hdr.flags |=
11044				                      CTL_FLAG_IS_WAS_ON_RTR;
11045				ctl_enqueue_rtr(cur_blocked);
11046			} else
11047				ctl_done(cur_blocked);
11048			break;
11049		}
11050		default:
11051			/*
11052			 * This probably shouldn't happen -- we shouldn't
11053			 * get CTL_ACTION_ERROR, or anything else.
11054			 */
11055			break;
11056		}
11057	}
11058
11059	return (CTL_RETVAL_COMPLETE);
11060}
11061
11062/*
11063 * This routine (with one exception) checks LUN flags that can be set by
11064 * commands ahead of us in the OOA queue.  These flags have to be checked
11065 * when a command initially comes in, and when we pull a command off the
11066 * blocked queue and are preparing to execute it.  The reason we have to
11067 * check these flags for commands on the blocked queue is that the LUN
11068 * state may have been changed by a command ahead of us while we're on the
11069 * blocked queue.
11070 *
11071 * Ordering is somewhat important with these checks, so please pay
11072 * careful attention to the placement of any new checks.
11073 */
11074static int
11075ctl_scsiio_lun_check(struct ctl_lun *lun,
11076    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11077{
11078	struct ctl_softc *softc = lun->ctl_softc;
11079	int retval;
11080	uint32_t residx;
11081
11082	retval = 0;
11083
11084	mtx_assert(&lun->lun_lock, MA_OWNED);
11085
11086	/*
11087	 * If this shelf is a secondary shelf controller, we may have to
11088	 * reject some commands disallowed by HA mode and link state.
11089	 */
11090	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11091		if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11092		    (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11093			ctl_set_lun_unavail(ctsio);
11094			retval = 1;
11095			goto bailout;
11096		}
11097		if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11098		    (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11099			ctl_set_lun_transit(ctsio);
11100			retval = 1;
11101			goto bailout;
11102		}
11103		if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11104		    (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11105			ctl_set_lun_standby(ctsio);
11106			retval = 1;
11107			goto bailout;
11108		}
11109
11110		/* The rest of checks are only done on executing side */
11111		if (softc->ha_mode == CTL_HA_MODE_XFER)
11112			goto bailout;
11113	}
11114
11115	if (entry->pattern & CTL_LUN_PAT_WRITE) {
11116		if (lun->be_lun &&
11117		    lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11118			ctl_set_hw_write_protected(ctsio);
11119			retval = 1;
11120			goto bailout;
11121		}
11122		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11123		    .eca_and_aen & SCP_SWP) != 0) {
11124			ctl_set_sense(ctsio, /*current_error*/ 1,
11125			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11126			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11127			retval = 1;
11128			goto bailout;
11129		}
11130	}
11131
11132	/*
11133	 * Check for a reservation conflict.  If this command isn't allowed
11134	 * even on reserved LUNs, and if this initiator isn't the one who
11135	 * reserved us, reject the command with a reservation conflict.
11136	 */
11137	residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11138	if ((lun->flags & CTL_LUN_RESERVED)
11139	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11140		if (lun->res_idx != residx) {
11141			ctl_set_reservation_conflict(ctsio);
11142			retval = 1;
11143			goto bailout;
11144		}
11145	}
11146
11147	if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11148	    (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11149		/* No reservation or command is allowed. */;
11150	} else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11151	    (lun->res_type == SPR_TYPE_WR_EX ||
11152	     lun->res_type == SPR_TYPE_WR_EX_RO ||
11153	     lun->res_type == SPR_TYPE_WR_EX_AR)) {
11154		/* The command is allowed for Write Exclusive resv. */;
11155	} else {
11156		/*
11157		 * if we aren't registered or it's a res holder type
11158		 * reservation and this isn't the res holder then set a
11159		 * conflict.
11160		 */
11161		if (ctl_get_prkey(lun, residx) == 0
11162		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11163			ctl_set_reservation_conflict(ctsio);
11164			retval = 1;
11165			goto bailout;
11166		}
11167	}
11168
11169	if ((lun->flags & CTL_LUN_OFFLINE)
11170	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0)) {
11171		ctl_set_lun_not_ready(ctsio);
11172		retval = 1;
11173		goto bailout;
11174	}
11175
11176	if ((lun->flags & CTL_LUN_STOPPED)
11177	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11178		/* "Logical unit not ready, initializing cmd. required" */
11179		ctl_set_lun_stopped(ctsio);
11180		retval = 1;
11181		goto bailout;
11182	}
11183
11184	if ((lun->flags & CTL_LUN_INOPERABLE)
11185	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11186		/* "Medium format corrupted" */
11187		ctl_set_medium_format_corrupted(ctsio);
11188		retval = 1;
11189		goto bailout;
11190	}
11191
11192bailout:
11193	return (retval);
11194}
11195
11196static void
11197ctl_failover_io(union ctl_io *io, int have_lock)
11198{
11199	ctl_set_busy(&io->scsiio);
11200	ctl_done(io);
11201}
11202
11203static void
11204ctl_failover_lun(union ctl_io *rio)
11205{
11206	struct ctl_softc *softc = control_softc;
11207	struct ctl_lun *lun;
11208	struct ctl_io_hdr *io, *next_io;
11209	uint32_t targ_lun;
11210
11211	targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11212	CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11213
11214	/* Find and lock the LUN. */
11215	mtx_lock(&softc->ctl_lock);
11216	if ((targ_lun < CTL_MAX_LUNS) &&
11217	    ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11218		mtx_lock(&lun->lun_lock);
11219		mtx_unlock(&softc->ctl_lock);
11220		if (lun->flags & CTL_LUN_DISABLED) {
11221			mtx_unlock(&lun->lun_lock);
11222			return;
11223		}
11224	} else {
11225		mtx_unlock(&softc->ctl_lock);
11226		return;
11227	}
11228
11229	if (softc->ha_mode == CTL_HA_MODE_XFER) {
11230		TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11231			/* We are master */
11232			if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11233				if (io->flags & CTL_FLAG_IO_ACTIVE) {
11234					io->flags |= CTL_FLAG_ABORT;
11235					io->flags |= CTL_FLAG_FAILOVER;
11236				} else { /* This can be only due to DATAMOVE */
11237					io->msg_type = CTL_MSG_DATAMOVE_DONE;
11238					io->flags &= ~CTL_FLAG_DMA_INPROG;
11239					io->flags |= CTL_FLAG_IO_ACTIVE;
11240					io->port_status = 31340;
11241					ctl_enqueue_isc((union ctl_io *)io);
11242				}
11243			}
11244			/* We are slave */
11245			if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11246				io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11247				if (io->flags & CTL_FLAG_IO_ACTIVE) {
11248					io->flags |= CTL_FLAG_FAILOVER;
11249				} else {
11250					ctl_set_busy(&((union ctl_io *)io)->
11251					    scsiio);
11252					ctl_done((union ctl_io *)io);
11253				}
11254			}
11255		}
11256	} else { /* SERIALIZE modes */
11257		TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11258		    next_io) {
11259			/* We are master */
11260			if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11261				TAILQ_REMOVE(&lun->blocked_queue, io,
11262				    blocked_links);
11263				io->flags &= ~CTL_FLAG_BLOCKED;
11264				TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11265				ctl_free_io((union ctl_io *)io);
11266			}
11267		}
11268		TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11269			/* We are master */
11270			if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11271				TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11272				ctl_free_io((union ctl_io *)io);
11273			}
11274			/* We are slave */
11275			if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11276				io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11277				if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11278					ctl_set_busy(&((union ctl_io *)io)->
11279					    scsiio);
11280					ctl_done((union ctl_io *)io);
11281				}
11282			}
11283		}
11284		ctl_check_blocked(lun);
11285	}
11286	mtx_unlock(&lun->lun_lock);
11287}
11288
11289static int
11290ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11291{
11292	struct ctl_lun *lun;
11293	const struct ctl_cmd_entry *entry;
11294	uint32_t initidx, targ_lun;
11295	int retval;
11296
11297	retval = 0;
11298
11299	lun = NULL;
11300
11301	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11302	if ((targ_lun < CTL_MAX_LUNS)
11303	 && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11304		/*
11305		 * If the LUN is invalid, pretend that it doesn't exist.
11306		 * It will go away as soon as all pending I/O has been
11307		 * completed.
11308		 */
11309		mtx_lock(&lun->lun_lock);
11310		if (lun->flags & CTL_LUN_DISABLED) {
11311			mtx_unlock(&lun->lun_lock);
11312			lun = NULL;
11313			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11314			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11315		} else {
11316			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11317			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11318				lun->be_lun;
11319
11320			/*
11321			 * Every I/O goes into the OOA queue for a
11322			 * particular LUN, and stays there until completion.
11323			 */
11324#ifdef CTL_TIME_IO
11325			if (TAILQ_EMPTY(&lun->ooa_queue)) {
11326				lun->idle_time += getsbinuptime() -
11327				    lun->last_busy;
11328			}
11329#endif
11330			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11331			    ooa_links);
11332		}
11333	} else {
11334		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11335		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11336	}
11337
11338	/* Get command entry and return error if it is unsuppotyed. */
11339	entry = ctl_validate_command(ctsio);
11340	if (entry == NULL) {
11341		if (lun)
11342			mtx_unlock(&lun->lun_lock);
11343		return (retval);
11344	}
11345
11346	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11347	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11348
11349	/*
11350	 * Check to see whether we can send this command to LUNs that don't
11351	 * exist.  This should pretty much only be the case for inquiry
11352	 * and request sense.  Further checks, below, really require having
11353	 * a LUN, so we can't really check the command anymore.  Just put
11354	 * it on the rtr queue.
11355	 */
11356	if (lun == NULL) {
11357		if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11358			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11359			ctl_enqueue_rtr((union ctl_io *)ctsio);
11360			return (retval);
11361		}
11362
11363		ctl_set_unsupported_lun(ctsio);
11364		ctl_done((union ctl_io *)ctsio);
11365		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11366		return (retval);
11367	} else {
11368		/*
11369		 * Make sure we support this particular command on this LUN.
11370		 * e.g., we don't support writes to the control LUN.
11371		 */
11372		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11373			mtx_unlock(&lun->lun_lock);
11374			ctl_set_invalid_opcode(ctsio);
11375			ctl_done((union ctl_io *)ctsio);
11376			return (retval);
11377		}
11378	}
11379
11380	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11381
11382#ifdef CTL_WITH_CA
11383	/*
11384	 * If we've got a request sense, it'll clear the contingent
11385	 * allegiance condition.  Otherwise, if we have a CA condition for
11386	 * this initiator, clear it, because it sent down a command other
11387	 * than request sense.
11388	 */
11389	if ((ctsio->cdb[0] != REQUEST_SENSE)
11390	 && (ctl_is_set(lun->have_ca, initidx)))
11391		ctl_clear_mask(lun->have_ca, initidx);
11392#endif
11393
11394	/*
11395	 * If the command has this flag set, it handles its own unit
11396	 * attention reporting, we shouldn't do anything.  Otherwise we
11397	 * check for any pending unit attentions, and send them back to the
11398	 * initiator.  We only do this when a command initially comes in,
11399	 * not when we pull it off the blocked queue.
11400	 *
11401	 * According to SAM-3, section 5.3.2, the order that things get
11402	 * presented back to the host is basically unit attentions caused
11403	 * by some sort of reset event, busy status, reservation conflicts
11404	 * or task set full, and finally any other status.
11405	 *
11406	 * One issue here is that some of the unit attentions we report
11407	 * don't fall into the "reset" category (e.g. "reported luns data
11408	 * has changed").  So reporting it here, before the reservation
11409	 * check, may be technically wrong.  I guess the only thing to do
11410	 * would be to check for and report the reset events here, and then
11411	 * check for the other unit attention types after we check for a
11412	 * reservation conflict.
11413	 *
11414	 * XXX KDM need to fix this
11415	 */
11416	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11417		ctl_ua_type ua_type;
11418
11419		ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11420		    SSD_TYPE_NONE);
11421		if (ua_type != CTL_UA_NONE) {
11422			mtx_unlock(&lun->lun_lock);
11423			ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11424			ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11425			ctsio->sense_len = SSD_FULL_SIZE;
11426			ctl_done((union ctl_io *)ctsio);
11427			return (retval);
11428		}
11429	}
11430
11431
11432	if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11433		mtx_unlock(&lun->lun_lock);
11434		ctl_done((union ctl_io *)ctsio);
11435		return (retval);
11436	}
11437
11438	/*
11439	 * XXX CHD this is where we want to send IO to other side if
11440	 * this LUN is secondary on this SC. We will need to make a copy
11441	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11442	 * the copy we send as FROM_OTHER.
11443	 * We also need to stuff the address of the original IO so we can
11444	 * find it easily. Something similar will need be done on the other
11445	 * side so when we are done we can find the copy.
11446	 */
11447	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11448	    (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11449	    (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11450		union ctl_ha_msg msg_info;
11451		int isc_retval;
11452
11453		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11454		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11455		mtx_unlock(&lun->lun_lock);
11456
11457		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11458		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11459		msg_info.hdr.serializing_sc = NULL;
11460		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11461		msg_info.scsi.tag_num = ctsio->tag_num;
11462		msg_info.scsi.tag_type = ctsio->tag_type;
11463		msg_info.scsi.cdb_len = ctsio->cdb_len;
11464		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11465
11466		if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11467		    sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11468		    M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11469			ctl_set_busy(ctsio);
11470			ctl_done((union ctl_io *)ctsio);
11471			return (retval);
11472		}
11473		return (retval);
11474	}
11475
11476	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11477			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11478			      ctl_ooaq, ooa_links))) {
11479	case CTL_ACTION_BLOCK:
11480		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11481		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11482				  blocked_links);
11483		mtx_unlock(&lun->lun_lock);
11484		return (retval);
11485	case CTL_ACTION_PASS:
11486	case CTL_ACTION_SKIP:
11487		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11488		mtx_unlock(&lun->lun_lock);
11489		ctl_enqueue_rtr((union ctl_io *)ctsio);
11490		break;
11491	case CTL_ACTION_OVERLAP:
11492		mtx_unlock(&lun->lun_lock);
11493		ctl_set_overlapped_cmd(ctsio);
11494		ctl_done((union ctl_io *)ctsio);
11495		break;
11496	case CTL_ACTION_OVERLAP_TAG:
11497		mtx_unlock(&lun->lun_lock);
11498		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11499		ctl_done((union ctl_io *)ctsio);
11500		break;
11501	case CTL_ACTION_ERROR:
11502	default:
11503		mtx_unlock(&lun->lun_lock);
11504		ctl_set_internal_failure(ctsio,
11505					 /*sks_valid*/ 0,
11506					 /*retry_count*/ 0);
11507		ctl_done((union ctl_io *)ctsio);
11508		break;
11509	}
11510	return (retval);
11511}
11512
11513const struct ctl_cmd_entry *
11514ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11515{
11516	const struct ctl_cmd_entry *entry;
11517	int service_action;
11518
11519	entry = &ctl_cmd_table[ctsio->cdb[0]];
11520	if (sa)
11521		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11522	if (entry->flags & CTL_CMD_FLAG_SA5) {
11523		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11524		entry = &((const struct ctl_cmd_entry *)
11525		    entry->execute)[service_action];
11526	}
11527	return (entry);
11528}
11529
11530const struct ctl_cmd_entry *
11531ctl_validate_command(struct ctl_scsiio *ctsio)
11532{
11533	const struct ctl_cmd_entry *entry;
11534	int i, sa;
11535	uint8_t diff;
11536
11537	entry = ctl_get_cmd_entry(ctsio, &sa);
11538	if (entry->execute == NULL) {
11539		if (sa)
11540			ctl_set_invalid_field(ctsio,
11541					      /*sks_valid*/ 1,
11542					      /*command*/ 1,
11543					      /*field*/ 1,
11544					      /*bit_valid*/ 1,
11545					      /*bit*/ 4);
11546		else
11547			ctl_set_invalid_opcode(ctsio);
11548		ctl_done((union ctl_io *)ctsio);
11549		return (NULL);
11550	}
11551	KASSERT(entry->length > 0,
11552	    ("Not defined length for command 0x%02x/0x%02x",
11553	     ctsio->cdb[0], ctsio->cdb[1]));
11554	for (i = 1; i < entry->length; i++) {
11555		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11556		if (diff == 0)
11557			continue;
11558		ctl_set_invalid_field(ctsio,
11559				      /*sks_valid*/ 1,
11560				      /*command*/ 1,
11561				      /*field*/ i,
11562				      /*bit_valid*/ 1,
11563				      /*bit*/ fls(diff) - 1);
11564		ctl_done((union ctl_io *)ctsio);
11565		return (NULL);
11566	}
11567	return (entry);
11568}
11569
11570static int
11571ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11572{
11573
11574	switch (lun_type) {
11575	case T_PROCESSOR:
11576		if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11577			return (0);
11578		break;
11579	case T_DIRECT:
11580		if ((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
11581			return (0);
11582		break;
11583	default:
11584		return (0);
11585	}
11586	return (1);
11587}
11588
11589static int
11590ctl_scsiio(struct ctl_scsiio *ctsio)
11591{
11592	int retval;
11593	const struct ctl_cmd_entry *entry;
11594
11595	retval = CTL_RETVAL_COMPLETE;
11596
11597	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11598
11599	entry = ctl_get_cmd_entry(ctsio, NULL);
11600
11601	/*
11602	 * If this I/O has been aborted, just send it straight to
11603	 * ctl_done() without executing it.
11604	 */
11605	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11606		ctl_done((union ctl_io *)ctsio);
11607		goto bailout;
11608	}
11609
11610	/*
11611	 * All the checks should have been handled by ctl_scsiio_precheck().
11612	 * We should be clear now to just execute the I/O.
11613	 */
11614	retval = entry->execute(ctsio);
11615
11616bailout:
11617	return (retval);
11618}
11619
11620/*
11621 * Since we only implement one target right now, a bus reset simply resets
11622 * our single target.
11623 */
11624static int
11625ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11626{
11627	return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11628}
11629
11630static int
11631ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11632		 ctl_ua_type ua_type)
11633{
11634	struct ctl_port *port;
11635	struct ctl_lun *lun;
11636	int retval;
11637
11638	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11639		union ctl_ha_msg msg_info;
11640
11641		msg_info.hdr.nexus = io->io_hdr.nexus;
11642		if (ua_type==CTL_UA_TARG_RESET)
11643			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11644		else
11645			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11646		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11647		msg_info.hdr.original_sc = NULL;
11648		msg_info.hdr.serializing_sc = NULL;
11649		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11650		    sizeof(msg_info.task), M_WAITOK);
11651	}
11652	retval = 0;
11653
11654	mtx_lock(&softc->ctl_lock);
11655	port = ctl_io_port(&io->io_hdr);
11656	STAILQ_FOREACH(lun, &softc->lun_list, links) {
11657		if (port != NULL &&
11658		    ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
11659			continue;
11660		retval += ctl_do_lun_reset(lun, io, ua_type);
11661	}
11662	mtx_unlock(&softc->ctl_lock);
11663	io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11664	return (retval);
11665}
11666
11667/*
11668 * The LUN should always be set.  The I/O is optional, and is used to
11669 * distinguish between I/Os sent by this initiator, and by other
11670 * initiators.  We set unit attention for initiators other than this one.
11671 * SAM-3 is vague on this point.  It does say that a unit attention should
11672 * be established for other initiators when a LUN is reset (see section
11673 * 5.7.3), but it doesn't specifically say that the unit attention should
11674 * be established for this particular initiator when a LUN is reset.  Here
11675 * is the relevant text, from SAM-3 rev 8:
11676 *
11677 * 5.7.2 When a SCSI initiator port aborts its own tasks
11678 *
11679 * When a SCSI initiator port causes its own task(s) to be aborted, no
11680 * notification that the task(s) have been aborted shall be returned to
11681 * the SCSI initiator port other than the completion response for the
11682 * command or task management function action that caused the task(s) to
11683 * be aborted and notification(s) associated with related effects of the
11684 * action (e.g., a reset unit attention condition).
11685 *
11686 * XXX KDM for now, we're setting unit attention for all initiators.
11687 */
11688static int
11689ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11690{
11691	union ctl_io *xio;
11692#if 0
11693	uint32_t initidx;
11694#endif
11695#ifdef CTL_WITH_CA
11696	int i;
11697#endif
11698
11699	mtx_lock(&lun->lun_lock);
11700	/*
11701	 * Run through the OOA queue and abort each I/O.
11702	 */
11703	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11704	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11705		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11706	}
11707
11708	/*
11709	 * This version sets unit attention for every
11710	 */
11711#if 0
11712	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11713	ctl_est_ua_all(lun, initidx, ua_type);
11714#else
11715	ctl_est_ua_all(lun, -1, ua_type);
11716#endif
11717
11718	/*
11719	 * A reset (any kind, really) clears reservations established with
11720	 * RESERVE/RELEASE.  It does not clear reservations established
11721	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11722	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11723	 * reservations made with the RESERVE/RELEASE commands, because
11724	 * those commands are obsolete in SPC-3.
11725	 */
11726	lun->flags &= ~CTL_LUN_RESERVED;
11727
11728#ifdef CTL_WITH_CA
11729	for (i = 0; i < CTL_MAX_INITIATORS; i++)
11730		ctl_clear_mask(lun->have_ca, i);
11731#endif
11732	mtx_unlock(&lun->lun_lock);
11733
11734	return (0);
11735}
11736
11737static int
11738ctl_lun_reset(struct ctl_softc *softc, union ctl_io *io)
11739{
11740	struct ctl_lun *lun;
11741	uint32_t targ_lun;
11742	int retval;
11743
11744	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11745	mtx_lock(&softc->ctl_lock);
11746	if ((targ_lun >= CTL_MAX_LUNS) ||
11747	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
11748		mtx_unlock(&softc->ctl_lock);
11749		io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11750		return (1);
11751	}
11752	retval = ctl_do_lun_reset(lun, io, CTL_UA_LUN_RESET);
11753	mtx_unlock(&softc->ctl_lock);
11754	io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11755
11756	if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11757		union ctl_ha_msg msg_info;
11758
11759		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11760		msg_info.hdr.nexus = io->io_hdr.nexus;
11761		msg_info.task.task_action = CTL_TASK_LUN_RESET;
11762		msg_info.hdr.original_sc = NULL;
11763		msg_info.hdr.serializing_sc = NULL;
11764		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11765		    sizeof(msg_info.task), M_WAITOK);
11766	}
11767	return (retval);
11768}
11769
11770static void
11771ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11772    int other_sc)
11773{
11774	union ctl_io *xio;
11775
11776	mtx_assert(&lun->lun_lock, MA_OWNED);
11777
11778	/*
11779	 * Run through the OOA queue and attempt to find the given I/O.
11780	 * The target port, initiator ID, tag type and tag number have to
11781	 * match the values that we got from the initiator.  If we have an
11782	 * untagged command to abort, simply abort the first untagged command
11783	 * we come to.  We only allow one untagged command at a time of course.
11784	 */
11785	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11786	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11787
11788		if ((targ_port == UINT32_MAX ||
11789		     targ_port == xio->io_hdr.nexus.targ_port) &&
11790		    (init_id == UINT32_MAX ||
11791		     init_id == xio->io_hdr.nexus.initid)) {
11792			if (targ_port != xio->io_hdr.nexus.targ_port ||
11793			    init_id != xio->io_hdr.nexus.initid)
11794				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11795			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11796			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11797				union ctl_ha_msg msg_info;
11798
11799				msg_info.hdr.nexus = xio->io_hdr.nexus;
11800				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11801				msg_info.task.tag_num = xio->scsiio.tag_num;
11802				msg_info.task.tag_type = xio->scsiio.tag_type;
11803				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11804				msg_info.hdr.original_sc = NULL;
11805				msg_info.hdr.serializing_sc = NULL;
11806				ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11807				    sizeof(msg_info.task), M_NOWAIT);
11808			}
11809		}
11810	}
11811}
11812
11813static int
11814ctl_abort_task_set(union ctl_io *io)
11815{
11816	struct ctl_softc *softc = control_softc;
11817	struct ctl_lun *lun;
11818	uint32_t targ_lun;
11819
11820	/*
11821	 * Look up the LUN.
11822	 */
11823	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11824	mtx_lock(&softc->ctl_lock);
11825	if ((targ_lun >= CTL_MAX_LUNS) ||
11826	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
11827		mtx_unlock(&softc->ctl_lock);
11828		io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11829		return (1);
11830	}
11831
11832	mtx_lock(&lun->lun_lock);
11833	mtx_unlock(&softc->ctl_lock);
11834	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11835		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11836		    io->io_hdr.nexus.initid,
11837		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11838	} else { /* CTL_TASK_CLEAR_TASK_SET */
11839		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11840		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11841	}
11842	mtx_unlock(&lun->lun_lock);
11843	io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11844	return (0);
11845}
11846
11847static int
11848ctl_i_t_nexus_reset(union ctl_io *io)
11849{
11850	struct ctl_softc *softc = control_softc;
11851	struct ctl_lun *lun;
11852	uint32_t initidx;
11853
11854	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11855		union ctl_ha_msg msg_info;
11856
11857		msg_info.hdr.nexus = io->io_hdr.nexus;
11858		msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
11859		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11860		msg_info.hdr.original_sc = NULL;
11861		msg_info.hdr.serializing_sc = NULL;
11862		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11863		    sizeof(msg_info.task), M_WAITOK);
11864	}
11865
11866	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11867	mtx_lock(&softc->ctl_lock);
11868	STAILQ_FOREACH(lun, &softc->lun_list, links) {
11869		mtx_lock(&lun->lun_lock);
11870		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11871		    io->io_hdr.nexus.initid, 1);
11872#ifdef CTL_WITH_CA
11873		ctl_clear_mask(lun->have_ca, initidx);
11874#endif
11875		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11876			lun->flags &= ~CTL_LUN_RESERVED;
11877		ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
11878		mtx_unlock(&lun->lun_lock);
11879	}
11880	mtx_unlock(&softc->ctl_lock);
11881	io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11882	return (0);
11883}
11884
11885static int
11886ctl_abort_task(union ctl_io *io)
11887{
11888	union ctl_io *xio;
11889	struct ctl_lun *lun;
11890	struct ctl_softc *softc;
11891#if 0
11892	struct sbuf sb;
11893	char printbuf[128];
11894#endif
11895	int found;
11896	uint32_t targ_lun;
11897
11898	softc = control_softc;
11899	found = 0;
11900
11901	/*
11902	 * Look up the LUN.
11903	 */
11904	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11905	mtx_lock(&softc->ctl_lock);
11906	if ((targ_lun >= CTL_MAX_LUNS) ||
11907	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
11908		mtx_unlock(&softc->ctl_lock);
11909		io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11910		return (1);
11911	}
11912
11913#if 0
11914	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11915	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11916#endif
11917
11918	mtx_lock(&lun->lun_lock);
11919	mtx_unlock(&softc->ctl_lock);
11920	/*
11921	 * Run through the OOA queue and attempt to find the given I/O.
11922	 * The target port, initiator ID, tag type and tag number have to
11923	 * match the values that we got from the initiator.  If we have an
11924	 * untagged command to abort, simply abort the first untagged command
11925	 * we come to.  We only allow one untagged command at a time of course.
11926	 */
11927	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11928	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11929#if 0
11930		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11931
11932		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11933			    lun->lun, xio->scsiio.tag_num,
11934			    xio->scsiio.tag_type,
11935			    (xio->io_hdr.blocked_links.tqe_prev
11936			    == NULL) ? "" : " BLOCKED",
11937			    (xio->io_hdr.flags &
11938			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11939			    (xio->io_hdr.flags &
11940			    CTL_FLAG_ABORT) ? " ABORT" : "",
11941			    (xio->io_hdr.flags &
11942			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11943		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11944		sbuf_finish(&sb);
11945		printf("%s\n", sbuf_data(&sb));
11946#endif
11947
11948		if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11949		 || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11950		 || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11951			continue;
11952
11953		/*
11954		 * If the abort says that the task is untagged, the
11955		 * task in the queue must be untagged.  Otherwise,
11956		 * we just check to see whether the tag numbers
11957		 * match.  This is because the QLogic firmware
11958		 * doesn't pass back the tag type in an abort
11959		 * request.
11960		 */
11961#if 0
11962		if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11963		  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11964		 || (xio->scsiio.tag_num == io->taskio.tag_num))
11965#endif
11966		/*
11967		 * XXX KDM we've got problems with FC, because it
11968		 * doesn't send down a tag type with aborts.  So we
11969		 * can only really go by the tag number...
11970		 * This may cause problems with parallel SCSI.
11971		 * Need to figure that out!!
11972		 */
11973		if (xio->scsiio.tag_num == io->taskio.tag_num) {
11974			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11975			found = 1;
11976			if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
11977			    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11978				union ctl_ha_msg msg_info;
11979
11980				msg_info.hdr.nexus = io->io_hdr.nexus;
11981				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11982				msg_info.task.tag_num = io->taskio.tag_num;
11983				msg_info.task.tag_type = io->taskio.tag_type;
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 0
11988				printf("Sent Abort to other side\n");
11989#endif
11990				ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11991				    sizeof(msg_info.task), M_NOWAIT);
11992			}
11993#if 0
11994			printf("ctl_abort_task: found I/O to abort\n");
11995#endif
11996		}
11997	}
11998	mtx_unlock(&lun->lun_lock);
11999
12000	if (found == 0) {
12001		/*
12002		 * This isn't really an error.  It's entirely possible for
12003		 * the abort and command completion to cross on the wire.
12004		 * This is more of an informative/diagnostic error.
12005		 */
12006#if 0
12007		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12008		       "%u:%u:%u tag %d type %d\n",
12009		       io->io_hdr.nexus.initid,
12010		       io->io_hdr.nexus.targ_port,
12011		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12012		       io->taskio.tag_type);
12013#endif
12014	}
12015	io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12016	return (0);
12017}
12018
12019static int
12020ctl_query_task(union ctl_io *io, int task_set)
12021{
12022	union ctl_io *xio;
12023	struct ctl_lun *lun;
12024	struct ctl_softc *softc;
12025	int found = 0;
12026	uint32_t targ_lun;
12027
12028	softc = control_softc;
12029	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12030	mtx_lock(&softc->ctl_lock);
12031	if ((targ_lun >= CTL_MAX_LUNS) ||
12032	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
12033		mtx_unlock(&softc->ctl_lock);
12034		io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12035		return (1);
12036	}
12037	mtx_lock(&lun->lun_lock);
12038	mtx_unlock(&softc->ctl_lock);
12039	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12040	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12041
12042		if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12043		 || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12044		 || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12045			continue;
12046
12047		if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12048			found = 1;
12049			break;
12050		}
12051	}
12052	mtx_unlock(&lun->lun_lock);
12053	if (found)
12054		io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12055	else
12056		io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12057	return (0);
12058}
12059
12060static int
12061ctl_query_async_event(union ctl_io *io)
12062{
12063	struct ctl_lun *lun;
12064	struct ctl_softc *softc;
12065	ctl_ua_type ua;
12066	uint32_t targ_lun, initidx;
12067
12068	softc = control_softc;
12069	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12070	mtx_lock(&softc->ctl_lock);
12071	if ((targ_lun >= CTL_MAX_LUNS) ||
12072	    (lun = softc->ctl_luns[targ_lun]) == NULL) {
12073		mtx_unlock(&softc->ctl_lock);
12074		io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12075		return (1);
12076	}
12077	mtx_lock(&lun->lun_lock);
12078	mtx_unlock(&softc->ctl_lock);
12079	initidx = ctl_get_initindex(&io->io_hdr.nexus);
12080	ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12081	mtx_unlock(&lun->lun_lock);
12082	if (ua != CTL_UA_NONE)
12083		io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12084	else
12085		io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12086	return (0);
12087}
12088
12089static void
12090ctl_run_task(union ctl_io *io)
12091{
12092	struct ctl_softc *softc = control_softc;
12093	int retval = 1;
12094
12095	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12096	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12097	    ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12098	io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12099	bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12100	switch (io->taskio.task_action) {
12101	case CTL_TASK_ABORT_TASK:
12102		retval = ctl_abort_task(io);
12103		break;
12104	case CTL_TASK_ABORT_TASK_SET:
12105	case CTL_TASK_CLEAR_TASK_SET:
12106		retval = ctl_abort_task_set(io);
12107		break;
12108	case CTL_TASK_CLEAR_ACA:
12109		break;
12110	case CTL_TASK_I_T_NEXUS_RESET:
12111		retval = ctl_i_t_nexus_reset(io);
12112		break;
12113	case CTL_TASK_LUN_RESET:
12114		retval = ctl_lun_reset(softc, io);
12115		break;
12116	case CTL_TASK_TARGET_RESET:
12117		retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
12118		break;
12119	case CTL_TASK_BUS_RESET:
12120		retval = ctl_bus_reset(softc, io);
12121		break;
12122	case CTL_TASK_PORT_LOGIN:
12123		break;
12124	case CTL_TASK_PORT_LOGOUT:
12125		break;
12126	case CTL_TASK_QUERY_TASK:
12127		retval = ctl_query_task(io, 0);
12128		break;
12129	case CTL_TASK_QUERY_TASK_SET:
12130		retval = ctl_query_task(io, 1);
12131		break;
12132	case CTL_TASK_QUERY_ASYNC_EVENT:
12133		retval = ctl_query_async_event(io);
12134		break;
12135	default:
12136		printf("%s: got unknown task management event %d\n",
12137		       __func__, io->taskio.task_action);
12138		break;
12139	}
12140	if (retval == 0)
12141		io->io_hdr.status = CTL_SUCCESS;
12142	else
12143		io->io_hdr.status = CTL_ERROR;
12144	ctl_done(io);
12145}
12146
12147/*
12148 * For HA operation.  Handle commands that come in from the other
12149 * controller.
12150 */
12151static void
12152ctl_handle_isc(union ctl_io *io)
12153{
12154	int free_io;
12155	struct ctl_lun *lun;
12156	struct ctl_softc *softc = control_softc;
12157	uint32_t targ_lun;
12158
12159	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12160	lun = softc->ctl_luns[targ_lun];
12161
12162	switch (io->io_hdr.msg_type) {
12163	case CTL_MSG_SERIALIZE:
12164		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12165		break;
12166	case CTL_MSG_R2R: {
12167		const struct ctl_cmd_entry *entry;
12168
12169		/*
12170		 * This is only used in SER_ONLY mode.
12171		 */
12172		free_io = 0;
12173		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12174		mtx_lock(&lun->lun_lock);
12175		if (ctl_scsiio_lun_check(lun,
12176		    entry, (struct ctl_scsiio *)io) != 0) {
12177			mtx_unlock(&lun->lun_lock);
12178			ctl_done(io);
12179			break;
12180		}
12181		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12182		mtx_unlock(&lun->lun_lock);
12183		ctl_enqueue_rtr(io);
12184		break;
12185	}
12186	case CTL_MSG_FINISH_IO:
12187		if (softc->ha_mode == CTL_HA_MODE_XFER) {
12188			free_io = 0;
12189			ctl_done(io);
12190		} else {
12191			free_io = 1;
12192			mtx_lock(&lun->lun_lock);
12193			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12194				     ooa_links);
12195			ctl_check_blocked(lun);
12196			mtx_unlock(&lun->lun_lock);
12197		}
12198		break;
12199	case CTL_MSG_PERS_ACTION:
12200		ctl_hndl_per_res_out_on_other_sc(
12201			(union ctl_ha_msg *)&io->presio.pr_msg);
12202		free_io = 1;
12203		break;
12204	case CTL_MSG_BAD_JUJU:
12205		free_io = 0;
12206		ctl_done(io);
12207		break;
12208	case CTL_MSG_DATAMOVE:
12209		/* Only used in XFER mode */
12210		free_io = 0;
12211		ctl_datamove_remote(io);
12212		break;
12213	case CTL_MSG_DATAMOVE_DONE:
12214		/* Only used in XFER mode */
12215		free_io = 0;
12216		io->scsiio.be_move_done(io);
12217		break;
12218	case CTL_MSG_FAILOVER:
12219		ctl_failover_lun(io);
12220		free_io = 1;
12221		break;
12222	default:
12223		free_io = 1;
12224		printf("%s: Invalid message type %d\n",
12225		       __func__, io->io_hdr.msg_type);
12226		break;
12227	}
12228	if (free_io)
12229		ctl_free_io(io);
12230
12231}
12232
12233
12234/*
12235 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12236 * there is no match.
12237 */
12238static ctl_lun_error_pattern
12239ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12240{
12241	const struct ctl_cmd_entry *entry;
12242	ctl_lun_error_pattern filtered_pattern, pattern;
12243
12244	pattern = desc->error_pattern;
12245
12246	/*
12247	 * XXX KDM we need more data passed into this function to match a
12248	 * custom pattern, and we actually need to implement custom pattern
12249	 * matching.
12250	 */
12251	if (pattern & CTL_LUN_PAT_CMD)
12252		return (CTL_LUN_PAT_CMD);
12253
12254	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12255		return (CTL_LUN_PAT_ANY);
12256
12257	entry = ctl_get_cmd_entry(ctsio, NULL);
12258
12259	filtered_pattern = entry->pattern & pattern;
12260
12261	/*
12262	 * If the user requested specific flags in the pattern (e.g.
12263	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12264	 * flags.
12265	 *
12266	 * If the user did not specify any flags, it doesn't matter whether
12267	 * or not the command supports the flags.
12268	 */
12269	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12270	     (pattern & ~CTL_LUN_PAT_MASK))
12271		return (CTL_LUN_PAT_NONE);
12272
12273	/*
12274	 * If the user asked for a range check, see if the requested LBA
12275	 * range overlaps with this command's LBA range.
12276	 */
12277	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12278		uint64_t lba1;
12279		uint64_t len1;
12280		ctl_action action;
12281		int retval;
12282
12283		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12284		if (retval != 0)
12285			return (CTL_LUN_PAT_NONE);
12286
12287		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12288					      desc->lba_range.len, FALSE);
12289		/*
12290		 * A "pass" means that the LBA ranges don't overlap, so
12291		 * this doesn't match the user's range criteria.
12292		 */
12293		if (action == CTL_ACTION_PASS)
12294			return (CTL_LUN_PAT_NONE);
12295	}
12296
12297	return (filtered_pattern);
12298}
12299
12300static void
12301ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12302{
12303	struct ctl_error_desc *desc, *desc2;
12304
12305	mtx_assert(&lun->lun_lock, MA_OWNED);
12306
12307	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12308		ctl_lun_error_pattern pattern;
12309		/*
12310		 * Check to see whether this particular command matches
12311		 * the pattern in the descriptor.
12312		 */
12313		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12314		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12315			continue;
12316
12317		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12318		case CTL_LUN_INJ_ABORTED:
12319			ctl_set_aborted(&io->scsiio);
12320			break;
12321		case CTL_LUN_INJ_MEDIUM_ERR:
12322			ctl_set_medium_error(&io->scsiio,
12323			    (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12324			     CTL_FLAG_DATA_OUT);
12325			break;
12326		case CTL_LUN_INJ_UA:
12327			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12328			 * OCCURRED */
12329			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12330			break;
12331		case CTL_LUN_INJ_CUSTOM:
12332			/*
12333			 * We're assuming the user knows what he is doing.
12334			 * Just copy the sense information without doing
12335			 * checks.
12336			 */
12337			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12338			      MIN(sizeof(desc->custom_sense),
12339				  sizeof(io->scsiio.sense_data)));
12340			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12341			io->scsiio.sense_len = SSD_FULL_SIZE;
12342			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12343			break;
12344		case CTL_LUN_INJ_NONE:
12345		default:
12346			/*
12347			 * If this is an error injection type we don't know
12348			 * about, clear the continuous flag (if it is set)
12349			 * so it will get deleted below.
12350			 */
12351			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12352			break;
12353		}
12354		/*
12355		 * By default, each error injection action is a one-shot
12356		 */
12357		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12358			continue;
12359
12360		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12361
12362		free(desc, M_CTL);
12363	}
12364}
12365
12366#ifdef CTL_IO_DELAY
12367static void
12368ctl_datamove_timer_wakeup(void *arg)
12369{
12370	union ctl_io *io;
12371
12372	io = (union ctl_io *)arg;
12373
12374	ctl_datamove(io);
12375}
12376#endif /* CTL_IO_DELAY */
12377
12378void
12379ctl_datamove(union ctl_io *io)
12380{
12381	struct ctl_lun *lun;
12382	void (*fe_datamove)(union ctl_io *io);
12383
12384	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12385
12386	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12387
12388	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12389#ifdef CTL_TIME_IO
12390	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12391		char str[256];
12392		char path_str[64];
12393		struct sbuf sb;
12394
12395		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12396		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12397
12398		sbuf_cat(&sb, path_str);
12399		switch (io->io_hdr.io_type) {
12400		case CTL_IO_SCSI:
12401			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12402			sbuf_printf(&sb, "\n");
12403			sbuf_cat(&sb, path_str);
12404			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12405				    io->scsiio.tag_num, io->scsiio.tag_type);
12406			break;
12407		case CTL_IO_TASK:
12408			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12409				    "Tag Type: %d\n", io->taskio.task_action,
12410				    io->taskio.tag_num, io->taskio.tag_type);
12411			break;
12412		default:
12413			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12414			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12415			break;
12416		}
12417		sbuf_cat(&sb, path_str);
12418		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12419			    (intmax_t)time_uptime - io->io_hdr.start_time);
12420		sbuf_finish(&sb);
12421		printf("%s", sbuf_data(&sb));
12422	}
12423#endif /* CTL_TIME_IO */
12424
12425#ifdef CTL_IO_DELAY
12426	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12427		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12428	} else {
12429		if ((lun != NULL)
12430		 && (lun->delay_info.datamove_delay > 0)) {
12431
12432			callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12433			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12434			callout_reset(&io->io_hdr.delay_callout,
12435				      lun->delay_info.datamove_delay * hz,
12436				      ctl_datamove_timer_wakeup, io);
12437			if (lun->delay_info.datamove_type ==
12438			    CTL_DELAY_TYPE_ONESHOT)
12439				lun->delay_info.datamove_delay = 0;
12440			return;
12441		}
12442	}
12443#endif
12444
12445	/*
12446	 * This command has been aborted.  Set the port status, so we fail
12447	 * the data move.
12448	 */
12449	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12450		printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12451		       io->scsiio.tag_num, io->io_hdr.nexus.initid,
12452		       io->io_hdr.nexus.targ_port,
12453		       io->io_hdr.nexus.targ_lun);
12454		io->io_hdr.port_status = 31337;
12455		/*
12456		 * Note that the backend, in this case, will get the
12457		 * callback in its context.  In other cases it may get
12458		 * called in the frontend's interrupt thread context.
12459		 */
12460		io->scsiio.be_move_done(io);
12461		return;
12462	}
12463
12464	/* Don't confuse frontend with zero length data move. */
12465	if (io->scsiio.kern_data_len == 0) {
12466		io->scsiio.be_move_done(io);
12467		return;
12468	}
12469
12470	fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12471	fe_datamove(io);
12472}
12473
12474static void
12475ctl_send_datamove_done(union ctl_io *io, int have_lock)
12476{
12477	union ctl_ha_msg msg;
12478#ifdef CTL_TIME_IO
12479	struct bintime cur_bt;
12480#endif
12481
12482	memset(&msg, 0, sizeof(msg));
12483	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12484	msg.hdr.original_sc = io;
12485	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12486	msg.hdr.nexus = io->io_hdr.nexus;
12487	msg.hdr.status = io->io_hdr.status;
12488	msg.scsi.tag_num = io->scsiio.tag_num;
12489	msg.scsi.tag_type = io->scsiio.tag_type;
12490	msg.scsi.scsi_status = io->scsiio.scsi_status;
12491	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12492	       io->scsiio.sense_len);
12493	msg.scsi.sense_len = io->scsiio.sense_len;
12494	msg.scsi.sense_residual = io->scsiio.sense_residual;
12495	msg.scsi.fetd_status = io->io_hdr.port_status;
12496	msg.scsi.residual = io->scsiio.residual;
12497	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12498	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12499		ctl_failover_io(io, /*have_lock*/ have_lock);
12500		return;
12501	}
12502	ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12503	    sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12504	    msg.scsi.sense_len, M_WAITOK);
12505
12506#ifdef CTL_TIME_IO
12507	getbinuptime(&cur_bt);
12508	bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12509	bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12510#endif
12511	io->io_hdr.num_dmas++;
12512}
12513
12514/*
12515 * The DMA to the remote side is done, now we need to tell the other side
12516 * we're done so it can continue with its data movement.
12517 */
12518static void
12519ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12520{
12521	union ctl_io *io;
12522	int i;
12523
12524	io = rq->context;
12525
12526	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12527		printf("%s: ISC DMA write failed with error %d", __func__,
12528		       rq->ret);
12529		ctl_set_internal_failure(&io->scsiio,
12530					 /*sks_valid*/ 1,
12531					 /*retry_count*/ rq->ret);
12532	}
12533
12534	ctl_dt_req_free(rq);
12535
12536	for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12537		free(io->io_hdr.local_sglist[i].addr, M_CTL);
12538	free(io->io_hdr.remote_sglist, M_CTL);
12539	io->io_hdr.remote_sglist = NULL;
12540	io->io_hdr.local_sglist = NULL;
12541
12542	/*
12543	 * The data is in local and remote memory, so now we need to send
12544	 * status (good or back) back to the other side.
12545	 */
12546	ctl_send_datamove_done(io, /*have_lock*/ 0);
12547}
12548
12549/*
12550 * We've moved the data from the host/controller into local memory.  Now we
12551 * need to push it over to the remote controller's memory.
12552 */
12553static int
12554ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12555{
12556	int retval;
12557
12558	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12559					  ctl_datamove_remote_write_cb);
12560	return (retval);
12561}
12562
12563static void
12564ctl_datamove_remote_write(union ctl_io *io)
12565{
12566	int retval;
12567	void (*fe_datamove)(union ctl_io *io);
12568
12569	/*
12570	 * - Get the data from the host/HBA into local memory.
12571	 * - DMA memory from the local controller to the remote controller.
12572	 * - Send status back to the remote controller.
12573	 */
12574
12575	retval = ctl_datamove_remote_sgl_setup(io);
12576	if (retval != 0)
12577		return;
12578
12579	/* Switch the pointer over so the FETD knows what to do */
12580	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12581
12582	/*
12583	 * Use a custom move done callback, since we need to send completion
12584	 * back to the other controller, not to the backend on this side.
12585	 */
12586	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12587
12588	fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12589	fe_datamove(io);
12590}
12591
12592static int
12593ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12594{
12595#if 0
12596	char str[256];
12597	char path_str[64];
12598	struct sbuf sb;
12599#endif
12600	int i;
12601
12602	for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12603		free(io->io_hdr.local_sglist[i].addr, M_CTL);
12604	free(io->io_hdr.remote_sglist, M_CTL);
12605	io->io_hdr.remote_sglist = NULL;
12606	io->io_hdr.local_sglist = NULL;
12607
12608#if 0
12609	scsi_path_string(io, path_str, sizeof(path_str));
12610	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12611	sbuf_cat(&sb, path_str);
12612	scsi_command_string(&io->scsiio, NULL, &sb);
12613	sbuf_printf(&sb, "\n");
12614	sbuf_cat(&sb, path_str);
12615	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12616		    io->scsiio.tag_num, io->scsiio.tag_type);
12617	sbuf_cat(&sb, path_str);
12618	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12619		    io->io_hdr.flags, io->io_hdr.status);
12620	sbuf_finish(&sb);
12621	printk("%s", sbuf_data(&sb));
12622#endif
12623
12624
12625	/*
12626	 * The read is done, now we need to send status (good or bad) back
12627	 * to the other side.
12628	 */
12629	ctl_send_datamove_done(io, /*have_lock*/ 0);
12630
12631	return (0);
12632}
12633
12634static void
12635ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12636{
12637	union ctl_io *io;
12638	void (*fe_datamove)(union ctl_io *io);
12639
12640	io = rq->context;
12641
12642	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12643		printf("%s: ISC DMA read failed with error %d\n", __func__,
12644		       rq->ret);
12645		ctl_set_internal_failure(&io->scsiio,
12646					 /*sks_valid*/ 1,
12647					 /*retry_count*/ rq->ret);
12648	}
12649
12650	ctl_dt_req_free(rq);
12651
12652	/* Switch the pointer over so the FETD knows what to do */
12653	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12654
12655	/*
12656	 * Use a custom move done callback, since we need to send completion
12657	 * back to the other controller, not to the backend on this side.
12658	 */
12659	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12660
12661	/* XXX KDM add checks like the ones in ctl_datamove? */
12662
12663	fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12664	fe_datamove(io);
12665}
12666
12667static int
12668ctl_datamove_remote_sgl_setup(union ctl_io *io)
12669{
12670	struct ctl_sg_entry *local_sglist;
12671	uint32_t len_to_go;
12672	int retval;
12673	int i;
12674
12675	retval = 0;
12676	local_sglist = io->io_hdr.local_sglist;
12677	len_to_go = io->scsiio.kern_data_len;
12678
12679	/*
12680	 * The difficult thing here is that the size of the various
12681	 * S/G segments may be different than the size from the
12682	 * remote controller.  That'll make it harder when DMAing
12683	 * the data back to the other side.
12684	 */
12685	for (i = 0; len_to_go > 0; i++) {
12686		local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12687		local_sglist[i].addr =
12688		    malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12689
12690		len_to_go -= local_sglist[i].len;
12691	}
12692	/*
12693	 * Reset the number of S/G entries accordingly.  The original
12694	 * number of S/G entries is available in rem_sg_entries.
12695	 */
12696	io->scsiio.kern_sg_entries = i;
12697
12698#if 0
12699	printf("%s: kern_sg_entries = %d\n", __func__,
12700	       io->scsiio.kern_sg_entries);
12701	for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12702		printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12703		       local_sglist[i].addr, local_sglist[i].len);
12704#endif
12705
12706	return (retval);
12707}
12708
12709static int
12710ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12711			 ctl_ha_dt_cb callback)
12712{
12713	struct ctl_ha_dt_req *rq;
12714	struct ctl_sg_entry *remote_sglist, *local_sglist;
12715	uint32_t local_used, remote_used, total_used;
12716	int i, j, isc_ret;
12717
12718	rq = ctl_dt_req_alloc();
12719
12720	/*
12721	 * If we failed to allocate the request, and if the DMA didn't fail
12722	 * anyway, set busy status.  This is just a resource allocation
12723	 * failure.
12724	 */
12725	if ((rq == NULL)
12726	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12727	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12728		ctl_set_busy(&io->scsiio);
12729
12730	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12731	    (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12732
12733		if (rq != NULL)
12734			ctl_dt_req_free(rq);
12735
12736		/*
12737		 * The data move failed.  We need to return status back
12738		 * to the other controller.  No point in trying to DMA
12739		 * data to the remote controller.
12740		 */
12741
12742		ctl_send_datamove_done(io, /*have_lock*/ 0);
12743
12744		return (1);
12745	}
12746
12747	local_sglist = io->io_hdr.local_sglist;
12748	remote_sglist = io->io_hdr.remote_sglist;
12749	local_used = 0;
12750	remote_used = 0;
12751	total_used = 0;
12752
12753	/*
12754	 * Pull/push the data over the wire from/to the other controller.
12755	 * This takes into account the possibility that the local and
12756	 * remote sglists may not be identical in terms of the size of
12757	 * the elements and the number of elements.
12758	 *
12759	 * One fundamental assumption here is that the length allocated for
12760	 * both the local and remote sglists is identical.  Otherwise, we've
12761	 * essentially got a coding error of some sort.
12762	 */
12763	isc_ret = CTL_HA_STATUS_SUCCESS;
12764	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12765		uint32_t cur_len;
12766		uint8_t *tmp_ptr;
12767
12768		rq->command = command;
12769		rq->context = io;
12770
12771		/*
12772		 * Both pointers should be aligned.  But it is possible
12773		 * that the allocation length is not.  They should both
12774		 * also have enough slack left over at the end, though,
12775		 * to round up to the next 8 byte boundary.
12776		 */
12777		cur_len = MIN(local_sglist[i].len - local_used,
12778			      remote_sglist[j].len - remote_used);
12779		rq->size = cur_len;
12780
12781		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12782		tmp_ptr += local_used;
12783
12784#if 0
12785		/* Use physical addresses when talking to ISC hardware */
12786		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12787			/* XXX KDM use busdma */
12788			rq->local = vtophys(tmp_ptr);
12789		} else
12790			rq->local = tmp_ptr;
12791#else
12792		KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12793		    ("HA does not support BUS_ADDR"));
12794		rq->local = tmp_ptr;
12795#endif
12796
12797		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12798		tmp_ptr += remote_used;
12799		rq->remote = tmp_ptr;
12800
12801		rq->callback = NULL;
12802
12803		local_used += cur_len;
12804		if (local_used >= local_sglist[i].len) {
12805			i++;
12806			local_used = 0;
12807		}
12808
12809		remote_used += cur_len;
12810		if (remote_used >= remote_sglist[j].len) {
12811			j++;
12812			remote_used = 0;
12813		}
12814		total_used += cur_len;
12815
12816		if (total_used >= io->scsiio.kern_data_len)
12817			rq->callback = callback;
12818
12819#if 0
12820		printf("%s: %s: local %p remote %p size %d\n", __func__,
12821		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12822		       rq->local, rq->remote, rq->size);
12823#endif
12824
12825		isc_ret = ctl_dt_single(rq);
12826		if (isc_ret > CTL_HA_STATUS_SUCCESS)
12827			break;
12828	}
12829	if (isc_ret != CTL_HA_STATUS_WAIT) {
12830		rq->ret = isc_ret;
12831		callback(rq);
12832	}
12833
12834	return (0);
12835}
12836
12837static void
12838ctl_datamove_remote_read(union ctl_io *io)
12839{
12840	int retval;
12841	int i;
12842
12843	/*
12844	 * This will send an error to the other controller in the case of a
12845	 * failure.
12846	 */
12847	retval = ctl_datamove_remote_sgl_setup(io);
12848	if (retval != 0)
12849		return;
12850
12851	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12852					  ctl_datamove_remote_read_cb);
12853	if (retval != 0) {
12854		/*
12855		 * Make sure we free memory if there was an error..  The
12856		 * ctl_datamove_remote_xfer() function will send the
12857		 * datamove done message, or call the callback with an
12858		 * error if there is a problem.
12859		 */
12860		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12861			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12862		free(io->io_hdr.remote_sglist, M_CTL);
12863		io->io_hdr.remote_sglist = NULL;
12864		io->io_hdr.local_sglist = NULL;
12865	}
12866}
12867
12868/*
12869 * Process a datamove request from the other controller.  This is used for
12870 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12871 * first.  Once that is complete, the data gets DMAed into the remote
12872 * controller's memory.  For reads, we DMA from the remote controller's
12873 * memory into our memory first, and then move it out to the FETD.
12874 */
12875static void
12876ctl_datamove_remote(union ctl_io *io)
12877{
12878
12879	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12880
12881	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12882		ctl_failover_io(io, /*have_lock*/ 0);
12883		return;
12884	}
12885
12886	/*
12887	 * Note that we look for an aborted I/O here, but don't do some of
12888	 * the other checks that ctl_datamove() normally does.
12889	 * We don't need to run the datamove delay code, since that should
12890	 * have been done if need be on the other controller.
12891	 */
12892	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12893		printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
12894		       io->scsiio.tag_num, io->io_hdr.nexus.initid,
12895		       io->io_hdr.nexus.targ_port,
12896		       io->io_hdr.nexus.targ_lun);
12897		io->io_hdr.port_status = 31338;
12898		ctl_send_datamove_done(io, /*have_lock*/ 0);
12899		return;
12900	}
12901
12902	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
12903		ctl_datamove_remote_write(io);
12904	else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
12905		ctl_datamove_remote_read(io);
12906	else {
12907		io->io_hdr.port_status = 31339;
12908		ctl_send_datamove_done(io, /*have_lock*/ 0);
12909	}
12910}
12911
12912static void
12913ctl_process_done(union ctl_io *io)
12914{
12915	struct ctl_lun *lun;
12916	struct ctl_softc *softc = control_softc;
12917	void (*fe_done)(union ctl_io *io);
12918	union ctl_ha_msg msg;
12919	uint32_t targ_port = io->io_hdr.nexus.targ_port;
12920
12921	CTL_DEBUG_PRINT(("ctl_process_done\n"));
12922	fe_done = softc->ctl_ports[targ_port]->fe_done;
12923
12924#ifdef CTL_TIME_IO
12925	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12926		char str[256];
12927		char path_str[64];
12928		struct sbuf sb;
12929
12930		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12931		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12932
12933		sbuf_cat(&sb, path_str);
12934		switch (io->io_hdr.io_type) {
12935		case CTL_IO_SCSI:
12936			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12937			sbuf_printf(&sb, "\n");
12938			sbuf_cat(&sb, path_str);
12939			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12940				    io->scsiio.tag_num, io->scsiio.tag_type);
12941			break;
12942		case CTL_IO_TASK:
12943			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12944				    "Tag Type: %d\n", io->taskio.task_action,
12945				    io->taskio.tag_num, io->taskio.tag_type);
12946			break;
12947		default:
12948			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12949			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12950			break;
12951		}
12952		sbuf_cat(&sb, path_str);
12953		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12954			    (intmax_t)time_uptime - io->io_hdr.start_time);
12955		sbuf_finish(&sb);
12956		printf("%s", sbuf_data(&sb));
12957	}
12958#endif /* CTL_TIME_IO */
12959
12960	switch (io->io_hdr.io_type) {
12961	case CTL_IO_SCSI:
12962		break;
12963	case CTL_IO_TASK:
12964		if (ctl_debug & CTL_DEBUG_INFO)
12965			ctl_io_error_print(io, NULL);
12966		fe_done(io);
12967		return;
12968	default:
12969		panic("ctl_process_done: invalid io type %d\n",
12970		      io->io_hdr.io_type);
12971		break; /* NOTREACHED */
12972	}
12973
12974	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12975	if (lun == NULL) {
12976		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12977				 io->io_hdr.nexus.targ_mapped_lun));
12978		goto bailout;
12979	}
12980
12981	mtx_lock(&lun->lun_lock);
12982
12983	/*
12984	 * Check to see if we have any errors to inject here.  We only
12985	 * inject errors for commands that don't already have errors set.
12986	 */
12987	if ((STAILQ_FIRST(&lun->error_list) != NULL) &&
12988	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
12989	    ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
12990		ctl_inject_error(lun, io);
12991
12992	/*
12993	 * XXX KDM how do we treat commands that aren't completed
12994	 * successfully?
12995	 *
12996	 * XXX KDM should we also track I/O latency?
12997	 */
12998	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
12999	    io->io_hdr.io_type == CTL_IO_SCSI) {
13000#ifdef CTL_TIME_IO
13001		struct bintime cur_bt;
13002#endif
13003		int type;
13004
13005		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13006		    CTL_FLAG_DATA_IN)
13007			type = CTL_STATS_READ;
13008		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13009		    CTL_FLAG_DATA_OUT)
13010			type = CTL_STATS_WRITE;
13011		else
13012			type = CTL_STATS_NO_IO;
13013
13014		lun->stats.ports[targ_port].bytes[type] +=
13015		    io->scsiio.kern_total_len;
13016		lun->stats.ports[targ_port].operations[type]++;
13017#ifdef CTL_TIME_IO
13018		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13019		   &io->io_hdr.dma_bt);
13020		getbinuptime(&cur_bt);
13021		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13022		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13023#endif
13024		lun->stats.ports[targ_port].num_dmas[type] +=
13025		    io->io_hdr.num_dmas;
13026	}
13027
13028	/*
13029	 * Remove this from the OOA queue.
13030	 */
13031	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13032#ifdef CTL_TIME_IO
13033	if (TAILQ_EMPTY(&lun->ooa_queue))
13034		lun->last_busy = getsbinuptime();
13035#endif
13036
13037	/*
13038	 * Run through the blocked queue on this LUN and see if anything
13039	 * has become unblocked, now that this transaction is done.
13040	 */
13041	ctl_check_blocked(lun);
13042
13043	/*
13044	 * If the LUN has been invalidated, free it if there is nothing
13045	 * left on its OOA queue.
13046	 */
13047	if ((lun->flags & CTL_LUN_INVALID)
13048	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13049		mtx_unlock(&lun->lun_lock);
13050		mtx_lock(&softc->ctl_lock);
13051		ctl_free_lun(lun);
13052		mtx_unlock(&softc->ctl_lock);
13053	} else
13054		mtx_unlock(&lun->lun_lock);
13055
13056bailout:
13057
13058	/*
13059	 * If this command has been aborted, make sure we set the status
13060	 * properly.  The FETD is responsible for freeing the I/O and doing
13061	 * whatever it needs to do to clean up its state.
13062	 */
13063	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13064		ctl_set_task_aborted(&io->scsiio);
13065
13066	/*
13067	 * If enabled, print command error status.
13068	 */
13069	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13070	    (ctl_debug & CTL_DEBUG_INFO) != 0)
13071		ctl_io_error_print(io, NULL);
13072
13073	/*
13074	 * Tell the FETD or the other shelf controller we're done with this
13075	 * command.  Note that only SCSI commands get to this point.  Task
13076	 * management commands are completed above.
13077	 */
13078	if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13079	    (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13080		memset(&msg, 0, sizeof(msg));
13081		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13082		msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13083		msg.hdr.nexus = io->io_hdr.nexus;
13084		ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13085		    sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13086		    M_WAITOK);
13087	}
13088
13089	fe_done(io);
13090}
13091
13092#ifdef CTL_WITH_CA
13093/*
13094 * Front end should call this if it doesn't do autosense.  When the request
13095 * sense comes back in from the initiator, we'll dequeue this and send it.
13096 */
13097int
13098ctl_queue_sense(union ctl_io *io)
13099{
13100	struct ctl_lun *lun;
13101	struct ctl_port *port;
13102	struct ctl_softc *softc;
13103	uint32_t initidx, targ_lun;
13104
13105	softc = control_softc;
13106
13107	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13108
13109	/*
13110	 * LUN lookup will likely move to the ctl_work_thread() once we
13111	 * have our new queueing infrastructure (that doesn't put things on
13112	 * a per-LUN queue initially).  That is so that we can handle
13113	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13114	 * can't deal with that right now.
13115	 */
13116	mtx_lock(&softc->ctl_lock);
13117
13118	/*
13119	 * If we don't have a LUN for this, just toss the sense
13120	 * information.
13121	 */
13122	port = ctl_io_port(&ctsio->io_hdr);
13123	targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13124	if ((targ_lun < CTL_MAX_LUNS)
13125	 && (softc->ctl_luns[targ_lun] != NULL))
13126		lun = softc->ctl_luns[targ_lun];
13127	else
13128		goto bailout;
13129
13130	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13131
13132	mtx_lock(&lun->lun_lock);
13133	/*
13134	 * Already have CA set for this LUN...toss the sense information.
13135	 */
13136	if (ctl_is_set(lun->have_ca, initidx)) {
13137		mtx_unlock(&lun->lun_lock);
13138		goto bailout;
13139	}
13140
13141	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13142	       MIN(sizeof(lun->pending_sense[initidx]),
13143	       sizeof(io->scsiio.sense_data)));
13144	ctl_set_mask(lun->have_ca, initidx);
13145	mtx_unlock(&lun->lun_lock);
13146
13147bailout:
13148	mtx_unlock(&softc->ctl_lock);
13149
13150	ctl_free_io(io);
13151
13152	return (CTL_RETVAL_COMPLETE);
13153}
13154#endif
13155
13156/*
13157 * Primary command inlet from frontend ports.  All SCSI and task I/O
13158 * requests must go through this function.
13159 */
13160int
13161ctl_queue(union ctl_io *io)
13162{
13163	struct ctl_port *port;
13164
13165	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13166
13167#ifdef CTL_TIME_IO
13168	io->io_hdr.start_time = time_uptime;
13169	getbinuptime(&io->io_hdr.start_bt);
13170#endif /* CTL_TIME_IO */
13171
13172	/* Map FE-specific LUN ID into global one. */
13173	port = ctl_io_port(&io->io_hdr);
13174	io->io_hdr.nexus.targ_mapped_lun =
13175	    ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13176
13177	switch (io->io_hdr.io_type) {
13178	case CTL_IO_SCSI:
13179	case CTL_IO_TASK:
13180		if (ctl_debug & CTL_DEBUG_CDB)
13181			ctl_io_print(io);
13182		ctl_enqueue_incoming(io);
13183		break;
13184	default:
13185		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13186		return (EINVAL);
13187	}
13188
13189	return (CTL_RETVAL_COMPLETE);
13190}
13191
13192#ifdef CTL_IO_DELAY
13193static void
13194ctl_done_timer_wakeup(void *arg)
13195{
13196	union ctl_io *io;
13197
13198	io = (union ctl_io *)arg;
13199	ctl_done(io);
13200}
13201#endif /* CTL_IO_DELAY */
13202
13203void
13204ctl_serseq_done(union ctl_io *io)
13205{
13206	struct ctl_lun *lun;
13207
13208	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13209	if (lun->be_lun == NULL ||
13210	    lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13211		return;
13212	mtx_lock(&lun->lun_lock);
13213	io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13214	ctl_check_blocked(lun);
13215	mtx_unlock(&lun->lun_lock);
13216}
13217
13218void
13219ctl_done(union ctl_io *io)
13220{
13221
13222	/*
13223	 * Enable this to catch duplicate completion issues.
13224	 */
13225#if 0
13226	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13227		printf("%s: type %d msg %d cdb %x iptl: "
13228		       "%u:%u:%u tag 0x%04x "
13229		       "flag %#x status %x\n",
13230			__func__,
13231			io->io_hdr.io_type,
13232			io->io_hdr.msg_type,
13233			io->scsiio.cdb[0],
13234			io->io_hdr.nexus.initid,
13235			io->io_hdr.nexus.targ_port,
13236			io->io_hdr.nexus.targ_lun,
13237			(io->io_hdr.io_type ==
13238			CTL_IO_TASK) ?
13239			io->taskio.tag_num :
13240			io->scsiio.tag_num,
13241		        io->io_hdr.flags,
13242			io->io_hdr.status);
13243	} else
13244		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13245#endif
13246
13247	/*
13248	 * This is an internal copy of an I/O, and should not go through
13249	 * the normal done processing logic.
13250	 */
13251	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13252		return;
13253
13254#ifdef CTL_IO_DELAY
13255	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13256		struct ctl_lun *lun;
13257
13258		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13259
13260		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13261	} else {
13262		struct ctl_lun *lun;
13263
13264		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13265
13266		if ((lun != NULL)
13267		 && (lun->delay_info.done_delay > 0)) {
13268
13269			callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13270			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13271			callout_reset(&io->io_hdr.delay_callout,
13272				      lun->delay_info.done_delay * hz,
13273				      ctl_done_timer_wakeup, io);
13274			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13275				lun->delay_info.done_delay = 0;
13276			return;
13277		}
13278	}
13279#endif /* CTL_IO_DELAY */
13280
13281	ctl_enqueue_done(io);
13282}
13283
13284static void
13285ctl_work_thread(void *arg)
13286{
13287	struct ctl_thread *thr = (struct ctl_thread *)arg;
13288	struct ctl_softc *softc = thr->ctl_softc;
13289	union ctl_io *io;
13290	int retval;
13291
13292	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13293
13294	for (;;) {
13295		/*
13296		 * We handle the queues in this order:
13297		 * - ISC
13298		 * - done queue (to free up resources, unblock other commands)
13299		 * - RtR queue
13300		 * - incoming queue
13301		 *
13302		 * If those queues are empty, we break out of the loop and
13303		 * go to sleep.
13304		 */
13305		mtx_lock(&thr->queue_lock);
13306		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13307		if (io != NULL) {
13308			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13309			mtx_unlock(&thr->queue_lock);
13310			ctl_handle_isc(io);
13311			continue;
13312		}
13313		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13314		if (io != NULL) {
13315			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13316			/* clear any blocked commands, call fe_done */
13317			mtx_unlock(&thr->queue_lock);
13318			ctl_process_done(io);
13319			continue;
13320		}
13321		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13322		if (io != NULL) {
13323			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13324			mtx_unlock(&thr->queue_lock);
13325			if (io->io_hdr.io_type == CTL_IO_TASK)
13326				ctl_run_task(io);
13327			else
13328				ctl_scsiio_precheck(softc, &io->scsiio);
13329			continue;
13330		}
13331		io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13332		if (io != NULL) {
13333			STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13334			mtx_unlock(&thr->queue_lock);
13335			retval = ctl_scsiio(&io->scsiio);
13336			if (retval != CTL_RETVAL_COMPLETE)
13337				CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13338			continue;
13339		}
13340
13341		/* Sleep until we have something to do. */
13342		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13343	}
13344}
13345
13346static void
13347ctl_lun_thread(void *arg)
13348{
13349	struct ctl_softc *softc = (struct ctl_softc *)arg;
13350	struct ctl_be_lun *be_lun;
13351
13352	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13353
13354	for (;;) {
13355		mtx_lock(&softc->ctl_lock);
13356		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13357		if (be_lun != NULL) {
13358			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13359			mtx_unlock(&softc->ctl_lock);
13360			ctl_create_lun(be_lun);
13361			continue;
13362		}
13363
13364		/* Sleep until we have something to do. */
13365		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13366		    PDROP | PRIBIO, "-", 0);
13367	}
13368}
13369
13370static void
13371ctl_thresh_thread(void *arg)
13372{
13373	struct ctl_softc *softc = (struct ctl_softc *)arg;
13374	struct ctl_lun *lun;
13375	struct scsi_da_rw_recovery_page *rwpage;
13376	struct ctl_logical_block_provisioning_page *page;
13377	const char *attr;
13378	union ctl_ha_msg msg;
13379	uint64_t thres, val;
13380	int i, e, set;
13381
13382	CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13383
13384	for (;;) {
13385		mtx_lock(&softc->ctl_lock);
13386		STAILQ_FOREACH(lun, &softc->lun_list, links) {
13387			if ((lun->flags & CTL_LUN_DISABLED) ||
13388			    (lun->flags & CTL_LUN_OFFLINE) ||
13389			    lun->backend->lun_attr == NULL)
13390				continue;
13391			if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13392			    softc->ha_mode == CTL_HA_MODE_XFER)
13393				continue;
13394			rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
13395			if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
13396				continue;
13397			e = 0;
13398			page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
13399			for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13400				if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13401					continue;
13402				thres = scsi_4btoul(page->descr[i].count);
13403				thres <<= CTL_LBP_EXPONENT;
13404				switch (page->descr[i].resource) {
13405				case 0x01:
13406					attr = "blocksavail";
13407					break;
13408				case 0x02:
13409					attr = "blocksused";
13410					break;
13411				case 0xf1:
13412					attr = "poolblocksavail";
13413					break;
13414				case 0xf2:
13415					attr = "poolblocksused";
13416					break;
13417				default:
13418					continue;
13419				}
13420				mtx_unlock(&softc->ctl_lock); // XXX
13421				val = lun->backend->lun_attr(
13422				    lun->be_lun->be_lun, attr);
13423				mtx_lock(&softc->ctl_lock);
13424				if (val == UINT64_MAX)
13425					continue;
13426				if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13427				    == SLBPPD_ARMING_INC)
13428					e = (val >= thres);
13429				else
13430					e = (val <= thres);
13431				if (e)
13432					break;
13433			}
13434			mtx_lock(&lun->lun_lock);
13435			if (e) {
13436				scsi_u64to8b((uint8_t *)&page->descr[i] -
13437				    (uint8_t *)page, lun->ua_tpt_info);
13438				if (lun->lasttpt == 0 ||
13439				    time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13440					lun->lasttpt = time_uptime;
13441					ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13442					set = 1;
13443				} else
13444					set = 0;
13445			} else {
13446				lun->lasttpt = 0;
13447				ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13448				set = -1;
13449			}
13450			mtx_unlock(&lun->lun_lock);
13451			if (set != 0 &&
13452			    lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13453				/* Send msg to other side. */
13454				bzero(&msg.ua, sizeof(msg.ua));
13455				msg.hdr.msg_type = CTL_MSG_UA;
13456				msg.hdr.nexus.initid = -1;
13457				msg.hdr.nexus.targ_port = -1;
13458				msg.hdr.nexus.targ_lun = lun->lun;
13459				msg.hdr.nexus.targ_mapped_lun = lun->lun;
13460				msg.ua.ua_all = 1;
13461				msg.ua.ua_set = (set > 0);
13462				msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13463				memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13464				mtx_unlock(&softc->ctl_lock); // XXX
13465				ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13466				    sizeof(msg.ua), M_WAITOK);
13467				mtx_lock(&softc->ctl_lock);
13468			}
13469		}
13470		mtx_unlock(&softc->ctl_lock);
13471		pause("-", CTL_LBP_PERIOD * hz);
13472	}
13473}
13474
13475static void
13476ctl_enqueue_incoming(union ctl_io *io)
13477{
13478	struct ctl_softc *softc = control_softc;
13479	struct ctl_thread *thr;
13480	u_int idx;
13481
13482	idx = (io->io_hdr.nexus.targ_port * 127 +
13483	       io->io_hdr.nexus.initid) % worker_threads;
13484	thr = &softc->threads[idx];
13485	mtx_lock(&thr->queue_lock);
13486	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13487	mtx_unlock(&thr->queue_lock);
13488	wakeup(thr);
13489}
13490
13491static void
13492ctl_enqueue_rtr(union ctl_io *io)
13493{
13494	struct ctl_softc *softc = control_softc;
13495	struct ctl_thread *thr;
13496
13497	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13498	mtx_lock(&thr->queue_lock);
13499	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13500	mtx_unlock(&thr->queue_lock);
13501	wakeup(thr);
13502}
13503
13504static void
13505ctl_enqueue_done(union ctl_io *io)
13506{
13507	struct ctl_softc *softc = control_softc;
13508	struct ctl_thread *thr;
13509
13510	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13511	mtx_lock(&thr->queue_lock);
13512	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13513	mtx_unlock(&thr->queue_lock);
13514	wakeup(thr);
13515}
13516
13517static void
13518ctl_enqueue_isc(union ctl_io *io)
13519{
13520	struct ctl_softc *softc = control_softc;
13521	struct ctl_thread *thr;
13522
13523	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13524	mtx_lock(&thr->queue_lock);
13525	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13526	mtx_unlock(&thr->queue_lock);
13527	wakeup(thr);
13528}
13529
13530/*
13531 *  vim: ts=8
13532 */
13533