ctl.c revision 288725
1/*-
2 * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3 * Copyright (c) 2012 The FreeBSD Foundation
4 * All rights reserved.
5 *
6 * Portions of this software were developed by Edward Tomasz Napierala
7 * under sponsorship from the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions, and the following disclaimer,
14 *    without modification.
15 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16 *    substantially similar to the "NO WARRANTY" disclaimer below
17 *    ("Disclaimer") and any redistribution must be conditioned upon
18 *    including a substantially similar Disclaimer requirement for further
19 *    binary redistribution.
20 *
21 * NO WARRANTY
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGES.
33 *
34 * $Id$
35 */
36/*
37 * CAM Target Layer, a SCSI device emulation subsystem.
38 *
39 * Author: Ken Merry <ken@FreeBSD.org>
40 */
41
42#define _CTL_C
43
44#include <sys/cdefs.h>
45__FBSDID("$FreeBSD: stable/10/sys/cam/ctl/ctl.c 288725 2015-10-05 08:48:47Z mav $");
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/ctype.h>
50#include <sys/kernel.h>
51#include <sys/types.h>
52#include <sys/kthread.h>
53#include <sys/bio.h>
54#include <sys/fcntl.h>
55#include <sys/lock.h>
56#include <sys/module.h>
57#include <sys/mutex.h>
58#include <sys/condvar.h>
59#include <sys/malloc.h>
60#include <sys/conf.h>
61#include <sys/ioccom.h>
62#include <sys/queue.h>
63#include <sys/sbuf.h>
64#include <sys/smp.h>
65#include <sys/endian.h>
66#include <sys/sysctl.h>
67#include <vm/uma.h>
68
69#include <cam/cam.h>
70#include <cam/scsi/scsi_all.h>
71#include <cam/scsi/scsi_da.h>
72#include <cam/ctl/ctl_io.h>
73#include <cam/ctl/ctl.h>
74#include <cam/ctl/ctl_frontend.h>
75#include <cam/ctl/ctl_util.h>
76#include <cam/ctl/ctl_backend.h>
77#include <cam/ctl/ctl_ioctl.h>
78#include <cam/ctl/ctl_ha.h>
79#include <cam/ctl/ctl_private.h>
80#include <cam/ctl/ctl_debug.h>
81#include <cam/ctl/ctl_scsi_all.h>
82#include <cam/ctl/ctl_error.h>
83
84struct ctl_softc *control_softc = NULL;
85
86/*
87 * Size and alignment macros needed for Copan-specific HA hardware.  These
88 * can go away when the HA code is re-written, and uses busdma for any
89 * hardware.
90 */
91#define	CTL_ALIGN_8B(target, source, type)				\
92	if (((uint32_t)source & 0x7) != 0)				\
93		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
94	else								\
95		target = (type)source;
96
97#define	CTL_SIZE_8B(target, size)					\
98	if ((size & 0x7) != 0)						\
99		target = size + (0x8 - (size & 0x7));			\
100	else								\
101		target = size;
102
103#define CTL_ALIGN_8B_MARGIN	16
104
105/*
106 * Template mode pages.
107 */
108
109/*
110 * Note that these are default values only.  The actual values will be
111 * filled in when the user does a mode sense.
112 */
113const static struct copan_debugconf_subpage debugconf_page_default = {
114	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
115	DBGCNF_SUBPAGE_CODE,		/* subpage */
116	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
117	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
118	DBGCNF_VERSION,			/* page_version */
119	{CTL_TIME_IO_DEFAULT_SECS>>8,
120	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
121};
122
123const static struct copan_debugconf_subpage debugconf_page_changeable = {
124	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
125	DBGCNF_SUBPAGE_CODE,		/* subpage */
126	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
127	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
128	0,				/* page_version */
129	{0xff,0xff},			/* ctl_time_io_secs */
130};
131
132const static struct scsi_da_rw_recovery_page rw_er_page_default = {
133	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
134	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
135	/*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
136	/*read_retry_count*/0,
137	/*correction_span*/0,
138	/*head_offset_count*/0,
139	/*data_strobe_offset_cnt*/0,
140	/*byte8*/SMS_RWER_LBPERE,
141	/*write_retry_count*/0,
142	/*reserved2*/0,
143	/*recovery_time_limit*/{0, 0},
144};
145
146const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
147	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
148	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
149	/*byte3*/0,
150	/*read_retry_count*/0,
151	/*correction_span*/0,
152	/*head_offset_count*/0,
153	/*data_strobe_offset_cnt*/0,
154	/*byte8*/0,
155	/*write_retry_count*/0,
156	/*reserved2*/0,
157	/*recovery_time_limit*/{0, 0},
158};
159
160const static struct scsi_format_page format_page_default = {
161	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
162	/*page_length*/sizeof(struct scsi_format_page) - 2,
163	/*tracks_per_zone*/ {0, 0},
164	/*alt_sectors_per_zone*/ {0, 0},
165	/*alt_tracks_per_zone*/ {0, 0},
166	/*alt_tracks_per_lun*/ {0, 0},
167	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
168			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
169	/*bytes_per_sector*/ {0, 0},
170	/*interleave*/ {0, 0},
171	/*track_skew*/ {0, 0},
172	/*cylinder_skew*/ {0, 0},
173	/*flags*/ SFP_HSEC,
174	/*reserved*/ {0, 0, 0}
175};
176
177const static struct scsi_format_page format_page_changeable = {
178	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
179	/*page_length*/sizeof(struct scsi_format_page) - 2,
180	/*tracks_per_zone*/ {0, 0},
181	/*alt_sectors_per_zone*/ {0, 0},
182	/*alt_tracks_per_zone*/ {0, 0},
183	/*alt_tracks_per_lun*/ {0, 0},
184	/*sectors_per_track*/ {0, 0},
185	/*bytes_per_sector*/ {0, 0},
186	/*interleave*/ {0, 0},
187	/*track_skew*/ {0, 0},
188	/*cylinder_skew*/ {0, 0},
189	/*flags*/ 0,
190	/*reserved*/ {0, 0, 0}
191};
192
193const static struct scsi_rigid_disk_page rigid_disk_page_default = {
194	/*page_code*/SMS_RIGID_DISK_PAGE,
195	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
196	/*cylinders*/ {0, 0, 0},
197	/*heads*/ CTL_DEFAULT_HEADS,
198	/*start_write_precomp*/ {0, 0, 0},
199	/*start_reduced_current*/ {0, 0, 0},
200	/*step_rate*/ {0, 0},
201	/*landing_zone_cylinder*/ {0, 0, 0},
202	/*rpl*/ SRDP_RPL_DISABLED,
203	/*rotational_offset*/ 0,
204	/*reserved1*/ 0,
205	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
206			   CTL_DEFAULT_ROTATION_RATE & 0xff},
207	/*reserved2*/ {0, 0}
208};
209
210const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
211	/*page_code*/SMS_RIGID_DISK_PAGE,
212	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
213	/*cylinders*/ {0, 0, 0},
214	/*heads*/ 0,
215	/*start_write_precomp*/ {0, 0, 0},
216	/*start_reduced_current*/ {0, 0, 0},
217	/*step_rate*/ {0, 0},
218	/*landing_zone_cylinder*/ {0, 0, 0},
219	/*rpl*/ 0,
220	/*rotational_offset*/ 0,
221	/*reserved1*/ 0,
222	/*rotation_rate*/ {0, 0},
223	/*reserved2*/ {0, 0}
224};
225
226const static struct scsi_caching_page caching_page_default = {
227	/*page_code*/SMS_CACHING_PAGE,
228	/*page_length*/sizeof(struct scsi_caching_page) - 2,
229	/*flags1*/ SCP_DISC | SCP_WCE,
230	/*ret_priority*/ 0,
231	/*disable_pf_transfer_len*/ {0xff, 0xff},
232	/*min_prefetch*/ {0, 0},
233	/*max_prefetch*/ {0xff, 0xff},
234	/*max_pf_ceiling*/ {0xff, 0xff},
235	/*flags2*/ 0,
236	/*cache_segments*/ 0,
237	/*cache_seg_size*/ {0, 0},
238	/*reserved*/ 0,
239	/*non_cache_seg_size*/ {0, 0, 0}
240};
241
242const static struct scsi_caching_page caching_page_changeable = {
243	/*page_code*/SMS_CACHING_PAGE,
244	/*page_length*/sizeof(struct scsi_caching_page) - 2,
245	/*flags1*/ SCP_WCE | SCP_RCD,
246	/*ret_priority*/ 0,
247	/*disable_pf_transfer_len*/ {0, 0},
248	/*min_prefetch*/ {0, 0},
249	/*max_prefetch*/ {0, 0},
250	/*max_pf_ceiling*/ {0, 0},
251	/*flags2*/ 0,
252	/*cache_segments*/ 0,
253	/*cache_seg_size*/ {0, 0},
254	/*reserved*/ 0,
255	/*non_cache_seg_size*/ {0, 0, 0}
256};
257
258const static struct scsi_control_page control_page_default = {
259	/*page_code*/SMS_CONTROL_MODE_PAGE,
260	/*page_length*/sizeof(struct scsi_control_page) - 2,
261	/*rlec*/0,
262	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
263	/*eca_and_aen*/0,
264	/*flags4*/SCP_TAS,
265	/*aen_holdoff_period*/{0, 0},
266	/*busy_timeout_period*/{0, 0},
267	/*extended_selftest_completion_time*/{0, 0}
268};
269
270const static struct scsi_control_page control_page_changeable = {
271	/*page_code*/SMS_CONTROL_MODE_PAGE,
272	/*page_length*/sizeof(struct scsi_control_page) - 2,
273	/*rlec*/SCP_DSENSE,
274	/*queue_flags*/SCP_QUEUE_ALG_MASK,
275	/*eca_and_aen*/SCP_SWP,
276	/*flags4*/0,
277	/*aen_holdoff_period*/{0, 0},
278	/*busy_timeout_period*/{0, 0},
279	/*extended_selftest_completion_time*/{0, 0}
280};
281
282const static struct scsi_info_exceptions_page ie_page_default = {
283	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
284	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
285	/*info_flags*/SIEP_FLAGS_DEXCPT,
286	/*mrie*/0,
287	/*interval_timer*/{0, 0, 0, 0},
288	/*report_count*/{0, 0, 0, 0}
289};
290
291const static struct scsi_info_exceptions_page ie_page_changeable = {
292	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
293	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
294	/*info_flags*/0,
295	/*mrie*/0,
296	/*interval_timer*/{0, 0, 0, 0},
297	/*report_count*/{0, 0, 0, 0}
298};
299
300#define CTL_LBPM_LEN	(sizeof(struct ctl_logical_block_provisioning_page) - 4)
301
302const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
303	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
304	/*subpage_code*/0x02,
305	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
306	/*flags*/0,
307	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
308	/*descr*/{}},
309	{{/*flags*/0,
310	  /*resource*/0x01,
311	  /*reserved*/{0, 0},
312	  /*count*/{0, 0, 0, 0}},
313	 {/*flags*/0,
314	  /*resource*/0x02,
315	  /*reserved*/{0, 0},
316	  /*count*/{0, 0, 0, 0}},
317	 {/*flags*/0,
318	  /*resource*/0xf1,
319	  /*reserved*/{0, 0},
320	  /*count*/{0, 0, 0, 0}},
321	 {/*flags*/0,
322	  /*resource*/0xf2,
323	  /*reserved*/{0, 0},
324	  /*count*/{0, 0, 0, 0}}
325	}
326};
327
328const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
329	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
330	/*subpage_code*/0x02,
331	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
332	/*flags*/0,
333	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
334	/*descr*/{}},
335	{{/*flags*/0,
336	  /*resource*/0,
337	  /*reserved*/{0, 0},
338	  /*count*/{0, 0, 0, 0}},
339	 {/*flags*/0,
340	  /*resource*/0,
341	  /*reserved*/{0, 0},
342	  /*count*/{0, 0, 0, 0}},
343	 {/*flags*/0,
344	  /*resource*/0,
345	  /*reserved*/{0, 0},
346	  /*count*/{0, 0, 0, 0}},
347	 {/*flags*/0,
348	  /*resource*/0,
349	  /*reserved*/{0, 0},
350	  /*count*/{0, 0, 0, 0}}
351	}
352};
353
354/*
355 * XXX KDM move these into the softc.
356 */
357static int rcv_sync_msg;
358static uint8_t ctl_pause_rtr;
359
360SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
361static int worker_threads = -1;
362TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
363SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
364    &worker_threads, 1, "Number of worker threads");
365static int ctl_debug = CTL_DEBUG_NONE;
366TUNABLE_INT("kern.cam.ctl.debug", &ctl_debug);
367SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
368    &ctl_debug, 0, "Enabled debug flags");
369
370/*
371 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
372 * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
373 * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
374 * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
375 */
376#define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
377
378#ifdef notyet
379static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
380				  int param);
381static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
382#endif
383static int ctl_init(void);
384void ctl_shutdown(void);
385static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
386static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
387static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
388static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
389			      struct ctl_ooa *ooa_hdr,
390			      struct ctl_ooa_entry *kern_entries);
391static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
392		     struct thread *td);
393static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
394			 struct ctl_be_lun *be_lun);
395static int ctl_free_lun(struct ctl_lun *lun);
396static void ctl_create_lun(struct ctl_be_lun *be_lun);
397static struct ctl_port * ctl_io_port(struct ctl_io_hdr *io_hdr);
398/**
399static void ctl_failover_change_pages(struct ctl_softc *softc,
400				      struct ctl_scsiio *ctsio, int master);
401**/
402
403static int ctl_do_mode_select(union ctl_io *io);
404static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
405			   uint64_t res_key, uint64_t sa_res_key,
406			   uint8_t type, uint32_t residx,
407			   struct ctl_scsiio *ctsio,
408			   struct scsi_per_res_out *cdb,
409			   struct scsi_per_res_out_parms* param);
410static void ctl_pro_preempt_other(struct ctl_lun *lun,
411				  union ctl_ha_msg *msg);
412static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
413static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
414static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
415static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
416static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
417static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
418static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
419					 int alloc_len);
420static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
421					 int alloc_len);
422static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
423static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
424static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
425static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
426static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
427static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
428    bool seq);
429static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
430static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
431    union ctl_io *pending_io, union ctl_io *ooa_io);
432static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
433				union ctl_io *starting_io);
434static int ctl_check_blocked(struct ctl_lun *lun);
435static int ctl_scsiio_lun_check(struct ctl_lun *lun,
436				const struct ctl_cmd_entry *entry,
437				struct ctl_scsiio *ctsio);
438//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
439#ifdef notyet
440static void ctl_failover(void);
441#endif
442static void ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
443			 ctl_ua_type ua_type);
444static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
445			       struct ctl_scsiio *ctsio);
446static int ctl_scsiio(struct ctl_scsiio *ctsio);
447
448static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
449static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
450			    ctl_ua_type ua_type);
451static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
452			 ctl_ua_type ua_type);
453static int ctl_abort_task(union ctl_io *io);
454static int ctl_abort_task_set(union ctl_io *io);
455static int ctl_i_t_nexus_reset(union ctl_io *io);
456static void ctl_run_task(union ctl_io *io);
457#ifdef CTL_IO_DELAY
458static void ctl_datamove_timer_wakeup(void *arg);
459static void ctl_done_timer_wakeup(void *arg);
460#endif /* CTL_IO_DELAY */
461
462static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
463static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
464static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
465static void ctl_datamove_remote_write(union ctl_io *io);
466static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
467static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
468static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
469static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
470				    ctl_ha_dt_cb callback);
471static void ctl_datamove_remote_read(union ctl_io *io);
472static void ctl_datamove_remote(union ctl_io *io);
473static int ctl_process_done(union ctl_io *io);
474static void ctl_lun_thread(void *arg);
475static void ctl_thresh_thread(void *arg);
476static void ctl_work_thread(void *arg);
477static void ctl_enqueue_incoming(union ctl_io *io);
478static void ctl_enqueue_rtr(union ctl_io *io);
479static void ctl_enqueue_done(union ctl_io *io);
480#ifdef notyet
481static void ctl_enqueue_isc(union ctl_io *io);
482#endif
483static const struct ctl_cmd_entry *
484    ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
485static const struct ctl_cmd_entry *
486    ctl_validate_command(struct ctl_scsiio *ctsio);
487static int ctl_cmd_applicable(uint8_t lun_type,
488    const struct ctl_cmd_entry *entry);
489
490/*
491 * Load the serialization table.  This isn't very pretty, but is probably
492 * the easiest way to do it.
493 */
494#include "ctl_ser_table.c"
495
496/*
497 * We only need to define open, close and ioctl routines for this driver.
498 */
499static struct cdevsw ctl_cdevsw = {
500	.d_version =	D_VERSION,
501	.d_flags =	0,
502	.d_open =	ctl_open,
503	.d_close =	ctl_close,
504	.d_ioctl =	ctl_ioctl,
505	.d_name =	"ctl",
506};
507
508
509MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
510
511static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
512
513static moduledata_t ctl_moduledata = {
514	"ctl",
515	ctl_module_event_handler,
516	NULL
517};
518
519DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
520MODULE_VERSION(ctl, 1);
521
522#ifdef notyet
523static void
524ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
525			    union ctl_ha_msg *msg_info)
526{
527	struct ctl_scsiio *ctsio;
528
529	if (msg_info->hdr.original_sc == NULL) {
530		printf("%s: original_sc == NULL!\n", __func__);
531		/* XXX KDM now what? */
532		return;
533	}
534
535	ctsio = &msg_info->hdr.original_sc->scsiio;
536	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
537	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
538	ctsio->io_hdr.status = msg_info->hdr.status;
539	ctsio->scsi_status = msg_info->scsi.scsi_status;
540	ctsio->sense_len = msg_info->scsi.sense_len;
541	ctsio->sense_residual = msg_info->scsi.sense_residual;
542	ctsio->residual = msg_info->scsi.residual;
543	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
544	       sizeof(ctsio->sense_data));
545	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
546	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
547	ctl_enqueue_isc((union ctl_io *)ctsio);
548}
549
550static void
551ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
552				union ctl_ha_msg *msg_info)
553{
554	struct ctl_scsiio *ctsio;
555
556	if (msg_info->hdr.serializing_sc == NULL) {
557		printf("%s: serializing_sc == NULL!\n", __func__);
558		/* XXX KDM now what? */
559		return;
560	}
561
562	ctsio = &msg_info->hdr.serializing_sc->scsiio;
563#if 0
564	/*
565	 * Attempt to catch the situation where an I/O has
566	 * been freed, and we're using it again.
567	 */
568	if (ctsio->io_hdr.io_type == 0xff) {
569		union ctl_io *tmp_io;
570		tmp_io = (union ctl_io *)ctsio;
571		printf("%s: %p use after free!\n", __func__,
572		       ctsio);
573		printf("%s: type %d msg %d cdb %x iptl: "
574		       "%d:%d:%d:%d tag 0x%04x "
575		       "flag %#x status %x\n",
576			__func__,
577			tmp_io->io_hdr.io_type,
578			tmp_io->io_hdr.msg_type,
579			tmp_io->scsiio.cdb[0],
580			tmp_io->io_hdr.nexus.initid.id,
581			tmp_io->io_hdr.nexus.targ_port,
582			tmp_io->io_hdr.nexus.targ_target.id,
583			tmp_io->io_hdr.nexus.targ_lun,
584			(tmp_io->io_hdr.io_type ==
585			CTL_IO_TASK) ?
586			tmp_io->taskio.tag_num :
587			tmp_io->scsiio.tag_num,
588		        tmp_io->io_hdr.flags,
589			tmp_io->io_hdr.status);
590	}
591#endif
592	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
593	ctl_enqueue_isc((union ctl_io *)ctsio);
594}
595
596/*
597 * ISC (Inter Shelf Communication) event handler.  Events from the HA
598 * subsystem come in here.
599 */
600static void
601ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
602{
603	struct ctl_softc *softc;
604	union ctl_io *io;
605	struct ctl_prio *presio;
606	ctl_ha_status isc_status;
607
608	softc = control_softc;
609	io = NULL;
610
611
612#if 0
613	printf("CTL: Isc Msg event %d\n", event);
614#endif
615	if (event == CTL_HA_EVT_MSG_RECV) {
616		union ctl_ha_msg msg_info;
617
618		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
619					     sizeof(msg_info), /*wait*/ 0);
620#if 0
621		printf("CTL: msg_type %d\n", msg_info.msg_type);
622#endif
623		if (isc_status != 0) {
624			printf("Error receiving message, status = %d\n",
625			       isc_status);
626			return;
627		}
628
629		switch (msg_info.hdr.msg_type) {
630		case CTL_MSG_SERIALIZE:
631#if 0
632			printf("Serialize\n");
633#endif
634			io = ctl_alloc_io_nowait(softc->othersc_pool);
635			if (io == NULL) {
636				printf("ctl_isc_event_handler: can't allocate "
637				       "ctl_io!\n");
638				/* Bad Juju */
639				/* Need to set busy and send msg back */
640				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
641				msg_info.hdr.status = CTL_SCSI_ERROR;
642				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
643				msg_info.scsi.sense_len = 0;
644			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
645				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
646				}
647				goto bailout;
648			}
649			ctl_zero_io(io);
650			// populate ctsio from msg_info
651			io->io_hdr.io_type = CTL_IO_SCSI;
652			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
653			io->io_hdr.original_sc = msg_info.hdr.original_sc;
654#if 0
655			printf("pOrig %x\n", (int)msg_info.original_sc);
656#endif
657			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
658					    CTL_FLAG_IO_ACTIVE;
659			/*
660			 * If we're in serialization-only mode, we don't
661			 * want to go through full done processing.  Thus
662			 * the COPY flag.
663			 *
664			 * XXX KDM add another flag that is more specific.
665			 */
666			if (softc->ha_mode == CTL_HA_MODE_SER_ONLY)
667				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
668			io->io_hdr.nexus = msg_info.hdr.nexus;
669#if 0
670			printf("targ %d, port %d, iid %d, lun %d\n",
671			       io->io_hdr.nexus.targ_target.id,
672			       io->io_hdr.nexus.targ_port,
673			       io->io_hdr.nexus.initid.id,
674			       io->io_hdr.nexus.targ_lun);
675#endif
676			io->scsiio.tag_num = msg_info.scsi.tag_num;
677			io->scsiio.tag_type = msg_info.scsi.tag_type;
678			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
679			       CTL_MAX_CDBLEN);
680			if (softc->ha_mode == CTL_HA_MODE_XFER) {
681				const struct ctl_cmd_entry *entry;
682
683				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
684				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
685				io->io_hdr.flags |=
686					entry->flags & CTL_FLAG_DATA_MASK;
687			}
688			ctl_enqueue_isc(io);
689			break;
690
691		/* Performed on the Originating SC, XFER mode only */
692		case CTL_MSG_DATAMOVE: {
693			struct ctl_sg_entry *sgl;
694			int i, j;
695
696			io = msg_info.hdr.original_sc;
697			if (io == NULL) {
698				printf("%s: original_sc == NULL!\n", __func__);
699				/* XXX KDM do something here */
700				break;
701			}
702			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
703			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
704			/*
705			 * Keep track of this, we need to send it back over
706			 * when the datamove is complete.
707			 */
708			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
709
710			if (msg_info.dt.sg_sequence == 0) {
711				/*
712				 * XXX KDM we use the preallocated S/G list
713				 * here, but we'll need to change this to
714				 * dynamic allocation if we need larger S/G
715				 * lists.
716				 */
717				if (msg_info.dt.kern_sg_entries >
718				    sizeof(io->io_hdr.remote_sglist) /
719				    sizeof(io->io_hdr.remote_sglist[0])) {
720					printf("%s: number of S/G entries "
721					    "needed %u > allocated num %zd\n",
722					    __func__,
723					    msg_info.dt.kern_sg_entries,
724					    sizeof(io->io_hdr.remote_sglist)/
725					    sizeof(io->io_hdr.remote_sglist[0]));
726
727					/*
728					 * XXX KDM send a message back to
729					 * the other side to shut down the
730					 * DMA.  The error will come back
731					 * through via the normal channel.
732					 */
733					break;
734				}
735				sgl = io->io_hdr.remote_sglist;
736				memset(sgl, 0,
737				       sizeof(io->io_hdr.remote_sglist));
738
739				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
740
741				io->scsiio.kern_sg_entries =
742					msg_info.dt.kern_sg_entries;
743				io->scsiio.rem_sg_entries =
744					msg_info.dt.kern_sg_entries;
745				io->scsiio.kern_data_len =
746					msg_info.dt.kern_data_len;
747				io->scsiio.kern_total_len =
748					msg_info.dt.kern_total_len;
749				io->scsiio.kern_data_resid =
750					msg_info.dt.kern_data_resid;
751				io->scsiio.kern_rel_offset =
752					msg_info.dt.kern_rel_offset;
753				/*
754				 * Clear out per-DMA flags.
755				 */
756				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
757				/*
758				 * Add per-DMA flags that are set for this
759				 * particular DMA request.
760				 */
761				io->io_hdr.flags |= msg_info.dt.flags &
762						    CTL_FLAG_RDMA_MASK;
763			} else
764				sgl = (struct ctl_sg_entry *)
765					io->scsiio.kern_data_ptr;
766
767			for (i = msg_info.dt.sent_sg_entries, j = 0;
768			     i < (msg_info.dt.sent_sg_entries +
769			     msg_info.dt.cur_sg_entries); i++, j++) {
770				sgl[i].addr = msg_info.dt.sg_list[j].addr;
771				sgl[i].len = msg_info.dt.sg_list[j].len;
772
773#if 0
774				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
775				       __func__,
776				       msg_info.dt.sg_list[j].addr,
777				       msg_info.dt.sg_list[j].len,
778				       sgl[i].addr, sgl[i].len, j, i);
779#endif
780			}
781#if 0
782			memcpy(&sgl[msg_info.dt.sent_sg_entries],
783			       msg_info.dt.sg_list,
784			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
785#endif
786
787			/*
788			 * If this is the last piece of the I/O, we've got
789			 * the full S/G list.  Queue processing in the thread.
790			 * Otherwise wait for the next piece.
791			 */
792			if (msg_info.dt.sg_last != 0)
793				ctl_enqueue_isc(io);
794			break;
795		}
796		/* Performed on the Serializing (primary) SC, XFER mode only */
797		case CTL_MSG_DATAMOVE_DONE: {
798			if (msg_info.hdr.serializing_sc == NULL) {
799				printf("%s: serializing_sc == NULL!\n",
800				       __func__);
801				/* XXX KDM now what? */
802				break;
803			}
804			/*
805			 * We grab the sense information here in case
806			 * there was a failure, so we can return status
807			 * back to the initiator.
808			 */
809			io = msg_info.hdr.serializing_sc;
810			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
811			io->io_hdr.status = msg_info.hdr.status;
812			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
813			io->scsiio.sense_len = msg_info.scsi.sense_len;
814			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
815			io->io_hdr.port_status = msg_info.scsi.fetd_status;
816			io->scsiio.residual = msg_info.scsi.residual;
817			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
818			       sizeof(io->scsiio.sense_data));
819			ctl_enqueue_isc(io);
820			break;
821		}
822
823		/* Preformed on Originating SC, SER_ONLY mode */
824		case CTL_MSG_R2R:
825			io = msg_info.hdr.original_sc;
826			if (io == NULL) {
827				printf("%s: Major Bummer\n", __func__);
828				return;
829			} else {
830#if 0
831				printf("pOrig %x\n",(int) ctsio);
832#endif
833			}
834			io->io_hdr.msg_type = CTL_MSG_R2R;
835			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
836			ctl_enqueue_isc(io);
837			break;
838
839		/*
840		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
841		 * mode.
842		 * Performed on the Originating (i.e. secondary) SC in XFER
843		 * mode
844		 */
845		case CTL_MSG_FINISH_IO:
846			if (softc->ha_mode == CTL_HA_MODE_XFER)
847				ctl_isc_handler_finish_xfer(softc,
848							    &msg_info);
849			else
850				ctl_isc_handler_finish_ser_only(softc,
851								&msg_info);
852			break;
853
854		/* Preformed on Originating SC */
855		case CTL_MSG_BAD_JUJU:
856			io = msg_info.hdr.original_sc;
857			if (io == NULL) {
858				printf("%s: Bad JUJU!, original_sc is NULL!\n",
859				       __func__);
860				break;
861			}
862			ctl_copy_sense_data(&msg_info, io);
863			/*
864			 * IO should have already been cleaned up on other
865			 * SC so clear this flag so we won't send a message
866			 * back to finish the IO there.
867			 */
868			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
869			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
870
871			/* io = msg_info.hdr.serializing_sc; */
872			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
873			ctl_enqueue_isc(io);
874			break;
875
876		/* Handle resets sent from the other side */
877		case CTL_MSG_MANAGE_TASKS: {
878			struct ctl_taskio *taskio;
879			taskio = (struct ctl_taskio *)ctl_alloc_io_nowait(
880			    softc->othersc_pool);
881			if (taskio == NULL) {
882				printf("ctl_isc_event_handler: can't allocate "
883				       "ctl_io!\n");
884				/* Bad Juju */
885				/* should I just call the proper reset func
886				   here??? */
887				goto bailout;
888			}
889			ctl_zero_io((union ctl_io *)taskio);
890			taskio->io_hdr.io_type = CTL_IO_TASK;
891			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
892			taskio->io_hdr.nexus = msg_info.hdr.nexus;
893			taskio->task_action = msg_info.task.task_action;
894			taskio->tag_num = msg_info.task.tag_num;
895			taskio->tag_type = msg_info.task.tag_type;
896#ifdef CTL_TIME_IO
897			taskio->io_hdr.start_time = time_uptime;
898			getbintime(&taskio->io_hdr.start_bt);
899#if 0
900			cs_prof_gettime(&taskio->io_hdr.start_ticks);
901#endif
902#endif /* CTL_TIME_IO */
903			ctl_run_task((union ctl_io *)taskio);
904			break;
905		}
906		/* Persistent Reserve action which needs attention */
907		case CTL_MSG_PERS_ACTION:
908			presio = (struct ctl_prio *)ctl_alloc_io_nowait(
909			    softc->othersc_pool);
910			if (presio == NULL) {
911				printf("ctl_isc_event_handler: can't allocate "
912				       "ctl_io!\n");
913				/* Bad Juju */
914				/* Need to set busy and send msg back */
915				goto bailout;
916			}
917			ctl_zero_io((union ctl_io *)presio);
918			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
919			presio->pr_msg = msg_info.pr;
920			ctl_enqueue_isc((union ctl_io *)presio);
921			break;
922		case CTL_MSG_SYNC_FE:
923			rcv_sync_msg = 1;
924			break;
925		default:
926		        printf("How did I get here?\n");
927		}
928	} else if (event == CTL_HA_EVT_MSG_SENT) {
929		if (param != CTL_HA_STATUS_SUCCESS) {
930			printf("Bad status from ctl_ha_msg_send status %d\n",
931			       param);
932		}
933		return;
934	} else if (event == CTL_HA_EVT_DISCONNECT) {
935		printf("CTL: Got a disconnect from Isc\n");
936		return;
937	} else {
938		printf("ctl_isc_event_handler: Unknown event %d\n", event);
939		return;
940	}
941
942bailout:
943	return;
944}
945
946static void
947ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
948{
949	struct scsi_sense_data *sense;
950
951	sense = &dest->scsiio.sense_data;
952	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
953	dest->scsiio.scsi_status = src->scsi.scsi_status;
954	dest->scsiio.sense_len = src->scsi.sense_len;
955	dest->io_hdr.status = src->hdr.status;
956}
957#endif
958
959static void
960ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
961{
962	ctl_ua_type *pu;
963
964	mtx_assert(&lun->lun_lock, MA_OWNED);
965	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
966	if (pu == NULL)
967		return;
968	pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
969}
970
971static void
972ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
973{
974	int i, j;
975
976	mtx_assert(&lun->lun_lock, MA_OWNED);
977	for (i = 0; i < CTL_MAX_PORTS; i++) {
978		if (lun->pending_ua[i] == NULL)
979			continue;
980		for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
981			if (i * CTL_MAX_INIT_PER_PORT + j == except)
982				continue;
983			lun->pending_ua[i][j] |= ua;
984		}
985	}
986}
987
988static void
989ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
990{
991	ctl_ua_type *pu;
992
993	mtx_assert(&lun->lun_lock, MA_OWNED);
994	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
995	if (pu == NULL)
996		return;
997	pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
998}
999
1000static void
1001ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1002{
1003	int i, j;
1004
1005	mtx_assert(&lun->lun_lock, MA_OWNED);
1006	for (i = 0; i < CTL_MAX_PORTS; i++) {
1007		if (lun->pending_ua[i] == NULL)
1008			continue;
1009		for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1010			if (i * CTL_MAX_INIT_PER_PORT + j == except)
1011				continue;
1012			lun->pending_ua[i][j] &= ~ua;
1013		}
1014	}
1015}
1016
1017static void
1018ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1019    ctl_ua_type ua_type)
1020{
1021	struct ctl_lun *lun;
1022
1023	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1024	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1025		mtx_lock(&lun->lun_lock);
1026		ctl_clr_ua(lun, initidx, ua_type);
1027		mtx_unlock(&lun->lun_lock);
1028	}
1029}
1030
1031static int
1032ctl_ha_state_sysctl(SYSCTL_HANDLER_ARGS)
1033{
1034	struct ctl_softc *softc = (struct ctl_softc *)arg1;
1035	struct ctl_lun *lun;
1036	int error, value;
1037
1038	if (softc->flags & CTL_FLAG_ACTIVE_SHELF)
1039		value = 0;
1040	else
1041		value = 1;
1042
1043	error = sysctl_handle_int(oidp, &value, 0, req);
1044	if ((error != 0) || (req->newptr == NULL))
1045		return (error);
1046
1047	mtx_lock(&softc->ctl_lock);
1048	if (value == 0)
1049		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1050	else
1051		softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1052	STAILQ_FOREACH(lun, &softc->lun_list, links) {
1053		mtx_lock(&lun->lun_lock);
1054		ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1055		mtx_unlock(&lun->lun_lock);
1056	}
1057	mtx_unlock(&softc->ctl_lock);
1058	return (0);
1059}
1060
1061static int
1062ctl_init(void)
1063{
1064	struct ctl_softc *softc;
1065	void *other_pool;
1066	int i, error, retval;
1067	//int isc_retval;
1068
1069	retval = 0;
1070	ctl_pause_rtr = 0;
1071        rcv_sync_msg = 0;
1072
1073	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1074			       M_WAITOK | M_ZERO);
1075	softc = control_softc;
1076
1077	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
1078			      "cam/ctl");
1079
1080	softc->dev->si_drv1 = softc;
1081
1082	/*
1083	 * By default, return a "bad LUN" peripheral qualifier for unknown
1084	 * LUNs.  The user can override this default using the tunable or
1085	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
1086	 */
1087	softc->inquiry_pq_no_lun = 1;
1088	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
1089			  &softc->inquiry_pq_no_lun);
1090	sysctl_ctx_init(&softc->sysctl_ctx);
1091	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1092		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1093		CTLFLAG_RD, 0, "CAM Target Layer");
1094
1095	if (softc->sysctl_tree == NULL) {
1096		printf("%s: unable to allocate sysctl tree\n", __func__);
1097		destroy_dev(softc->dev);
1098		free(control_softc, M_DEVBUF);
1099		control_softc = NULL;
1100		return (ENOMEM);
1101	}
1102
1103	SYSCTL_ADD_INT(&softc->sysctl_ctx,
1104		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
1105		       "inquiry_pq_no_lun", CTLFLAG_RW,
1106		       &softc->inquiry_pq_no_lun, 0,
1107		       "Report no lun possible for invalid LUNs");
1108
1109	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1110	softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1111	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1112	softc->open_count = 0;
1113
1114	/*
1115	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1116	 * the drive.
1117	 */
1118	softc->flags = CTL_FLAG_REAL_SYNC;
1119
1120	/*
1121	 * In Copan's HA scheme, the "master" and "slave" roles are
1122	 * figured out through the slot the controller is in.  Although it
1123	 * is an active/active system, someone has to be in charge.
1124	 */
1125	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1126	    OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1127	    "HA head ID (0 - no HA)");
1128	if (softc->ha_id == 0) {
1129		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1130		softc->is_single = 1;
1131		softc->port_offset = 0;
1132	} else
1133		softc->port_offset = (softc->ha_id - 1) * CTL_MAX_PORTS;
1134	softc->persis_offset = softc->port_offset * CTL_MAX_INIT_PER_PORT;
1135
1136	STAILQ_INIT(&softc->lun_list);
1137	STAILQ_INIT(&softc->pending_lun_queue);
1138	STAILQ_INIT(&softc->fe_list);
1139	STAILQ_INIT(&softc->port_list);
1140	STAILQ_INIT(&softc->be_list);
1141	ctl_tpc_init(softc);
1142
1143	if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1144	                    &other_pool) != 0)
1145	{
1146		printf("ctl: can't allocate %d entry other SC pool, "
1147		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1148		return (ENOMEM);
1149	}
1150	softc->othersc_pool = other_pool;
1151
1152	if (worker_threads <= 0)
1153		worker_threads = max(1, mp_ncpus / 4);
1154	if (worker_threads > CTL_MAX_THREADS)
1155		worker_threads = CTL_MAX_THREADS;
1156
1157	for (i = 0; i < worker_threads; i++) {
1158		struct ctl_thread *thr = &softc->threads[i];
1159
1160		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1161		thr->ctl_softc = softc;
1162		STAILQ_INIT(&thr->incoming_queue);
1163		STAILQ_INIT(&thr->rtr_queue);
1164		STAILQ_INIT(&thr->done_queue);
1165		STAILQ_INIT(&thr->isc_queue);
1166
1167		error = kproc_kthread_add(ctl_work_thread, thr,
1168		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1169		if (error != 0) {
1170			printf("error creating CTL work thread!\n");
1171			ctl_pool_free(other_pool);
1172			return (error);
1173		}
1174	}
1175	error = kproc_kthread_add(ctl_lun_thread, softc,
1176	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1177	if (error != 0) {
1178		printf("error creating CTL lun thread!\n");
1179		ctl_pool_free(other_pool);
1180		return (error);
1181	}
1182	error = kproc_kthread_add(ctl_thresh_thread, softc,
1183	    &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1184	if (error != 0) {
1185		printf("error creating CTL threshold thread!\n");
1186		ctl_pool_free(other_pool);
1187		return (error);
1188	}
1189
1190	SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1191	    OID_AUTO, "ha_state", CTLTYPE_INT | CTLFLAG_RWTUN,
1192	    softc, 0, ctl_ha_state_sysctl, "I", "HA state for this head");
1193
1194#ifdef CTL_IO_DELAY
1195	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1196		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1197		       sizeof(struct callout), CTL_TIMER_BYTES);
1198		return (EINVAL);
1199	}
1200#endif /* CTL_IO_DELAY */
1201
1202	return (0);
1203}
1204
1205void
1206ctl_shutdown(void)
1207{
1208	struct ctl_softc *softc;
1209	struct ctl_lun *lun, *next_lun;
1210
1211	softc = (struct ctl_softc *)control_softc;
1212
1213	mtx_lock(&softc->ctl_lock);
1214
1215	/*
1216	 * Free up each LUN.
1217	 */
1218	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1219		next_lun = STAILQ_NEXT(lun, links);
1220		ctl_free_lun(lun);
1221	}
1222
1223	mtx_unlock(&softc->ctl_lock);
1224
1225#if 0
1226	ctl_shutdown_thread(softc->work_thread);
1227	mtx_destroy(&softc->queue_lock);
1228#endif
1229
1230	ctl_tpc_shutdown(softc);
1231	uma_zdestroy(softc->io_zone);
1232	mtx_destroy(&softc->ctl_lock);
1233
1234	destroy_dev(softc->dev);
1235
1236	sysctl_ctx_free(&softc->sysctl_ctx);
1237
1238	free(control_softc, M_DEVBUF);
1239	control_softc = NULL;
1240}
1241
1242static int
1243ctl_module_event_handler(module_t mod, int what, void *arg)
1244{
1245
1246	switch (what) {
1247	case MOD_LOAD:
1248		return (ctl_init());
1249	case MOD_UNLOAD:
1250		return (EBUSY);
1251	default:
1252		return (EOPNOTSUPP);
1253	}
1254}
1255
1256/*
1257 * XXX KDM should we do some access checks here?  Bump a reference count to
1258 * prevent a CTL module from being unloaded while someone has it open?
1259 */
1260static int
1261ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1262{
1263	return (0);
1264}
1265
1266static int
1267ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1268{
1269	return (0);
1270}
1271
1272int
1273ctl_port_enable(ctl_port_type port_type)
1274{
1275	struct ctl_softc *softc = control_softc;
1276	struct ctl_port *port;
1277
1278	if (softc->is_single == 0) {
1279		union ctl_ha_msg msg_info;
1280		int isc_retval;
1281
1282#if 0
1283		printf("%s: HA mode, synchronizing frontend enable\n",
1284		        __func__);
1285#endif
1286		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1287	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1288		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1289			printf("Sync msg send error retval %d\n", isc_retval);
1290		}
1291		if (!rcv_sync_msg) {
1292			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1293			        sizeof(msg_info), 1);
1294		}
1295#if 0
1296        	printf("CTL:Frontend Enable\n");
1297	} else {
1298		printf("%s: single mode, skipping frontend synchronization\n",
1299		        __func__);
1300#endif
1301	}
1302
1303	STAILQ_FOREACH(port, &softc->port_list, links) {
1304		if (port_type & port->port_type)
1305		{
1306#if 0
1307			printf("port %d\n", port->targ_port);
1308#endif
1309			ctl_port_online(port);
1310		}
1311	}
1312
1313	return (0);
1314}
1315
1316int
1317ctl_port_disable(ctl_port_type port_type)
1318{
1319	struct ctl_softc *softc;
1320	struct ctl_port *port;
1321
1322	softc = control_softc;
1323
1324	STAILQ_FOREACH(port, &softc->port_list, links) {
1325		if (port_type & port->port_type)
1326			ctl_port_offline(port);
1327	}
1328
1329	return (0);
1330}
1331
1332/*
1333 * Returns 0 for success, 1 for failure.
1334 * Currently the only failure mode is if there aren't enough entries
1335 * allocated.  So, in case of a failure, look at num_entries_dropped,
1336 * reallocate and try again.
1337 */
1338int
1339ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1340	      int *num_entries_filled, int *num_entries_dropped,
1341	      ctl_port_type port_type, int no_virtual)
1342{
1343	struct ctl_softc *softc;
1344	struct ctl_port *port;
1345	int entries_dropped, entries_filled;
1346	int retval;
1347	int i;
1348
1349	softc = control_softc;
1350
1351	retval = 0;
1352	entries_filled = 0;
1353	entries_dropped = 0;
1354
1355	i = 0;
1356	mtx_lock(&softc->ctl_lock);
1357	STAILQ_FOREACH(port, &softc->port_list, links) {
1358		struct ctl_port_entry *entry;
1359
1360		if ((port->port_type & port_type) == 0)
1361			continue;
1362
1363		if ((no_virtual != 0)
1364		 && (port->virtual_port != 0))
1365			continue;
1366
1367		if (entries_filled >= num_entries_alloced) {
1368			entries_dropped++;
1369			continue;
1370		}
1371		entry = &entries[i];
1372
1373		entry->port_type = port->port_type;
1374		strlcpy(entry->port_name, port->port_name,
1375			sizeof(entry->port_name));
1376		entry->physical_port = port->physical_port;
1377		entry->virtual_port = port->virtual_port;
1378		entry->wwnn = port->wwnn;
1379		entry->wwpn = port->wwpn;
1380
1381		i++;
1382		entries_filled++;
1383	}
1384
1385	mtx_unlock(&softc->ctl_lock);
1386
1387	if (entries_dropped > 0)
1388		retval = 1;
1389
1390	*num_entries_dropped = entries_dropped;
1391	*num_entries_filled = entries_filled;
1392
1393	return (retval);
1394}
1395
1396/*
1397 * Remove an initiator by port number and initiator ID.
1398 * Returns 0 for success, -1 for failure.
1399 */
1400int
1401ctl_remove_initiator(struct ctl_port *port, int iid)
1402{
1403	struct ctl_softc *softc = control_softc;
1404
1405	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1406
1407	if (iid > CTL_MAX_INIT_PER_PORT) {
1408		printf("%s: initiator ID %u > maximun %u!\n",
1409		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1410		return (-1);
1411	}
1412
1413	mtx_lock(&softc->ctl_lock);
1414	port->wwpn_iid[iid].in_use--;
1415	port->wwpn_iid[iid].last_use = time_uptime;
1416	mtx_unlock(&softc->ctl_lock);
1417
1418	return (0);
1419}
1420
1421/*
1422 * Add an initiator to the initiator map.
1423 * Returns iid for success, < 0 for failure.
1424 */
1425int
1426ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1427{
1428	struct ctl_softc *softc = control_softc;
1429	time_t best_time;
1430	int i, best;
1431
1432	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1433
1434	if (iid >= CTL_MAX_INIT_PER_PORT) {
1435		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1436		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1437		free(name, M_CTL);
1438		return (-1);
1439	}
1440
1441	mtx_lock(&softc->ctl_lock);
1442
1443	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1444		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1445			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1446				iid = i;
1447				break;
1448			}
1449			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1450			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1451				iid = i;
1452				break;
1453			}
1454		}
1455	}
1456
1457	if (iid < 0) {
1458		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1459			if (port->wwpn_iid[i].in_use == 0 &&
1460			    port->wwpn_iid[i].wwpn == 0 &&
1461			    port->wwpn_iid[i].name == NULL) {
1462				iid = i;
1463				break;
1464			}
1465		}
1466	}
1467
1468	if (iid < 0) {
1469		best = -1;
1470		best_time = INT32_MAX;
1471		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1472			if (port->wwpn_iid[i].in_use == 0) {
1473				if (port->wwpn_iid[i].last_use < best_time) {
1474					best = i;
1475					best_time = port->wwpn_iid[i].last_use;
1476				}
1477			}
1478		}
1479		iid = best;
1480	}
1481
1482	if (iid < 0) {
1483		mtx_unlock(&softc->ctl_lock);
1484		free(name, M_CTL);
1485		return (-2);
1486	}
1487
1488	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1489		/*
1490		 * This is not an error yet.
1491		 */
1492		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1493#if 0
1494			printf("%s: port %d iid %u WWPN %#jx arrived"
1495			    " again\n", __func__, port->targ_port,
1496			    iid, (uintmax_t)wwpn);
1497#endif
1498			goto take;
1499		}
1500		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1501		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1502#if 0
1503			printf("%s: port %d iid %u name '%s' arrived"
1504			    " again\n", __func__, port->targ_port,
1505			    iid, name);
1506#endif
1507			goto take;
1508		}
1509
1510		/*
1511		 * This is an error, but what do we do about it?  The
1512		 * driver is telling us we have a new WWPN for this
1513		 * initiator ID, so we pretty much need to use it.
1514		 */
1515		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1516		    " but WWPN %#jx '%s' is still at that address\n",
1517		    __func__, port->targ_port, iid, wwpn, name,
1518		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1519		    port->wwpn_iid[iid].name);
1520
1521		/*
1522		 * XXX KDM clear have_ca and ua_pending on each LUN for
1523		 * this initiator.
1524		 */
1525	}
1526take:
1527	free(port->wwpn_iid[iid].name, M_CTL);
1528	port->wwpn_iid[iid].name = name;
1529	port->wwpn_iid[iid].wwpn = wwpn;
1530	port->wwpn_iid[iid].in_use++;
1531	mtx_unlock(&softc->ctl_lock);
1532
1533	return (iid);
1534}
1535
1536static int
1537ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1538{
1539	int len;
1540
1541	switch (port->port_type) {
1542	case CTL_PORT_FC:
1543	{
1544		struct scsi_transportid_fcp *id =
1545		    (struct scsi_transportid_fcp *)buf;
1546		if (port->wwpn_iid[iid].wwpn == 0)
1547			return (0);
1548		memset(id, 0, sizeof(*id));
1549		id->format_protocol = SCSI_PROTO_FC;
1550		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1551		return (sizeof(*id));
1552	}
1553	case CTL_PORT_ISCSI:
1554	{
1555		struct scsi_transportid_iscsi_port *id =
1556		    (struct scsi_transportid_iscsi_port *)buf;
1557		if (port->wwpn_iid[iid].name == NULL)
1558			return (0);
1559		memset(id, 0, 256);
1560		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1561		    SCSI_PROTO_ISCSI;
1562		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1563		len = roundup2(min(len, 252), 4);
1564		scsi_ulto2b(len, id->additional_length);
1565		return (sizeof(*id) + len);
1566	}
1567	case CTL_PORT_SAS:
1568	{
1569		struct scsi_transportid_sas *id =
1570		    (struct scsi_transportid_sas *)buf;
1571		if (port->wwpn_iid[iid].wwpn == 0)
1572			return (0);
1573		memset(id, 0, sizeof(*id));
1574		id->format_protocol = SCSI_PROTO_SAS;
1575		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1576		return (sizeof(*id));
1577	}
1578	default:
1579	{
1580		struct scsi_transportid_spi *id =
1581		    (struct scsi_transportid_spi *)buf;
1582		memset(id, 0, sizeof(*id));
1583		id->format_protocol = SCSI_PROTO_SPI;
1584		scsi_ulto2b(iid, id->scsi_addr);
1585		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1586		return (sizeof(*id));
1587	}
1588	}
1589}
1590
1591/*
1592 * Serialize a command that went down the "wrong" side, and so was sent to
1593 * this controller for execution.  The logic is a little different than the
1594 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1595 * sent back to the other side, but in the success case, we execute the
1596 * command on this side (XFER mode) or tell the other side to execute it
1597 * (SER_ONLY mode).
1598 */
1599static int
1600ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1601{
1602	struct ctl_softc *softc;
1603	union ctl_ha_msg msg_info;
1604	struct ctl_lun *lun;
1605	int retval = 0;
1606	uint32_t targ_lun;
1607
1608	softc = control_softc;
1609
1610	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1611	lun = softc->ctl_luns[targ_lun];
1612	if (lun==NULL)
1613	{
1614		/*
1615		 * Why isn't LUN defined? The other side wouldn't
1616		 * send a cmd if the LUN is undefined.
1617		 */
1618		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1619
1620		/* "Logical unit not supported" */
1621		ctl_set_sense_data(&msg_info.scsi.sense_data,
1622				   lun,
1623				   /*sense_format*/SSD_TYPE_NONE,
1624				   /*current_error*/ 1,
1625				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1626				   /*asc*/ 0x25,
1627				   /*ascq*/ 0x00,
1628				   SSD_ELEM_NONE);
1629
1630		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1631		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1632		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1633		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1634		msg_info.hdr.serializing_sc = NULL;
1635		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1636	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1637				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1638		}
1639		return(1);
1640
1641	}
1642
1643	mtx_lock(&lun->lun_lock);
1644    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1645
1646	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1647		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1648		 ooa_links))) {
1649	case CTL_ACTION_BLOCK:
1650		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1651		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1652				  blocked_links);
1653		break;
1654	case CTL_ACTION_PASS:
1655	case CTL_ACTION_SKIP:
1656		if (softc->ha_mode == CTL_HA_MODE_XFER) {
1657			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1658			ctl_enqueue_rtr((union ctl_io *)ctsio);
1659		} else {
1660
1661			/* send msg back to other side */
1662			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1663			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1664			msg_info.hdr.msg_type = CTL_MSG_R2R;
1665#if 0
1666			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1667#endif
1668		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1669			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1670			}
1671		}
1672		break;
1673	case CTL_ACTION_OVERLAP:
1674		/* OVERLAPPED COMMANDS ATTEMPTED */
1675		ctl_set_sense_data(&msg_info.scsi.sense_data,
1676				   lun,
1677				   /*sense_format*/SSD_TYPE_NONE,
1678				   /*current_error*/ 1,
1679				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1680				   /*asc*/ 0x4E,
1681				   /*ascq*/ 0x00,
1682				   SSD_ELEM_NONE);
1683
1684		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1685		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1686		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1687		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1688		msg_info.hdr.serializing_sc = NULL;
1689		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1690#if 0
1691		printf("BAD JUJU:Major Bummer Overlap\n");
1692#endif
1693		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1694		retval = 1;
1695		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1696		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1697		}
1698		break;
1699	case CTL_ACTION_OVERLAP_TAG:
1700		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1701		ctl_set_sense_data(&msg_info.scsi.sense_data,
1702				   lun,
1703				   /*sense_format*/SSD_TYPE_NONE,
1704				   /*current_error*/ 1,
1705				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1706				   /*asc*/ 0x4D,
1707				   /*ascq*/ ctsio->tag_num & 0xff,
1708				   SSD_ELEM_NONE);
1709
1710		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1711		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1712		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1713		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1714		msg_info.hdr.serializing_sc = NULL;
1715		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1716#if 0
1717		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1718#endif
1719		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1720		retval = 1;
1721		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1722		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1723		}
1724		break;
1725	case CTL_ACTION_ERROR:
1726	default:
1727		/* "Internal target failure" */
1728		ctl_set_sense_data(&msg_info.scsi.sense_data,
1729				   lun,
1730				   /*sense_format*/SSD_TYPE_NONE,
1731				   /*current_error*/ 1,
1732				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1733				   /*asc*/ 0x44,
1734				   /*ascq*/ 0x00,
1735				   SSD_ELEM_NONE);
1736
1737		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1738		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1739		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1740		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1741		msg_info.hdr.serializing_sc = NULL;
1742		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1743#if 0
1744		printf("BAD JUJU:Major Bummer HW Error\n");
1745#endif
1746		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1747		retval = 1;
1748		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1749		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1750		}
1751		break;
1752	}
1753	mtx_unlock(&lun->lun_lock);
1754	return (retval);
1755}
1756
1757/*
1758 * Returns 0 for success, errno for failure.
1759 */
1760static int
1761ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1762		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1763{
1764	union ctl_io *io;
1765	int retval;
1766
1767	retval = 0;
1768
1769	mtx_lock(&lun->lun_lock);
1770	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
1771	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
1772	     ooa_links)) {
1773		struct ctl_ooa_entry *entry;
1774
1775		/*
1776		 * If we've got more than we can fit, just count the
1777		 * remaining entries.
1778		 */
1779		if (*cur_fill_num >= ooa_hdr->alloc_num)
1780			continue;
1781
1782		entry = &kern_entries[*cur_fill_num];
1783
1784		entry->tag_num = io->scsiio.tag_num;
1785		entry->lun_num = lun->lun;
1786#ifdef CTL_TIME_IO
1787		entry->start_bt = io->io_hdr.start_bt;
1788#endif
1789		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
1790		entry->cdb_len = io->scsiio.cdb_len;
1791		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
1792			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
1793
1794		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
1795			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
1796
1797		if (io->io_hdr.flags & CTL_FLAG_ABORT)
1798			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
1799
1800		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
1801			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
1802
1803		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
1804			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
1805	}
1806	mtx_unlock(&lun->lun_lock);
1807
1808	return (retval);
1809}
1810
1811static void *
1812ctl_copyin_alloc(void *user_addr, int len, char *error_str,
1813		 size_t error_str_len)
1814{
1815	void *kptr;
1816
1817	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
1818
1819	if (copyin(user_addr, kptr, len) != 0) {
1820		snprintf(error_str, error_str_len, "Error copying %d bytes "
1821			 "from user address %p to kernel address %p", len,
1822			 user_addr, kptr);
1823		free(kptr, M_CTL);
1824		return (NULL);
1825	}
1826
1827	return (kptr);
1828}
1829
1830static void
1831ctl_free_args(int num_args, struct ctl_be_arg *args)
1832{
1833	int i;
1834
1835	if (args == NULL)
1836		return;
1837
1838	for (i = 0; i < num_args; i++) {
1839		free(args[i].kname, M_CTL);
1840		free(args[i].kvalue, M_CTL);
1841	}
1842
1843	free(args, M_CTL);
1844}
1845
1846static struct ctl_be_arg *
1847ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
1848		char *error_str, size_t error_str_len)
1849{
1850	struct ctl_be_arg *args;
1851	int i;
1852
1853	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
1854				error_str, error_str_len);
1855
1856	if (args == NULL)
1857		goto bailout;
1858
1859	for (i = 0; i < num_args; i++) {
1860		args[i].kname = NULL;
1861		args[i].kvalue = NULL;
1862	}
1863
1864	for (i = 0; i < num_args; i++) {
1865		uint8_t *tmpptr;
1866
1867		args[i].kname = ctl_copyin_alloc(args[i].name,
1868			args[i].namelen, error_str, error_str_len);
1869		if (args[i].kname == NULL)
1870			goto bailout;
1871
1872		if (args[i].kname[args[i].namelen - 1] != '\0') {
1873			snprintf(error_str, error_str_len, "Argument %d "
1874				 "name is not NUL-terminated", i);
1875			goto bailout;
1876		}
1877
1878		if (args[i].flags & CTL_BEARG_RD) {
1879			tmpptr = ctl_copyin_alloc(args[i].value,
1880				args[i].vallen, error_str, error_str_len);
1881			if (tmpptr == NULL)
1882				goto bailout;
1883			if ((args[i].flags & CTL_BEARG_ASCII)
1884			 && (tmpptr[args[i].vallen - 1] != '\0')) {
1885				snprintf(error_str, error_str_len, "Argument "
1886				    "%d value is not NUL-terminated", i);
1887				goto bailout;
1888			}
1889			args[i].kvalue = tmpptr;
1890		} else {
1891			args[i].kvalue = malloc(args[i].vallen,
1892			    M_CTL, M_WAITOK | M_ZERO);
1893		}
1894	}
1895
1896	return (args);
1897bailout:
1898
1899	ctl_free_args(num_args, args);
1900
1901	return (NULL);
1902}
1903
1904static void
1905ctl_copyout_args(int num_args, struct ctl_be_arg *args)
1906{
1907	int i;
1908
1909	for (i = 0; i < num_args; i++) {
1910		if (args[i].flags & CTL_BEARG_WR)
1911			copyout(args[i].kvalue, args[i].value, args[i].vallen);
1912	}
1913}
1914
1915/*
1916 * Escape characters that are illegal or not recommended in XML.
1917 */
1918int
1919ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
1920{
1921	char *end = str + size;
1922	int retval;
1923
1924	retval = 0;
1925
1926	for (; *str && str < end; str++) {
1927		switch (*str) {
1928		case '&':
1929			retval = sbuf_printf(sb, "&amp;");
1930			break;
1931		case '>':
1932			retval = sbuf_printf(sb, "&gt;");
1933			break;
1934		case '<':
1935			retval = sbuf_printf(sb, "&lt;");
1936			break;
1937		default:
1938			retval = sbuf_putc(sb, *str);
1939			break;
1940		}
1941
1942		if (retval != 0)
1943			break;
1944
1945	}
1946
1947	return (retval);
1948}
1949
1950static void
1951ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
1952{
1953	struct scsi_vpd_id_descriptor *desc;
1954	int i;
1955
1956	if (id == NULL || id->len < 4)
1957		return;
1958	desc = (struct scsi_vpd_id_descriptor *)id->data;
1959	switch (desc->id_type & SVPD_ID_TYPE_MASK) {
1960	case SVPD_ID_TYPE_T10:
1961		sbuf_printf(sb, "t10.");
1962		break;
1963	case SVPD_ID_TYPE_EUI64:
1964		sbuf_printf(sb, "eui.");
1965		break;
1966	case SVPD_ID_TYPE_NAA:
1967		sbuf_printf(sb, "naa.");
1968		break;
1969	case SVPD_ID_TYPE_SCSI_NAME:
1970		break;
1971	}
1972	switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
1973	case SVPD_ID_CODESET_BINARY:
1974		for (i = 0; i < desc->length; i++)
1975			sbuf_printf(sb, "%02x", desc->identifier[i]);
1976		break;
1977	case SVPD_ID_CODESET_ASCII:
1978		sbuf_printf(sb, "%.*s", (int)desc->length,
1979		    (char *)desc->identifier);
1980		break;
1981	case SVPD_ID_CODESET_UTF8:
1982		sbuf_printf(sb, "%s", (char *)desc->identifier);
1983		break;
1984	}
1985}
1986
1987static int
1988ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
1989	  struct thread *td)
1990{
1991	struct ctl_softc *softc;
1992	int retval;
1993
1994	softc = control_softc;
1995
1996	retval = 0;
1997
1998	switch (cmd) {
1999	case CTL_IO:
2000		retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2001		break;
2002	case CTL_ENABLE_PORT:
2003	case CTL_DISABLE_PORT:
2004	case CTL_SET_PORT_WWNS: {
2005		struct ctl_port *port;
2006		struct ctl_port_entry *entry;
2007
2008		entry = (struct ctl_port_entry *)addr;
2009
2010		mtx_lock(&softc->ctl_lock);
2011		STAILQ_FOREACH(port, &softc->port_list, links) {
2012			int action, done;
2013
2014			action = 0;
2015			done = 0;
2016
2017			if ((entry->port_type == CTL_PORT_NONE)
2018			 && (entry->targ_port == port->targ_port)) {
2019				/*
2020				 * If the user only wants to enable or
2021				 * disable or set WWNs on a specific port,
2022				 * do the operation and we're done.
2023				 */
2024				action = 1;
2025				done = 1;
2026			} else if (entry->port_type & port->port_type) {
2027				/*
2028				 * Compare the user's type mask with the
2029				 * particular frontend type to see if we
2030				 * have a match.
2031				 */
2032				action = 1;
2033				done = 0;
2034
2035				/*
2036				 * Make sure the user isn't trying to set
2037				 * WWNs on multiple ports at the same time.
2038				 */
2039				if (cmd == CTL_SET_PORT_WWNS) {
2040					printf("%s: Can't set WWNs on "
2041					       "multiple ports\n", __func__);
2042					retval = EINVAL;
2043					break;
2044				}
2045			}
2046			if (action != 0) {
2047				/*
2048				 * XXX KDM we have to drop the lock here,
2049				 * because the online/offline operations
2050				 * can potentially block.  We need to
2051				 * reference count the frontends so they
2052				 * can't go away,
2053				 */
2054				mtx_unlock(&softc->ctl_lock);
2055
2056				if (cmd == CTL_ENABLE_PORT) {
2057					ctl_port_online(port);
2058				} else if (cmd == CTL_DISABLE_PORT) {
2059					ctl_port_offline(port);
2060				}
2061
2062				mtx_lock(&softc->ctl_lock);
2063
2064				if (cmd == CTL_SET_PORT_WWNS)
2065					ctl_port_set_wwns(port,
2066					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2067					    1 : 0, entry->wwnn,
2068					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2069					    1 : 0, entry->wwpn);
2070			}
2071			if (done != 0)
2072				break;
2073		}
2074		mtx_unlock(&softc->ctl_lock);
2075		break;
2076	}
2077	case CTL_GET_PORT_LIST: {
2078		struct ctl_port *port;
2079		struct ctl_port_list *list;
2080		int i;
2081
2082		list = (struct ctl_port_list *)addr;
2083
2084		if (list->alloc_len != (list->alloc_num *
2085		    sizeof(struct ctl_port_entry))) {
2086			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2087			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2088			       "%zu\n", __func__, list->alloc_len,
2089			       list->alloc_num, sizeof(struct ctl_port_entry));
2090			retval = EINVAL;
2091			break;
2092		}
2093		list->fill_len = 0;
2094		list->fill_num = 0;
2095		list->dropped_num = 0;
2096		i = 0;
2097		mtx_lock(&softc->ctl_lock);
2098		STAILQ_FOREACH(port, &softc->port_list, links) {
2099			struct ctl_port_entry entry, *list_entry;
2100
2101			if (list->fill_num >= list->alloc_num) {
2102				list->dropped_num++;
2103				continue;
2104			}
2105
2106			entry.port_type = port->port_type;
2107			strlcpy(entry.port_name, port->port_name,
2108				sizeof(entry.port_name));
2109			entry.targ_port = port->targ_port;
2110			entry.physical_port = port->physical_port;
2111			entry.virtual_port = port->virtual_port;
2112			entry.wwnn = port->wwnn;
2113			entry.wwpn = port->wwpn;
2114			if (port->status & CTL_PORT_STATUS_ONLINE)
2115				entry.online = 1;
2116			else
2117				entry.online = 0;
2118
2119			list_entry = &list->entries[i];
2120
2121			retval = copyout(&entry, list_entry, sizeof(entry));
2122			if (retval != 0) {
2123				printf("%s: CTL_GET_PORT_LIST: copyout "
2124				       "returned %d\n", __func__, retval);
2125				break;
2126			}
2127			i++;
2128			list->fill_num++;
2129			list->fill_len += sizeof(entry);
2130		}
2131		mtx_unlock(&softc->ctl_lock);
2132
2133		/*
2134		 * If this is non-zero, we had a copyout fault, so there's
2135		 * probably no point in attempting to set the status inside
2136		 * the structure.
2137		 */
2138		if (retval != 0)
2139			break;
2140
2141		if (list->dropped_num > 0)
2142			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2143		else
2144			list->status = CTL_PORT_LIST_OK;
2145		break;
2146	}
2147	case CTL_DUMP_OOA: {
2148		struct ctl_lun *lun;
2149		union ctl_io *io;
2150		char printbuf[128];
2151		struct sbuf sb;
2152
2153		mtx_lock(&softc->ctl_lock);
2154		printf("Dumping OOA queues:\n");
2155		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2156			mtx_lock(&lun->lun_lock);
2157			for (io = (union ctl_io *)TAILQ_FIRST(
2158			     &lun->ooa_queue); io != NULL;
2159			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2160			     ooa_links)) {
2161				sbuf_new(&sb, printbuf, sizeof(printbuf),
2162					 SBUF_FIXEDLEN);
2163				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2164					    (intmax_t)lun->lun,
2165					    io->scsiio.tag_num,
2166					    (io->io_hdr.flags &
2167					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2168					    (io->io_hdr.flags &
2169					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2170					    (io->io_hdr.flags &
2171					    CTL_FLAG_ABORT) ? " ABORT" : "",
2172			                    (io->io_hdr.flags &
2173		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2174				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2175				sbuf_finish(&sb);
2176				printf("%s\n", sbuf_data(&sb));
2177			}
2178			mtx_unlock(&lun->lun_lock);
2179		}
2180		printf("OOA queues dump done\n");
2181		mtx_unlock(&softc->ctl_lock);
2182		break;
2183	}
2184	case CTL_GET_OOA: {
2185		struct ctl_lun *lun;
2186		struct ctl_ooa *ooa_hdr;
2187		struct ctl_ooa_entry *entries;
2188		uint32_t cur_fill_num;
2189
2190		ooa_hdr = (struct ctl_ooa *)addr;
2191
2192		if ((ooa_hdr->alloc_len == 0)
2193		 || (ooa_hdr->alloc_num == 0)) {
2194			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2195			       "must be non-zero\n", __func__,
2196			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2197			retval = EINVAL;
2198			break;
2199		}
2200
2201		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2202		    sizeof(struct ctl_ooa_entry))) {
2203			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2204			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2205			       __func__, ooa_hdr->alloc_len,
2206			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2207			retval = EINVAL;
2208			break;
2209		}
2210
2211		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2212		if (entries == NULL) {
2213			printf("%s: could not allocate %d bytes for OOA "
2214			       "dump\n", __func__, ooa_hdr->alloc_len);
2215			retval = ENOMEM;
2216			break;
2217		}
2218
2219		mtx_lock(&softc->ctl_lock);
2220		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2221		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2222		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2223			mtx_unlock(&softc->ctl_lock);
2224			free(entries, M_CTL);
2225			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2226			       __func__, (uintmax_t)ooa_hdr->lun_num);
2227			retval = EINVAL;
2228			break;
2229		}
2230
2231		cur_fill_num = 0;
2232
2233		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2234			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2235				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2236					ooa_hdr, entries);
2237				if (retval != 0)
2238					break;
2239			}
2240			if (retval != 0) {
2241				mtx_unlock(&softc->ctl_lock);
2242				free(entries, M_CTL);
2243				break;
2244			}
2245		} else {
2246			lun = softc->ctl_luns[ooa_hdr->lun_num];
2247
2248			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2249						    entries);
2250		}
2251		mtx_unlock(&softc->ctl_lock);
2252
2253		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2254		ooa_hdr->fill_len = ooa_hdr->fill_num *
2255			sizeof(struct ctl_ooa_entry);
2256		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2257		if (retval != 0) {
2258			printf("%s: error copying out %d bytes for OOA dump\n",
2259			       __func__, ooa_hdr->fill_len);
2260		}
2261
2262		getbintime(&ooa_hdr->cur_bt);
2263
2264		if (cur_fill_num > ooa_hdr->alloc_num) {
2265			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2266			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2267		} else {
2268			ooa_hdr->dropped_num = 0;
2269			ooa_hdr->status = CTL_OOA_OK;
2270		}
2271
2272		free(entries, M_CTL);
2273		break;
2274	}
2275	case CTL_CHECK_OOA: {
2276		union ctl_io *io;
2277		struct ctl_lun *lun;
2278		struct ctl_ooa_info *ooa_info;
2279
2280
2281		ooa_info = (struct ctl_ooa_info *)addr;
2282
2283		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2284			ooa_info->status = CTL_OOA_INVALID_LUN;
2285			break;
2286		}
2287		mtx_lock(&softc->ctl_lock);
2288		lun = softc->ctl_luns[ooa_info->lun_id];
2289		if (lun == NULL) {
2290			mtx_unlock(&softc->ctl_lock);
2291			ooa_info->status = CTL_OOA_INVALID_LUN;
2292			break;
2293		}
2294		mtx_lock(&lun->lun_lock);
2295		mtx_unlock(&softc->ctl_lock);
2296		ooa_info->num_entries = 0;
2297		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2298		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2299		     &io->io_hdr, ooa_links)) {
2300			ooa_info->num_entries++;
2301		}
2302		mtx_unlock(&lun->lun_lock);
2303
2304		ooa_info->status = CTL_OOA_SUCCESS;
2305
2306		break;
2307	}
2308	case CTL_DELAY_IO: {
2309		struct ctl_io_delay_info *delay_info;
2310#ifdef CTL_IO_DELAY
2311		struct ctl_lun *lun;
2312#endif /* CTL_IO_DELAY */
2313
2314		delay_info = (struct ctl_io_delay_info *)addr;
2315
2316#ifdef CTL_IO_DELAY
2317		mtx_lock(&softc->ctl_lock);
2318
2319		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2320		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2321			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2322		} else {
2323			lun = softc->ctl_luns[delay_info->lun_id];
2324			mtx_lock(&lun->lun_lock);
2325
2326			delay_info->status = CTL_DELAY_STATUS_OK;
2327
2328			switch (delay_info->delay_type) {
2329			case CTL_DELAY_TYPE_CONT:
2330				break;
2331			case CTL_DELAY_TYPE_ONESHOT:
2332				break;
2333			default:
2334				delay_info->status =
2335					CTL_DELAY_STATUS_INVALID_TYPE;
2336				break;
2337			}
2338
2339			switch (delay_info->delay_loc) {
2340			case CTL_DELAY_LOC_DATAMOVE:
2341				lun->delay_info.datamove_type =
2342					delay_info->delay_type;
2343				lun->delay_info.datamove_delay =
2344					delay_info->delay_secs;
2345				break;
2346			case CTL_DELAY_LOC_DONE:
2347				lun->delay_info.done_type =
2348					delay_info->delay_type;
2349				lun->delay_info.done_delay =
2350					delay_info->delay_secs;
2351				break;
2352			default:
2353				delay_info->status =
2354					CTL_DELAY_STATUS_INVALID_LOC;
2355				break;
2356			}
2357			mtx_unlock(&lun->lun_lock);
2358		}
2359
2360		mtx_unlock(&softc->ctl_lock);
2361#else
2362		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2363#endif /* CTL_IO_DELAY */
2364		break;
2365	}
2366	case CTL_REALSYNC_SET: {
2367		int *syncstate;
2368
2369		syncstate = (int *)addr;
2370
2371		mtx_lock(&softc->ctl_lock);
2372		switch (*syncstate) {
2373		case 0:
2374			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2375			break;
2376		case 1:
2377			softc->flags |= CTL_FLAG_REAL_SYNC;
2378			break;
2379		default:
2380			retval = EINVAL;
2381			break;
2382		}
2383		mtx_unlock(&softc->ctl_lock);
2384		break;
2385	}
2386	case CTL_REALSYNC_GET: {
2387		int *syncstate;
2388
2389		syncstate = (int*)addr;
2390
2391		mtx_lock(&softc->ctl_lock);
2392		if (softc->flags & CTL_FLAG_REAL_SYNC)
2393			*syncstate = 1;
2394		else
2395			*syncstate = 0;
2396		mtx_unlock(&softc->ctl_lock);
2397
2398		break;
2399	}
2400	case CTL_SETSYNC:
2401	case CTL_GETSYNC: {
2402		struct ctl_sync_info *sync_info;
2403		struct ctl_lun *lun;
2404
2405		sync_info = (struct ctl_sync_info *)addr;
2406
2407		mtx_lock(&softc->ctl_lock);
2408		lun = softc->ctl_luns[sync_info->lun_id];
2409		if (lun == NULL) {
2410			mtx_unlock(&softc->ctl_lock);
2411			sync_info->status = CTL_GS_SYNC_NO_LUN;
2412		}
2413		/*
2414		 * Get or set the sync interval.  We're not bounds checking
2415		 * in the set case, hopefully the user won't do something
2416		 * silly.
2417		 */
2418		mtx_lock(&lun->lun_lock);
2419		mtx_unlock(&softc->ctl_lock);
2420		if (cmd == CTL_GETSYNC)
2421			sync_info->sync_interval = lun->sync_interval;
2422		else
2423			lun->sync_interval = sync_info->sync_interval;
2424		mtx_unlock(&lun->lun_lock);
2425
2426		sync_info->status = CTL_GS_SYNC_OK;
2427
2428		break;
2429	}
2430	case CTL_GETSTATS: {
2431		struct ctl_stats *stats;
2432		struct ctl_lun *lun;
2433		int i;
2434
2435		stats = (struct ctl_stats *)addr;
2436
2437		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2438		     stats->alloc_len) {
2439			stats->status = CTL_SS_NEED_MORE_SPACE;
2440			stats->num_luns = softc->num_luns;
2441			break;
2442		}
2443		/*
2444		 * XXX KDM no locking here.  If the LUN list changes,
2445		 * things can blow up.
2446		 */
2447		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2448		     i++, lun = STAILQ_NEXT(lun, links)) {
2449			retval = copyout(&lun->stats, &stats->lun_stats[i],
2450					 sizeof(lun->stats));
2451			if (retval != 0)
2452				break;
2453		}
2454		stats->num_luns = softc->num_luns;
2455		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2456				 softc->num_luns;
2457		stats->status = CTL_SS_OK;
2458#ifdef CTL_TIME_IO
2459		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2460#else
2461		stats->flags = CTL_STATS_FLAG_NONE;
2462#endif
2463		getnanouptime(&stats->timestamp);
2464		break;
2465	}
2466	case CTL_ERROR_INJECT: {
2467		struct ctl_error_desc *err_desc, *new_err_desc;
2468		struct ctl_lun *lun;
2469
2470		err_desc = (struct ctl_error_desc *)addr;
2471
2472		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2473				      M_WAITOK | M_ZERO);
2474		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2475
2476		mtx_lock(&softc->ctl_lock);
2477		lun = softc->ctl_luns[err_desc->lun_id];
2478		if (lun == NULL) {
2479			mtx_unlock(&softc->ctl_lock);
2480			free(new_err_desc, M_CTL);
2481			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2482			       __func__, (uintmax_t)err_desc->lun_id);
2483			retval = EINVAL;
2484			break;
2485		}
2486		mtx_lock(&lun->lun_lock);
2487		mtx_unlock(&softc->ctl_lock);
2488
2489		/*
2490		 * We could do some checking here to verify the validity
2491		 * of the request, but given the complexity of error
2492		 * injection requests, the checking logic would be fairly
2493		 * complex.
2494		 *
2495		 * For now, if the request is invalid, it just won't get
2496		 * executed and might get deleted.
2497		 */
2498		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2499
2500		/*
2501		 * XXX KDM check to make sure the serial number is unique,
2502		 * in case we somehow manage to wrap.  That shouldn't
2503		 * happen for a very long time, but it's the right thing to
2504		 * do.
2505		 */
2506		new_err_desc->serial = lun->error_serial;
2507		err_desc->serial = lun->error_serial;
2508		lun->error_serial++;
2509
2510		mtx_unlock(&lun->lun_lock);
2511		break;
2512	}
2513	case CTL_ERROR_INJECT_DELETE: {
2514		struct ctl_error_desc *delete_desc, *desc, *desc2;
2515		struct ctl_lun *lun;
2516		int delete_done;
2517
2518		delete_desc = (struct ctl_error_desc *)addr;
2519		delete_done = 0;
2520
2521		mtx_lock(&softc->ctl_lock);
2522		lun = softc->ctl_luns[delete_desc->lun_id];
2523		if (lun == NULL) {
2524			mtx_unlock(&softc->ctl_lock);
2525			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2526			       __func__, (uintmax_t)delete_desc->lun_id);
2527			retval = EINVAL;
2528			break;
2529		}
2530		mtx_lock(&lun->lun_lock);
2531		mtx_unlock(&softc->ctl_lock);
2532		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2533			if (desc->serial != delete_desc->serial)
2534				continue;
2535
2536			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2537				      links);
2538			free(desc, M_CTL);
2539			delete_done = 1;
2540		}
2541		mtx_unlock(&lun->lun_lock);
2542		if (delete_done == 0) {
2543			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2544			       "error serial %ju on LUN %u\n", __func__,
2545			       delete_desc->serial, delete_desc->lun_id);
2546			retval = EINVAL;
2547			break;
2548		}
2549		break;
2550	}
2551	case CTL_DUMP_STRUCTS: {
2552		int i, j, k;
2553		struct ctl_port *port;
2554		struct ctl_frontend *fe;
2555
2556		mtx_lock(&softc->ctl_lock);
2557		printf("CTL Persistent Reservation information start:\n");
2558		for (i = 0; i < CTL_MAX_LUNS; i++) {
2559			struct ctl_lun *lun;
2560
2561			lun = softc->ctl_luns[i];
2562
2563			if ((lun == NULL)
2564			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2565				continue;
2566
2567			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2568				if (lun->pr_keys[j] == NULL)
2569					continue;
2570				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2571					if (lun->pr_keys[j][k] == 0)
2572						continue;
2573					printf("  LUN %d port %d iid %d key "
2574					       "%#jx\n", i, j, k,
2575					       (uintmax_t)lun->pr_keys[j][k]);
2576				}
2577			}
2578		}
2579		printf("CTL Persistent Reservation information end\n");
2580		printf("CTL Ports:\n");
2581		STAILQ_FOREACH(port, &softc->port_list, links) {
2582			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2583			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2584			       port->frontend->name, port->port_type,
2585			       port->physical_port, port->virtual_port,
2586			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2587			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2588				if (port->wwpn_iid[j].in_use == 0 &&
2589				    port->wwpn_iid[j].wwpn == 0 &&
2590				    port->wwpn_iid[j].name == NULL)
2591					continue;
2592
2593				printf("    iid %u use %d WWPN %#jx '%s'\n",
2594				    j, port->wwpn_iid[j].in_use,
2595				    (uintmax_t)port->wwpn_iid[j].wwpn,
2596				    port->wwpn_iid[j].name);
2597			}
2598		}
2599		printf("CTL Port information end\n");
2600		mtx_unlock(&softc->ctl_lock);
2601		/*
2602		 * XXX KDM calling this without a lock.  We'd likely want
2603		 * to drop the lock before calling the frontend's dump
2604		 * routine anyway.
2605		 */
2606		printf("CTL Frontends:\n");
2607		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2608			printf("  Frontend '%s'\n", fe->name);
2609			if (fe->fe_dump != NULL)
2610				fe->fe_dump();
2611		}
2612		printf("CTL Frontend information end\n");
2613		break;
2614	}
2615	case CTL_LUN_REQ: {
2616		struct ctl_lun_req *lun_req;
2617		struct ctl_backend_driver *backend;
2618
2619		lun_req = (struct ctl_lun_req *)addr;
2620
2621		backend = ctl_backend_find(lun_req->backend);
2622		if (backend == NULL) {
2623			lun_req->status = CTL_LUN_ERROR;
2624			snprintf(lun_req->error_str,
2625				 sizeof(lun_req->error_str),
2626				 "Backend \"%s\" not found.",
2627				 lun_req->backend);
2628			break;
2629		}
2630		if (lun_req->num_be_args > 0) {
2631			lun_req->kern_be_args = ctl_copyin_args(
2632				lun_req->num_be_args,
2633				lun_req->be_args,
2634				lun_req->error_str,
2635				sizeof(lun_req->error_str));
2636			if (lun_req->kern_be_args == NULL) {
2637				lun_req->status = CTL_LUN_ERROR;
2638				break;
2639			}
2640		}
2641
2642		retval = backend->ioctl(dev, cmd, addr, flag, td);
2643
2644		if (lun_req->num_be_args > 0) {
2645			ctl_copyout_args(lun_req->num_be_args,
2646				      lun_req->kern_be_args);
2647			ctl_free_args(lun_req->num_be_args,
2648				      lun_req->kern_be_args);
2649		}
2650		break;
2651	}
2652	case CTL_LUN_LIST: {
2653		struct sbuf *sb;
2654		struct ctl_lun *lun;
2655		struct ctl_lun_list *list;
2656		struct ctl_option *opt;
2657
2658		list = (struct ctl_lun_list *)addr;
2659
2660		/*
2661		 * Allocate a fixed length sbuf here, based on the length
2662		 * of the user's buffer.  We could allocate an auto-extending
2663		 * buffer, and then tell the user how much larger our
2664		 * amount of data is than his buffer, but that presents
2665		 * some problems:
2666		 *
2667		 * 1.  The sbuf(9) routines use a blocking malloc, and so
2668		 *     we can't hold a lock while calling them with an
2669		 *     auto-extending buffer.
2670 		 *
2671		 * 2.  There is not currently a LUN reference counting
2672		 *     mechanism, outside of outstanding transactions on
2673		 *     the LUN's OOA queue.  So a LUN could go away on us
2674		 *     while we're getting the LUN number, backend-specific
2675		 *     information, etc.  Thus, given the way things
2676		 *     currently work, we need to hold the CTL lock while
2677		 *     grabbing LUN information.
2678		 *
2679		 * So, from the user's standpoint, the best thing to do is
2680		 * allocate what he thinks is a reasonable buffer length,
2681		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
2682		 * double the buffer length and try again.  (And repeat
2683		 * that until he succeeds.)
2684		 */
2685		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
2686		if (sb == NULL) {
2687			list->status = CTL_LUN_LIST_ERROR;
2688			snprintf(list->error_str, sizeof(list->error_str),
2689				 "Unable to allocate %d bytes for LUN list",
2690				 list->alloc_len);
2691			break;
2692		}
2693
2694		sbuf_printf(sb, "<ctllunlist>\n");
2695
2696		mtx_lock(&softc->ctl_lock);
2697		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2698			mtx_lock(&lun->lun_lock);
2699			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
2700					     (uintmax_t)lun->lun);
2701
2702			/*
2703			 * Bail out as soon as we see that we've overfilled
2704			 * the buffer.
2705			 */
2706			if (retval != 0)
2707				break;
2708
2709			retval = sbuf_printf(sb, "\t<backend_type>%s"
2710					     "</backend_type>\n",
2711					     (lun->backend == NULL) ?  "none" :
2712					     lun->backend->name);
2713
2714			if (retval != 0)
2715				break;
2716
2717			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
2718					     lun->be_lun->lun_type);
2719
2720			if (retval != 0)
2721				break;
2722
2723			if (lun->backend == NULL) {
2724				retval = sbuf_printf(sb, "</lun>\n");
2725				if (retval != 0)
2726					break;
2727				continue;
2728			}
2729
2730			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
2731					     (lun->be_lun->maxlba > 0) ?
2732					     lun->be_lun->maxlba + 1 : 0);
2733
2734			if (retval != 0)
2735				break;
2736
2737			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
2738					     lun->be_lun->blocksize);
2739
2740			if (retval != 0)
2741				break;
2742
2743			retval = sbuf_printf(sb, "\t<serial_number>");
2744
2745			if (retval != 0)
2746				break;
2747
2748			retval = ctl_sbuf_printf_esc(sb,
2749			    lun->be_lun->serial_num,
2750			    sizeof(lun->be_lun->serial_num));
2751
2752			if (retval != 0)
2753				break;
2754
2755			retval = sbuf_printf(sb, "</serial_number>\n");
2756
2757			if (retval != 0)
2758				break;
2759
2760			retval = sbuf_printf(sb, "\t<device_id>");
2761
2762			if (retval != 0)
2763				break;
2764
2765			retval = ctl_sbuf_printf_esc(sb,
2766			    lun->be_lun->device_id,
2767			    sizeof(lun->be_lun->device_id));
2768
2769			if (retval != 0)
2770				break;
2771
2772			retval = sbuf_printf(sb, "</device_id>\n");
2773
2774			if (retval != 0)
2775				break;
2776
2777			if (lun->backend->lun_info != NULL) {
2778				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
2779				if (retval != 0)
2780					break;
2781			}
2782			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
2783				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
2784				    opt->name, opt->value, opt->name);
2785				if (retval != 0)
2786					break;
2787			}
2788
2789			retval = sbuf_printf(sb, "</lun>\n");
2790
2791			if (retval != 0)
2792				break;
2793			mtx_unlock(&lun->lun_lock);
2794		}
2795		if (lun != NULL)
2796			mtx_unlock(&lun->lun_lock);
2797		mtx_unlock(&softc->ctl_lock);
2798
2799		if ((retval != 0)
2800		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
2801			retval = 0;
2802			sbuf_delete(sb);
2803			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
2804			snprintf(list->error_str, sizeof(list->error_str),
2805				 "Out of space, %d bytes is too small",
2806				 list->alloc_len);
2807			break;
2808		}
2809
2810		sbuf_finish(sb);
2811
2812		retval = copyout(sbuf_data(sb), list->lun_xml,
2813				 sbuf_len(sb) + 1);
2814
2815		list->fill_len = sbuf_len(sb) + 1;
2816		list->status = CTL_LUN_LIST_OK;
2817		sbuf_delete(sb);
2818		break;
2819	}
2820	case CTL_ISCSI: {
2821		struct ctl_iscsi *ci;
2822		struct ctl_frontend *fe;
2823
2824		ci = (struct ctl_iscsi *)addr;
2825
2826		fe = ctl_frontend_find("iscsi");
2827		if (fe == NULL) {
2828			ci->status = CTL_ISCSI_ERROR;
2829			snprintf(ci->error_str, sizeof(ci->error_str),
2830			    "Frontend \"iscsi\" not found.");
2831			break;
2832		}
2833
2834		retval = fe->ioctl(dev, cmd, addr, flag, td);
2835		break;
2836	}
2837	case CTL_PORT_REQ: {
2838		struct ctl_req *req;
2839		struct ctl_frontend *fe;
2840
2841		req = (struct ctl_req *)addr;
2842
2843		fe = ctl_frontend_find(req->driver);
2844		if (fe == NULL) {
2845			req->status = CTL_LUN_ERROR;
2846			snprintf(req->error_str, sizeof(req->error_str),
2847			    "Frontend \"%s\" not found.", req->driver);
2848			break;
2849		}
2850		if (req->num_args > 0) {
2851			req->kern_args = ctl_copyin_args(req->num_args,
2852			    req->args, req->error_str, sizeof(req->error_str));
2853			if (req->kern_args == NULL) {
2854				req->status = CTL_LUN_ERROR;
2855				break;
2856			}
2857		}
2858
2859		retval = fe->ioctl(dev, cmd, addr, flag, td);
2860
2861		if (req->num_args > 0) {
2862			ctl_copyout_args(req->num_args, req->kern_args);
2863			ctl_free_args(req->num_args, req->kern_args);
2864		}
2865		break;
2866	}
2867	case CTL_PORT_LIST: {
2868		struct sbuf *sb;
2869		struct ctl_port *port;
2870		struct ctl_lun_list *list;
2871		struct ctl_option *opt;
2872		int j;
2873		uint32_t plun;
2874
2875		list = (struct ctl_lun_list *)addr;
2876
2877		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
2878		if (sb == NULL) {
2879			list->status = CTL_LUN_LIST_ERROR;
2880			snprintf(list->error_str, sizeof(list->error_str),
2881				 "Unable to allocate %d bytes for LUN list",
2882				 list->alloc_len);
2883			break;
2884		}
2885
2886		sbuf_printf(sb, "<ctlportlist>\n");
2887
2888		mtx_lock(&softc->ctl_lock);
2889		STAILQ_FOREACH(port, &softc->port_list, links) {
2890			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
2891					     (uintmax_t)port->targ_port);
2892
2893			/*
2894			 * Bail out as soon as we see that we've overfilled
2895			 * the buffer.
2896			 */
2897			if (retval != 0)
2898				break;
2899
2900			retval = sbuf_printf(sb, "\t<frontend_type>%s"
2901			    "</frontend_type>\n", port->frontend->name);
2902			if (retval != 0)
2903				break;
2904
2905			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
2906					     port->port_type);
2907			if (retval != 0)
2908				break;
2909
2910			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
2911			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
2912			if (retval != 0)
2913				break;
2914
2915			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
2916			    port->port_name);
2917			if (retval != 0)
2918				break;
2919
2920			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
2921			    port->physical_port);
2922			if (retval != 0)
2923				break;
2924
2925			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
2926			    port->virtual_port);
2927			if (retval != 0)
2928				break;
2929
2930			if (port->target_devid != NULL) {
2931				sbuf_printf(sb, "\t<target>");
2932				ctl_id_sbuf(port->target_devid, sb);
2933				sbuf_printf(sb, "</target>\n");
2934			}
2935
2936			if (port->port_devid != NULL) {
2937				sbuf_printf(sb, "\t<port>");
2938				ctl_id_sbuf(port->port_devid, sb);
2939				sbuf_printf(sb, "</port>\n");
2940			}
2941
2942			if (port->port_info != NULL) {
2943				retval = port->port_info(port->onoff_arg, sb);
2944				if (retval != 0)
2945					break;
2946			}
2947			STAILQ_FOREACH(opt, &port->options, links) {
2948				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
2949				    opt->name, opt->value, opt->name);
2950				if (retval != 0)
2951					break;
2952			}
2953
2954			if (port->lun_map != NULL) {
2955				sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
2956				for (j = 0; j < CTL_MAX_LUNS; j++) {
2957					plun = ctl_lun_map_from_port(port, j);
2958					if (plun >= CTL_MAX_LUNS)
2959						continue;
2960					sbuf_printf(sb,
2961					    "\t<lun id=\"%u\">%u</lun>\n",
2962					    j, plun);
2963				}
2964			}
2965
2966			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2967				if (port->wwpn_iid[j].in_use == 0 ||
2968				    (port->wwpn_iid[j].wwpn == 0 &&
2969				     port->wwpn_iid[j].name == NULL))
2970					continue;
2971
2972				if (port->wwpn_iid[j].name != NULL)
2973					retval = sbuf_printf(sb,
2974					    "\t<initiator id=\"%u\">%s</initiator>\n",
2975					    j, port->wwpn_iid[j].name);
2976				else
2977					retval = sbuf_printf(sb,
2978					    "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
2979					    j, port->wwpn_iid[j].wwpn);
2980				if (retval != 0)
2981					break;
2982			}
2983			if (retval != 0)
2984				break;
2985
2986			retval = sbuf_printf(sb, "</targ_port>\n");
2987			if (retval != 0)
2988				break;
2989		}
2990		mtx_unlock(&softc->ctl_lock);
2991
2992		if ((retval != 0)
2993		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
2994			retval = 0;
2995			sbuf_delete(sb);
2996			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
2997			snprintf(list->error_str, sizeof(list->error_str),
2998				 "Out of space, %d bytes is too small",
2999				 list->alloc_len);
3000			break;
3001		}
3002
3003		sbuf_finish(sb);
3004
3005		retval = copyout(sbuf_data(sb), list->lun_xml,
3006				 sbuf_len(sb) + 1);
3007
3008		list->fill_len = sbuf_len(sb) + 1;
3009		list->status = CTL_LUN_LIST_OK;
3010		sbuf_delete(sb);
3011		break;
3012	}
3013	case CTL_LUN_MAP: {
3014		struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3015		struct ctl_port *port;
3016
3017		mtx_lock(&softc->ctl_lock);
3018		if (lm->port >= CTL_MAX_PORTS ||
3019		    (port = softc->ctl_ports[lm->port]) == NULL) {
3020			mtx_unlock(&softc->ctl_lock);
3021			return (ENXIO);
3022		}
3023		mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3024		if (lm->plun < CTL_MAX_LUNS) {
3025			if (lm->lun == UINT32_MAX)
3026				retval = ctl_lun_map_unset(port, lm->plun);
3027			else if (lm->lun < CTL_MAX_LUNS &&
3028			    softc->ctl_luns[lm->lun] != NULL)
3029				retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3030			else
3031				return (ENXIO);
3032		} else if (lm->plun == UINT32_MAX) {
3033			if (lm->lun == UINT32_MAX)
3034				retval = ctl_lun_map_deinit(port);
3035			else
3036				retval = ctl_lun_map_init(port);
3037		} else
3038			return (ENXIO);
3039		break;
3040	}
3041	default: {
3042		/* XXX KDM should we fix this? */
3043#if 0
3044		struct ctl_backend_driver *backend;
3045		unsigned int type;
3046		int found;
3047
3048		found = 0;
3049
3050		/*
3051		 * We encode the backend type as the ioctl type for backend
3052		 * ioctls.  So parse it out here, and then search for a
3053		 * backend of this type.
3054		 */
3055		type = _IOC_TYPE(cmd);
3056
3057		STAILQ_FOREACH(backend, &softc->be_list, links) {
3058			if (backend->type == type) {
3059				found = 1;
3060				break;
3061			}
3062		}
3063		if (found == 0) {
3064			printf("ctl: unknown ioctl command %#lx or backend "
3065			       "%d\n", cmd, type);
3066			retval = EINVAL;
3067			break;
3068		}
3069		retval = backend->ioctl(dev, cmd, addr, flag, td);
3070#endif
3071		retval = ENOTTY;
3072		break;
3073	}
3074	}
3075	return (retval);
3076}
3077
3078uint32_t
3079ctl_get_initindex(struct ctl_nexus *nexus)
3080{
3081	if (nexus->targ_port < CTL_MAX_PORTS)
3082		return (nexus->initid.id +
3083			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3084	else
3085		return (nexus->initid.id +
3086		       ((nexus->targ_port - CTL_MAX_PORTS) *
3087			CTL_MAX_INIT_PER_PORT));
3088}
3089
3090uint32_t
3091ctl_get_resindex(struct ctl_nexus *nexus)
3092{
3093	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3094}
3095
3096uint32_t
3097ctl_port_idx(int port_num)
3098{
3099	if (port_num < CTL_MAX_PORTS)
3100		return(port_num);
3101	else
3102		return(port_num - CTL_MAX_PORTS);
3103}
3104
3105int
3106ctl_lun_map_init(struct ctl_port *port)
3107{
3108	struct ctl_softc *softc = control_softc;
3109	struct ctl_lun *lun;
3110	uint32_t i;
3111
3112	if (port->lun_map == NULL)
3113		port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
3114		    M_CTL, M_NOWAIT);
3115	if (port->lun_map == NULL)
3116		return (ENOMEM);
3117	for (i = 0; i < CTL_MAX_LUNS; i++)
3118		port->lun_map[i] = UINT32_MAX;
3119	if (port->status & CTL_PORT_STATUS_ONLINE &&
3120	    port->lun_disable != NULL) {
3121		STAILQ_FOREACH(lun, &softc->lun_list, links)
3122			port->lun_disable(port->targ_lun_arg, lun->lun);
3123	}
3124	return (0);
3125}
3126
3127int
3128ctl_lun_map_deinit(struct ctl_port *port)
3129{
3130	struct ctl_softc *softc = control_softc;
3131	struct ctl_lun *lun;
3132
3133	if (port->lun_map == NULL)
3134		return (0);
3135	free(port->lun_map, M_CTL);
3136	port->lun_map = NULL;
3137	if (port->status & CTL_PORT_STATUS_ONLINE &&
3138	    port->lun_enable != NULL) {
3139		STAILQ_FOREACH(lun, &softc->lun_list, links)
3140			port->lun_enable(port->targ_lun_arg, lun->lun);
3141	}
3142	return (0);
3143}
3144
3145int
3146ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3147{
3148	int status;
3149	uint32_t old;
3150
3151	if (port->lun_map == NULL) {
3152		status = ctl_lun_map_init(port);
3153		if (status != 0)
3154			return (status);
3155	}
3156	old = port->lun_map[plun];
3157	port->lun_map[plun] = glun;
3158	if ((port->status & CTL_PORT_STATUS_ONLINE) && old >= CTL_MAX_LUNS &&
3159	    port->lun_enable != NULL)
3160		port->lun_enable(port->targ_lun_arg, plun);
3161	return (0);
3162}
3163
3164int
3165ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3166{
3167	uint32_t old;
3168
3169	if (port->lun_map == NULL)
3170		return (0);
3171	old = port->lun_map[plun];
3172	port->lun_map[plun] = UINT32_MAX;
3173	if ((port->status & CTL_PORT_STATUS_ONLINE) && old < CTL_MAX_LUNS &&
3174	    port->lun_disable != NULL)
3175		port->lun_disable(port->targ_lun_arg, plun);
3176	return (0);
3177}
3178
3179uint32_t
3180ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3181{
3182
3183	if (port == NULL)
3184		return (UINT32_MAX);
3185	if (port->lun_map == NULL || lun_id >= CTL_MAX_LUNS)
3186		return (lun_id);
3187	return (port->lun_map[lun_id]);
3188}
3189
3190uint32_t
3191ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3192{
3193	uint32_t i;
3194
3195	if (port == NULL)
3196		return (UINT32_MAX);
3197	if (port->lun_map == NULL)
3198		return (lun_id);
3199	for (i = 0; i < CTL_MAX_LUNS; i++) {
3200		if (port->lun_map[i] == lun_id)
3201			return (i);
3202	}
3203	return (UINT32_MAX);
3204}
3205
3206static struct ctl_port *
3207ctl_io_port(struct ctl_io_hdr *io_hdr)
3208{
3209	int port_num;
3210
3211	port_num = io_hdr->nexus.targ_port;
3212	return (control_softc->ctl_ports[ctl_port_idx(port_num)]);
3213}
3214
3215/*
3216 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3217 * that are a power of 2.
3218 */
3219int
3220ctl_ffz(uint32_t *mask, uint32_t size)
3221{
3222	uint32_t num_chunks, num_pieces;
3223	int i, j;
3224
3225	num_chunks = (size >> 5);
3226	if (num_chunks == 0)
3227		num_chunks++;
3228	num_pieces = MIN((sizeof(uint32_t) * 8), size);
3229
3230	for (i = 0; i < num_chunks; i++) {
3231		for (j = 0; j < num_pieces; j++) {
3232			if ((mask[i] & (1 << j)) == 0)
3233				return ((i << 5) + j);
3234		}
3235	}
3236
3237	return (-1);
3238}
3239
3240int
3241ctl_set_mask(uint32_t *mask, uint32_t bit)
3242{
3243	uint32_t chunk, piece;
3244
3245	chunk = bit >> 5;
3246	piece = bit % (sizeof(uint32_t) * 8);
3247
3248	if ((mask[chunk] & (1 << piece)) != 0)
3249		return (-1);
3250	else
3251		mask[chunk] |= (1 << piece);
3252
3253	return (0);
3254}
3255
3256int
3257ctl_clear_mask(uint32_t *mask, uint32_t bit)
3258{
3259	uint32_t chunk, piece;
3260
3261	chunk = bit >> 5;
3262	piece = bit % (sizeof(uint32_t) * 8);
3263
3264	if ((mask[chunk] & (1 << piece)) == 0)
3265		return (-1);
3266	else
3267		mask[chunk] &= ~(1 << piece);
3268
3269	return (0);
3270}
3271
3272int
3273ctl_is_set(uint32_t *mask, uint32_t bit)
3274{
3275	uint32_t chunk, piece;
3276
3277	chunk = bit >> 5;
3278	piece = bit % (sizeof(uint32_t) * 8);
3279
3280	if ((mask[chunk] & (1 << piece)) == 0)
3281		return (0);
3282	else
3283		return (1);
3284}
3285
3286static uint64_t
3287ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3288{
3289	uint64_t *t;
3290
3291	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3292	if (t == NULL)
3293		return (0);
3294	return (t[residx % CTL_MAX_INIT_PER_PORT]);
3295}
3296
3297static void
3298ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3299{
3300	uint64_t *t;
3301
3302	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3303	if (t == NULL)
3304		return;
3305	t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3306}
3307
3308static void
3309ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3310{
3311	uint64_t *p;
3312	u_int i;
3313
3314	i = residx/CTL_MAX_INIT_PER_PORT;
3315	if (lun->pr_keys[i] != NULL)
3316		return;
3317	mtx_unlock(&lun->lun_lock);
3318	p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3319	    M_WAITOK | M_ZERO);
3320	mtx_lock(&lun->lun_lock);
3321	if (lun->pr_keys[i] == NULL)
3322		lun->pr_keys[i] = p;
3323	else
3324		free(p, M_CTL);
3325}
3326
3327static void
3328ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3329{
3330	uint64_t *t;
3331
3332	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3333	KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3334	t[residx % CTL_MAX_INIT_PER_PORT] = key;
3335}
3336
3337/*
3338 * ctl_softc, pool_name, total_ctl_io are passed in.
3339 * npool is passed out.
3340 */
3341int
3342ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3343		uint32_t total_ctl_io, void **npool)
3344{
3345#ifdef IO_POOLS
3346	struct ctl_io_pool *pool;
3347
3348	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3349					    M_NOWAIT | M_ZERO);
3350	if (pool == NULL)
3351		return (ENOMEM);
3352
3353	snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3354	pool->ctl_softc = ctl_softc;
3355	pool->zone = uma_zsecond_create(pool->name, NULL,
3356	    NULL, NULL, NULL, ctl_softc->io_zone);
3357	/* uma_prealloc(pool->zone, total_ctl_io); */
3358
3359	*npool = pool;
3360#else
3361	*npool = ctl_softc->io_zone;
3362#endif
3363	return (0);
3364}
3365
3366void
3367ctl_pool_free(struct ctl_io_pool *pool)
3368{
3369
3370	if (pool == NULL)
3371		return;
3372
3373#ifdef IO_POOLS
3374	uma_zdestroy(pool->zone);
3375	free(pool, M_CTL);
3376#endif
3377}
3378
3379union ctl_io *
3380ctl_alloc_io(void *pool_ref)
3381{
3382	union ctl_io *io;
3383#ifdef IO_POOLS
3384	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3385
3386	io = uma_zalloc(pool->zone, M_WAITOK);
3387#else
3388	io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3389#endif
3390	if (io != NULL)
3391		io->io_hdr.pool = pool_ref;
3392	return (io);
3393}
3394
3395union ctl_io *
3396ctl_alloc_io_nowait(void *pool_ref)
3397{
3398	union ctl_io *io;
3399#ifdef IO_POOLS
3400	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3401
3402	io = uma_zalloc(pool->zone, M_NOWAIT);
3403#else
3404	io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3405#endif
3406	if (io != NULL)
3407		io->io_hdr.pool = pool_ref;
3408	return (io);
3409}
3410
3411void
3412ctl_free_io(union ctl_io *io)
3413{
3414#ifdef IO_POOLS
3415	struct ctl_io_pool *pool;
3416#endif
3417
3418	if (io == NULL)
3419		return;
3420
3421#ifdef IO_POOLS
3422	pool = (struct ctl_io_pool *)io->io_hdr.pool;
3423	uma_zfree(pool->zone, io);
3424#else
3425	uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3426#endif
3427}
3428
3429void
3430ctl_zero_io(union ctl_io *io)
3431{
3432	void *pool_ref;
3433
3434	if (io == NULL)
3435		return;
3436
3437	/*
3438	 * May need to preserve linked list pointers at some point too.
3439	 */
3440	pool_ref = io->io_hdr.pool;
3441	memset(io, 0, sizeof(*io));
3442	io->io_hdr.pool = pool_ref;
3443}
3444
3445/*
3446 * This routine is currently used for internal copies of ctl_ios that need
3447 * to persist for some reason after we've already returned status to the
3448 * FETD.  (Thus the flag set.)
3449 *
3450 * XXX XXX
3451 * Note that this makes a blind copy of all fields in the ctl_io, except
3452 * for the pool reference.  This includes any memory that has been
3453 * allocated!  That memory will no longer be valid after done has been
3454 * called, so this would be VERY DANGEROUS for command that actually does
3455 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3456 * start and stop commands, which don't transfer any data, so this is not a
3457 * problem.  If it is used for anything else, the caller would also need to
3458 * allocate data buffer space and this routine would need to be modified to
3459 * copy the data buffer(s) as well.
3460 */
3461void
3462ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3463{
3464	void *pool_ref;
3465
3466	if ((src == NULL)
3467	 || (dest == NULL))
3468		return;
3469
3470	/*
3471	 * May need to preserve linked list pointers at some point too.
3472	 */
3473	pool_ref = dest->io_hdr.pool;
3474
3475	memcpy(dest, src, MIN(sizeof(*src), sizeof(*dest)));
3476
3477	dest->io_hdr.pool = pool_ref;
3478	/*
3479	 * We need to know that this is an internal copy, and doesn't need
3480	 * to get passed back to the FETD that allocated it.
3481	 */
3482	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3483}
3484
3485int
3486ctl_expand_number(const char *buf, uint64_t *num)
3487{
3488	char *endptr;
3489	uint64_t number;
3490	unsigned shift;
3491
3492	number = strtoq(buf, &endptr, 0);
3493
3494	switch (tolower((unsigned char)*endptr)) {
3495	case 'e':
3496		shift = 60;
3497		break;
3498	case 'p':
3499		shift = 50;
3500		break;
3501	case 't':
3502		shift = 40;
3503		break;
3504	case 'g':
3505		shift = 30;
3506		break;
3507	case 'm':
3508		shift = 20;
3509		break;
3510	case 'k':
3511		shift = 10;
3512		break;
3513	case 'b':
3514	case '\0': /* No unit. */
3515		*num = number;
3516		return (0);
3517	default:
3518		/* Unrecognized unit. */
3519		return (-1);
3520	}
3521
3522	if ((number << shift) >> shift != number) {
3523		/* Overflow */
3524		return (-1);
3525	}
3526	*num = number << shift;
3527	return (0);
3528}
3529
3530
3531/*
3532 * This routine could be used in the future to load default and/or saved
3533 * mode page parameters for a particuar lun.
3534 */
3535static int
3536ctl_init_page_index(struct ctl_lun *lun)
3537{
3538	int i;
3539	struct ctl_page_index *page_index;
3540	const char *value;
3541	uint64_t ival;
3542
3543	memcpy(&lun->mode_pages.index, page_index_template,
3544	       sizeof(page_index_template));
3545
3546	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3547
3548		page_index = &lun->mode_pages.index[i];
3549		/*
3550		 * If this is a disk-only mode page, there's no point in
3551		 * setting it up.  For some pages, we have to have some
3552		 * basic information about the disk in order to calculate the
3553		 * mode page data.
3554		 */
3555		if ((lun->be_lun->lun_type != T_DIRECT)
3556		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3557			continue;
3558
3559		switch (page_index->page_code & SMPH_PC_MASK) {
3560		case SMS_RW_ERROR_RECOVERY_PAGE: {
3561			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3562				panic("subpage is incorrect!");
3563			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3564			       &rw_er_page_default,
3565			       sizeof(rw_er_page_default));
3566			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3567			       &rw_er_page_changeable,
3568			       sizeof(rw_er_page_changeable));
3569			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3570			       &rw_er_page_default,
3571			       sizeof(rw_er_page_default));
3572			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3573			       &rw_er_page_default,
3574			       sizeof(rw_er_page_default));
3575			page_index->page_data =
3576				(uint8_t *)lun->mode_pages.rw_er_page;
3577			break;
3578		}
3579		case SMS_FORMAT_DEVICE_PAGE: {
3580			struct scsi_format_page *format_page;
3581
3582			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3583				panic("subpage is incorrect!");
3584
3585			/*
3586			 * Sectors per track are set above.  Bytes per
3587			 * sector need to be set here on a per-LUN basis.
3588			 */
3589			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3590			       &format_page_default,
3591			       sizeof(format_page_default));
3592			memcpy(&lun->mode_pages.format_page[
3593			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3594			       sizeof(format_page_changeable));
3595			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3596			       &format_page_default,
3597			       sizeof(format_page_default));
3598			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3599			       &format_page_default,
3600			       sizeof(format_page_default));
3601
3602			format_page = &lun->mode_pages.format_page[
3603				CTL_PAGE_CURRENT];
3604			scsi_ulto2b(lun->be_lun->blocksize,
3605				    format_page->bytes_per_sector);
3606
3607			format_page = &lun->mode_pages.format_page[
3608				CTL_PAGE_DEFAULT];
3609			scsi_ulto2b(lun->be_lun->blocksize,
3610				    format_page->bytes_per_sector);
3611
3612			format_page = &lun->mode_pages.format_page[
3613				CTL_PAGE_SAVED];
3614			scsi_ulto2b(lun->be_lun->blocksize,
3615				    format_page->bytes_per_sector);
3616
3617			page_index->page_data =
3618				(uint8_t *)lun->mode_pages.format_page;
3619			break;
3620		}
3621		case SMS_RIGID_DISK_PAGE: {
3622			struct scsi_rigid_disk_page *rigid_disk_page;
3623			uint32_t sectors_per_cylinder;
3624			uint64_t cylinders;
3625#ifndef	__XSCALE__
3626			int shift;
3627#endif /* !__XSCALE__ */
3628
3629			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3630				panic("invalid subpage value %d",
3631				      page_index->subpage);
3632
3633			/*
3634			 * Rotation rate and sectors per track are set
3635			 * above.  We calculate the cylinders here based on
3636			 * capacity.  Due to the number of heads and
3637			 * sectors per track we're using, smaller arrays
3638			 * may turn out to have 0 cylinders.  Linux and
3639			 * FreeBSD don't pay attention to these mode pages
3640			 * to figure out capacity, but Solaris does.  It
3641			 * seems to deal with 0 cylinders just fine, and
3642			 * works out a fake geometry based on the capacity.
3643			 */
3644			memcpy(&lun->mode_pages.rigid_disk_page[
3645			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3646			       sizeof(rigid_disk_page_default));
3647			memcpy(&lun->mode_pages.rigid_disk_page[
3648			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3649			       sizeof(rigid_disk_page_changeable));
3650
3651			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3652				CTL_DEFAULT_HEADS;
3653
3654			/*
3655			 * The divide method here will be more accurate,
3656			 * probably, but results in floating point being
3657			 * used in the kernel on i386 (__udivdi3()).  On the
3658			 * XScale, though, __udivdi3() is implemented in
3659			 * software.
3660			 *
3661			 * The shift method for cylinder calculation is
3662			 * accurate if sectors_per_cylinder is a power of
3663			 * 2.  Otherwise it might be slightly off -- you
3664			 * might have a bit of a truncation problem.
3665			 */
3666#ifdef	__XSCALE__
3667			cylinders = (lun->be_lun->maxlba + 1) /
3668				sectors_per_cylinder;
3669#else
3670			for (shift = 31; shift > 0; shift--) {
3671				if (sectors_per_cylinder & (1 << shift))
3672					break;
3673			}
3674			cylinders = (lun->be_lun->maxlba + 1) >> shift;
3675#endif
3676
3677			/*
3678			 * We've basically got 3 bytes, or 24 bits for the
3679			 * cylinder size in the mode page.  If we're over,
3680			 * just round down to 2^24.
3681			 */
3682			if (cylinders > 0xffffff)
3683				cylinders = 0xffffff;
3684
3685			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3686				CTL_PAGE_DEFAULT];
3687			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3688
3689			if ((value = ctl_get_opt(&lun->be_lun->options,
3690			    "rpm")) != NULL) {
3691				scsi_ulto2b(strtol(value, NULL, 0),
3692				     rigid_disk_page->rotation_rate);
3693			}
3694
3695			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
3696			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
3697			       sizeof(rigid_disk_page_default));
3698			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
3699			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
3700			       sizeof(rigid_disk_page_default));
3701
3702			page_index->page_data =
3703				(uint8_t *)lun->mode_pages.rigid_disk_page;
3704			break;
3705		}
3706		case SMS_CACHING_PAGE: {
3707			struct scsi_caching_page *caching_page;
3708
3709			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3710				panic("invalid subpage value %d",
3711				      page_index->subpage);
3712			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
3713			       &caching_page_default,
3714			       sizeof(caching_page_default));
3715			memcpy(&lun->mode_pages.caching_page[
3716			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
3717			       sizeof(caching_page_changeable));
3718			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
3719			       &caching_page_default,
3720			       sizeof(caching_page_default));
3721			caching_page = &lun->mode_pages.caching_page[
3722			    CTL_PAGE_SAVED];
3723			value = ctl_get_opt(&lun->be_lun->options, "writecache");
3724			if (value != NULL && strcmp(value, "off") == 0)
3725				caching_page->flags1 &= ~SCP_WCE;
3726			value = ctl_get_opt(&lun->be_lun->options, "readcache");
3727			if (value != NULL && strcmp(value, "off") == 0)
3728				caching_page->flags1 |= SCP_RCD;
3729			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
3730			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
3731			       sizeof(caching_page_default));
3732			page_index->page_data =
3733				(uint8_t *)lun->mode_pages.caching_page;
3734			break;
3735		}
3736		case SMS_CONTROL_MODE_PAGE: {
3737			struct scsi_control_page *control_page;
3738
3739			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3740				panic("invalid subpage value %d",
3741				      page_index->subpage);
3742
3743			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
3744			       &control_page_default,
3745			       sizeof(control_page_default));
3746			memcpy(&lun->mode_pages.control_page[
3747			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
3748			       sizeof(control_page_changeable));
3749			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
3750			       &control_page_default,
3751			       sizeof(control_page_default));
3752			control_page = &lun->mode_pages.control_page[
3753			    CTL_PAGE_SAVED];
3754			value = ctl_get_opt(&lun->be_lun->options, "reordering");
3755			if (value != NULL && strcmp(value, "unrestricted") == 0) {
3756				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
3757				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
3758			}
3759			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
3760			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
3761			       sizeof(control_page_default));
3762			page_index->page_data =
3763				(uint8_t *)lun->mode_pages.control_page;
3764			break;
3765
3766		}
3767		case SMS_INFO_EXCEPTIONS_PAGE: {
3768			switch (page_index->subpage) {
3769			case SMS_SUBPAGE_PAGE_0:
3770				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
3771				       &ie_page_default,
3772				       sizeof(ie_page_default));
3773				memcpy(&lun->mode_pages.ie_page[
3774				       CTL_PAGE_CHANGEABLE], &ie_page_changeable,
3775				       sizeof(ie_page_changeable));
3776				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
3777				       &ie_page_default,
3778				       sizeof(ie_page_default));
3779				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
3780				       &ie_page_default,
3781				       sizeof(ie_page_default));
3782				page_index->page_data =
3783					(uint8_t *)lun->mode_pages.ie_page;
3784				break;
3785			case 0x02: {
3786				struct ctl_logical_block_provisioning_page *page;
3787
3788				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
3789				       &lbp_page_default,
3790				       sizeof(lbp_page_default));
3791				memcpy(&lun->mode_pages.lbp_page[
3792				       CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
3793				       sizeof(lbp_page_changeable));
3794				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
3795				       &lbp_page_default,
3796				       sizeof(lbp_page_default));
3797				page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
3798				value = ctl_get_opt(&lun->be_lun->options,
3799				    "avail-threshold");
3800				if (value != NULL &&
3801				    ctl_expand_number(value, &ival) == 0) {
3802					page->descr[0].flags |= SLBPPD_ENABLED |
3803					    SLBPPD_ARMING_DEC;
3804					if (lun->be_lun->blocksize)
3805						ival /= lun->be_lun->blocksize;
3806					else
3807						ival /= 512;
3808					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
3809					    page->descr[0].count);
3810				}
3811				value = ctl_get_opt(&lun->be_lun->options,
3812				    "used-threshold");
3813				if (value != NULL &&
3814				    ctl_expand_number(value, &ival) == 0) {
3815					page->descr[1].flags |= SLBPPD_ENABLED |
3816					    SLBPPD_ARMING_INC;
3817					if (lun->be_lun->blocksize)
3818						ival /= lun->be_lun->blocksize;
3819					else
3820						ival /= 512;
3821					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
3822					    page->descr[1].count);
3823				}
3824				value = ctl_get_opt(&lun->be_lun->options,
3825				    "pool-avail-threshold");
3826				if (value != NULL &&
3827				    ctl_expand_number(value, &ival) == 0) {
3828					page->descr[2].flags |= SLBPPD_ENABLED |
3829					    SLBPPD_ARMING_DEC;
3830					if (lun->be_lun->blocksize)
3831						ival /= lun->be_lun->blocksize;
3832					else
3833						ival /= 512;
3834					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
3835					    page->descr[2].count);
3836				}
3837				value = ctl_get_opt(&lun->be_lun->options,
3838				    "pool-used-threshold");
3839				if (value != NULL &&
3840				    ctl_expand_number(value, &ival) == 0) {
3841					page->descr[3].flags |= SLBPPD_ENABLED |
3842					    SLBPPD_ARMING_INC;
3843					if (lun->be_lun->blocksize)
3844						ival /= lun->be_lun->blocksize;
3845					else
3846						ival /= 512;
3847					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
3848					    page->descr[3].count);
3849				}
3850				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
3851				       &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
3852				       sizeof(lbp_page_default));
3853				page_index->page_data =
3854					(uint8_t *)lun->mode_pages.lbp_page;
3855			}}
3856			break;
3857		}
3858		case SMS_VENDOR_SPECIFIC_PAGE:{
3859			switch (page_index->subpage) {
3860			case DBGCNF_SUBPAGE_CODE: {
3861				struct copan_debugconf_subpage *current_page,
3862							       *saved_page;
3863
3864				memcpy(&lun->mode_pages.debugconf_subpage[
3865				       CTL_PAGE_CURRENT],
3866				       &debugconf_page_default,
3867				       sizeof(debugconf_page_default));
3868				memcpy(&lun->mode_pages.debugconf_subpage[
3869				       CTL_PAGE_CHANGEABLE],
3870				       &debugconf_page_changeable,
3871				       sizeof(debugconf_page_changeable));
3872				memcpy(&lun->mode_pages.debugconf_subpage[
3873				       CTL_PAGE_DEFAULT],
3874				       &debugconf_page_default,
3875				       sizeof(debugconf_page_default));
3876				memcpy(&lun->mode_pages.debugconf_subpage[
3877				       CTL_PAGE_SAVED],
3878				       &debugconf_page_default,
3879				       sizeof(debugconf_page_default));
3880				page_index->page_data =
3881					(uint8_t *)lun->mode_pages.debugconf_subpage;
3882
3883				current_page = (struct copan_debugconf_subpage *)
3884					(page_index->page_data +
3885					 (page_index->page_len *
3886					  CTL_PAGE_CURRENT));
3887				saved_page = (struct copan_debugconf_subpage *)
3888					(page_index->page_data +
3889					 (page_index->page_len *
3890					  CTL_PAGE_SAVED));
3891				break;
3892			}
3893			default:
3894				panic("invalid subpage value %d",
3895				      page_index->subpage);
3896				break;
3897			}
3898   			break;
3899		}
3900		default:
3901			panic("invalid page value %d",
3902			      page_index->page_code & SMPH_PC_MASK);
3903			break;
3904    	}
3905	}
3906
3907	return (CTL_RETVAL_COMPLETE);
3908}
3909
3910static int
3911ctl_init_log_page_index(struct ctl_lun *lun)
3912{
3913	struct ctl_page_index *page_index;
3914	int i, j, k, prev;
3915
3916	memcpy(&lun->log_pages.index, log_page_index_template,
3917	       sizeof(log_page_index_template));
3918
3919	prev = -1;
3920	for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
3921
3922		page_index = &lun->log_pages.index[i];
3923		/*
3924		 * If this is a disk-only mode page, there's no point in
3925		 * setting it up.  For some pages, we have to have some
3926		 * basic information about the disk in order to calculate the
3927		 * mode page data.
3928		 */
3929		if ((lun->be_lun->lun_type != T_DIRECT)
3930		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3931			continue;
3932
3933		if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
3934		     lun->backend->lun_attr == NULL)
3935			continue;
3936
3937		if (page_index->page_code != prev) {
3938			lun->log_pages.pages_page[j] = page_index->page_code;
3939			prev = page_index->page_code;
3940			j++;
3941		}
3942		lun->log_pages.subpages_page[k*2] = page_index->page_code;
3943		lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
3944		k++;
3945	}
3946	lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
3947	lun->log_pages.index[0].page_len = j;
3948	lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
3949	lun->log_pages.index[1].page_len = k * 2;
3950	lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
3951	lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
3952	lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
3953	lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
3954
3955	return (CTL_RETVAL_COMPLETE);
3956}
3957
3958static int
3959hex2bin(const char *str, uint8_t *buf, int buf_size)
3960{
3961	int i;
3962	u_char c;
3963
3964	memset(buf, 0, buf_size);
3965	while (isspace(str[0]))
3966		str++;
3967	if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
3968		str += 2;
3969	buf_size *= 2;
3970	for (i = 0; str[i] != 0 && i < buf_size; i++) {
3971		c = str[i];
3972		if (isdigit(c))
3973			c -= '0';
3974		else if (isalpha(c))
3975			c -= isupper(c) ? 'A' - 10 : 'a' - 10;
3976		else
3977			break;
3978		if (c >= 16)
3979			break;
3980		if ((i & 1) == 0)
3981			buf[i / 2] |= (c << 4);
3982		else
3983			buf[i / 2] |= c;
3984	}
3985	return ((i + 1) / 2);
3986}
3987
3988/*
3989 * LUN allocation.
3990 *
3991 * Requirements:
3992 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
3993 *   wants us to allocate the LUN and he can block.
3994 * - ctl_softc is always set
3995 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
3996 *
3997 * Returns 0 for success, non-zero (errno) for failure.
3998 */
3999static int
4000ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4001	      struct ctl_be_lun *const be_lun)
4002{
4003	struct ctl_lun *nlun, *lun;
4004	struct scsi_vpd_id_descriptor *desc;
4005	struct scsi_vpd_id_t10 *t10id;
4006	const char *eui, *naa, *scsiname, *vendor, *value;
4007	int lun_number, i, lun_malloced;
4008	int devidlen, idlen1, idlen2 = 0, len;
4009
4010	if (be_lun == NULL)
4011		return (EINVAL);
4012
4013	/*
4014	 * We currently only support Direct Access or Processor LUN types.
4015	 */
4016	switch (be_lun->lun_type) {
4017	case T_DIRECT:
4018		break;
4019	case T_PROCESSOR:
4020		break;
4021	case T_SEQUENTIAL:
4022	case T_CHANGER:
4023	default:
4024		be_lun->lun_config_status(be_lun->be_lun,
4025					  CTL_LUN_CONFIG_FAILURE);
4026		break;
4027	}
4028	if (ctl_lun == NULL) {
4029		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4030		lun_malloced = 1;
4031	} else {
4032		lun_malloced = 0;
4033		lun = ctl_lun;
4034	}
4035
4036	memset(lun, 0, sizeof(*lun));
4037	if (lun_malloced)
4038		lun->flags = CTL_LUN_MALLOCED;
4039
4040	/* Generate LUN ID. */
4041	devidlen = max(CTL_DEVID_MIN_LEN,
4042	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4043	idlen1 = sizeof(*t10id) + devidlen;
4044	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4045	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4046	if (scsiname != NULL) {
4047		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4048		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4049	}
4050	eui = ctl_get_opt(&be_lun->options, "eui");
4051	if (eui != NULL) {
4052		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4053	}
4054	naa = ctl_get_opt(&be_lun->options, "naa");
4055	if (naa != NULL) {
4056		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4057	}
4058	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4059	    M_CTL, M_WAITOK | M_ZERO);
4060	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4061	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4062	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4063	desc->length = idlen1;
4064	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4065	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4066	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4067		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4068	} else {
4069		strncpy(t10id->vendor, vendor,
4070		    min(sizeof(t10id->vendor), strlen(vendor)));
4071	}
4072	strncpy((char *)t10id->vendor_spec_id,
4073	    (char *)be_lun->device_id, devidlen);
4074	if (scsiname != NULL) {
4075		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4076		    desc->length);
4077		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4078		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4079		    SVPD_ID_TYPE_SCSI_NAME;
4080		desc->length = idlen2;
4081		strlcpy(desc->identifier, scsiname, idlen2);
4082	}
4083	if (eui != NULL) {
4084		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4085		    desc->length);
4086		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4087		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4088		    SVPD_ID_TYPE_EUI64;
4089		desc->length = hex2bin(eui, desc->identifier, 16);
4090		desc->length = desc->length > 12 ? 16 :
4091		    (desc->length > 8 ? 12 : 8);
4092		len -= 16 - desc->length;
4093	}
4094	if (naa != NULL) {
4095		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4096		    desc->length);
4097		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4098		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4099		    SVPD_ID_TYPE_NAA;
4100		desc->length = hex2bin(naa, desc->identifier, 16);
4101		desc->length = desc->length > 8 ? 16 : 8;
4102		len -= 16 - desc->length;
4103	}
4104	lun->lun_devid->len = len;
4105
4106	mtx_lock(&ctl_softc->ctl_lock);
4107	/*
4108	 * See if the caller requested a particular LUN number.  If so, see
4109	 * if it is available.  Otherwise, allocate the first available LUN.
4110	 */
4111	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4112		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4113		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4114			mtx_unlock(&ctl_softc->ctl_lock);
4115			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4116				printf("ctl: requested LUN ID %d is higher "
4117				       "than CTL_MAX_LUNS - 1 (%d)\n",
4118				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4119			} else {
4120				/*
4121				 * XXX KDM return an error, or just assign
4122				 * another LUN ID in this case??
4123				 */
4124				printf("ctl: requested LUN ID %d is already "
4125				       "in use\n", be_lun->req_lun_id);
4126			}
4127			if (lun->flags & CTL_LUN_MALLOCED)
4128				free(lun, M_CTL);
4129			be_lun->lun_config_status(be_lun->be_lun,
4130						  CTL_LUN_CONFIG_FAILURE);
4131			return (ENOSPC);
4132		}
4133		lun_number = be_lun->req_lun_id;
4134	} else {
4135		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4136		if (lun_number == -1) {
4137			mtx_unlock(&ctl_softc->ctl_lock);
4138			printf("ctl: can't allocate LUN, out of LUNs\n");
4139			if (lun->flags & CTL_LUN_MALLOCED)
4140				free(lun, M_CTL);
4141			be_lun->lun_config_status(be_lun->be_lun,
4142						  CTL_LUN_CONFIG_FAILURE);
4143			return (ENOSPC);
4144		}
4145	}
4146	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4147
4148	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4149	lun->lun = lun_number;
4150	lun->be_lun = be_lun;
4151	/*
4152	 * The processor LUN is always enabled.  Disk LUNs come on line
4153	 * disabled, and must be enabled by the backend.
4154	 */
4155	lun->flags |= CTL_LUN_DISABLED;
4156	lun->backend = be_lun->be;
4157	be_lun->ctl_lun = lun;
4158	be_lun->lun_id = lun_number;
4159	atomic_add_int(&be_lun->be->num_luns, 1);
4160	if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4161		lun->flags |= CTL_LUN_OFFLINE;
4162
4163	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4164		lun->flags |= CTL_LUN_STOPPED;
4165
4166	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4167		lun->flags |= CTL_LUN_INOPERABLE;
4168
4169	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4170		lun->flags |= CTL_LUN_PRIMARY_SC;
4171
4172	value = ctl_get_opt(&be_lun->options, "readonly");
4173	if (value != NULL && strcmp(value, "on") == 0)
4174		lun->flags |= CTL_LUN_READONLY;
4175
4176	lun->serseq = CTL_LUN_SERSEQ_OFF;
4177	if (be_lun->flags & CTL_LUN_FLAG_SERSEQ_READ)
4178		lun->serseq = CTL_LUN_SERSEQ_READ;
4179	value = ctl_get_opt(&be_lun->options, "serseq");
4180	if (value != NULL && strcmp(value, "on") == 0)
4181		lun->serseq = CTL_LUN_SERSEQ_ON;
4182	else if (value != NULL && strcmp(value, "read") == 0)
4183		lun->serseq = CTL_LUN_SERSEQ_READ;
4184	else if (value != NULL && strcmp(value, "off") == 0)
4185		lun->serseq = CTL_LUN_SERSEQ_OFF;
4186
4187	lun->ctl_softc = ctl_softc;
4188#ifdef CTL_TIME_IO
4189	lun->last_busy = getsbinuptime();
4190#endif
4191	TAILQ_INIT(&lun->ooa_queue);
4192	TAILQ_INIT(&lun->blocked_queue);
4193	STAILQ_INIT(&lun->error_list);
4194	ctl_tpc_lun_init(lun);
4195
4196	/*
4197	 * Initialize the mode and log page index.
4198	 */
4199	ctl_init_page_index(lun);
4200	ctl_init_log_page_index(lun);
4201
4202	/*
4203	 * Now, before we insert this lun on the lun list, set the lun
4204	 * inventory changed UA for all other luns.
4205	 */
4206	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4207		mtx_lock(&nlun->lun_lock);
4208		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4209		mtx_unlock(&nlun->lun_lock);
4210	}
4211
4212	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4213
4214	ctl_softc->ctl_luns[lun_number] = lun;
4215
4216	ctl_softc->num_luns++;
4217
4218	/* Setup statistics gathering */
4219	lun->stats.device_type = be_lun->lun_type;
4220	lun->stats.lun_number = lun_number;
4221	if (lun->stats.device_type == T_DIRECT)
4222		lun->stats.blocksize = be_lun->blocksize;
4223	else
4224		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4225	for (i = 0;i < CTL_MAX_PORTS;i++)
4226		lun->stats.ports[i].targ_port = i;
4227
4228	mtx_unlock(&ctl_softc->ctl_lock);
4229
4230	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4231	return (0);
4232}
4233
4234/*
4235 * Delete a LUN.
4236 * Assumptions:
4237 * - LUN has already been marked invalid and any pending I/O has been taken
4238 *   care of.
4239 */
4240static int
4241ctl_free_lun(struct ctl_lun *lun)
4242{
4243	struct ctl_softc *softc;
4244	struct ctl_lun *nlun;
4245	int i;
4246
4247	softc = lun->ctl_softc;
4248
4249	mtx_assert(&softc->ctl_lock, MA_OWNED);
4250
4251	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4252
4253	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4254
4255	softc->ctl_luns[lun->lun] = NULL;
4256
4257	if (!TAILQ_EMPTY(&lun->ooa_queue))
4258		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4259
4260	softc->num_luns--;
4261
4262	/*
4263	 * Tell the backend to free resources, if this LUN has a backend.
4264	 */
4265	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4266	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4267
4268	ctl_tpc_lun_shutdown(lun);
4269	mtx_destroy(&lun->lun_lock);
4270	free(lun->lun_devid, M_CTL);
4271	for (i = 0; i < CTL_MAX_PORTS; i++)
4272		free(lun->pending_ua[i], M_CTL);
4273	for (i = 0; i < 2 * CTL_MAX_PORTS; i++)
4274		free(lun->pr_keys[i], M_CTL);
4275	free(lun->write_buffer, M_CTL);
4276	if (lun->flags & CTL_LUN_MALLOCED)
4277		free(lun, M_CTL);
4278
4279	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4280		mtx_lock(&nlun->lun_lock);
4281		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4282		mtx_unlock(&nlun->lun_lock);
4283	}
4284
4285	return (0);
4286}
4287
4288static void
4289ctl_create_lun(struct ctl_be_lun *be_lun)
4290{
4291	struct ctl_softc *softc;
4292
4293	softc = control_softc;
4294
4295	/*
4296	 * ctl_alloc_lun() should handle all potential failure cases.
4297	 */
4298	ctl_alloc_lun(softc, NULL, be_lun);
4299}
4300
4301int
4302ctl_add_lun(struct ctl_be_lun *be_lun)
4303{
4304	struct ctl_softc *softc = control_softc;
4305
4306	mtx_lock(&softc->ctl_lock);
4307	STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4308	mtx_unlock(&softc->ctl_lock);
4309	wakeup(&softc->pending_lun_queue);
4310
4311	return (0);
4312}
4313
4314int
4315ctl_enable_lun(struct ctl_be_lun *be_lun)
4316{
4317	struct ctl_softc *softc;
4318	struct ctl_port *port, *nport;
4319	struct ctl_lun *lun;
4320	int retval;
4321
4322	lun = (struct ctl_lun *)be_lun->ctl_lun;
4323	softc = lun->ctl_softc;
4324
4325	mtx_lock(&softc->ctl_lock);
4326	mtx_lock(&lun->lun_lock);
4327	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4328		/*
4329		 * eh?  Why did we get called if the LUN is already
4330		 * enabled?
4331		 */
4332		mtx_unlock(&lun->lun_lock);
4333		mtx_unlock(&softc->ctl_lock);
4334		return (0);
4335	}
4336	lun->flags &= ~CTL_LUN_DISABLED;
4337	mtx_unlock(&lun->lun_lock);
4338
4339	for (port = STAILQ_FIRST(&softc->port_list); port != NULL; port = nport) {
4340		nport = STAILQ_NEXT(port, links);
4341		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4342		    port->lun_map != NULL || port->lun_enable == NULL)
4343			continue;
4344
4345		/*
4346		 * Drop the lock while we call the FETD's enable routine.
4347		 * This can lead to a callback into CTL (at least in the
4348		 * case of the internal initiator frontend.
4349		 */
4350		mtx_unlock(&softc->ctl_lock);
4351		retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4352		mtx_lock(&softc->ctl_lock);
4353		if (retval != 0) {
4354			printf("%s: FETD %s port %d returned error "
4355			       "%d for lun_enable on lun %jd\n",
4356			       __func__, port->port_name, port->targ_port,
4357			       retval, (intmax_t)lun->lun);
4358		}
4359	}
4360
4361	mtx_unlock(&softc->ctl_lock);
4362
4363	return (0);
4364}
4365
4366int
4367ctl_disable_lun(struct ctl_be_lun *be_lun)
4368{
4369	struct ctl_softc *softc;
4370	struct ctl_port *port;
4371	struct ctl_lun *lun;
4372	int retval;
4373
4374	lun = (struct ctl_lun *)be_lun->ctl_lun;
4375	softc = lun->ctl_softc;
4376
4377	mtx_lock(&softc->ctl_lock);
4378	mtx_lock(&lun->lun_lock);
4379	if (lun->flags & CTL_LUN_DISABLED) {
4380		mtx_unlock(&lun->lun_lock);
4381		mtx_unlock(&softc->ctl_lock);
4382		return (0);
4383	}
4384	lun->flags |= CTL_LUN_DISABLED;
4385	mtx_unlock(&lun->lun_lock);
4386
4387	STAILQ_FOREACH(port, &softc->port_list, links) {
4388		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4389		    port->lun_map != NULL || port->lun_disable == NULL)
4390			continue;
4391
4392		/*
4393		 * Drop the lock before we call the frontend's disable
4394		 * routine, to avoid lock order reversals.
4395		 *
4396		 * XXX KDM what happens if the frontend list changes while
4397		 * we're traversing it?  It's unlikely, but should be handled.
4398		 */
4399		mtx_unlock(&softc->ctl_lock);
4400		retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4401		mtx_lock(&softc->ctl_lock);
4402		if (retval != 0) {
4403			printf("%s: FETD %s port %d returned error "
4404			       "%d for lun_disable on lun %jd\n",
4405			       __func__, port->port_name, port->targ_port,
4406			       retval, (intmax_t)lun->lun);
4407		}
4408	}
4409
4410	mtx_unlock(&softc->ctl_lock);
4411
4412	return (0);
4413}
4414
4415int
4416ctl_start_lun(struct ctl_be_lun *be_lun)
4417{
4418	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4419
4420	mtx_lock(&lun->lun_lock);
4421	lun->flags &= ~CTL_LUN_STOPPED;
4422	mtx_unlock(&lun->lun_lock);
4423	return (0);
4424}
4425
4426int
4427ctl_stop_lun(struct ctl_be_lun *be_lun)
4428{
4429	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4430
4431	mtx_lock(&lun->lun_lock);
4432	lun->flags |= CTL_LUN_STOPPED;
4433	mtx_unlock(&lun->lun_lock);
4434	return (0);
4435}
4436
4437int
4438ctl_lun_offline(struct ctl_be_lun *be_lun)
4439{
4440	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4441
4442	mtx_lock(&lun->lun_lock);
4443	lun->flags |= CTL_LUN_OFFLINE;
4444	mtx_unlock(&lun->lun_lock);
4445	return (0);
4446}
4447
4448int
4449ctl_lun_online(struct ctl_be_lun *be_lun)
4450{
4451	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4452
4453	mtx_lock(&lun->lun_lock);
4454	lun->flags &= ~CTL_LUN_OFFLINE;
4455	mtx_unlock(&lun->lun_lock);
4456	return (0);
4457}
4458
4459int
4460ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4461{
4462	struct ctl_softc *softc;
4463	struct ctl_lun *lun;
4464
4465	lun = (struct ctl_lun *)be_lun->ctl_lun;
4466	softc = lun->ctl_softc;
4467
4468	mtx_lock(&lun->lun_lock);
4469
4470	/*
4471	 * The LUN needs to be disabled before it can be marked invalid.
4472	 */
4473	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4474		mtx_unlock(&lun->lun_lock);
4475		return (-1);
4476	}
4477	/*
4478	 * Mark the LUN invalid.
4479	 */
4480	lun->flags |= CTL_LUN_INVALID;
4481
4482	/*
4483	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4484	 * If we have something in the OOA queue, we'll free it when the
4485	 * last I/O completes.
4486	 */
4487	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4488		mtx_unlock(&lun->lun_lock);
4489		mtx_lock(&softc->ctl_lock);
4490		ctl_free_lun(lun);
4491		mtx_unlock(&softc->ctl_lock);
4492	} else
4493		mtx_unlock(&lun->lun_lock);
4494
4495	return (0);
4496}
4497
4498int
4499ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4500{
4501	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4502
4503	mtx_lock(&lun->lun_lock);
4504	lun->flags |= CTL_LUN_INOPERABLE;
4505	mtx_unlock(&lun->lun_lock);
4506	return (0);
4507}
4508
4509int
4510ctl_lun_operable(struct ctl_be_lun *be_lun)
4511{
4512	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4513
4514	mtx_lock(&lun->lun_lock);
4515	lun->flags &= ~CTL_LUN_INOPERABLE;
4516	mtx_unlock(&lun->lun_lock);
4517	return (0);
4518}
4519
4520void
4521ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4522{
4523	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4524
4525	mtx_lock(&lun->lun_lock);
4526	ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGED);
4527	mtx_unlock(&lun->lun_lock);
4528}
4529
4530/*
4531 * Backend "memory move is complete" callback for requests that never
4532 * make it down to say RAIDCore's configuration code.
4533 */
4534int
4535ctl_config_move_done(union ctl_io *io)
4536{
4537	int retval;
4538
4539	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4540	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4541	    ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
4542
4543	if ((io->io_hdr.port_status != 0) &&
4544	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4545	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4546		/*
4547		 * For hardware error sense keys, the sense key
4548		 * specific value is defined to be a retry count,
4549		 * but we use it to pass back an internal FETD
4550		 * error code.  XXX KDM  Hopefully the FETD is only
4551		 * using 16 bits for an error code, since that's
4552		 * all the space we have in the sks field.
4553		 */
4554		ctl_set_internal_failure(&io->scsiio,
4555					 /*sks_valid*/ 1,
4556					 /*retry_count*/
4557					 io->io_hdr.port_status);
4558	}
4559
4560	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
4561	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4562	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
4563	    ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4564		/*
4565		 * XXX KDM just assuming a single pointer here, and not a
4566		 * S/G list.  If we start using S/G lists for config data,
4567		 * we'll need to know how to clean them up here as well.
4568		 */
4569		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4570			free(io->scsiio.kern_data_ptr, M_CTL);
4571		ctl_done(io);
4572		retval = CTL_RETVAL_COMPLETE;
4573	} else {
4574		/*
4575		 * XXX KDM now we need to continue data movement.  Some
4576		 * options:
4577		 * - call ctl_scsiio() again?  We don't do this for data
4578		 *   writes, because for those at least we know ahead of
4579		 *   time where the write will go and how long it is.  For
4580		 *   config writes, though, that information is largely
4581		 *   contained within the write itself, thus we need to
4582		 *   parse out the data again.
4583		 *
4584		 * - Call some other function once the data is in?
4585		 */
4586		if (ctl_debug & CTL_DEBUG_CDB_DATA)
4587			ctl_data_print(io);
4588
4589		/*
4590		 * XXX KDM call ctl_scsiio() again for now, and check flag
4591		 * bits to see whether we're allocated or not.
4592		 */
4593		retval = ctl_scsiio(&io->scsiio);
4594	}
4595	return (retval);
4596}
4597
4598/*
4599 * This gets called by a backend driver when it is done with a
4600 * data_submit method.
4601 */
4602void
4603ctl_data_submit_done(union ctl_io *io)
4604{
4605	/*
4606	 * If the IO_CONT flag is set, we need to call the supplied
4607	 * function to continue processing the I/O, instead of completing
4608	 * the I/O just yet.
4609	 *
4610	 * If there is an error, though, we don't want to keep processing.
4611	 * Instead, just send status back to the initiator.
4612	 */
4613	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4614	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4615	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4616	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4617		io->scsiio.io_cont(io);
4618		return;
4619	}
4620	ctl_done(io);
4621}
4622
4623/*
4624 * This gets called by a backend driver when it is done with a
4625 * configuration write.
4626 */
4627void
4628ctl_config_write_done(union ctl_io *io)
4629{
4630	uint8_t *buf;
4631
4632	/*
4633	 * If the IO_CONT flag is set, we need to call the supplied
4634	 * function to continue processing the I/O, instead of completing
4635	 * the I/O just yet.
4636	 *
4637	 * If there is an error, though, we don't want to keep processing.
4638	 * Instead, just send status back to the initiator.
4639	 */
4640	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4641	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4642	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4643	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4644		io->scsiio.io_cont(io);
4645		return;
4646	}
4647	/*
4648	 * Since a configuration write can be done for commands that actually
4649	 * have data allocated, like write buffer, and commands that have
4650	 * no data, like start/stop unit, we need to check here.
4651	 */
4652	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4653		buf = io->scsiio.kern_data_ptr;
4654	else
4655		buf = NULL;
4656	ctl_done(io);
4657	if (buf)
4658		free(buf, M_CTL);
4659}
4660
4661void
4662ctl_config_read_done(union ctl_io *io)
4663{
4664	uint8_t *buf;
4665
4666	/*
4667	 * If there is some error -- we are done, skip data transfer.
4668	 */
4669	if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
4670	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4671	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
4672		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4673			buf = io->scsiio.kern_data_ptr;
4674		else
4675			buf = NULL;
4676		ctl_done(io);
4677		if (buf)
4678			free(buf, M_CTL);
4679		return;
4680	}
4681
4682	/*
4683	 * If the IO_CONT flag is set, we need to call the supplied
4684	 * function to continue processing the I/O, instead of completing
4685	 * the I/O just yet.
4686	 */
4687	if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
4688		io->scsiio.io_cont(io);
4689		return;
4690	}
4691
4692	ctl_datamove(io);
4693}
4694
4695/*
4696 * SCSI release command.
4697 */
4698int
4699ctl_scsi_release(struct ctl_scsiio *ctsio)
4700{
4701	int length, longid, thirdparty_id, resv_id;
4702	struct ctl_lun *lun;
4703	uint32_t residx;
4704
4705	length = 0;
4706	resv_id = 0;
4707
4708	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
4709
4710	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
4711	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4712
4713	switch (ctsio->cdb[0]) {
4714	case RELEASE_10: {
4715		struct scsi_release_10 *cdb;
4716
4717		cdb = (struct scsi_release_10 *)ctsio->cdb;
4718
4719		if (cdb->byte2 & SR10_LONGID)
4720			longid = 1;
4721		else
4722			thirdparty_id = cdb->thirdparty_id;
4723
4724		resv_id = cdb->resv_id;
4725		length = scsi_2btoul(cdb->length);
4726		break;
4727	}
4728	}
4729
4730
4731	/*
4732	 * XXX KDM right now, we only support LUN reservation.  We don't
4733	 * support 3rd party reservations, or extent reservations, which
4734	 * might actually need the parameter list.  If we've gotten this
4735	 * far, we've got a LUN reservation.  Anything else got kicked out
4736	 * above.  So, according to SPC, ignore the length.
4737	 */
4738	length = 0;
4739
4740	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
4741	 && (length > 0)) {
4742		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
4743		ctsio->kern_data_len = length;
4744		ctsio->kern_total_len = length;
4745		ctsio->kern_data_resid = 0;
4746		ctsio->kern_rel_offset = 0;
4747		ctsio->kern_sg_entries = 0;
4748		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
4749		ctsio->be_move_done = ctl_config_move_done;
4750		ctl_datamove((union ctl_io *)ctsio);
4751
4752		return (CTL_RETVAL_COMPLETE);
4753	}
4754
4755	if (length > 0)
4756		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
4757
4758	mtx_lock(&lun->lun_lock);
4759
4760	/*
4761	 * According to SPC, it is not an error for an intiator to attempt
4762	 * to release a reservation on a LUN that isn't reserved, or that
4763	 * is reserved by another initiator.  The reservation can only be
4764	 * released, though, by the initiator who made it or by one of
4765	 * several reset type events.
4766	 */
4767	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
4768			lun->flags &= ~CTL_LUN_RESERVED;
4769
4770	mtx_unlock(&lun->lun_lock);
4771
4772	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
4773		free(ctsio->kern_data_ptr, M_CTL);
4774		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
4775	}
4776
4777	ctl_set_success(ctsio);
4778	ctl_done((union ctl_io *)ctsio);
4779	return (CTL_RETVAL_COMPLETE);
4780}
4781
4782int
4783ctl_scsi_reserve(struct ctl_scsiio *ctsio)
4784{
4785	int extent, thirdparty, longid;
4786	int resv_id, length;
4787	uint64_t thirdparty_id;
4788	struct ctl_lun *lun;
4789	uint32_t residx;
4790
4791	extent = 0;
4792	thirdparty = 0;
4793	longid = 0;
4794	resv_id = 0;
4795	length = 0;
4796	thirdparty_id = 0;
4797
4798	CTL_DEBUG_PRINT(("ctl_reserve\n"));
4799
4800	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
4801	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4802
4803	switch (ctsio->cdb[0]) {
4804	case RESERVE_10: {
4805		struct scsi_reserve_10 *cdb;
4806
4807		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
4808
4809		if (cdb->byte2 & SR10_LONGID)
4810			longid = 1;
4811		else
4812			thirdparty_id = cdb->thirdparty_id;
4813
4814		resv_id = cdb->resv_id;
4815		length = scsi_2btoul(cdb->length);
4816		break;
4817	}
4818	}
4819
4820	/*
4821	 * XXX KDM right now, we only support LUN reservation.  We don't
4822	 * support 3rd party reservations, or extent reservations, which
4823	 * might actually need the parameter list.  If we've gotten this
4824	 * far, we've got a LUN reservation.  Anything else got kicked out
4825	 * above.  So, according to SPC, ignore the length.
4826	 */
4827	length = 0;
4828
4829	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
4830	 && (length > 0)) {
4831		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
4832		ctsio->kern_data_len = length;
4833		ctsio->kern_total_len = length;
4834		ctsio->kern_data_resid = 0;
4835		ctsio->kern_rel_offset = 0;
4836		ctsio->kern_sg_entries = 0;
4837		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
4838		ctsio->be_move_done = ctl_config_move_done;
4839		ctl_datamove((union ctl_io *)ctsio);
4840
4841		return (CTL_RETVAL_COMPLETE);
4842	}
4843
4844	if (length > 0)
4845		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
4846
4847	mtx_lock(&lun->lun_lock);
4848	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
4849		ctl_set_reservation_conflict(ctsio);
4850		goto bailout;
4851	}
4852
4853	lun->flags |= CTL_LUN_RESERVED;
4854	lun->res_idx = residx;
4855
4856	ctl_set_success(ctsio);
4857
4858bailout:
4859	mtx_unlock(&lun->lun_lock);
4860
4861	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
4862		free(ctsio->kern_data_ptr, M_CTL);
4863		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
4864	}
4865
4866	ctl_done((union ctl_io *)ctsio);
4867	return (CTL_RETVAL_COMPLETE);
4868}
4869
4870int
4871ctl_start_stop(struct ctl_scsiio *ctsio)
4872{
4873	struct scsi_start_stop_unit *cdb;
4874	struct ctl_lun *lun;
4875	int retval;
4876
4877	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
4878
4879	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4880	retval = 0;
4881
4882	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
4883
4884	/*
4885	 * XXX KDM
4886	 * We don't support the immediate bit on a stop unit.  In order to
4887	 * do that, we would need to code up a way to know that a stop is
4888	 * pending, and hold off any new commands until it completes, one
4889	 * way or another.  Then we could accept or reject those commands
4890	 * depending on its status.  We would almost need to do the reverse
4891	 * of what we do below for an immediate start -- return the copy of
4892	 * the ctl_io to the FETD with status to send to the host (and to
4893	 * free the copy!) and then free the original I/O once the stop
4894	 * actually completes.  That way, the OOA queue mechanism can work
4895	 * to block commands that shouldn't proceed.  Another alternative
4896	 * would be to put the copy in the queue in place of the original,
4897	 * and return the original back to the caller.  That could be
4898	 * slightly safer..
4899	 */
4900	if ((cdb->byte2 & SSS_IMMED)
4901	 && ((cdb->how & SSS_START) == 0)) {
4902		ctl_set_invalid_field(ctsio,
4903				      /*sks_valid*/ 1,
4904				      /*command*/ 1,
4905				      /*field*/ 1,
4906				      /*bit_valid*/ 1,
4907				      /*bit*/ 0);
4908		ctl_done((union ctl_io *)ctsio);
4909		return (CTL_RETVAL_COMPLETE);
4910	}
4911
4912	if ((lun->flags & CTL_LUN_PR_RESERVED)
4913	 && ((cdb->how & SSS_START)==0)) {
4914		uint32_t residx;
4915
4916		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
4917		if (ctl_get_prkey(lun, residx) == 0
4918		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
4919
4920			ctl_set_reservation_conflict(ctsio);
4921			ctl_done((union ctl_io *)ctsio);
4922			return (CTL_RETVAL_COMPLETE);
4923		}
4924	}
4925
4926	/*
4927	 * If there is no backend on this device, we can't start or stop
4928	 * it.  In theory we shouldn't get any start/stop commands in the
4929	 * first place at this level if the LUN doesn't have a backend.
4930	 * That should get stopped by the command decode code.
4931	 */
4932	if (lun->backend == NULL) {
4933		ctl_set_invalid_opcode(ctsio);
4934		ctl_done((union ctl_io *)ctsio);
4935		return (CTL_RETVAL_COMPLETE);
4936	}
4937
4938	/*
4939	 * XXX KDM Copan-specific offline behavior.
4940	 * Figure out a reasonable way to port this?
4941	 */
4942#ifdef NEEDTOPORT
4943	mtx_lock(&lun->lun_lock);
4944
4945	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
4946	 && (lun->flags & CTL_LUN_OFFLINE)) {
4947		/*
4948		 * If the LUN is offline, and the on/offline bit isn't set,
4949		 * reject the start or stop.  Otherwise, let it through.
4950		 */
4951		mtx_unlock(&lun->lun_lock);
4952		ctl_set_lun_not_ready(ctsio);
4953		ctl_done((union ctl_io *)ctsio);
4954	} else {
4955		mtx_unlock(&lun->lun_lock);
4956#endif /* NEEDTOPORT */
4957		/*
4958		 * This could be a start or a stop when we're online,
4959		 * or a stop/offline or start/online.  A start or stop when
4960		 * we're offline is covered in the case above.
4961		 */
4962		/*
4963		 * In the non-immediate case, we send the request to
4964		 * the backend and return status to the user when
4965		 * it is done.
4966		 *
4967		 * In the immediate case, we allocate a new ctl_io
4968		 * to hold a copy of the request, and send that to
4969		 * the backend.  We then set good status on the
4970		 * user's request and return it immediately.
4971		 */
4972		if (cdb->byte2 & SSS_IMMED) {
4973			union ctl_io *new_io;
4974
4975			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
4976			ctl_copy_io((union ctl_io *)ctsio, new_io);
4977			retval = lun->backend->config_write(new_io);
4978			ctl_set_success(ctsio);
4979			ctl_done((union ctl_io *)ctsio);
4980		} else {
4981			retval = lun->backend->config_write(
4982				(union ctl_io *)ctsio);
4983		}
4984#ifdef NEEDTOPORT
4985	}
4986#endif
4987	return (retval);
4988}
4989
4990/*
4991 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
4992 * we don't really do anything with the LBA and length fields if the user
4993 * passes them in.  Instead we'll just flush out the cache for the entire
4994 * LUN.
4995 */
4996int
4997ctl_sync_cache(struct ctl_scsiio *ctsio)
4998{
4999	struct ctl_lun *lun;
5000	struct ctl_softc *softc;
5001	struct ctl_lba_len_flags *lbalen;
5002	uint64_t starting_lba;
5003	uint32_t block_count;
5004	int retval;
5005	uint8_t byte2;
5006
5007	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5008
5009	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5010	softc = lun->ctl_softc;
5011	retval = 0;
5012
5013	switch (ctsio->cdb[0]) {
5014	case SYNCHRONIZE_CACHE: {
5015		struct scsi_sync_cache *cdb;
5016		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5017
5018		starting_lba = scsi_4btoul(cdb->begin_lba);
5019		block_count = scsi_2btoul(cdb->lb_count);
5020		byte2 = cdb->byte2;
5021		break;
5022	}
5023	case SYNCHRONIZE_CACHE_16: {
5024		struct scsi_sync_cache_16 *cdb;
5025		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5026
5027		starting_lba = scsi_8btou64(cdb->begin_lba);
5028		block_count = scsi_4btoul(cdb->lb_count);
5029		byte2 = cdb->byte2;
5030		break;
5031	}
5032	default:
5033		ctl_set_invalid_opcode(ctsio);
5034		ctl_done((union ctl_io *)ctsio);
5035		goto bailout;
5036		break; /* NOTREACHED */
5037	}
5038
5039	/*
5040	 * We check the LBA and length, but don't do anything with them.
5041	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5042	 * get flushed.  This check will just help satisfy anyone who wants
5043	 * to see an error for an out of range LBA.
5044	 */
5045	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5046		ctl_set_lba_out_of_range(ctsio);
5047		ctl_done((union ctl_io *)ctsio);
5048		goto bailout;
5049	}
5050
5051	/*
5052	 * If this LUN has no backend, we can't flush the cache anyway.
5053	 */
5054	if (lun->backend == NULL) {
5055		ctl_set_invalid_opcode(ctsio);
5056		ctl_done((union ctl_io *)ctsio);
5057		goto bailout;
5058	}
5059
5060	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5061	lbalen->lba = starting_lba;
5062	lbalen->len = block_count;
5063	lbalen->flags = byte2;
5064
5065	/*
5066	 * Check to see whether we're configured to send the SYNCHRONIZE
5067	 * CACHE command directly to the back end.
5068	 */
5069	mtx_lock(&lun->lun_lock);
5070	if ((softc->flags & CTL_FLAG_REAL_SYNC)
5071	 && (++(lun->sync_count) >= lun->sync_interval)) {
5072		lun->sync_count = 0;
5073		mtx_unlock(&lun->lun_lock);
5074		retval = lun->backend->config_write((union ctl_io *)ctsio);
5075	} else {
5076		mtx_unlock(&lun->lun_lock);
5077		ctl_set_success(ctsio);
5078		ctl_done((union ctl_io *)ctsio);
5079	}
5080
5081bailout:
5082
5083	return (retval);
5084}
5085
5086int
5087ctl_format(struct ctl_scsiio *ctsio)
5088{
5089	struct scsi_format *cdb;
5090	struct ctl_lun *lun;
5091	int length, defect_list_len;
5092
5093	CTL_DEBUG_PRINT(("ctl_format\n"));
5094
5095	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5096
5097	cdb = (struct scsi_format *)ctsio->cdb;
5098
5099	length = 0;
5100	if (cdb->byte2 & SF_FMTDATA) {
5101		if (cdb->byte2 & SF_LONGLIST)
5102			length = sizeof(struct scsi_format_header_long);
5103		else
5104			length = sizeof(struct scsi_format_header_short);
5105	}
5106
5107	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5108	 && (length > 0)) {
5109		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5110		ctsio->kern_data_len = length;
5111		ctsio->kern_total_len = length;
5112		ctsio->kern_data_resid = 0;
5113		ctsio->kern_rel_offset = 0;
5114		ctsio->kern_sg_entries = 0;
5115		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5116		ctsio->be_move_done = ctl_config_move_done;
5117		ctl_datamove((union ctl_io *)ctsio);
5118
5119		return (CTL_RETVAL_COMPLETE);
5120	}
5121
5122	defect_list_len = 0;
5123
5124	if (cdb->byte2 & SF_FMTDATA) {
5125		if (cdb->byte2 & SF_LONGLIST) {
5126			struct scsi_format_header_long *header;
5127
5128			header = (struct scsi_format_header_long *)
5129				ctsio->kern_data_ptr;
5130
5131			defect_list_len = scsi_4btoul(header->defect_list_len);
5132			if (defect_list_len != 0) {
5133				ctl_set_invalid_field(ctsio,
5134						      /*sks_valid*/ 1,
5135						      /*command*/ 0,
5136						      /*field*/ 2,
5137						      /*bit_valid*/ 0,
5138						      /*bit*/ 0);
5139				goto bailout;
5140			}
5141		} else {
5142			struct scsi_format_header_short *header;
5143
5144			header = (struct scsi_format_header_short *)
5145				ctsio->kern_data_ptr;
5146
5147			defect_list_len = scsi_2btoul(header->defect_list_len);
5148			if (defect_list_len != 0) {
5149				ctl_set_invalid_field(ctsio,
5150						      /*sks_valid*/ 1,
5151						      /*command*/ 0,
5152						      /*field*/ 2,
5153						      /*bit_valid*/ 0,
5154						      /*bit*/ 0);
5155				goto bailout;
5156			}
5157		}
5158	}
5159
5160	/*
5161	 * The format command will clear out the "Medium format corrupted"
5162	 * status if set by the configuration code.  That status is really
5163	 * just a way to notify the host that we have lost the media, and
5164	 * get them to issue a command that will basically make them think
5165	 * they're blowing away the media.
5166	 */
5167	mtx_lock(&lun->lun_lock);
5168	lun->flags &= ~CTL_LUN_INOPERABLE;
5169	mtx_unlock(&lun->lun_lock);
5170
5171	ctl_set_success(ctsio);
5172bailout:
5173
5174	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5175		free(ctsio->kern_data_ptr, M_CTL);
5176		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5177	}
5178
5179	ctl_done((union ctl_io *)ctsio);
5180	return (CTL_RETVAL_COMPLETE);
5181}
5182
5183int
5184ctl_read_buffer(struct ctl_scsiio *ctsio)
5185{
5186	struct scsi_read_buffer *cdb;
5187	struct ctl_lun *lun;
5188	int buffer_offset, len;
5189	static uint8_t descr[4];
5190	static uint8_t echo_descr[4] = { 0 };
5191
5192	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5193
5194	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5195	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5196
5197	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5198	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5199	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5200		ctl_set_invalid_field(ctsio,
5201				      /*sks_valid*/ 1,
5202				      /*command*/ 1,
5203				      /*field*/ 1,
5204				      /*bit_valid*/ 1,
5205				      /*bit*/ 4);
5206		ctl_done((union ctl_io *)ctsio);
5207		return (CTL_RETVAL_COMPLETE);
5208	}
5209
5210	len = scsi_3btoul(cdb->length);
5211	buffer_offset = scsi_3btoul(cdb->offset);
5212
5213	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5214		ctl_set_invalid_field(ctsio,
5215				      /*sks_valid*/ 1,
5216				      /*command*/ 1,
5217				      /*field*/ 6,
5218				      /*bit_valid*/ 0,
5219				      /*bit*/ 0);
5220		ctl_done((union ctl_io *)ctsio);
5221		return (CTL_RETVAL_COMPLETE);
5222	}
5223
5224	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5225		descr[0] = 0;
5226		scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5227		ctsio->kern_data_ptr = descr;
5228		len = min(len, sizeof(descr));
5229	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5230		ctsio->kern_data_ptr = echo_descr;
5231		len = min(len, sizeof(echo_descr));
5232	} else {
5233		if (lun->write_buffer == NULL) {
5234			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5235			    M_CTL, M_WAITOK);
5236		}
5237		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5238	}
5239	ctsio->kern_data_len = len;
5240	ctsio->kern_total_len = len;
5241	ctsio->kern_data_resid = 0;
5242	ctsio->kern_rel_offset = 0;
5243	ctsio->kern_sg_entries = 0;
5244	ctl_set_success(ctsio);
5245	ctsio->be_move_done = ctl_config_move_done;
5246	ctl_datamove((union ctl_io *)ctsio);
5247	return (CTL_RETVAL_COMPLETE);
5248}
5249
5250int
5251ctl_write_buffer(struct ctl_scsiio *ctsio)
5252{
5253	struct scsi_write_buffer *cdb;
5254	struct ctl_lun *lun;
5255	int buffer_offset, len;
5256
5257	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5258
5259	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5260	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5261
5262	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5263		ctl_set_invalid_field(ctsio,
5264				      /*sks_valid*/ 1,
5265				      /*command*/ 1,
5266				      /*field*/ 1,
5267				      /*bit_valid*/ 1,
5268				      /*bit*/ 4);
5269		ctl_done((union ctl_io *)ctsio);
5270		return (CTL_RETVAL_COMPLETE);
5271	}
5272
5273	len = scsi_3btoul(cdb->length);
5274	buffer_offset = scsi_3btoul(cdb->offset);
5275
5276	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5277		ctl_set_invalid_field(ctsio,
5278				      /*sks_valid*/ 1,
5279				      /*command*/ 1,
5280				      /*field*/ 6,
5281				      /*bit_valid*/ 0,
5282				      /*bit*/ 0);
5283		ctl_done((union ctl_io *)ctsio);
5284		return (CTL_RETVAL_COMPLETE);
5285	}
5286
5287	/*
5288	 * If we've got a kernel request that hasn't been malloced yet,
5289	 * malloc it and tell the caller the data buffer is here.
5290	 */
5291	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5292		if (lun->write_buffer == NULL) {
5293			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5294			    M_CTL, M_WAITOK);
5295		}
5296		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5297		ctsio->kern_data_len = len;
5298		ctsio->kern_total_len = len;
5299		ctsio->kern_data_resid = 0;
5300		ctsio->kern_rel_offset = 0;
5301		ctsio->kern_sg_entries = 0;
5302		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5303		ctsio->be_move_done = ctl_config_move_done;
5304		ctl_datamove((union ctl_io *)ctsio);
5305
5306		return (CTL_RETVAL_COMPLETE);
5307	}
5308
5309	ctl_set_success(ctsio);
5310	ctl_done((union ctl_io *)ctsio);
5311	return (CTL_RETVAL_COMPLETE);
5312}
5313
5314int
5315ctl_write_same(struct ctl_scsiio *ctsio)
5316{
5317	struct ctl_lun *lun;
5318	struct ctl_lba_len_flags *lbalen;
5319	uint64_t lba;
5320	uint32_t num_blocks;
5321	int len, retval;
5322	uint8_t byte2;
5323
5324	retval = CTL_RETVAL_COMPLETE;
5325
5326	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5327
5328	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5329
5330	switch (ctsio->cdb[0]) {
5331	case WRITE_SAME_10: {
5332		struct scsi_write_same_10 *cdb;
5333
5334		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5335
5336		lba = scsi_4btoul(cdb->addr);
5337		num_blocks = scsi_2btoul(cdb->length);
5338		byte2 = cdb->byte2;
5339		break;
5340	}
5341	case WRITE_SAME_16: {
5342		struct scsi_write_same_16 *cdb;
5343
5344		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5345
5346		lba = scsi_8btou64(cdb->addr);
5347		num_blocks = scsi_4btoul(cdb->length);
5348		byte2 = cdb->byte2;
5349		break;
5350	}
5351	default:
5352		/*
5353		 * We got a command we don't support.  This shouldn't
5354		 * happen, commands should be filtered out above us.
5355		 */
5356		ctl_set_invalid_opcode(ctsio);
5357		ctl_done((union ctl_io *)ctsio);
5358
5359		return (CTL_RETVAL_COMPLETE);
5360		break; /* NOTREACHED */
5361	}
5362
5363	/* NDOB and ANCHOR flags can be used only together with UNMAP */
5364	if ((byte2 & SWS_UNMAP) == 0 &&
5365	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5366		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5367		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5368		ctl_done((union ctl_io *)ctsio);
5369		return (CTL_RETVAL_COMPLETE);
5370	}
5371
5372	/*
5373	 * The first check is to make sure we're in bounds, the second
5374	 * check is to catch wrap-around problems.  If the lba + num blocks
5375	 * is less than the lba, then we've wrapped around and the block
5376	 * range is invalid anyway.
5377	 */
5378	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5379	 || ((lba + num_blocks) < lba)) {
5380		ctl_set_lba_out_of_range(ctsio);
5381		ctl_done((union ctl_io *)ctsio);
5382		return (CTL_RETVAL_COMPLETE);
5383	}
5384
5385	/* Zero number of blocks means "to the last logical block" */
5386	if (num_blocks == 0) {
5387		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5388			ctl_set_invalid_field(ctsio,
5389					      /*sks_valid*/ 0,
5390					      /*command*/ 1,
5391					      /*field*/ 0,
5392					      /*bit_valid*/ 0,
5393					      /*bit*/ 0);
5394			ctl_done((union ctl_io *)ctsio);
5395			return (CTL_RETVAL_COMPLETE);
5396		}
5397		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5398	}
5399
5400	len = lun->be_lun->blocksize;
5401
5402	/*
5403	 * If we've got a kernel request that hasn't been malloced yet,
5404	 * malloc it and tell the caller the data buffer is here.
5405	 */
5406	if ((byte2 & SWS_NDOB) == 0 &&
5407	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5408		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5409		ctsio->kern_data_len = len;
5410		ctsio->kern_total_len = len;
5411		ctsio->kern_data_resid = 0;
5412		ctsio->kern_rel_offset = 0;
5413		ctsio->kern_sg_entries = 0;
5414		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5415		ctsio->be_move_done = ctl_config_move_done;
5416		ctl_datamove((union ctl_io *)ctsio);
5417
5418		return (CTL_RETVAL_COMPLETE);
5419	}
5420
5421	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5422	lbalen->lba = lba;
5423	lbalen->len = num_blocks;
5424	lbalen->flags = byte2;
5425	retval = lun->backend->config_write((union ctl_io *)ctsio);
5426
5427	return (retval);
5428}
5429
5430int
5431ctl_unmap(struct ctl_scsiio *ctsio)
5432{
5433	struct ctl_lun *lun;
5434	struct scsi_unmap *cdb;
5435	struct ctl_ptr_len_flags *ptrlen;
5436	struct scsi_unmap_header *hdr;
5437	struct scsi_unmap_desc *buf, *end, *endnz, *range;
5438	uint64_t lba;
5439	uint32_t num_blocks;
5440	int len, retval;
5441	uint8_t byte2;
5442
5443	retval = CTL_RETVAL_COMPLETE;
5444
5445	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5446
5447	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5448	cdb = (struct scsi_unmap *)ctsio->cdb;
5449
5450	len = scsi_2btoul(cdb->length);
5451	byte2 = cdb->byte2;
5452
5453	/*
5454	 * If we've got a kernel request that hasn't been malloced yet,
5455	 * malloc it and tell the caller the data buffer is here.
5456	 */
5457	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5458		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5459		ctsio->kern_data_len = len;
5460		ctsio->kern_total_len = len;
5461		ctsio->kern_data_resid = 0;
5462		ctsio->kern_rel_offset = 0;
5463		ctsio->kern_sg_entries = 0;
5464		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5465		ctsio->be_move_done = ctl_config_move_done;
5466		ctl_datamove((union ctl_io *)ctsio);
5467
5468		return (CTL_RETVAL_COMPLETE);
5469	}
5470
5471	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5472	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5473	if (len < sizeof (*hdr) ||
5474	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5475	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5476	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5477		ctl_set_invalid_field(ctsio,
5478				      /*sks_valid*/ 0,
5479				      /*command*/ 0,
5480				      /*field*/ 0,
5481				      /*bit_valid*/ 0,
5482				      /*bit*/ 0);
5483		goto done;
5484	}
5485	len = scsi_2btoul(hdr->desc_length);
5486	buf = (struct scsi_unmap_desc *)(hdr + 1);
5487	end = buf + len / sizeof(*buf);
5488
5489	endnz = buf;
5490	for (range = buf; range < end; range++) {
5491		lba = scsi_8btou64(range->lba);
5492		num_blocks = scsi_4btoul(range->length);
5493		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5494		 || ((lba + num_blocks) < lba)) {
5495			ctl_set_lba_out_of_range(ctsio);
5496			ctl_done((union ctl_io *)ctsio);
5497			return (CTL_RETVAL_COMPLETE);
5498		}
5499		if (num_blocks != 0)
5500			endnz = range + 1;
5501	}
5502
5503	/*
5504	 * Block backend can not handle zero last range.
5505	 * Filter it out and return if there is nothing left.
5506	 */
5507	len = (uint8_t *)endnz - (uint8_t *)buf;
5508	if (len == 0) {
5509		ctl_set_success(ctsio);
5510		goto done;
5511	}
5512
5513	mtx_lock(&lun->lun_lock);
5514	ptrlen = (struct ctl_ptr_len_flags *)
5515	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5516	ptrlen->ptr = (void *)buf;
5517	ptrlen->len = len;
5518	ptrlen->flags = byte2;
5519	ctl_check_blocked(lun);
5520	mtx_unlock(&lun->lun_lock);
5521
5522	retval = lun->backend->config_write((union ctl_io *)ctsio);
5523	return (retval);
5524
5525done:
5526	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5527		free(ctsio->kern_data_ptr, M_CTL);
5528		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5529	}
5530	ctl_done((union ctl_io *)ctsio);
5531	return (CTL_RETVAL_COMPLETE);
5532}
5533
5534/*
5535 * Note that this function currently doesn't actually do anything inside
5536 * CTL to enforce things if the DQue bit is turned on.
5537 *
5538 * Also note that this function can't be used in the default case, because
5539 * the DQue bit isn't set in the changeable mask for the control mode page
5540 * anyway.  This is just here as an example for how to implement a page
5541 * handler, and a placeholder in case we want to allow the user to turn
5542 * tagged queueing on and off.
5543 *
5544 * The D_SENSE bit handling is functional, however, and will turn
5545 * descriptor sense on and off for a given LUN.
5546 */
5547int
5548ctl_control_page_handler(struct ctl_scsiio *ctsio,
5549			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5550{
5551	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5552	struct ctl_lun *lun;
5553	int set_ua;
5554	uint32_t initidx;
5555
5556	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5557	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5558	set_ua = 0;
5559
5560	user_cp = (struct scsi_control_page *)page_ptr;
5561	current_cp = (struct scsi_control_page *)
5562		(page_index->page_data + (page_index->page_len *
5563		CTL_PAGE_CURRENT));
5564	saved_cp = (struct scsi_control_page *)
5565		(page_index->page_data + (page_index->page_len *
5566		CTL_PAGE_SAVED));
5567
5568	mtx_lock(&lun->lun_lock);
5569	if (((current_cp->rlec & SCP_DSENSE) == 0)
5570	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5571		/*
5572		 * Descriptor sense is currently turned off and the user
5573		 * wants to turn it on.
5574		 */
5575		current_cp->rlec |= SCP_DSENSE;
5576		saved_cp->rlec |= SCP_DSENSE;
5577		lun->flags |= CTL_LUN_SENSE_DESC;
5578		set_ua = 1;
5579	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
5580		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
5581		/*
5582		 * Descriptor sense is currently turned on, and the user
5583		 * wants to turn it off.
5584		 */
5585		current_cp->rlec &= ~SCP_DSENSE;
5586		saved_cp->rlec &= ~SCP_DSENSE;
5587		lun->flags &= ~CTL_LUN_SENSE_DESC;
5588		set_ua = 1;
5589	}
5590	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
5591	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
5592		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5593		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5594		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5595		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5596		set_ua = 1;
5597	}
5598	if ((current_cp->eca_and_aen & SCP_SWP) !=
5599	    (user_cp->eca_and_aen & SCP_SWP)) {
5600		current_cp->eca_and_aen &= ~SCP_SWP;
5601		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5602		saved_cp->eca_and_aen &= ~SCP_SWP;
5603		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5604		set_ua = 1;
5605	}
5606	if (set_ua != 0)
5607		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5608	mtx_unlock(&lun->lun_lock);
5609
5610	return (0);
5611}
5612
5613int
5614ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
5615		     struct ctl_page_index *page_index, uint8_t *page_ptr)
5616{
5617	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
5618	struct ctl_lun *lun;
5619	int set_ua;
5620	uint32_t initidx;
5621
5622	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5623	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5624	set_ua = 0;
5625
5626	user_cp = (struct scsi_caching_page *)page_ptr;
5627	current_cp = (struct scsi_caching_page *)
5628		(page_index->page_data + (page_index->page_len *
5629		CTL_PAGE_CURRENT));
5630	saved_cp = (struct scsi_caching_page *)
5631		(page_index->page_data + (page_index->page_len *
5632		CTL_PAGE_SAVED));
5633
5634	mtx_lock(&lun->lun_lock);
5635	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
5636	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
5637		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
5638		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
5639		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
5640		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
5641		set_ua = 1;
5642	}
5643	if (set_ua != 0)
5644		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5645	mtx_unlock(&lun->lun_lock);
5646
5647	return (0);
5648}
5649
5650int
5651ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
5652				struct ctl_page_index *page_index,
5653				uint8_t *page_ptr)
5654{
5655	uint8_t *c;
5656	int i;
5657
5658	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
5659	ctl_time_io_secs =
5660		(c[0] << 8) |
5661		(c[1] << 0) |
5662		0;
5663	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
5664	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
5665	printf("page data:");
5666	for (i=0; i<8; i++)
5667		printf(" %.2x",page_ptr[i]);
5668	printf("\n");
5669	return (0);
5670}
5671
5672int
5673ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
5674			       struct ctl_page_index *page_index,
5675			       int pc)
5676{
5677	struct copan_debugconf_subpage *page;
5678
5679	page = (struct copan_debugconf_subpage *)page_index->page_data +
5680		(page_index->page_len * pc);
5681
5682	switch (pc) {
5683	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5684	case SMS_PAGE_CTRL_DEFAULT >> 6:
5685	case SMS_PAGE_CTRL_SAVED >> 6:
5686		/*
5687		 * We don't update the changable or default bits for this page.
5688		 */
5689		break;
5690	case SMS_PAGE_CTRL_CURRENT >> 6:
5691		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
5692		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
5693		break;
5694	default:
5695#ifdef NEEDTOPORT
5696		EPRINT(0, "Invalid PC %d!!", pc);
5697#endif /* NEEDTOPORT */
5698		break;
5699	}
5700	return (0);
5701}
5702
5703
5704static int
5705ctl_do_mode_select(union ctl_io *io)
5706{
5707	struct scsi_mode_page_header *page_header;
5708	struct ctl_page_index *page_index;
5709	struct ctl_scsiio *ctsio;
5710	int control_dev, page_len;
5711	int page_len_offset, page_len_size;
5712	union ctl_modepage_info *modepage_info;
5713	struct ctl_lun *lun;
5714	int *len_left, *len_used;
5715	int retval, i;
5716
5717	ctsio = &io->scsiio;
5718	page_index = NULL;
5719	page_len = 0;
5720	retval = CTL_RETVAL_COMPLETE;
5721
5722	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5723
5724	if (lun->be_lun->lun_type != T_DIRECT)
5725		control_dev = 1;
5726	else
5727		control_dev = 0;
5728
5729	modepage_info = (union ctl_modepage_info *)
5730		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5731	len_left = &modepage_info->header.len_left;
5732	len_used = &modepage_info->header.len_used;
5733
5734do_next_page:
5735
5736	page_header = (struct scsi_mode_page_header *)
5737		(ctsio->kern_data_ptr + *len_used);
5738
5739	if (*len_left == 0) {
5740		free(ctsio->kern_data_ptr, M_CTL);
5741		ctl_set_success(ctsio);
5742		ctl_done((union ctl_io *)ctsio);
5743		return (CTL_RETVAL_COMPLETE);
5744	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
5745
5746		free(ctsio->kern_data_ptr, M_CTL);
5747		ctl_set_param_len_error(ctsio);
5748		ctl_done((union ctl_io *)ctsio);
5749		return (CTL_RETVAL_COMPLETE);
5750
5751	} else if ((page_header->page_code & SMPH_SPF)
5752		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
5753
5754		free(ctsio->kern_data_ptr, M_CTL);
5755		ctl_set_param_len_error(ctsio);
5756		ctl_done((union ctl_io *)ctsio);
5757		return (CTL_RETVAL_COMPLETE);
5758	}
5759
5760
5761	/*
5762	 * XXX KDM should we do something with the block descriptor?
5763	 */
5764	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5765
5766		if ((control_dev != 0)
5767		 && (lun->mode_pages.index[i].page_flags &
5768		     CTL_PAGE_FLAG_DISK_ONLY))
5769			continue;
5770
5771		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5772		    (page_header->page_code & SMPH_PC_MASK))
5773			continue;
5774
5775		/*
5776		 * If neither page has a subpage code, then we've got a
5777		 * match.
5778		 */
5779		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
5780		 && ((page_header->page_code & SMPH_SPF) == 0)) {
5781			page_index = &lun->mode_pages.index[i];
5782			page_len = page_header->page_length;
5783			break;
5784		}
5785
5786		/*
5787		 * If both pages have subpages, then the subpage numbers
5788		 * have to match.
5789		 */
5790		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
5791		  && (page_header->page_code & SMPH_SPF)) {
5792			struct scsi_mode_page_header_sp *sph;
5793
5794			sph = (struct scsi_mode_page_header_sp *)page_header;
5795
5796			if (lun->mode_pages.index[i].subpage ==
5797			    sph->subpage) {
5798				page_index = &lun->mode_pages.index[i];
5799				page_len = scsi_2btoul(sph->page_length);
5800				break;
5801			}
5802		}
5803	}
5804
5805	/*
5806	 * If we couldn't find the page, or if we don't have a mode select
5807	 * handler for it, send back an error to the user.
5808	 */
5809	if ((page_index == NULL)
5810	 || (page_index->select_handler == NULL)) {
5811		ctl_set_invalid_field(ctsio,
5812				      /*sks_valid*/ 1,
5813				      /*command*/ 0,
5814				      /*field*/ *len_used,
5815				      /*bit_valid*/ 0,
5816				      /*bit*/ 0);
5817		free(ctsio->kern_data_ptr, M_CTL);
5818		ctl_done((union ctl_io *)ctsio);
5819		return (CTL_RETVAL_COMPLETE);
5820	}
5821
5822	if (page_index->page_code & SMPH_SPF) {
5823		page_len_offset = 2;
5824		page_len_size = 2;
5825	} else {
5826		page_len_size = 1;
5827		page_len_offset = 1;
5828	}
5829
5830	/*
5831	 * If the length the initiator gives us isn't the one we specify in
5832	 * the mode page header, or if they didn't specify enough data in
5833	 * the CDB to avoid truncating this page, kick out the request.
5834	 */
5835	if ((page_len != (page_index->page_len - page_len_offset -
5836			  page_len_size))
5837	 || (*len_left < page_index->page_len)) {
5838
5839
5840		ctl_set_invalid_field(ctsio,
5841				      /*sks_valid*/ 1,
5842				      /*command*/ 0,
5843				      /*field*/ *len_used + page_len_offset,
5844				      /*bit_valid*/ 0,
5845				      /*bit*/ 0);
5846		free(ctsio->kern_data_ptr, M_CTL);
5847		ctl_done((union ctl_io *)ctsio);
5848		return (CTL_RETVAL_COMPLETE);
5849	}
5850
5851	/*
5852	 * Run through the mode page, checking to make sure that the bits
5853	 * the user changed are actually legal for him to change.
5854	 */
5855	for (i = 0; i < page_index->page_len; i++) {
5856		uint8_t *user_byte, *change_mask, *current_byte;
5857		int bad_bit;
5858		int j;
5859
5860		user_byte = (uint8_t *)page_header + i;
5861		change_mask = page_index->page_data +
5862			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
5863		current_byte = page_index->page_data +
5864			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
5865
5866		/*
5867		 * Check to see whether the user set any bits in this byte
5868		 * that he is not allowed to set.
5869		 */
5870		if ((*user_byte & ~(*change_mask)) ==
5871		    (*current_byte & ~(*change_mask)))
5872			continue;
5873
5874		/*
5875		 * Go through bit by bit to determine which one is illegal.
5876		 */
5877		bad_bit = 0;
5878		for (j = 7; j >= 0; j--) {
5879			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
5880			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
5881				bad_bit = i;
5882				break;
5883			}
5884		}
5885		ctl_set_invalid_field(ctsio,
5886				      /*sks_valid*/ 1,
5887				      /*command*/ 0,
5888				      /*field*/ *len_used + i,
5889				      /*bit_valid*/ 1,
5890				      /*bit*/ bad_bit);
5891		free(ctsio->kern_data_ptr, M_CTL);
5892		ctl_done((union ctl_io *)ctsio);
5893		return (CTL_RETVAL_COMPLETE);
5894	}
5895
5896	/*
5897	 * Decrement these before we call the page handler, since we may
5898	 * end up getting called back one way or another before the handler
5899	 * returns to this context.
5900	 */
5901	*len_left -= page_index->page_len;
5902	*len_used += page_index->page_len;
5903
5904	retval = page_index->select_handler(ctsio, page_index,
5905					    (uint8_t *)page_header);
5906
5907	/*
5908	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
5909	 * wait until this queued command completes to finish processing
5910	 * the mode page.  If it returns anything other than
5911	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
5912	 * already set the sense information, freed the data pointer, and
5913	 * completed the io for us.
5914	 */
5915	if (retval != CTL_RETVAL_COMPLETE)
5916		goto bailout_no_done;
5917
5918	/*
5919	 * If the initiator sent us more than one page, parse the next one.
5920	 */
5921	if (*len_left > 0)
5922		goto do_next_page;
5923
5924	ctl_set_success(ctsio);
5925	free(ctsio->kern_data_ptr, M_CTL);
5926	ctl_done((union ctl_io *)ctsio);
5927
5928bailout_no_done:
5929
5930	return (CTL_RETVAL_COMPLETE);
5931
5932}
5933
5934int
5935ctl_mode_select(struct ctl_scsiio *ctsio)
5936{
5937	int param_len, pf, sp;
5938	int header_size, bd_len;
5939	int len_left, len_used;
5940	struct ctl_page_index *page_index;
5941	struct ctl_lun *lun;
5942	int control_dev, page_len;
5943	union ctl_modepage_info *modepage_info;
5944	int retval;
5945
5946	pf = 0;
5947	sp = 0;
5948	page_len = 0;
5949	len_used = 0;
5950	len_left = 0;
5951	retval = 0;
5952	bd_len = 0;
5953	page_index = NULL;
5954
5955	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5956
5957	if (lun->be_lun->lun_type != T_DIRECT)
5958		control_dev = 1;
5959	else
5960		control_dev = 0;
5961
5962	switch (ctsio->cdb[0]) {
5963	case MODE_SELECT_6: {
5964		struct scsi_mode_select_6 *cdb;
5965
5966		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
5967
5968		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
5969		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
5970
5971		param_len = cdb->length;
5972		header_size = sizeof(struct scsi_mode_header_6);
5973		break;
5974	}
5975	case MODE_SELECT_10: {
5976		struct scsi_mode_select_10 *cdb;
5977
5978		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
5979
5980		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
5981		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
5982
5983		param_len = scsi_2btoul(cdb->length);
5984		header_size = sizeof(struct scsi_mode_header_10);
5985		break;
5986	}
5987	default:
5988		ctl_set_invalid_opcode(ctsio);
5989		ctl_done((union ctl_io *)ctsio);
5990		return (CTL_RETVAL_COMPLETE);
5991		break; /* NOTREACHED */
5992	}
5993
5994	/*
5995	 * From SPC-3:
5996	 * "A parameter list length of zero indicates that the Data-Out Buffer
5997	 * shall be empty. This condition shall not be considered as an error."
5998	 */
5999	if (param_len == 0) {
6000		ctl_set_success(ctsio);
6001		ctl_done((union ctl_io *)ctsio);
6002		return (CTL_RETVAL_COMPLETE);
6003	}
6004
6005	/*
6006	 * Since we'll hit this the first time through, prior to
6007	 * allocation, we don't need to free a data buffer here.
6008	 */
6009	if (param_len < header_size) {
6010		ctl_set_param_len_error(ctsio);
6011		ctl_done((union ctl_io *)ctsio);
6012		return (CTL_RETVAL_COMPLETE);
6013	}
6014
6015	/*
6016	 * Allocate the data buffer and grab the user's data.  In theory,
6017	 * we shouldn't have to sanity check the parameter list length here
6018	 * because the maximum size is 64K.  We should be able to malloc
6019	 * that much without too many problems.
6020	 */
6021	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6022		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6023		ctsio->kern_data_len = param_len;
6024		ctsio->kern_total_len = param_len;
6025		ctsio->kern_data_resid = 0;
6026		ctsio->kern_rel_offset = 0;
6027		ctsio->kern_sg_entries = 0;
6028		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6029		ctsio->be_move_done = ctl_config_move_done;
6030		ctl_datamove((union ctl_io *)ctsio);
6031
6032		return (CTL_RETVAL_COMPLETE);
6033	}
6034
6035	switch (ctsio->cdb[0]) {
6036	case MODE_SELECT_6: {
6037		struct scsi_mode_header_6 *mh6;
6038
6039		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6040		bd_len = mh6->blk_desc_len;
6041		break;
6042	}
6043	case MODE_SELECT_10: {
6044		struct scsi_mode_header_10 *mh10;
6045
6046		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6047		bd_len = scsi_2btoul(mh10->blk_desc_len);
6048		break;
6049	}
6050	default:
6051		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6052		break;
6053	}
6054
6055	if (param_len < (header_size + bd_len)) {
6056		free(ctsio->kern_data_ptr, M_CTL);
6057		ctl_set_param_len_error(ctsio);
6058		ctl_done((union ctl_io *)ctsio);
6059		return (CTL_RETVAL_COMPLETE);
6060	}
6061
6062	/*
6063	 * Set the IO_CONT flag, so that if this I/O gets passed to
6064	 * ctl_config_write_done(), it'll get passed back to
6065	 * ctl_do_mode_select() for further processing, or completion if
6066	 * we're all done.
6067	 */
6068	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6069	ctsio->io_cont = ctl_do_mode_select;
6070
6071	modepage_info = (union ctl_modepage_info *)
6072		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6073
6074	memset(modepage_info, 0, sizeof(*modepage_info));
6075
6076	len_left = param_len - header_size - bd_len;
6077	len_used = header_size + bd_len;
6078
6079	modepage_info->header.len_left = len_left;
6080	modepage_info->header.len_used = len_used;
6081
6082	return (ctl_do_mode_select((union ctl_io *)ctsio));
6083}
6084
6085int
6086ctl_mode_sense(struct ctl_scsiio *ctsio)
6087{
6088	struct ctl_lun *lun;
6089	int pc, page_code, dbd, llba, subpage;
6090	int alloc_len, page_len, header_len, total_len;
6091	struct scsi_mode_block_descr *block_desc;
6092	struct ctl_page_index *page_index;
6093	int control_dev;
6094
6095	dbd = 0;
6096	llba = 0;
6097	block_desc = NULL;
6098	page_index = NULL;
6099
6100	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6101
6102	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6103
6104	if (lun->be_lun->lun_type != T_DIRECT)
6105		control_dev = 1;
6106	else
6107		control_dev = 0;
6108
6109	switch (ctsio->cdb[0]) {
6110	case MODE_SENSE_6: {
6111		struct scsi_mode_sense_6 *cdb;
6112
6113		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6114
6115		header_len = sizeof(struct scsi_mode_hdr_6);
6116		if (cdb->byte2 & SMS_DBD)
6117			dbd = 1;
6118		else
6119			header_len += sizeof(struct scsi_mode_block_descr);
6120
6121		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6122		page_code = cdb->page & SMS_PAGE_CODE;
6123		subpage = cdb->subpage;
6124		alloc_len = cdb->length;
6125		break;
6126	}
6127	case MODE_SENSE_10: {
6128		struct scsi_mode_sense_10 *cdb;
6129
6130		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6131
6132		header_len = sizeof(struct scsi_mode_hdr_10);
6133
6134		if (cdb->byte2 & SMS_DBD)
6135			dbd = 1;
6136		else
6137			header_len += sizeof(struct scsi_mode_block_descr);
6138		if (cdb->byte2 & SMS10_LLBAA)
6139			llba = 1;
6140		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6141		page_code = cdb->page & SMS_PAGE_CODE;
6142		subpage = cdb->subpage;
6143		alloc_len = scsi_2btoul(cdb->length);
6144		break;
6145	}
6146	default:
6147		ctl_set_invalid_opcode(ctsio);
6148		ctl_done((union ctl_io *)ctsio);
6149		return (CTL_RETVAL_COMPLETE);
6150		break; /* NOTREACHED */
6151	}
6152
6153	/*
6154	 * We have to make a first pass through to calculate the size of
6155	 * the pages that match the user's query.  Then we allocate enough
6156	 * memory to hold it, and actually copy the data into the buffer.
6157	 */
6158	switch (page_code) {
6159	case SMS_ALL_PAGES_PAGE: {
6160		int i;
6161
6162		page_len = 0;
6163
6164		/*
6165		 * At the moment, values other than 0 and 0xff here are
6166		 * reserved according to SPC-3.
6167		 */
6168		if ((subpage != SMS_SUBPAGE_PAGE_0)
6169		 && (subpage != SMS_SUBPAGE_ALL)) {
6170			ctl_set_invalid_field(ctsio,
6171					      /*sks_valid*/ 1,
6172					      /*command*/ 1,
6173					      /*field*/ 3,
6174					      /*bit_valid*/ 0,
6175					      /*bit*/ 0);
6176			ctl_done((union ctl_io *)ctsio);
6177			return (CTL_RETVAL_COMPLETE);
6178		}
6179
6180		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6181			if ((control_dev != 0)
6182			 && (lun->mode_pages.index[i].page_flags &
6183			     CTL_PAGE_FLAG_DISK_ONLY))
6184				continue;
6185
6186			/*
6187			 * We don't use this subpage if the user didn't
6188			 * request all subpages.
6189			 */
6190			if ((lun->mode_pages.index[i].subpage != 0)
6191			 && (subpage == SMS_SUBPAGE_PAGE_0))
6192				continue;
6193
6194#if 0
6195			printf("found page %#x len %d\n",
6196			       lun->mode_pages.index[i].page_code &
6197			       SMPH_PC_MASK,
6198			       lun->mode_pages.index[i].page_len);
6199#endif
6200			page_len += lun->mode_pages.index[i].page_len;
6201		}
6202		break;
6203	}
6204	default: {
6205		int i;
6206
6207		page_len = 0;
6208
6209		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6210			/* Look for the right page code */
6211			if ((lun->mode_pages.index[i].page_code &
6212			     SMPH_PC_MASK) != page_code)
6213				continue;
6214
6215			/* Look for the right subpage or the subpage wildcard*/
6216			if ((lun->mode_pages.index[i].subpage != subpage)
6217			 && (subpage != SMS_SUBPAGE_ALL))
6218				continue;
6219
6220			/* Make sure the page is supported for this dev type */
6221			if ((control_dev != 0)
6222			 && (lun->mode_pages.index[i].page_flags &
6223			     CTL_PAGE_FLAG_DISK_ONLY))
6224				continue;
6225
6226#if 0
6227			printf("found page %#x len %d\n",
6228			       lun->mode_pages.index[i].page_code &
6229			       SMPH_PC_MASK,
6230			       lun->mode_pages.index[i].page_len);
6231#endif
6232
6233			page_len += lun->mode_pages.index[i].page_len;
6234		}
6235
6236		if (page_len == 0) {
6237			ctl_set_invalid_field(ctsio,
6238					      /*sks_valid*/ 1,
6239					      /*command*/ 1,
6240					      /*field*/ 2,
6241					      /*bit_valid*/ 1,
6242					      /*bit*/ 5);
6243			ctl_done((union ctl_io *)ctsio);
6244			return (CTL_RETVAL_COMPLETE);
6245		}
6246		break;
6247	}
6248	}
6249
6250	total_len = header_len + page_len;
6251#if 0
6252	printf("header_len = %d, page_len = %d, total_len = %d\n",
6253	       header_len, page_len, total_len);
6254#endif
6255
6256	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6257	ctsio->kern_sg_entries = 0;
6258	ctsio->kern_data_resid = 0;
6259	ctsio->kern_rel_offset = 0;
6260	if (total_len < alloc_len) {
6261		ctsio->residual = alloc_len - total_len;
6262		ctsio->kern_data_len = total_len;
6263		ctsio->kern_total_len = total_len;
6264	} else {
6265		ctsio->residual = 0;
6266		ctsio->kern_data_len = alloc_len;
6267		ctsio->kern_total_len = alloc_len;
6268	}
6269
6270	switch (ctsio->cdb[0]) {
6271	case MODE_SENSE_6: {
6272		struct scsi_mode_hdr_6 *header;
6273
6274		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6275
6276		header->datalen = MIN(total_len - 1, 254);
6277		if (control_dev == 0) {
6278			header->dev_specific = 0x10; /* DPOFUA */
6279			if ((lun->flags & CTL_LUN_READONLY) ||
6280			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6281			    .eca_and_aen & SCP_SWP) != 0)
6282				    header->dev_specific |= 0x80; /* WP */
6283		}
6284		if (dbd)
6285			header->block_descr_len = 0;
6286		else
6287			header->block_descr_len =
6288				sizeof(struct scsi_mode_block_descr);
6289		block_desc = (struct scsi_mode_block_descr *)&header[1];
6290		break;
6291	}
6292	case MODE_SENSE_10: {
6293		struct scsi_mode_hdr_10 *header;
6294		int datalen;
6295
6296		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6297
6298		datalen = MIN(total_len - 2, 65533);
6299		scsi_ulto2b(datalen, header->datalen);
6300		if (control_dev == 0) {
6301			header->dev_specific = 0x10; /* DPOFUA */
6302			if ((lun->flags & CTL_LUN_READONLY) ||
6303			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6304			    .eca_and_aen & SCP_SWP) != 0)
6305				    header->dev_specific |= 0x80; /* WP */
6306		}
6307		if (dbd)
6308			scsi_ulto2b(0, header->block_descr_len);
6309		else
6310			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6311				    header->block_descr_len);
6312		block_desc = (struct scsi_mode_block_descr *)&header[1];
6313		break;
6314	}
6315	default:
6316		panic("invalid CDB type %#x", ctsio->cdb[0]);
6317		break; /* NOTREACHED */
6318	}
6319
6320	/*
6321	 * If we've got a disk, use its blocksize in the block
6322	 * descriptor.  Otherwise, just set it to 0.
6323	 */
6324	if (dbd == 0) {
6325		if (control_dev == 0)
6326			scsi_ulto3b(lun->be_lun->blocksize,
6327				    block_desc->block_len);
6328		else
6329			scsi_ulto3b(0, block_desc->block_len);
6330	}
6331
6332	switch (page_code) {
6333	case SMS_ALL_PAGES_PAGE: {
6334		int i, data_used;
6335
6336		data_used = header_len;
6337		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6338			struct ctl_page_index *page_index;
6339
6340			page_index = &lun->mode_pages.index[i];
6341
6342			if ((control_dev != 0)
6343			 && (page_index->page_flags &
6344			    CTL_PAGE_FLAG_DISK_ONLY))
6345				continue;
6346
6347			/*
6348			 * We don't use this subpage if the user didn't
6349			 * request all subpages.  We already checked (above)
6350			 * to make sure the user only specified a subpage
6351			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6352			 */
6353			if ((page_index->subpage != 0)
6354			 && (subpage == SMS_SUBPAGE_PAGE_0))
6355				continue;
6356
6357			/*
6358			 * Call the handler, if it exists, to update the
6359			 * page to the latest values.
6360			 */
6361			if (page_index->sense_handler != NULL)
6362				page_index->sense_handler(ctsio, page_index,pc);
6363
6364			memcpy(ctsio->kern_data_ptr + data_used,
6365			       page_index->page_data +
6366			       (page_index->page_len * pc),
6367			       page_index->page_len);
6368			data_used += page_index->page_len;
6369		}
6370		break;
6371	}
6372	default: {
6373		int i, data_used;
6374
6375		data_used = header_len;
6376
6377		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6378			struct ctl_page_index *page_index;
6379
6380			page_index = &lun->mode_pages.index[i];
6381
6382			/* Look for the right page code */
6383			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6384				continue;
6385
6386			/* Look for the right subpage or the subpage wildcard*/
6387			if ((page_index->subpage != subpage)
6388			 && (subpage != SMS_SUBPAGE_ALL))
6389				continue;
6390
6391			/* Make sure the page is supported for this dev type */
6392			if ((control_dev != 0)
6393			 && (page_index->page_flags &
6394			     CTL_PAGE_FLAG_DISK_ONLY))
6395				continue;
6396
6397			/*
6398			 * Call the handler, if it exists, to update the
6399			 * page to the latest values.
6400			 */
6401			if (page_index->sense_handler != NULL)
6402				page_index->sense_handler(ctsio, page_index,pc);
6403
6404			memcpy(ctsio->kern_data_ptr + data_used,
6405			       page_index->page_data +
6406			       (page_index->page_len * pc),
6407			       page_index->page_len);
6408			data_used += page_index->page_len;
6409		}
6410		break;
6411	}
6412	}
6413
6414	ctl_set_success(ctsio);
6415	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6416	ctsio->be_move_done = ctl_config_move_done;
6417	ctl_datamove((union ctl_io *)ctsio);
6418	return (CTL_RETVAL_COMPLETE);
6419}
6420
6421int
6422ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6423			       struct ctl_page_index *page_index,
6424			       int pc)
6425{
6426	struct ctl_lun *lun;
6427	struct scsi_log_param_header *phdr;
6428	uint8_t *data;
6429	uint64_t val;
6430
6431	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6432	data = page_index->page_data;
6433
6434	if (lun->backend->lun_attr != NULL &&
6435	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6436	     != UINT64_MAX) {
6437		phdr = (struct scsi_log_param_header *)data;
6438		scsi_ulto2b(0x0001, phdr->param_code);
6439		phdr->param_control = SLP_LBIN | SLP_LP;
6440		phdr->param_len = 8;
6441		data = (uint8_t *)(phdr + 1);
6442		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6443		data[4] = 0x02; /* per-pool */
6444		data += phdr->param_len;
6445	}
6446
6447	if (lun->backend->lun_attr != NULL &&
6448	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6449	     != UINT64_MAX) {
6450		phdr = (struct scsi_log_param_header *)data;
6451		scsi_ulto2b(0x0002, phdr->param_code);
6452		phdr->param_control = SLP_LBIN | SLP_LP;
6453		phdr->param_len = 8;
6454		data = (uint8_t *)(phdr + 1);
6455		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6456		data[4] = 0x01; /* per-LUN */
6457		data += phdr->param_len;
6458	}
6459
6460	if (lun->backend->lun_attr != NULL &&
6461	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6462	     != UINT64_MAX) {
6463		phdr = (struct scsi_log_param_header *)data;
6464		scsi_ulto2b(0x00f1, phdr->param_code);
6465		phdr->param_control = SLP_LBIN | SLP_LP;
6466		phdr->param_len = 8;
6467		data = (uint8_t *)(phdr + 1);
6468		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6469		data[4] = 0x02; /* per-pool */
6470		data += phdr->param_len;
6471	}
6472
6473	if (lun->backend->lun_attr != NULL &&
6474	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6475	     != UINT64_MAX) {
6476		phdr = (struct scsi_log_param_header *)data;
6477		scsi_ulto2b(0x00f2, phdr->param_code);
6478		phdr->param_control = SLP_LBIN | SLP_LP;
6479		phdr->param_len = 8;
6480		data = (uint8_t *)(phdr + 1);
6481		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6482		data[4] = 0x02; /* per-pool */
6483		data += phdr->param_len;
6484	}
6485
6486	page_index->page_len = data - page_index->page_data;
6487	return (0);
6488}
6489
6490int
6491ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6492			       struct ctl_page_index *page_index,
6493			       int pc)
6494{
6495	struct ctl_lun *lun;
6496	struct stat_page *data;
6497	uint64_t rn, wn, rb, wb;
6498	struct bintime rt, wt;
6499	int i;
6500
6501	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6502	data = (struct stat_page *)page_index->page_data;
6503
6504	scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6505	data->sap.hdr.param_control = SLP_LBIN;
6506	data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6507	    sizeof(struct scsi_log_param_header);
6508	rn = wn = rb = wb = 0;
6509	bintime_clear(&rt);
6510	bintime_clear(&wt);
6511	for (i = 0; i < CTL_MAX_PORTS; i++) {
6512		rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6513		wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6514		rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6515		wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6516		bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6517		bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6518	}
6519	scsi_u64to8b(rn, data->sap.read_num);
6520	scsi_u64to8b(wn, data->sap.write_num);
6521	if (lun->stats.blocksize > 0) {
6522		scsi_u64to8b(wb / lun->stats.blocksize,
6523		    data->sap.recvieved_lba);
6524		scsi_u64to8b(rb / lun->stats.blocksize,
6525		    data->sap.transmitted_lba);
6526	}
6527	scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6528	    data->sap.read_int);
6529	scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6530	    data->sap.write_int);
6531	scsi_u64to8b(0, data->sap.weighted_num);
6532	scsi_u64to8b(0, data->sap.weighted_int);
6533	scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6534	data->it.hdr.param_control = SLP_LBIN;
6535	data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6536	    sizeof(struct scsi_log_param_header);
6537#ifdef CTL_TIME_IO
6538	scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6539#endif
6540	scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6541	data->it.hdr.param_control = SLP_LBIN;
6542	data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6543	    sizeof(struct scsi_log_param_header);
6544	scsi_ulto4b(3, data->ti.exponent);
6545	scsi_ulto4b(1, data->ti.integer);
6546
6547	page_index->page_len = sizeof(*data);
6548	return (0);
6549}
6550
6551int
6552ctl_log_sense(struct ctl_scsiio *ctsio)
6553{
6554	struct ctl_lun *lun;
6555	int i, pc, page_code, subpage;
6556	int alloc_len, total_len;
6557	struct ctl_page_index *page_index;
6558	struct scsi_log_sense *cdb;
6559	struct scsi_log_header *header;
6560
6561	CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6562
6563	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6564	cdb = (struct scsi_log_sense *)ctsio->cdb;
6565	pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6566	page_code = cdb->page & SLS_PAGE_CODE;
6567	subpage = cdb->subpage;
6568	alloc_len = scsi_2btoul(cdb->length);
6569
6570	page_index = NULL;
6571	for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6572		page_index = &lun->log_pages.index[i];
6573
6574		/* Look for the right page code */
6575		if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6576			continue;
6577
6578		/* Look for the right subpage or the subpage wildcard*/
6579		if (page_index->subpage != subpage)
6580			continue;
6581
6582		break;
6583	}
6584	if (i >= CTL_NUM_LOG_PAGES) {
6585		ctl_set_invalid_field(ctsio,
6586				      /*sks_valid*/ 1,
6587				      /*command*/ 1,
6588				      /*field*/ 2,
6589				      /*bit_valid*/ 0,
6590				      /*bit*/ 0);
6591		ctl_done((union ctl_io *)ctsio);
6592		return (CTL_RETVAL_COMPLETE);
6593	}
6594
6595	total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6596
6597	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6598	ctsio->kern_sg_entries = 0;
6599	ctsio->kern_data_resid = 0;
6600	ctsio->kern_rel_offset = 0;
6601	if (total_len < alloc_len) {
6602		ctsio->residual = alloc_len - total_len;
6603		ctsio->kern_data_len = total_len;
6604		ctsio->kern_total_len = total_len;
6605	} else {
6606		ctsio->residual = 0;
6607		ctsio->kern_data_len = alloc_len;
6608		ctsio->kern_total_len = alloc_len;
6609	}
6610
6611	header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6612	header->page = page_index->page_code;
6613	if (page_index->subpage) {
6614		header->page |= SL_SPF;
6615		header->subpage = page_index->subpage;
6616	}
6617	scsi_ulto2b(page_index->page_len, header->datalen);
6618
6619	/*
6620	 * Call the handler, if it exists, to update the
6621	 * page to the latest values.
6622	 */
6623	if (page_index->sense_handler != NULL)
6624		page_index->sense_handler(ctsio, page_index, pc);
6625
6626	memcpy(header + 1, page_index->page_data, page_index->page_len);
6627
6628	ctl_set_success(ctsio);
6629	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6630	ctsio->be_move_done = ctl_config_move_done;
6631	ctl_datamove((union ctl_io *)ctsio);
6632	return (CTL_RETVAL_COMPLETE);
6633}
6634
6635int
6636ctl_read_capacity(struct ctl_scsiio *ctsio)
6637{
6638	struct scsi_read_capacity *cdb;
6639	struct scsi_read_capacity_data *data;
6640	struct ctl_lun *lun;
6641	uint32_t lba;
6642
6643	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6644
6645	cdb = (struct scsi_read_capacity *)ctsio->cdb;
6646
6647	lba = scsi_4btoul(cdb->addr);
6648	if (((cdb->pmi & SRC_PMI) == 0)
6649	 && (lba != 0)) {
6650		ctl_set_invalid_field(/*ctsio*/ ctsio,
6651				      /*sks_valid*/ 1,
6652				      /*command*/ 1,
6653				      /*field*/ 2,
6654				      /*bit_valid*/ 0,
6655				      /*bit*/ 0);
6656		ctl_done((union ctl_io *)ctsio);
6657		return (CTL_RETVAL_COMPLETE);
6658	}
6659
6660	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6661
6662	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6663	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6664	ctsio->residual = 0;
6665	ctsio->kern_data_len = sizeof(*data);
6666	ctsio->kern_total_len = sizeof(*data);
6667	ctsio->kern_data_resid = 0;
6668	ctsio->kern_rel_offset = 0;
6669	ctsio->kern_sg_entries = 0;
6670
6671	/*
6672	 * If the maximum LBA is greater than 0xfffffffe, the user must
6673	 * issue a SERVICE ACTION IN (16) command, with the read capacity
6674	 * serivce action set.
6675	 */
6676	if (lun->be_lun->maxlba > 0xfffffffe)
6677		scsi_ulto4b(0xffffffff, data->addr);
6678	else
6679		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6680
6681	/*
6682	 * XXX KDM this may not be 512 bytes...
6683	 */
6684	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6685
6686	ctl_set_success(ctsio);
6687	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6688	ctsio->be_move_done = ctl_config_move_done;
6689	ctl_datamove((union ctl_io *)ctsio);
6690	return (CTL_RETVAL_COMPLETE);
6691}
6692
6693int
6694ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6695{
6696	struct scsi_read_capacity_16 *cdb;
6697	struct scsi_read_capacity_data_long *data;
6698	struct ctl_lun *lun;
6699	uint64_t lba;
6700	uint32_t alloc_len;
6701
6702	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6703
6704	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6705
6706	alloc_len = scsi_4btoul(cdb->alloc_len);
6707	lba = scsi_8btou64(cdb->addr);
6708
6709	if ((cdb->reladr & SRC16_PMI)
6710	 && (lba != 0)) {
6711		ctl_set_invalid_field(/*ctsio*/ ctsio,
6712				      /*sks_valid*/ 1,
6713				      /*command*/ 1,
6714				      /*field*/ 2,
6715				      /*bit_valid*/ 0,
6716				      /*bit*/ 0);
6717		ctl_done((union ctl_io *)ctsio);
6718		return (CTL_RETVAL_COMPLETE);
6719	}
6720
6721	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6722
6723	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6724	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6725
6726	if (sizeof(*data) < alloc_len) {
6727		ctsio->residual = alloc_len - sizeof(*data);
6728		ctsio->kern_data_len = sizeof(*data);
6729		ctsio->kern_total_len = sizeof(*data);
6730	} else {
6731		ctsio->residual = 0;
6732		ctsio->kern_data_len = alloc_len;
6733		ctsio->kern_total_len = alloc_len;
6734	}
6735	ctsio->kern_data_resid = 0;
6736	ctsio->kern_rel_offset = 0;
6737	ctsio->kern_sg_entries = 0;
6738
6739	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6740	/* XXX KDM this may not be 512 bytes... */
6741	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6742	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6743	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6744	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6745		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6746
6747	ctl_set_success(ctsio);
6748	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6749	ctsio->be_move_done = ctl_config_move_done;
6750	ctl_datamove((union ctl_io *)ctsio);
6751	return (CTL_RETVAL_COMPLETE);
6752}
6753
6754int
6755ctl_get_lba_status(struct ctl_scsiio *ctsio)
6756{
6757	struct scsi_get_lba_status *cdb;
6758	struct scsi_get_lba_status_data *data;
6759	struct ctl_lun *lun;
6760	struct ctl_lba_len_flags *lbalen;
6761	uint64_t lba;
6762	uint32_t alloc_len, total_len;
6763	int retval;
6764
6765	CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
6766
6767	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6768	cdb = (struct scsi_get_lba_status *)ctsio->cdb;
6769	lba = scsi_8btou64(cdb->addr);
6770	alloc_len = scsi_4btoul(cdb->alloc_len);
6771
6772	if (lba > lun->be_lun->maxlba) {
6773		ctl_set_lba_out_of_range(ctsio);
6774		ctl_done((union ctl_io *)ctsio);
6775		return (CTL_RETVAL_COMPLETE);
6776	}
6777
6778	total_len = sizeof(*data) + sizeof(data->descr[0]);
6779	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6780	data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
6781
6782	if (total_len < alloc_len) {
6783		ctsio->residual = alloc_len - total_len;
6784		ctsio->kern_data_len = total_len;
6785		ctsio->kern_total_len = total_len;
6786	} else {
6787		ctsio->residual = 0;
6788		ctsio->kern_data_len = alloc_len;
6789		ctsio->kern_total_len = alloc_len;
6790	}
6791	ctsio->kern_data_resid = 0;
6792	ctsio->kern_rel_offset = 0;
6793	ctsio->kern_sg_entries = 0;
6794
6795	/* Fill dummy data in case backend can't tell anything. */
6796	scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
6797	scsi_u64to8b(lba, data->descr[0].addr);
6798	scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
6799	    data->descr[0].length);
6800	data->descr[0].status = 0; /* Mapped or unknown. */
6801
6802	ctl_set_success(ctsio);
6803	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6804	ctsio->be_move_done = ctl_config_move_done;
6805
6806	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6807	lbalen->lba = lba;
6808	lbalen->len = total_len;
6809	lbalen->flags = 0;
6810	retval = lun->backend->config_read((union ctl_io *)ctsio);
6811	return (CTL_RETVAL_COMPLETE);
6812}
6813
6814int
6815ctl_read_defect(struct ctl_scsiio *ctsio)
6816{
6817	struct scsi_read_defect_data_10 *ccb10;
6818	struct scsi_read_defect_data_12 *ccb12;
6819	struct scsi_read_defect_data_hdr_10 *data10;
6820	struct scsi_read_defect_data_hdr_12 *data12;
6821	uint32_t alloc_len, data_len;
6822	uint8_t format;
6823
6824	CTL_DEBUG_PRINT(("ctl_read_defect\n"));
6825
6826	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
6827		ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
6828		format = ccb10->format;
6829		alloc_len = scsi_2btoul(ccb10->alloc_length);
6830		data_len = sizeof(*data10);
6831	} else {
6832		ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
6833		format = ccb12->format;
6834		alloc_len = scsi_4btoul(ccb12->alloc_length);
6835		data_len = sizeof(*data12);
6836	}
6837	if (alloc_len == 0) {
6838		ctl_set_success(ctsio);
6839		ctl_done((union ctl_io *)ctsio);
6840		return (CTL_RETVAL_COMPLETE);
6841	}
6842
6843	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
6844	if (data_len < alloc_len) {
6845		ctsio->residual = alloc_len - data_len;
6846		ctsio->kern_data_len = data_len;
6847		ctsio->kern_total_len = data_len;
6848	} else {
6849		ctsio->residual = 0;
6850		ctsio->kern_data_len = alloc_len;
6851		ctsio->kern_total_len = alloc_len;
6852	}
6853	ctsio->kern_data_resid = 0;
6854	ctsio->kern_rel_offset = 0;
6855	ctsio->kern_sg_entries = 0;
6856
6857	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
6858		data10 = (struct scsi_read_defect_data_hdr_10 *)
6859		    ctsio->kern_data_ptr;
6860		data10->format = format;
6861		scsi_ulto2b(0, data10->length);
6862	} else {
6863		data12 = (struct scsi_read_defect_data_hdr_12 *)
6864		    ctsio->kern_data_ptr;
6865		data12->format = format;
6866		scsi_ulto2b(0, data12->generation);
6867		scsi_ulto4b(0, data12->length);
6868	}
6869
6870	ctl_set_success(ctsio);
6871	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6872	ctsio->be_move_done = ctl_config_move_done;
6873	ctl_datamove((union ctl_io *)ctsio);
6874	return (CTL_RETVAL_COMPLETE);
6875}
6876
6877int
6878ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
6879{
6880	struct scsi_maintenance_in *cdb;
6881	int retval;
6882	int alloc_len, ext, total_len = 0, g, p, pc, pg, gs, os;
6883	int num_target_port_groups, num_target_ports;
6884	struct ctl_lun *lun;
6885	struct ctl_softc *softc;
6886	struct ctl_port *port;
6887	struct scsi_target_group_data *rtg_ptr;
6888	struct scsi_target_group_data_extended *rtg_ext_ptr;
6889	struct scsi_target_port_group_descriptor *tpg_desc;
6890
6891	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
6892
6893	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
6894	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6895	softc = lun->ctl_softc;
6896
6897	retval = CTL_RETVAL_COMPLETE;
6898
6899	switch (cdb->byte2 & STG_PDF_MASK) {
6900	case STG_PDF_LENGTH:
6901		ext = 0;
6902		break;
6903	case STG_PDF_EXTENDED:
6904		ext = 1;
6905		break;
6906	default:
6907		ctl_set_invalid_field(/*ctsio*/ ctsio,
6908				      /*sks_valid*/ 1,
6909				      /*command*/ 1,
6910				      /*field*/ 2,
6911				      /*bit_valid*/ 1,
6912				      /*bit*/ 5);
6913		ctl_done((union ctl_io *)ctsio);
6914		return(retval);
6915	}
6916
6917	if (softc->is_single)
6918		num_target_port_groups = 1;
6919	else
6920		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
6921	num_target_ports = 0;
6922	mtx_lock(&softc->ctl_lock);
6923	STAILQ_FOREACH(port, &softc->port_list, links) {
6924		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
6925			continue;
6926		if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
6927			continue;
6928		num_target_ports++;
6929	}
6930	mtx_unlock(&softc->ctl_lock);
6931
6932	if (ext)
6933		total_len = sizeof(struct scsi_target_group_data_extended);
6934	else
6935		total_len = sizeof(struct scsi_target_group_data);
6936	total_len += sizeof(struct scsi_target_port_group_descriptor) *
6937		num_target_port_groups +
6938	    sizeof(struct scsi_target_port_descriptor) *
6939		num_target_ports * num_target_port_groups;
6940
6941	alloc_len = scsi_4btoul(cdb->length);
6942
6943	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6944
6945	ctsio->kern_sg_entries = 0;
6946
6947	if (total_len < alloc_len) {
6948		ctsio->residual = alloc_len - total_len;
6949		ctsio->kern_data_len = total_len;
6950		ctsio->kern_total_len = total_len;
6951	} else {
6952		ctsio->residual = 0;
6953		ctsio->kern_data_len = alloc_len;
6954		ctsio->kern_total_len = alloc_len;
6955	}
6956	ctsio->kern_data_resid = 0;
6957	ctsio->kern_rel_offset = 0;
6958
6959	if (ext) {
6960		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
6961		    ctsio->kern_data_ptr;
6962		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
6963		rtg_ext_ptr->format_type = 0x10;
6964		rtg_ext_ptr->implicit_transition_time = 0;
6965		tpg_desc = &rtg_ext_ptr->groups[0];
6966	} else {
6967		rtg_ptr = (struct scsi_target_group_data *)
6968		    ctsio->kern_data_ptr;
6969		scsi_ulto4b(total_len - 4, rtg_ptr->length);
6970		tpg_desc = &rtg_ptr->groups[0];
6971	}
6972
6973	mtx_lock(&softc->ctl_lock);
6974	pg = softc->port_offset / CTL_MAX_PORTS;
6975	if (softc->flags & CTL_FLAG_ACTIVE_SHELF) {
6976		if (softc->ha_mode == CTL_HA_MODE_ACT_STBY) {
6977			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
6978			os = TPG_ASYMMETRIC_ACCESS_STANDBY;
6979		} else if (lun->flags & CTL_LUN_PRIMARY_SC) {
6980			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
6981			os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6982		} else {
6983			gs = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6984			os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
6985		}
6986	} else {
6987		gs = TPG_ASYMMETRIC_ACCESS_STANDBY;
6988		os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
6989	}
6990	for (g = 0; g < num_target_port_groups; g++) {
6991		tpg_desc->pref_state = (g == pg) ? gs : os;
6992		tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP;
6993		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
6994		tpg_desc->status = TPG_IMPLICIT;
6995		pc = 0;
6996		STAILQ_FOREACH(port, &softc->port_list, links) {
6997			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
6998				continue;
6999			if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7000				continue;
7001			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7002			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7003			    relative_target_port_identifier);
7004			pc++;
7005		}
7006		tpg_desc->target_port_count = pc;
7007		tpg_desc = (struct scsi_target_port_group_descriptor *)
7008		    &tpg_desc->descriptors[pc];
7009	}
7010	mtx_unlock(&softc->ctl_lock);
7011
7012	ctl_set_success(ctsio);
7013	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7014	ctsio->be_move_done = ctl_config_move_done;
7015	ctl_datamove((union ctl_io *)ctsio);
7016	return(retval);
7017}
7018
7019int
7020ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7021{
7022	struct ctl_lun *lun;
7023	struct scsi_report_supported_opcodes *cdb;
7024	const struct ctl_cmd_entry *entry, *sentry;
7025	struct scsi_report_supported_opcodes_all *all;
7026	struct scsi_report_supported_opcodes_descr *descr;
7027	struct scsi_report_supported_opcodes_one *one;
7028	int retval;
7029	int alloc_len, total_len;
7030	int opcode, service_action, i, j, num;
7031
7032	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7033
7034	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7035	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7036
7037	retval = CTL_RETVAL_COMPLETE;
7038
7039	opcode = cdb->requested_opcode;
7040	service_action = scsi_2btoul(cdb->requested_service_action);
7041	switch (cdb->options & RSO_OPTIONS_MASK) {
7042	case RSO_OPTIONS_ALL:
7043		num = 0;
7044		for (i = 0; i < 256; i++) {
7045			entry = &ctl_cmd_table[i];
7046			if (entry->flags & CTL_CMD_FLAG_SA5) {
7047				for (j = 0; j < 32; j++) {
7048					sentry = &((const struct ctl_cmd_entry *)
7049					    entry->execute)[j];
7050					if (ctl_cmd_applicable(
7051					    lun->be_lun->lun_type, sentry))
7052						num++;
7053				}
7054			} else {
7055				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7056				    entry))
7057					num++;
7058			}
7059		}
7060		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7061		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7062		break;
7063	case RSO_OPTIONS_OC:
7064		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7065			ctl_set_invalid_field(/*ctsio*/ ctsio,
7066					      /*sks_valid*/ 1,
7067					      /*command*/ 1,
7068					      /*field*/ 2,
7069					      /*bit_valid*/ 1,
7070					      /*bit*/ 2);
7071			ctl_done((union ctl_io *)ctsio);
7072			return (CTL_RETVAL_COMPLETE);
7073		}
7074		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7075		break;
7076	case RSO_OPTIONS_OC_SA:
7077		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7078		    service_action >= 32) {
7079			ctl_set_invalid_field(/*ctsio*/ ctsio,
7080					      /*sks_valid*/ 1,
7081					      /*command*/ 1,
7082					      /*field*/ 2,
7083					      /*bit_valid*/ 1,
7084					      /*bit*/ 2);
7085			ctl_done((union ctl_io *)ctsio);
7086			return (CTL_RETVAL_COMPLETE);
7087		}
7088		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7089		break;
7090	default:
7091		ctl_set_invalid_field(/*ctsio*/ ctsio,
7092				      /*sks_valid*/ 1,
7093				      /*command*/ 1,
7094				      /*field*/ 2,
7095				      /*bit_valid*/ 1,
7096				      /*bit*/ 2);
7097		ctl_done((union ctl_io *)ctsio);
7098		return (CTL_RETVAL_COMPLETE);
7099	}
7100
7101	alloc_len = scsi_4btoul(cdb->length);
7102
7103	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7104
7105	ctsio->kern_sg_entries = 0;
7106
7107	if (total_len < alloc_len) {
7108		ctsio->residual = alloc_len - total_len;
7109		ctsio->kern_data_len = total_len;
7110		ctsio->kern_total_len = total_len;
7111	} else {
7112		ctsio->residual = 0;
7113		ctsio->kern_data_len = alloc_len;
7114		ctsio->kern_total_len = alloc_len;
7115	}
7116	ctsio->kern_data_resid = 0;
7117	ctsio->kern_rel_offset = 0;
7118
7119	switch (cdb->options & RSO_OPTIONS_MASK) {
7120	case RSO_OPTIONS_ALL:
7121		all = (struct scsi_report_supported_opcodes_all *)
7122		    ctsio->kern_data_ptr;
7123		num = 0;
7124		for (i = 0; i < 256; i++) {
7125			entry = &ctl_cmd_table[i];
7126			if (entry->flags & CTL_CMD_FLAG_SA5) {
7127				for (j = 0; j < 32; j++) {
7128					sentry = &((const struct ctl_cmd_entry *)
7129					    entry->execute)[j];
7130					if (!ctl_cmd_applicable(
7131					    lun->be_lun->lun_type, sentry))
7132						continue;
7133					descr = &all->descr[num++];
7134					descr->opcode = i;
7135					scsi_ulto2b(j, descr->service_action);
7136					descr->flags = RSO_SERVACTV;
7137					scsi_ulto2b(sentry->length,
7138					    descr->cdb_length);
7139				}
7140			} else {
7141				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7142				    entry))
7143					continue;
7144				descr = &all->descr[num++];
7145				descr->opcode = i;
7146				scsi_ulto2b(0, descr->service_action);
7147				descr->flags = 0;
7148				scsi_ulto2b(entry->length, descr->cdb_length);
7149			}
7150		}
7151		scsi_ulto4b(
7152		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7153		    all->length);
7154		break;
7155	case RSO_OPTIONS_OC:
7156		one = (struct scsi_report_supported_opcodes_one *)
7157		    ctsio->kern_data_ptr;
7158		entry = &ctl_cmd_table[opcode];
7159		goto fill_one;
7160	case RSO_OPTIONS_OC_SA:
7161		one = (struct scsi_report_supported_opcodes_one *)
7162		    ctsio->kern_data_ptr;
7163		entry = &ctl_cmd_table[opcode];
7164		entry = &((const struct ctl_cmd_entry *)
7165		    entry->execute)[service_action];
7166fill_one:
7167		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7168			one->support = 3;
7169			scsi_ulto2b(entry->length, one->cdb_length);
7170			one->cdb_usage[0] = opcode;
7171			memcpy(&one->cdb_usage[1], entry->usage,
7172			    entry->length - 1);
7173		} else
7174			one->support = 1;
7175		break;
7176	}
7177
7178	ctl_set_success(ctsio);
7179	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7180	ctsio->be_move_done = ctl_config_move_done;
7181	ctl_datamove((union ctl_io *)ctsio);
7182	return(retval);
7183}
7184
7185int
7186ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7187{
7188	struct scsi_report_supported_tmf *cdb;
7189	struct scsi_report_supported_tmf_data *data;
7190	int retval;
7191	int alloc_len, total_len;
7192
7193	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7194
7195	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7196
7197	retval = CTL_RETVAL_COMPLETE;
7198
7199	total_len = sizeof(struct scsi_report_supported_tmf_data);
7200	alloc_len = scsi_4btoul(cdb->length);
7201
7202	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7203
7204	ctsio->kern_sg_entries = 0;
7205
7206	if (total_len < alloc_len) {
7207		ctsio->residual = alloc_len - total_len;
7208		ctsio->kern_data_len = total_len;
7209		ctsio->kern_total_len = total_len;
7210	} else {
7211		ctsio->residual = 0;
7212		ctsio->kern_data_len = alloc_len;
7213		ctsio->kern_total_len = alloc_len;
7214	}
7215	ctsio->kern_data_resid = 0;
7216	ctsio->kern_rel_offset = 0;
7217
7218	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7219	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7220	data->byte2 |= RST_ITNRS;
7221
7222	ctl_set_success(ctsio);
7223	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7224	ctsio->be_move_done = ctl_config_move_done;
7225	ctl_datamove((union ctl_io *)ctsio);
7226	return (retval);
7227}
7228
7229int
7230ctl_report_timestamp(struct ctl_scsiio *ctsio)
7231{
7232	struct scsi_report_timestamp *cdb;
7233	struct scsi_report_timestamp_data *data;
7234	struct timeval tv;
7235	int64_t timestamp;
7236	int retval;
7237	int alloc_len, total_len;
7238
7239	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7240
7241	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7242
7243	retval = CTL_RETVAL_COMPLETE;
7244
7245	total_len = sizeof(struct scsi_report_timestamp_data);
7246	alloc_len = scsi_4btoul(cdb->length);
7247
7248	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7249
7250	ctsio->kern_sg_entries = 0;
7251
7252	if (total_len < alloc_len) {
7253		ctsio->residual = alloc_len - total_len;
7254		ctsio->kern_data_len = total_len;
7255		ctsio->kern_total_len = total_len;
7256	} else {
7257		ctsio->residual = 0;
7258		ctsio->kern_data_len = alloc_len;
7259		ctsio->kern_total_len = alloc_len;
7260	}
7261	ctsio->kern_data_resid = 0;
7262	ctsio->kern_rel_offset = 0;
7263
7264	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7265	scsi_ulto2b(sizeof(*data) - 2, data->length);
7266	data->origin = RTS_ORIG_OUTSIDE;
7267	getmicrotime(&tv);
7268	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7269	scsi_ulto4b(timestamp >> 16, data->timestamp);
7270	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7271
7272	ctl_set_success(ctsio);
7273	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7274	ctsio->be_move_done = ctl_config_move_done;
7275	ctl_datamove((union ctl_io *)ctsio);
7276	return (retval);
7277}
7278
7279int
7280ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7281{
7282	struct scsi_per_res_in *cdb;
7283	int alloc_len, total_len = 0;
7284	/* struct scsi_per_res_in_rsrv in_data; */
7285	struct ctl_lun *lun;
7286	struct ctl_softc *softc;
7287	uint64_t key;
7288
7289	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7290
7291	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7292
7293	alloc_len = scsi_2btoul(cdb->length);
7294
7295	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7296	softc = lun->ctl_softc;
7297
7298retry:
7299	mtx_lock(&lun->lun_lock);
7300	switch (cdb->action) {
7301	case SPRI_RK: /* read keys */
7302		total_len = sizeof(struct scsi_per_res_in_keys) +
7303			lun->pr_key_count *
7304			sizeof(struct scsi_per_res_key);
7305		break;
7306	case SPRI_RR: /* read reservation */
7307		if (lun->flags & CTL_LUN_PR_RESERVED)
7308			total_len = sizeof(struct scsi_per_res_in_rsrv);
7309		else
7310			total_len = sizeof(struct scsi_per_res_in_header);
7311		break;
7312	case SPRI_RC: /* report capabilities */
7313		total_len = sizeof(struct scsi_per_res_cap);
7314		break;
7315	case SPRI_RS: /* read full status */
7316		total_len = sizeof(struct scsi_per_res_in_header) +
7317		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7318		    lun->pr_key_count;
7319		break;
7320	default:
7321		panic("Invalid PR type %x", cdb->action);
7322	}
7323	mtx_unlock(&lun->lun_lock);
7324
7325	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7326
7327	if (total_len < alloc_len) {
7328		ctsio->residual = alloc_len - total_len;
7329		ctsio->kern_data_len = total_len;
7330		ctsio->kern_total_len = total_len;
7331	} else {
7332		ctsio->residual = 0;
7333		ctsio->kern_data_len = alloc_len;
7334		ctsio->kern_total_len = alloc_len;
7335	}
7336
7337	ctsio->kern_data_resid = 0;
7338	ctsio->kern_rel_offset = 0;
7339	ctsio->kern_sg_entries = 0;
7340
7341	mtx_lock(&lun->lun_lock);
7342	switch (cdb->action) {
7343	case SPRI_RK: { // read keys
7344        struct scsi_per_res_in_keys *res_keys;
7345		int i, key_count;
7346
7347		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7348
7349		/*
7350		 * We had to drop the lock to allocate our buffer, which
7351		 * leaves time for someone to come in with another
7352		 * persistent reservation.  (That is unlikely, though,
7353		 * since this should be the only persistent reservation
7354		 * command active right now.)
7355		 */
7356		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7357		    (lun->pr_key_count *
7358		     sizeof(struct scsi_per_res_key)))){
7359			mtx_unlock(&lun->lun_lock);
7360			free(ctsio->kern_data_ptr, M_CTL);
7361			printf("%s: reservation length changed, retrying\n",
7362			       __func__);
7363			goto retry;
7364		}
7365
7366		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7367
7368		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7369			     lun->pr_key_count, res_keys->header.length);
7370
7371		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7372			if ((key = ctl_get_prkey(lun, i)) == 0)
7373				continue;
7374
7375			/*
7376			 * We used lun->pr_key_count to calculate the
7377			 * size to allocate.  If it turns out the number of
7378			 * initiators with the registered flag set is
7379			 * larger than that (i.e. they haven't been kept in
7380			 * sync), we've got a problem.
7381			 */
7382			if (key_count >= lun->pr_key_count) {
7383#ifdef NEEDTOPORT
7384				csevent_log(CSC_CTL | CSC_SHELF_SW |
7385					    CTL_PR_ERROR,
7386					    csevent_LogType_Fault,
7387					    csevent_AlertLevel_Yellow,
7388					    csevent_FRU_ShelfController,
7389					    csevent_FRU_Firmware,
7390				        csevent_FRU_Unknown,
7391					    "registered keys %d >= key "
7392					    "count %d", key_count,
7393					    lun->pr_key_count);
7394#endif
7395				key_count++;
7396				continue;
7397			}
7398			scsi_u64to8b(key, res_keys->keys[key_count].key);
7399			key_count++;
7400		}
7401		break;
7402	}
7403	case SPRI_RR: { // read reservation
7404		struct scsi_per_res_in_rsrv *res;
7405		int tmp_len, header_only;
7406
7407		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7408
7409		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7410
7411		if (lun->flags & CTL_LUN_PR_RESERVED)
7412		{
7413			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7414			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7415				    res->header.length);
7416			header_only = 0;
7417		} else {
7418			tmp_len = sizeof(struct scsi_per_res_in_header);
7419			scsi_ulto4b(0, res->header.length);
7420			header_only = 1;
7421		}
7422
7423		/*
7424		 * We had to drop the lock to allocate our buffer, which
7425		 * leaves time for someone to come in with another
7426		 * persistent reservation.  (That is unlikely, though,
7427		 * since this should be the only persistent reservation
7428		 * command active right now.)
7429		 */
7430		if (tmp_len != total_len) {
7431			mtx_unlock(&lun->lun_lock);
7432			free(ctsio->kern_data_ptr, M_CTL);
7433			printf("%s: reservation status changed, retrying\n",
7434			       __func__);
7435			goto retry;
7436		}
7437
7438		/*
7439		 * No reservation held, so we're done.
7440		 */
7441		if (header_only != 0)
7442			break;
7443
7444		/*
7445		 * If the registration is an All Registrants type, the key
7446		 * is 0, since it doesn't really matter.
7447		 */
7448		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7449			scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7450			    res->data.reservation);
7451		}
7452		res->data.scopetype = lun->res_type;
7453		break;
7454	}
7455	case SPRI_RC:     //report capabilities
7456	{
7457		struct scsi_per_res_cap *res_cap;
7458		uint16_t type_mask;
7459
7460		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7461		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7462		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7463		type_mask = SPRI_TM_WR_EX_AR |
7464			    SPRI_TM_EX_AC_RO |
7465			    SPRI_TM_WR_EX_RO |
7466			    SPRI_TM_EX_AC |
7467			    SPRI_TM_WR_EX |
7468			    SPRI_TM_EX_AC_AR;
7469		scsi_ulto2b(type_mask, res_cap->type_mask);
7470		break;
7471	}
7472	case SPRI_RS: { // read full status
7473		struct scsi_per_res_in_full *res_status;
7474		struct scsi_per_res_in_full_desc *res_desc;
7475		struct ctl_port *port;
7476		int i, len;
7477
7478		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7479
7480		/*
7481		 * We had to drop the lock to allocate our buffer, which
7482		 * leaves time for someone to come in with another
7483		 * persistent reservation.  (That is unlikely, though,
7484		 * since this should be the only persistent reservation
7485		 * command active right now.)
7486		 */
7487		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7488		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7489		     lun->pr_key_count)){
7490			mtx_unlock(&lun->lun_lock);
7491			free(ctsio->kern_data_ptr, M_CTL);
7492			printf("%s: reservation length changed, retrying\n",
7493			       __func__);
7494			goto retry;
7495		}
7496
7497		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7498
7499		res_desc = &res_status->desc[0];
7500		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7501			if ((key = ctl_get_prkey(lun, i)) == 0)
7502				continue;
7503
7504			scsi_u64to8b(key, res_desc->res_key.key);
7505			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7506			    (lun->pr_res_idx == i ||
7507			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7508				res_desc->flags = SPRI_FULL_R_HOLDER;
7509				res_desc->scopetype = lun->res_type;
7510			}
7511			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7512			    res_desc->rel_trgt_port_id);
7513			len = 0;
7514			port = softc->ctl_ports[
7515			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7516			if (port != NULL)
7517				len = ctl_create_iid(port,
7518				    i % CTL_MAX_INIT_PER_PORT,
7519				    res_desc->transport_id);
7520			scsi_ulto4b(len, res_desc->additional_length);
7521			res_desc = (struct scsi_per_res_in_full_desc *)
7522			    &res_desc->transport_id[len];
7523		}
7524		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7525		    res_status->header.length);
7526		break;
7527	}
7528	default:
7529		/*
7530		 * This is a bug, because we just checked for this above,
7531		 * and should have returned an error.
7532		 */
7533		panic("Invalid PR type %x", cdb->action);
7534		break; /* NOTREACHED */
7535	}
7536	mtx_unlock(&lun->lun_lock);
7537
7538	ctl_set_success(ctsio);
7539	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7540	ctsio->be_move_done = ctl_config_move_done;
7541	ctl_datamove((union ctl_io *)ctsio);
7542	return (CTL_RETVAL_COMPLETE);
7543}
7544
7545static void
7546ctl_est_res_ua(struct ctl_lun *lun, uint32_t residx, ctl_ua_type ua)
7547{
7548	int off = lun->ctl_softc->persis_offset;
7549
7550	if (residx >= off && residx < off + CTL_MAX_INITIATORS)
7551		ctl_est_ua(lun, residx - off, ua);
7552}
7553
7554/*
7555 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7556 * it should return.
7557 */
7558static int
7559ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7560		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7561		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7562		struct scsi_per_res_out_parms* param)
7563{
7564	union ctl_ha_msg persis_io;
7565	int retval, i;
7566	int isc_retval;
7567
7568	retval = 0;
7569
7570	mtx_lock(&lun->lun_lock);
7571	if (sa_res_key == 0) {
7572		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7573			/* validate scope and type */
7574			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7575			     SPR_LU_SCOPE) {
7576				mtx_unlock(&lun->lun_lock);
7577				ctl_set_invalid_field(/*ctsio*/ ctsio,
7578						      /*sks_valid*/ 1,
7579						      /*command*/ 1,
7580						      /*field*/ 2,
7581						      /*bit_valid*/ 1,
7582						      /*bit*/ 4);
7583				ctl_done((union ctl_io *)ctsio);
7584				return (1);
7585			}
7586
7587		        if (type>8 || type==2 || type==4 || type==0) {
7588				mtx_unlock(&lun->lun_lock);
7589				ctl_set_invalid_field(/*ctsio*/ ctsio,
7590       	           				      /*sks_valid*/ 1,
7591						      /*command*/ 1,
7592						      /*field*/ 2,
7593						      /*bit_valid*/ 1,
7594						      /*bit*/ 0);
7595				ctl_done((union ctl_io *)ctsio);
7596				return (1);
7597		        }
7598
7599			/*
7600			 * Unregister everybody else and build UA for
7601			 * them
7602			 */
7603			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7604				if (i == residx || ctl_get_prkey(lun, i) == 0)
7605					continue;
7606
7607				ctl_clr_prkey(lun, i);
7608				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7609			}
7610			lun->pr_key_count = 1;
7611			lun->res_type = type;
7612			if (lun->res_type != SPR_TYPE_WR_EX_AR
7613			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7614				lun->pr_res_idx = residx;
7615
7616			/* send msg to other side */
7617			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7618			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7619			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7620			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7621			persis_io.pr.pr_info.res_type = type;
7622			memcpy(persis_io.pr.pr_info.sa_res_key,
7623			       param->serv_act_res_key,
7624			       sizeof(param->serv_act_res_key));
7625			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7626			     &persis_io, sizeof(persis_io), 0)) >
7627			     CTL_HA_STATUS_SUCCESS) {
7628				printf("CTL:Persis Out error returned "
7629				       "from ctl_ha_msg_send %d\n",
7630				       isc_retval);
7631			}
7632		} else {
7633			/* not all registrants */
7634			mtx_unlock(&lun->lun_lock);
7635			free(ctsio->kern_data_ptr, M_CTL);
7636			ctl_set_invalid_field(ctsio,
7637					      /*sks_valid*/ 1,
7638					      /*command*/ 0,
7639					      /*field*/ 8,
7640					      /*bit_valid*/ 0,
7641					      /*bit*/ 0);
7642			ctl_done((union ctl_io *)ctsio);
7643			return (1);
7644		}
7645	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7646		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7647		int found = 0;
7648
7649		if (res_key == sa_res_key) {
7650			/* special case */
7651			/*
7652			 * The spec implies this is not good but doesn't
7653			 * say what to do. There are two choices either
7654			 * generate a res conflict or check condition
7655			 * with illegal field in parameter data. Since
7656			 * that is what is done when the sa_res_key is
7657			 * zero I'll take that approach since this has
7658			 * to do with the sa_res_key.
7659			 */
7660			mtx_unlock(&lun->lun_lock);
7661			free(ctsio->kern_data_ptr, M_CTL);
7662			ctl_set_invalid_field(ctsio,
7663					      /*sks_valid*/ 1,
7664					      /*command*/ 0,
7665					      /*field*/ 8,
7666					      /*bit_valid*/ 0,
7667					      /*bit*/ 0);
7668			ctl_done((union ctl_io *)ctsio);
7669			return (1);
7670		}
7671
7672		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7673			if (ctl_get_prkey(lun, i) != sa_res_key)
7674				continue;
7675
7676			found = 1;
7677			ctl_clr_prkey(lun, i);
7678			lun->pr_key_count--;
7679			ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7680		}
7681		if (!found) {
7682			mtx_unlock(&lun->lun_lock);
7683			free(ctsio->kern_data_ptr, M_CTL);
7684			ctl_set_reservation_conflict(ctsio);
7685			ctl_done((union ctl_io *)ctsio);
7686			return (CTL_RETVAL_COMPLETE);
7687		}
7688		/* send msg to other side */
7689		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7690		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7691		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7692		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7693		persis_io.pr.pr_info.res_type = type;
7694		memcpy(persis_io.pr.pr_info.sa_res_key,
7695		       param->serv_act_res_key,
7696		       sizeof(param->serv_act_res_key));
7697		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7698		     &persis_io, sizeof(persis_io), 0)) >
7699		     CTL_HA_STATUS_SUCCESS) {
7700			printf("CTL:Persis Out error returned from "
7701			       "ctl_ha_msg_send %d\n", isc_retval);
7702		}
7703	} else {
7704		/* Reserved but not all registrants */
7705		/* sa_res_key is res holder */
7706		if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7707			/* validate scope and type */
7708			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7709			     SPR_LU_SCOPE) {
7710				mtx_unlock(&lun->lun_lock);
7711				ctl_set_invalid_field(/*ctsio*/ ctsio,
7712						      /*sks_valid*/ 1,
7713						      /*command*/ 1,
7714						      /*field*/ 2,
7715						      /*bit_valid*/ 1,
7716						      /*bit*/ 4);
7717				ctl_done((union ctl_io *)ctsio);
7718				return (1);
7719			}
7720
7721			if (type>8 || type==2 || type==4 || type==0) {
7722				mtx_unlock(&lun->lun_lock);
7723				ctl_set_invalid_field(/*ctsio*/ ctsio,
7724						      /*sks_valid*/ 1,
7725						      /*command*/ 1,
7726						      /*field*/ 2,
7727						      /*bit_valid*/ 1,
7728						      /*bit*/ 0);
7729				ctl_done((union ctl_io *)ctsio);
7730				return (1);
7731			}
7732
7733			/*
7734			 * Do the following:
7735			 * if sa_res_key != res_key remove all
7736			 * registrants w/sa_res_key and generate UA
7737			 * for these registrants(Registrations
7738			 * Preempted) if it wasn't an exclusive
7739			 * reservation generate UA(Reservations
7740			 * Preempted) for all other registered nexuses
7741			 * if the type has changed. Establish the new
7742			 * reservation and holder. If res_key and
7743			 * sa_res_key are the same do the above
7744			 * except don't unregister the res holder.
7745			 */
7746
7747			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7748				if (i == residx || ctl_get_prkey(lun, i) == 0)
7749					continue;
7750
7751				if (sa_res_key == ctl_get_prkey(lun, i)) {
7752					ctl_clr_prkey(lun, i);
7753					lun->pr_key_count--;
7754					ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7755				} else if (type != lun->res_type
7756					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7757					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7758					ctl_est_res_ua(lun, i, CTL_UA_RES_RELEASE);
7759				}
7760			}
7761			lun->res_type = type;
7762			if (lun->res_type != SPR_TYPE_WR_EX_AR
7763			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7764				lun->pr_res_idx = residx;
7765			else
7766				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7767
7768			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7769			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7770			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7771			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7772			persis_io.pr.pr_info.res_type = type;
7773			memcpy(persis_io.pr.pr_info.sa_res_key,
7774			       param->serv_act_res_key,
7775			       sizeof(param->serv_act_res_key));
7776			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7777			     &persis_io, sizeof(persis_io), 0)) >
7778			     CTL_HA_STATUS_SUCCESS) {
7779				printf("CTL:Persis Out error returned "
7780				       "from ctl_ha_msg_send %d\n",
7781				       isc_retval);
7782			}
7783		} else {
7784			/*
7785			 * sa_res_key is not the res holder just
7786			 * remove registrants
7787			 */
7788			int found=0;
7789
7790			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7791				if (sa_res_key != ctl_get_prkey(lun, i))
7792					continue;
7793
7794				found = 1;
7795				ctl_clr_prkey(lun, i);
7796				lun->pr_key_count--;
7797				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7798			}
7799
7800			if (!found) {
7801				mtx_unlock(&lun->lun_lock);
7802				free(ctsio->kern_data_ptr, M_CTL);
7803				ctl_set_reservation_conflict(ctsio);
7804				ctl_done((union ctl_io *)ctsio);
7805		        	return (1);
7806			}
7807			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7808			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7809			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7810			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7811			persis_io.pr.pr_info.res_type = type;
7812			memcpy(persis_io.pr.pr_info.sa_res_key,
7813			       param->serv_act_res_key,
7814			       sizeof(param->serv_act_res_key));
7815			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7816			     &persis_io, sizeof(persis_io), 0)) >
7817			     CTL_HA_STATUS_SUCCESS) {
7818				printf("CTL:Persis Out error returned "
7819				       "from ctl_ha_msg_send %d\n",
7820				isc_retval);
7821			}
7822		}
7823	}
7824
7825	lun->PRGeneration++;
7826	mtx_unlock(&lun->lun_lock);
7827
7828	return (retval);
7829}
7830
7831static void
7832ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7833{
7834	uint64_t sa_res_key;
7835	int i;
7836
7837	sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7838
7839	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7840	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7841	 || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
7842		if (sa_res_key == 0) {
7843			/*
7844			 * Unregister everybody else and build UA for
7845			 * them
7846			 */
7847			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7848				if (i == msg->pr.pr_info.residx ||
7849				    ctl_get_prkey(lun, i) == 0)
7850					continue;
7851
7852				ctl_clr_prkey(lun, i);
7853				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7854			}
7855
7856			lun->pr_key_count = 1;
7857			lun->res_type = msg->pr.pr_info.res_type;
7858			if (lun->res_type != SPR_TYPE_WR_EX_AR
7859			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7860				lun->pr_res_idx = msg->pr.pr_info.residx;
7861		} else {
7862		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7863				if (sa_res_key == ctl_get_prkey(lun, i))
7864					continue;
7865
7866				ctl_clr_prkey(lun, i);
7867				lun->pr_key_count--;
7868				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7869			}
7870		}
7871	} else {
7872		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7873			if (i == msg->pr.pr_info.residx ||
7874			    ctl_get_prkey(lun, i) == 0)
7875				continue;
7876
7877			if (sa_res_key == ctl_get_prkey(lun, i)) {
7878				ctl_clr_prkey(lun, i);
7879				lun->pr_key_count--;
7880				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7881			} else if (msg->pr.pr_info.res_type != lun->res_type
7882				&& (lun->res_type == SPR_TYPE_WR_EX_RO
7883				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7884				ctl_est_res_ua(lun, i, CTL_UA_RES_RELEASE);
7885			}
7886		}
7887		lun->res_type = msg->pr.pr_info.res_type;
7888		if (lun->res_type != SPR_TYPE_WR_EX_AR
7889		 && lun->res_type != SPR_TYPE_EX_AC_AR)
7890			lun->pr_res_idx = msg->pr.pr_info.residx;
7891		else
7892			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7893	}
7894	lun->PRGeneration++;
7895
7896}
7897
7898
7899int
7900ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
7901{
7902	int retval;
7903	int isc_retval;
7904	u_int32_t param_len;
7905	struct scsi_per_res_out *cdb;
7906	struct ctl_lun *lun;
7907	struct scsi_per_res_out_parms* param;
7908	struct ctl_softc *softc;
7909	uint32_t residx;
7910	uint64_t res_key, sa_res_key, key;
7911	uint8_t type;
7912	union ctl_ha_msg persis_io;
7913	int    i;
7914
7915	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
7916
7917	retval = CTL_RETVAL_COMPLETE;
7918
7919	cdb = (struct scsi_per_res_out *)ctsio->cdb;
7920	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7921	softc = lun->ctl_softc;
7922
7923	/*
7924	 * We only support whole-LUN scope.  The scope & type are ignored for
7925	 * register, register and ignore existing key and clear.
7926	 * We sometimes ignore scope and type on preempts too!!
7927	 * Verify reservation type here as well.
7928	 */
7929	type = cdb->scope_type & SPR_TYPE_MASK;
7930	if ((cdb->action == SPRO_RESERVE)
7931	 || (cdb->action == SPRO_RELEASE)) {
7932		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
7933			ctl_set_invalid_field(/*ctsio*/ ctsio,
7934					      /*sks_valid*/ 1,
7935					      /*command*/ 1,
7936					      /*field*/ 2,
7937					      /*bit_valid*/ 1,
7938					      /*bit*/ 4);
7939			ctl_done((union ctl_io *)ctsio);
7940			return (CTL_RETVAL_COMPLETE);
7941		}
7942
7943		if (type>8 || type==2 || type==4 || type==0) {
7944			ctl_set_invalid_field(/*ctsio*/ ctsio,
7945					      /*sks_valid*/ 1,
7946					      /*command*/ 1,
7947					      /*field*/ 2,
7948					      /*bit_valid*/ 1,
7949					      /*bit*/ 0);
7950			ctl_done((union ctl_io *)ctsio);
7951			return (CTL_RETVAL_COMPLETE);
7952		}
7953	}
7954
7955	param_len = scsi_4btoul(cdb->length);
7956
7957	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
7958		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
7959		ctsio->kern_data_len = param_len;
7960		ctsio->kern_total_len = param_len;
7961		ctsio->kern_data_resid = 0;
7962		ctsio->kern_rel_offset = 0;
7963		ctsio->kern_sg_entries = 0;
7964		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7965		ctsio->be_move_done = ctl_config_move_done;
7966		ctl_datamove((union ctl_io *)ctsio);
7967
7968		return (CTL_RETVAL_COMPLETE);
7969	}
7970
7971	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
7972
7973	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
7974	res_key = scsi_8btou64(param->res_key.key);
7975	sa_res_key = scsi_8btou64(param->serv_act_res_key);
7976
7977	/*
7978	 * Validate the reservation key here except for SPRO_REG_IGNO
7979	 * This must be done for all other service actions
7980	 */
7981	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
7982		mtx_lock(&lun->lun_lock);
7983		if ((key = ctl_get_prkey(lun, residx)) != 0) {
7984			if (res_key != key) {
7985				/*
7986				 * The current key passed in doesn't match
7987				 * the one the initiator previously
7988				 * registered.
7989				 */
7990				mtx_unlock(&lun->lun_lock);
7991				free(ctsio->kern_data_ptr, M_CTL);
7992				ctl_set_reservation_conflict(ctsio);
7993				ctl_done((union ctl_io *)ctsio);
7994				return (CTL_RETVAL_COMPLETE);
7995			}
7996		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
7997			/*
7998			 * We are not registered
7999			 */
8000			mtx_unlock(&lun->lun_lock);
8001			free(ctsio->kern_data_ptr, M_CTL);
8002			ctl_set_reservation_conflict(ctsio);
8003			ctl_done((union ctl_io *)ctsio);
8004			return (CTL_RETVAL_COMPLETE);
8005		} else if (res_key != 0) {
8006			/*
8007			 * We are not registered and trying to register but
8008			 * the register key isn't zero.
8009			 */
8010			mtx_unlock(&lun->lun_lock);
8011			free(ctsio->kern_data_ptr, M_CTL);
8012			ctl_set_reservation_conflict(ctsio);
8013			ctl_done((union ctl_io *)ctsio);
8014			return (CTL_RETVAL_COMPLETE);
8015		}
8016		mtx_unlock(&lun->lun_lock);
8017	}
8018
8019	switch (cdb->action & SPRO_ACTION_MASK) {
8020	case SPRO_REGISTER:
8021	case SPRO_REG_IGNO: {
8022
8023#if 0
8024		printf("Registration received\n");
8025#endif
8026
8027		/*
8028		 * We don't support any of these options, as we report in
8029		 * the read capabilities request (see
8030		 * ctl_persistent_reserve_in(), above).
8031		 */
8032		if ((param->flags & SPR_SPEC_I_PT)
8033		 || (param->flags & SPR_ALL_TG_PT)
8034		 || (param->flags & SPR_APTPL)) {
8035			int bit_ptr;
8036
8037			if (param->flags & SPR_APTPL)
8038				bit_ptr = 0;
8039			else if (param->flags & SPR_ALL_TG_PT)
8040				bit_ptr = 2;
8041			else /* SPR_SPEC_I_PT */
8042				bit_ptr = 3;
8043
8044			free(ctsio->kern_data_ptr, M_CTL);
8045			ctl_set_invalid_field(ctsio,
8046					      /*sks_valid*/ 1,
8047					      /*command*/ 0,
8048					      /*field*/ 20,
8049					      /*bit_valid*/ 1,
8050					      /*bit*/ bit_ptr);
8051			ctl_done((union ctl_io *)ctsio);
8052			return (CTL_RETVAL_COMPLETE);
8053		}
8054
8055		mtx_lock(&lun->lun_lock);
8056
8057		/*
8058		 * The initiator wants to clear the
8059		 * key/unregister.
8060		 */
8061		if (sa_res_key == 0) {
8062			if ((res_key == 0
8063			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8064			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8065			  && ctl_get_prkey(lun, residx) == 0)) {
8066				mtx_unlock(&lun->lun_lock);
8067				goto done;
8068			}
8069
8070			ctl_clr_prkey(lun, residx);
8071			lun->pr_key_count--;
8072
8073			if (residx == lun->pr_res_idx) {
8074				lun->flags &= ~CTL_LUN_PR_RESERVED;
8075				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8076
8077				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8078				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8079				 && lun->pr_key_count) {
8080					/*
8081					 * If the reservation is a registrants
8082					 * only type we need to generate a UA
8083					 * for other registered inits.  The
8084					 * sense code should be RESERVATIONS
8085					 * RELEASED
8086					 */
8087
8088					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8089						if (ctl_get_prkey(lun, i +
8090						    softc->persis_offset) == 0)
8091							continue;
8092						ctl_est_ua(lun, i,
8093						    CTL_UA_RES_RELEASE);
8094					}
8095				}
8096				lun->res_type = 0;
8097			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8098				if (lun->pr_key_count==0) {
8099					lun->flags &= ~CTL_LUN_PR_RESERVED;
8100					lun->res_type = 0;
8101					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8102				}
8103			}
8104			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8105			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8106			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8107			persis_io.pr.pr_info.residx = residx;
8108			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8109			     &persis_io, sizeof(persis_io), 0 )) >
8110			     CTL_HA_STATUS_SUCCESS) {
8111				printf("CTL:Persis Out error returned from "
8112				       "ctl_ha_msg_send %d\n", isc_retval);
8113			}
8114		} else /* sa_res_key != 0 */ {
8115
8116			/*
8117			 * If we aren't registered currently then increment
8118			 * the key count and set the registered flag.
8119			 */
8120			ctl_alloc_prkey(lun, residx);
8121			if (ctl_get_prkey(lun, residx) == 0)
8122				lun->pr_key_count++;
8123			ctl_set_prkey(lun, residx, sa_res_key);
8124
8125			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8126			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8127			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8128			persis_io.pr.pr_info.residx = residx;
8129			memcpy(persis_io.pr.pr_info.sa_res_key,
8130			       param->serv_act_res_key,
8131			       sizeof(param->serv_act_res_key));
8132			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8133			     &persis_io, sizeof(persis_io), 0)) >
8134			     CTL_HA_STATUS_SUCCESS) {
8135				printf("CTL:Persis Out error returned from "
8136				       "ctl_ha_msg_send %d\n", isc_retval);
8137			}
8138		}
8139		lun->PRGeneration++;
8140		mtx_unlock(&lun->lun_lock);
8141
8142		break;
8143	}
8144	case SPRO_RESERVE:
8145#if 0
8146                printf("Reserve executed type %d\n", type);
8147#endif
8148		mtx_lock(&lun->lun_lock);
8149		if (lun->flags & CTL_LUN_PR_RESERVED) {
8150			/*
8151			 * if this isn't the reservation holder and it's
8152			 * not a "all registrants" type or if the type is
8153			 * different then we have a conflict
8154			 */
8155			if ((lun->pr_res_idx != residx
8156			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8157			 || lun->res_type != type) {
8158				mtx_unlock(&lun->lun_lock);
8159				free(ctsio->kern_data_ptr, M_CTL);
8160				ctl_set_reservation_conflict(ctsio);
8161				ctl_done((union ctl_io *)ctsio);
8162				return (CTL_RETVAL_COMPLETE);
8163			}
8164			mtx_unlock(&lun->lun_lock);
8165		} else /* create a reservation */ {
8166			/*
8167			 * If it's not an "all registrants" type record
8168			 * reservation holder
8169			 */
8170			if (type != SPR_TYPE_WR_EX_AR
8171			 && type != SPR_TYPE_EX_AC_AR)
8172				lun->pr_res_idx = residx; /* Res holder */
8173			else
8174				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8175
8176			lun->flags |= CTL_LUN_PR_RESERVED;
8177			lun->res_type = type;
8178
8179			mtx_unlock(&lun->lun_lock);
8180
8181			/* send msg to other side */
8182			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8183			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8184			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8185			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8186			persis_io.pr.pr_info.res_type = type;
8187			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8188			     &persis_io, sizeof(persis_io), 0)) >
8189			     CTL_HA_STATUS_SUCCESS) {
8190				printf("CTL:Persis Out error returned from "
8191				       "ctl_ha_msg_send %d\n", isc_retval);
8192			}
8193		}
8194		break;
8195
8196	case SPRO_RELEASE:
8197		mtx_lock(&lun->lun_lock);
8198		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8199			/* No reservation exists return good status */
8200			mtx_unlock(&lun->lun_lock);
8201			goto done;
8202		}
8203		/*
8204		 * Is this nexus a reservation holder?
8205		 */
8206		if (lun->pr_res_idx != residx
8207		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8208			/*
8209			 * not a res holder return good status but
8210			 * do nothing
8211			 */
8212			mtx_unlock(&lun->lun_lock);
8213			goto done;
8214		}
8215
8216		if (lun->res_type != type) {
8217			mtx_unlock(&lun->lun_lock);
8218			free(ctsio->kern_data_ptr, M_CTL);
8219			ctl_set_illegal_pr_release(ctsio);
8220			ctl_done((union ctl_io *)ctsio);
8221			return (CTL_RETVAL_COMPLETE);
8222		}
8223
8224		/* okay to release */
8225		lun->flags &= ~CTL_LUN_PR_RESERVED;
8226		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8227		lun->res_type = 0;
8228
8229		/*
8230		 * if this isn't an exclusive access
8231		 * res generate UA for all other
8232		 * registrants.
8233		 */
8234		if (type != SPR_TYPE_EX_AC
8235		 && type != SPR_TYPE_WR_EX) {
8236			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8237				if (i == residx ||
8238				    ctl_get_prkey(lun,
8239				     i + softc->persis_offset) == 0)
8240					continue;
8241				ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8242			}
8243		}
8244		mtx_unlock(&lun->lun_lock);
8245		/* Send msg to other side */
8246		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8247		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8248		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8249		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8250		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8251			printf("CTL:Persis Out error returned from "
8252			       "ctl_ha_msg_send %d\n", isc_retval);
8253		}
8254		break;
8255
8256	case SPRO_CLEAR:
8257		/* send msg to other side */
8258
8259		mtx_lock(&lun->lun_lock);
8260		lun->flags &= ~CTL_LUN_PR_RESERVED;
8261		lun->res_type = 0;
8262		lun->pr_key_count = 0;
8263		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8264
8265		ctl_clr_prkey(lun, residx);
8266		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8267			if (ctl_get_prkey(lun, i) != 0) {
8268				ctl_clr_prkey(lun, i);
8269				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8270			}
8271		lun->PRGeneration++;
8272		mtx_unlock(&lun->lun_lock);
8273		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8274		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8275		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8276		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8277		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8278			printf("CTL:Persis Out error returned from "
8279			       "ctl_ha_msg_send %d\n", isc_retval);
8280		}
8281		break;
8282
8283	case SPRO_PREEMPT:
8284	case SPRO_PRE_ABO: {
8285		int nretval;
8286
8287		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8288					  residx, ctsio, cdb, param);
8289		if (nretval != 0)
8290			return (CTL_RETVAL_COMPLETE);
8291		break;
8292	}
8293	default:
8294		panic("Invalid PR type %x", cdb->action);
8295	}
8296
8297done:
8298	free(ctsio->kern_data_ptr, M_CTL);
8299	ctl_set_success(ctsio);
8300	ctl_done((union ctl_io *)ctsio);
8301
8302	return (retval);
8303}
8304
8305/*
8306 * This routine is for handling a message from the other SC pertaining to
8307 * persistent reserve out. All the error checking will have been done
8308 * so only perorming the action need be done here to keep the two
8309 * in sync.
8310 */
8311static void
8312ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8313{
8314	struct ctl_lun *lun;
8315	struct ctl_softc *softc;
8316	int i;
8317	uint32_t targ_lun;
8318
8319	softc = control_softc;
8320
8321	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8322	lun = softc->ctl_luns[targ_lun];
8323	mtx_lock(&lun->lun_lock);
8324	switch(msg->pr.pr_info.action) {
8325	case CTL_PR_REG_KEY:
8326		ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8327		if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8328			lun->pr_key_count++;
8329		ctl_set_prkey(lun, msg->pr.pr_info.residx,
8330		    scsi_8btou64(msg->pr.pr_info.sa_res_key));
8331		lun->PRGeneration++;
8332		break;
8333
8334	case CTL_PR_UNREG_KEY:
8335		ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8336		lun->pr_key_count--;
8337
8338		/* XXX Need to see if the reservation has been released */
8339		/* if so do we need to generate UA? */
8340		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8341			lun->flags &= ~CTL_LUN_PR_RESERVED;
8342			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8343
8344			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8345			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8346			 && lun->pr_key_count) {
8347				/*
8348				 * If the reservation is a registrants
8349				 * only type we need to generate a UA
8350				 * for other registered inits.  The
8351				 * sense code should be RESERVATIONS
8352				 * RELEASED
8353				 */
8354
8355				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8356					if (ctl_get_prkey(lun, i +
8357					    softc->persis_offset) == 0)
8358						continue;
8359
8360					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8361				}
8362			}
8363			lun->res_type = 0;
8364		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8365			if (lun->pr_key_count==0) {
8366				lun->flags &= ~CTL_LUN_PR_RESERVED;
8367				lun->res_type = 0;
8368				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8369			}
8370		}
8371		lun->PRGeneration++;
8372		break;
8373
8374	case CTL_PR_RESERVE:
8375		lun->flags |= CTL_LUN_PR_RESERVED;
8376		lun->res_type = msg->pr.pr_info.res_type;
8377		lun->pr_res_idx = msg->pr.pr_info.residx;
8378
8379		break;
8380
8381	case CTL_PR_RELEASE:
8382		/*
8383		 * if this isn't an exclusive access res generate UA for all
8384		 * other registrants.
8385		 */
8386		if (lun->res_type != SPR_TYPE_EX_AC
8387		 && lun->res_type != SPR_TYPE_WR_EX) {
8388			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8389				if (ctl_get_prkey(lun, i + softc->persis_offset) != 0)
8390					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8391		}
8392
8393		lun->flags &= ~CTL_LUN_PR_RESERVED;
8394		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8395		lun->res_type = 0;
8396		break;
8397
8398	case CTL_PR_PREEMPT:
8399		ctl_pro_preempt_other(lun, msg);
8400		break;
8401	case CTL_PR_CLEAR:
8402		lun->flags &= ~CTL_LUN_PR_RESERVED;
8403		lun->res_type = 0;
8404		lun->pr_key_count = 0;
8405		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8406
8407		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8408			if (ctl_get_prkey(lun, i) == 0)
8409				continue;
8410			ctl_clr_prkey(lun, i);
8411			ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8412		}
8413		lun->PRGeneration++;
8414		break;
8415	}
8416
8417	mtx_unlock(&lun->lun_lock);
8418}
8419
8420int
8421ctl_read_write(struct ctl_scsiio *ctsio)
8422{
8423	struct ctl_lun *lun;
8424	struct ctl_lba_len_flags *lbalen;
8425	uint64_t lba;
8426	uint32_t num_blocks;
8427	int flags, retval;
8428	int isread;
8429
8430	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8431
8432	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8433
8434	flags = 0;
8435	retval = CTL_RETVAL_COMPLETE;
8436
8437	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8438	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8439	switch (ctsio->cdb[0]) {
8440	case READ_6:
8441	case WRITE_6: {
8442		struct scsi_rw_6 *cdb;
8443
8444		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8445
8446		lba = scsi_3btoul(cdb->addr);
8447		/* only 5 bits are valid in the most significant address byte */
8448		lba &= 0x1fffff;
8449		num_blocks = cdb->length;
8450		/*
8451		 * This is correct according to SBC-2.
8452		 */
8453		if (num_blocks == 0)
8454			num_blocks = 256;
8455		break;
8456	}
8457	case READ_10:
8458	case WRITE_10: {
8459		struct scsi_rw_10 *cdb;
8460
8461		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8462		if (cdb->byte2 & SRW10_FUA)
8463			flags |= CTL_LLF_FUA;
8464		if (cdb->byte2 & SRW10_DPO)
8465			flags |= CTL_LLF_DPO;
8466		lba = scsi_4btoul(cdb->addr);
8467		num_blocks = scsi_2btoul(cdb->length);
8468		break;
8469	}
8470	case WRITE_VERIFY_10: {
8471		struct scsi_write_verify_10 *cdb;
8472
8473		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8474		flags |= CTL_LLF_FUA;
8475		if (cdb->byte2 & SWV_DPO)
8476			flags |= CTL_LLF_DPO;
8477		lba = scsi_4btoul(cdb->addr);
8478		num_blocks = scsi_2btoul(cdb->length);
8479		break;
8480	}
8481	case READ_12:
8482	case WRITE_12: {
8483		struct scsi_rw_12 *cdb;
8484
8485		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8486		if (cdb->byte2 & SRW12_FUA)
8487			flags |= CTL_LLF_FUA;
8488		if (cdb->byte2 & SRW12_DPO)
8489			flags |= CTL_LLF_DPO;
8490		lba = scsi_4btoul(cdb->addr);
8491		num_blocks = scsi_4btoul(cdb->length);
8492		break;
8493	}
8494	case WRITE_VERIFY_12: {
8495		struct scsi_write_verify_12 *cdb;
8496
8497		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8498		flags |= CTL_LLF_FUA;
8499		if (cdb->byte2 & SWV_DPO)
8500			flags |= CTL_LLF_DPO;
8501		lba = scsi_4btoul(cdb->addr);
8502		num_blocks = scsi_4btoul(cdb->length);
8503		break;
8504	}
8505	case READ_16:
8506	case WRITE_16: {
8507		struct scsi_rw_16 *cdb;
8508
8509		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8510		if (cdb->byte2 & SRW12_FUA)
8511			flags |= CTL_LLF_FUA;
8512		if (cdb->byte2 & SRW12_DPO)
8513			flags |= CTL_LLF_DPO;
8514		lba = scsi_8btou64(cdb->addr);
8515		num_blocks = scsi_4btoul(cdb->length);
8516		break;
8517	}
8518	case WRITE_ATOMIC_16: {
8519		struct scsi_rw_16 *cdb;
8520
8521		if (lun->be_lun->atomicblock == 0) {
8522			ctl_set_invalid_opcode(ctsio);
8523			ctl_done((union ctl_io *)ctsio);
8524			return (CTL_RETVAL_COMPLETE);
8525		}
8526
8527		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8528		if (cdb->byte2 & SRW12_FUA)
8529			flags |= CTL_LLF_FUA;
8530		if (cdb->byte2 & SRW12_DPO)
8531			flags |= CTL_LLF_DPO;
8532		lba = scsi_8btou64(cdb->addr);
8533		num_blocks = scsi_4btoul(cdb->length);
8534		if (num_blocks > lun->be_lun->atomicblock) {
8535			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8536			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8537			    /*bit*/ 0);
8538			ctl_done((union ctl_io *)ctsio);
8539			return (CTL_RETVAL_COMPLETE);
8540		}
8541		break;
8542	}
8543	case WRITE_VERIFY_16: {
8544		struct scsi_write_verify_16 *cdb;
8545
8546		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8547		flags |= CTL_LLF_FUA;
8548		if (cdb->byte2 & SWV_DPO)
8549			flags |= CTL_LLF_DPO;
8550		lba = scsi_8btou64(cdb->addr);
8551		num_blocks = scsi_4btoul(cdb->length);
8552		break;
8553	}
8554	default:
8555		/*
8556		 * We got a command we don't support.  This shouldn't
8557		 * happen, commands should be filtered out above us.
8558		 */
8559		ctl_set_invalid_opcode(ctsio);
8560		ctl_done((union ctl_io *)ctsio);
8561
8562		return (CTL_RETVAL_COMPLETE);
8563		break; /* NOTREACHED */
8564	}
8565
8566	/*
8567	 * The first check is to make sure we're in bounds, the second
8568	 * check is to catch wrap-around problems.  If the lba + num blocks
8569	 * is less than the lba, then we've wrapped around and the block
8570	 * range is invalid anyway.
8571	 */
8572	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8573	 || ((lba + num_blocks) < lba)) {
8574		ctl_set_lba_out_of_range(ctsio);
8575		ctl_done((union ctl_io *)ctsio);
8576		return (CTL_RETVAL_COMPLETE);
8577	}
8578
8579	/*
8580	 * According to SBC-3, a transfer length of 0 is not an error.
8581	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8582	 * translates to 256 blocks for those commands.
8583	 */
8584	if (num_blocks == 0) {
8585		ctl_set_success(ctsio);
8586		ctl_done((union ctl_io *)ctsio);
8587		return (CTL_RETVAL_COMPLETE);
8588	}
8589
8590	/* Set FUA and/or DPO if caches are disabled. */
8591	if (isread) {
8592		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8593		    SCP_RCD) != 0)
8594			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8595	} else {
8596		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8597		    SCP_WCE) == 0)
8598			flags |= CTL_LLF_FUA;
8599	}
8600
8601	lbalen = (struct ctl_lba_len_flags *)
8602	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8603	lbalen->lba = lba;
8604	lbalen->len = num_blocks;
8605	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8606
8607	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8608	ctsio->kern_rel_offset = 0;
8609
8610	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8611
8612	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8613
8614	return (retval);
8615}
8616
8617static int
8618ctl_cnw_cont(union ctl_io *io)
8619{
8620	struct ctl_scsiio *ctsio;
8621	struct ctl_lun *lun;
8622	struct ctl_lba_len_flags *lbalen;
8623	int retval;
8624
8625	ctsio = &io->scsiio;
8626	ctsio->io_hdr.status = CTL_STATUS_NONE;
8627	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8628	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8629	lbalen = (struct ctl_lba_len_flags *)
8630	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8631	lbalen->flags &= ~CTL_LLF_COMPARE;
8632	lbalen->flags |= CTL_LLF_WRITE;
8633
8634	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8635	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8636	return (retval);
8637}
8638
8639int
8640ctl_cnw(struct ctl_scsiio *ctsio)
8641{
8642	struct ctl_lun *lun;
8643	struct ctl_lba_len_flags *lbalen;
8644	uint64_t lba;
8645	uint32_t num_blocks;
8646	int flags, retval;
8647
8648	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8649
8650	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8651
8652	flags = 0;
8653	retval = CTL_RETVAL_COMPLETE;
8654
8655	switch (ctsio->cdb[0]) {
8656	case COMPARE_AND_WRITE: {
8657		struct scsi_compare_and_write *cdb;
8658
8659		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8660		if (cdb->byte2 & SRW10_FUA)
8661			flags |= CTL_LLF_FUA;
8662		if (cdb->byte2 & SRW10_DPO)
8663			flags |= CTL_LLF_DPO;
8664		lba = scsi_8btou64(cdb->addr);
8665		num_blocks = cdb->length;
8666		break;
8667	}
8668	default:
8669		/*
8670		 * We got a command we don't support.  This shouldn't
8671		 * happen, commands should be filtered out above us.
8672		 */
8673		ctl_set_invalid_opcode(ctsio);
8674		ctl_done((union ctl_io *)ctsio);
8675
8676		return (CTL_RETVAL_COMPLETE);
8677		break; /* NOTREACHED */
8678	}
8679
8680	/*
8681	 * The first check is to make sure we're in bounds, the second
8682	 * check is to catch wrap-around problems.  If the lba + num blocks
8683	 * is less than the lba, then we've wrapped around and the block
8684	 * range is invalid anyway.
8685	 */
8686	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8687	 || ((lba + num_blocks) < lba)) {
8688		ctl_set_lba_out_of_range(ctsio);
8689		ctl_done((union ctl_io *)ctsio);
8690		return (CTL_RETVAL_COMPLETE);
8691	}
8692
8693	/*
8694	 * According to SBC-3, a transfer length of 0 is not an error.
8695	 */
8696	if (num_blocks == 0) {
8697		ctl_set_success(ctsio);
8698		ctl_done((union ctl_io *)ctsio);
8699		return (CTL_RETVAL_COMPLETE);
8700	}
8701
8702	/* Set FUA if write cache is disabled. */
8703	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8704	    SCP_WCE) == 0)
8705		flags |= CTL_LLF_FUA;
8706
8707	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8708	ctsio->kern_rel_offset = 0;
8709
8710	/*
8711	 * Set the IO_CONT flag, so that if this I/O gets passed to
8712	 * ctl_data_submit_done(), it'll get passed back to
8713	 * ctl_ctl_cnw_cont() for further processing.
8714	 */
8715	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8716	ctsio->io_cont = ctl_cnw_cont;
8717
8718	lbalen = (struct ctl_lba_len_flags *)
8719	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8720	lbalen->lba = lba;
8721	lbalen->len = num_blocks;
8722	lbalen->flags = CTL_LLF_COMPARE | flags;
8723
8724	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8725	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8726	return (retval);
8727}
8728
8729int
8730ctl_verify(struct ctl_scsiio *ctsio)
8731{
8732	struct ctl_lun *lun;
8733	struct ctl_lba_len_flags *lbalen;
8734	uint64_t lba;
8735	uint32_t num_blocks;
8736	int bytchk, flags;
8737	int retval;
8738
8739	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8740
8741	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8742
8743	bytchk = 0;
8744	flags = CTL_LLF_FUA;
8745	retval = CTL_RETVAL_COMPLETE;
8746
8747	switch (ctsio->cdb[0]) {
8748	case VERIFY_10: {
8749		struct scsi_verify_10 *cdb;
8750
8751		cdb = (struct scsi_verify_10 *)ctsio->cdb;
8752		if (cdb->byte2 & SVFY_BYTCHK)
8753			bytchk = 1;
8754		if (cdb->byte2 & SVFY_DPO)
8755			flags |= CTL_LLF_DPO;
8756		lba = scsi_4btoul(cdb->addr);
8757		num_blocks = scsi_2btoul(cdb->length);
8758		break;
8759	}
8760	case VERIFY_12: {
8761		struct scsi_verify_12 *cdb;
8762
8763		cdb = (struct scsi_verify_12 *)ctsio->cdb;
8764		if (cdb->byte2 & SVFY_BYTCHK)
8765			bytchk = 1;
8766		if (cdb->byte2 & SVFY_DPO)
8767			flags |= CTL_LLF_DPO;
8768		lba = scsi_4btoul(cdb->addr);
8769		num_blocks = scsi_4btoul(cdb->length);
8770		break;
8771	}
8772	case VERIFY_16: {
8773		struct scsi_rw_16 *cdb;
8774
8775		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8776		if (cdb->byte2 & SVFY_BYTCHK)
8777			bytchk = 1;
8778		if (cdb->byte2 & SVFY_DPO)
8779			flags |= CTL_LLF_DPO;
8780		lba = scsi_8btou64(cdb->addr);
8781		num_blocks = scsi_4btoul(cdb->length);
8782		break;
8783	}
8784	default:
8785		/*
8786		 * We got a command we don't support.  This shouldn't
8787		 * happen, commands should be filtered out above us.
8788		 */
8789		ctl_set_invalid_opcode(ctsio);
8790		ctl_done((union ctl_io *)ctsio);
8791		return (CTL_RETVAL_COMPLETE);
8792	}
8793
8794	/*
8795	 * The first check is to make sure we're in bounds, the second
8796	 * check is to catch wrap-around problems.  If the lba + num blocks
8797	 * is less than the lba, then we've wrapped around and the block
8798	 * range is invalid anyway.
8799	 */
8800	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8801	 || ((lba + num_blocks) < lba)) {
8802		ctl_set_lba_out_of_range(ctsio);
8803		ctl_done((union ctl_io *)ctsio);
8804		return (CTL_RETVAL_COMPLETE);
8805	}
8806
8807	/*
8808	 * According to SBC-3, a transfer length of 0 is not an error.
8809	 */
8810	if (num_blocks == 0) {
8811		ctl_set_success(ctsio);
8812		ctl_done((union ctl_io *)ctsio);
8813		return (CTL_RETVAL_COMPLETE);
8814	}
8815
8816	lbalen = (struct ctl_lba_len_flags *)
8817	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8818	lbalen->lba = lba;
8819	lbalen->len = num_blocks;
8820	if (bytchk) {
8821		lbalen->flags = CTL_LLF_COMPARE | flags;
8822		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8823	} else {
8824		lbalen->flags = CTL_LLF_VERIFY | flags;
8825		ctsio->kern_total_len = 0;
8826	}
8827	ctsio->kern_rel_offset = 0;
8828
8829	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
8830	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8831	return (retval);
8832}
8833
8834int
8835ctl_report_luns(struct ctl_scsiio *ctsio)
8836{
8837	struct ctl_softc *softc = control_softc;
8838	struct scsi_report_luns *cdb;
8839	struct scsi_report_luns_data *lun_data;
8840	struct ctl_lun *lun, *request_lun;
8841	struct ctl_port *port;
8842	int num_luns, retval;
8843	uint32_t alloc_len, lun_datalen;
8844	int num_filled, well_known;
8845	uint32_t initidx, targ_lun_id, lun_id;
8846
8847	retval = CTL_RETVAL_COMPLETE;
8848	well_known = 0;
8849
8850	cdb = (struct scsi_report_luns *)ctsio->cdb;
8851	port = ctl_io_port(&ctsio->io_hdr);
8852
8853	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
8854
8855	mtx_lock(&softc->ctl_lock);
8856	num_luns = 0;
8857	for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
8858		if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
8859			num_luns++;
8860	}
8861	mtx_unlock(&softc->ctl_lock);
8862
8863	switch (cdb->select_report) {
8864	case RPL_REPORT_DEFAULT:
8865	case RPL_REPORT_ALL:
8866		break;
8867	case RPL_REPORT_WELLKNOWN:
8868		well_known = 1;
8869		num_luns = 0;
8870		break;
8871	default:
8872		ctl_set_invalid_field(ctsio,
8873				      /*sks_valid*/ 1,
8874				      /*command*/ 1,
8875				      /*field*/ 2,
8876				      /*bit_valid*/ 0,
8877				      /*bit*/ 0);
8878		ctl_done((union ctl_io *)ctsio);
8879		return (retval);
8880		break; /* NOTREACHED */
8881	}
8882
8883	alloc_len = scsi_4btoul(cdb->length);
8884	/*
8885	 * The initiator has to allocate at least 16 bytes for this request,
8886	 * so he can at least get the header and the first LUN.  Otherwise
8887	 * we reject the request (per SPC-3 rev 14, section 6.21).
8888	 */
8889	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
8890	    sizeof(struct scsi_report_luns_lundata))) {
8891		ctl_set_invalid_field(ctsio,
8892				      /*sks_valid*/ 1,
8893				      /*command*/ 1,
8894				      /*field*/ 6,
8895				      /*bit_valid*/ 0,
8896				      /*bit*/ 0);
8897		ctl_done((union ctl_io *)ctsio);
8898		return (retval);
8899	}
8900
8901	request_lun = (struct ctl_lun *)
8902		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8903
8904	lun_datalen = sizeof(*lun_data) +
8905		(num_luns * sizeof(struct scsi_report_luns_lundata));
8906
8907	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
8908	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
8909	ctsio->kern_sg_entries = 0;
8910
8911	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8912
8913	mtx_lock(&softc->ctl_lock);
8914	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
8915		lun_id = ctl_lun_map_from_port(port, targ_lun_id);
8916		if (lun_id >= CTL_MAX_LUNS)
8917			continue;
8918		lun = softc->ctl_luns[lun_id];
8919		if (lun == NULL)
8920			continue;
8921
8922		if (targ_lun_id <= 0xff) {
8923			/*
8924			 * Peripheral addressing method, bus number 0.
8925			 */
8926			lun_data->luns[num_filled].lundata[0] =
8927				RPL_LUNDATA_ATYP_PERIPH;
8928			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
8929			num_filled++;
8930		} else if (targ_lun_id <= 0x3fff) {
8931			/*
8932			 * Flat addressing method.
8933			 */
8934			lun_data->luns[num_filled].lundata[0] =
8935				RPL_LUNDATA_ATYP_FLAT | (targ_lun_id >> 8);
8936			lun_data->luns[num_filled].lundata[1] =
8937				(targ_lun_id & 0xff);
8938			num_filled++;
8939		} else if (targ_lun_id <= 0xffffff) {
8940			/*
8941			 * Extended flat addressing method.
8942			 */
8943			lun_data->luns[num_filled].lundata[0] =
8944			    RPL_LUNDATA_ATYP_EXTLUN | 0x12;
8945			scsi_ulto3b(targ_lun_id,
8946			    &lun_data->luns[num_filled].lundata[1]);
8947			num_filled++;
8948		} else {
8949			printf("ctl_report_luns: bogus LUN number %jd, "
8950			       "skipping\n", (intmax_t)targ_lun_id);
8951		}
8952		/*
8953		 * According to SPC-3, rev 14 section 6.21:
8954		 *
8955		 * "The execution of a REPORT LUNS command to any valid and
8956		 * installed logical unit shall clear the REPORTED LUNS DATA
8957		 * HAS CHANGED unit attention condition for all logical
8958		 * units of that target with respect to the requesting
8959		 * initiator. A valid and installed logical unit is one
8960		 * having a PERIPHERAL QUALIFIER of 000b in the standard
8961		 * INQUIRY data (see 6.4.2)."
8962		 *
8963		 * If request_lun is NULL, the LUN this report luns command
8964		 * was issued to is either disabled or doesn't exist. In that
8965		 * case, we shouldn't clear any pending lun change unit
8966		 * attention.
8967		 */
8968		if (request_lun != NULL) {
8969			mtx_lock(&lun->lun_lock);
8970			ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
8971			mtx_unlock(&lun->lun_lock);
8972		}
8973	}
8974	mtx_unlock(&softc->ctl_lock);
8975
8976	/*
8977	 * It's quite possible that we've returned fewer LUNs than we allocated
8978	 * space for.  Trim it.
8979	 */
8980	lun_datalen = sizeof(*lun_data) +
8981		(num_filled * sizeof(struct scsi_report_luns_lundata));
8982
8983	if (lun_datalen < alloc_len) {
8984		ctsio->residual = alloc_len - lun_datalen;
8985		ctsio->kern_data_len = lun_datalen;
8986		ctsio->kern_total_len = lun_datalen;
8987	} else {
8988		ctsio->residual = 0;
8989		ctsio->kern_data_len = alloc_len;
8990		ctsio->kern_total_len = alloc_len;
8991	}
8992	ctsio->kern_data_resid = 0;
8993	ctsio->kern_rel_offset = 0;
8994	ctsio->kern_sg_entries = 0;
8995
8996	/*
8997	 * We set this to the actual data length, regardless of how much
8998	 * space we actually have to return results.  If the user looks at
8999	 * this value, he'll know whether or not he allocated enough space
9000	 * and reissue the command if necessary.  We don't support well
9001	 * known logical units, so if the user asks for that, return none.
9002	 */
9003	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9004
9005	/*
9006	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9007	 * this request.
9008	 */
9009	ctl_set_success(ctsio);
9010	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9011	ctsio->be_move_done = ctl_config_move_done;
9012	ctl_datamove((union ctl_io *)ctsio);
9013	return (retval);
9014}
9015
9016int
9017ctl_request_sense(struct ctl_scsiio *ctsio)
9018{
9019	struct scsi_request_sense *cdb;
9020	struct scsi_sense_data *sense_ptr;
9021	struct ctl_softc *ctl_softc;
9022	struct ctl_lun *lun;
9023	uint32_t initidx;
9024	int have_error;
9025	scsi_sense_data_type sense_format;
9026	ctl_ua_type ua_type;
9027
9028	cdb = (struct scsi_request_sense *)ctsio->cdb;
9029
9030	ctl_softc = control_softc;
9031	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9032
9033	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9034
9035	/*
9036	 * Determine which sense format the user wants.
9037	 */
9038	if (cdb->byte2 & SRS_DESC)
9039		sense_format = SSD_TYPE_DESC;
9040	else
9041		sense_format = SSD_TYPE_FIXED;
9042
9043	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9044	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9045	ctsio->kern_sg_entries = 0;
9046
9047	/*
9048	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9049	 * larger than the largest allowed value for the length field in the
9050	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9051	 */
9052	ctsio->residual = 0;
9053	ctsio->kern_data_len = cdb->length;
9054	ctsio->kern_total_len = cdb->length;
9055
9056	ctsio->kern_data_resid = 0;
9057	ctsio->kern_rel_offset = 0;
9058	ctsio->kern_sg_entries = 0;
9059
9060	/*
9061	 * If we don't have a LUN, we don't have any pending sense.
9062	 */
9063	if (lun == NULL)
9064		goto no_sense;
9065
9066	have_error = 0;
9067	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9068	/*
9069	 * Check for pending sense, and then for pending unit attentions.
9070	 * Pending sense gets returned first, then pending unit attentions.
9071	 */
9072	mtx_lock(&lun->lun_lock);
9073#ifdef CTL_WITH_CA
9074	if (ctl_is_set(lun->have_ca, initidx)) {
9075		scsi_sense_data_type stored_format;
9076
9077		/*
9078		 * Check to see which sense format was used for the stored
9079		 * sense data.
9080		 */
9081		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9082
9083		/*
9084		 * If the user requested a different sense format than the
9085		 * one we stored, then we need to convert it to the other
9086		 * format.  If we're going from descriptor to fixed format
9087		 * sense data, we may lose things in translation, depending
9088		 * on what options were used.
9089		 *
9090		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9091		 * for some reason we'll just copy it out as-is.
9092		 */
9093		if ((stored_format == SSD_TYPE_FIXED)
9094		 && (sense_format == SSD_TYPE_DESC))
9095			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9096			    &lun->pending_sense[initidx],
9097			    (struct scsi_sense_data_desc *)sense_ptr);
9098		else if ((stored_format == SSD_TYPE_DESC)
9099		      && (sense_format == SSD_TYPE_FIXED))
9100			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9101			    &lun->pending_sense[initidx],
9102			    (struct scsi_sense_data_fixed *)sense_ptr);
9103		else
9104			memcpy(sense_ptr, &lun->pending_sense[initidx],
9105			       MIN(sizeof(*sense_ptr),
9106			       sizeof(lun->pending_sense[initidx])));
9107
9108		ctl_clear_mask(lun->have_ca, initidx);
9109		have_error = 1;
9110	} else
9111#endif
9112	{
9113		ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9114		if (ua_type != CTL_UA_NONE)
9115			have_error = 1;
9116		if (ua_type == CTL_UA_LUN_CHANGE) {
9117			mtx_unlock(&lun->lun_lock);
9118			mtx_lock(&ctl_softc->ctl_lock);
9119			ctl_clr_ua_allluns(ctl_softc, initidx, ua_type);
9120			mtx_unlock(&ctl_softc->ctl_lock);
9121			mtx_lock(&lun->lun_lock);
9122		}
9123
9124	}
9125	mtx_unlock(&lun->lun_lock);
9126
9127	/*
9128	 * We already have a pending error, return it.
9129	 */
9130	if (have_error != 0) {
9131		/*
9132		 * We report the SCSI status as OK, since the status of the
9133		 * request sense command itself is OK.
9134		 * We report 0 for the sense length, because we aren't doing
9135		 * autosense in this case.  We're reporting sense as
9136		 * parameter data.
9137		 */
9138		ctl_set_success(ctsio);
9139		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9140		ctsio->be_move_done = ctl_config_move_done;
9141		ctl_datamove((union ctl_io *)ctsio);
9142		return (CTL_RETVAL_COMPLETE);
9143	}
9144
9145no_sense:
9146
9147	/*
9148	 * No sense information to report, so we report that everything is
9149	 * okay.
9150	 */
9151	ctl_set_sense_data(sense_ptr,
9152			   lun,
9153			   sense_format,
9154			   /*current_error*/ 1,
9155			   /*sense_key*/ SSD_KEY_NO_SENSE,
9156			   /*asc*/ 0x00,
9157			   /*ascq*/ 0x00,
9158			   SSD_ELEM_NONE);
9159
9160	/*
9161	 * We report 0 for the sense length, because we aren't doing
9162	 * autosense in this case.  We're reporting sense as parameter data.
9163	 */
9164	ctl_set_success(ctsio);
9165	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9166	ctsio->be_move_done = ctl_config_move_done;
9167	ctl_datamove((union ctl_io *)ctsio);
9168	return (CTL_RETVAL_COMPLETE);
9169}
9170
9171int
9172ctl_tur(struct ctl_scsiio *ctsio)
9173{
9174
9175	CTL_DEBUG_PRINT(("ctl_tur\n"));
9176
9177	ctl_set_success(ctsio);
9178	ctl_done((union ctl_io *)ctsio);
9179
9180	return (CTL_RETVAL_COMPLETE);
9181}
9182
9183#ifdef notyet
9184static int
9185ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9186{
9187
9188}
9189#endif
9190
9191/*
9192 * SCSI VPD page 0x00, the Supported VPD Pages page.
9193 */
9194static int
9195ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9196{
9197	struct scsi_vpd_supported_pages *pages;
9198	int sup_page_size;
9199	struct ctl_lun *lun;
9200	int p;
9201
9202	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9203
9204	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9205	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9206	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9207	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9208	ctsio->kern_sg_entries = 0;
9209
9210	if (sup_page_size < alloc_len) {
9211		ctsio->residual = alloc_len - sup_page_size;
9212		ctsio->kern_data_len = sup_page_size;
9213		ctsio->kern_total_len = sup_page_size;
9214	} else {
9215		ctsio->residual = 0;
9216		ctsio->kern_data_len = alloc_len;
9217		ctsio->kern_total_len = alloc_len;
9218	}
9219	ctsio->kern_data_resid = 0;
9220	ctsio->kern_rel_offset = 0;
9221	ctsio->kern_sg_entries = 0;
9222
9223	/*
9224	 * The control device is always connected.  The disk device, on the
9225	 * other hand, may not be online all the time.  Need to change this
9226	 * to figure out whether the disk device is actually online or not.
9227	 */
9228	if (lun != NULL)
9229		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9230				lun->be_lun->lun_type;
9231	else
9232		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9233
9234	p = 0;
9235	/* Supported VPD pages */
9236	pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9237	/* Serial Number */
9238	pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9239	/* Device Identification */
9240	pages->page_list[p++] = SVPD_DEVICE_ID;
9241	/* Extended INQUIRY Data */
9242	pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9243	/* Mode Page Policy */
9244	pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9245	/* SCSI Ports */
9246	pages->page_list[p++] = SVPD_SCSI_PORTS;
9247	/* Third-party Copy */
9248	pages->page_list[p++] = SVPD_SCSI_TPC;
9249	if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9250		/* Block limits */
9251		pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9252		/* Block Device Characteristics */
9253		pages->page_list[p++] = SVPD_BDC;
9254		/* Logical Block Provisioning */
9255		pages->page_list[p++] = SVPD_LBP;
9256	}
9257	pages->length = p;
9258
9259	ctl_set_success(ctsio);
9260	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9261	ctsio->be_move_done = ctl_config_move_done;
9262	ctl_datamove((union ctl_io *)ctsio);
9263	return (CTL_RETVAL_COMPLETE);
9264}
9265
9266/*
9267 * SCSI VPD page 0x80, the Unit Serial Number page.
9268 */
9269static int
9270ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9271{
9272	struct scsi_vpd_unit_serial_number *sn_ptr;
9273	struct ctl_lun *lun;
9274	int data_len;
9275
9276	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9277
9278	data_len = 4 + CTL_SN_LEN;
9279	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9280	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9281	if (data_len < alloc_len) {
9282		ctsio->residual = alloc_len - data_len;
9283		ctsio->kern_data_len = data_len;
9284		ctsio->kern_total_len = data_len;
9285	} else {
9286		ctsio->residual = 0;
9287		ctsio->kern_data_len = alloc_len;
9288		ctsio->kern_total_len = alloc_len;
9289	}
9290	ctsio->kern_data_resid = 0;
9291	ctsio->kern_rel_offset = 0;
9292	ctsio->kern_sg_entries = 0;
9293
9294	/*
9295	 * The control device is always connected.  The disk device, on the
9296	 * other hand, may not be online all the time.  Need to change this
9297	 * to figure out whether the disk device is actually online or not.
9298	 */
9299	if (lun != NULL)
9300		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9301				  lun->be_lun->lun_type;
9302	else
9303		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9304
9305	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9306	sn_ptr->length = CTL_SN_LEN;
9307	/*
9308	 * If we don't have a LUN, we just leave the serial number as
9309	 * all spaces.
9310	 */
9311	if (lun != NULL) {
9312		strncpy((char *)sn_ptr->serial_num,
9313			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9314	} else
9315		memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9316
9317	ctl_set_success(ctsio);
9318	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9319	ctsio->be_move_done = ctl_config_move_done;
9320	ctl_datamove((union ctl_io *)ctsio);
9321	return (CTL_RETVAL_COMPLETE);
9322}
9323
9324
9325/*
9326 * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9327 */
9328static int
9329ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9330{
9331	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9332	struct ctl_lun *lun;
9333	int data_len;
9334
9335	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9336
9337	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9338	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9339	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9340	ctsio->kern_sg_entries = 0;
9341
9342	if (data_len < alloc_len) {
9343		ctsio->residual = alloc_len - data_len;
9344		ctsio->kern_data_len = data_len;
9345		ctsio->kern_total_len = data_len;
9346	} else {
9347		ctsio->residual = 0;
9348		ctsio->kern_data_len = alloc_len;
9349		ctsio->kern_total_len = alloc_len;
9350	}
9351	ctsio->kern_data_resid = 0;
9352	ctsio->kern_rel_offset = 0;
9353	ctsio->kern_sg_entries = 0;
9354
9355	/*
9356	 * The control device is always connected.  The disk device, on the
9357	 * other hand, may not be online all the time.
9358	 */
9359	if (lun != NULL)
9360		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9361				     lun->be_lun->lun_type;
9362	else
9363		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9364	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9365	scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9366	/*
9367	 * We support head of queue, ordered and simple tags.
9368	 */
9369	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9370	/*
9371	 * Volatile cache supported.
9372	 */
9373	eid_ptr->flags3 = SVPD_EID_V_SUP;
9374
9375	/*
9376	 * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9377	 * attention for a particular IT nexus on all LUNs once we report
9378	 * it to that nexus once.  This bit is required as of SPC-4.
9379	 */
9380	eid_ptr->flags4 = SVPD_EID_LUICLT;
9381
9382	/*
9383	 * XXX KDM in order to correctly answer this, we would need
9384	 * information from the SIM to determine how much sense data it
9385	 * can send.  So this would really be a path inquiry field, most
9386	 * likely.  This can be set to a maximum of 252 according to SPC-4,
9387	 * but the hardware may or may not be able to support that much.
9388	 * 0 just means that the maximum sense data length is not reported.
9389	 */
9390	eid_ptr->max_sense_length = 0;
9391
9392	ctl_set_success(ctsio);
9393	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9394	ctsio->be_move_done = ctl_config_move_done;
9395	ctl_datamove((union ctl_io *)ctsio);
9396	return (CTL_RETVAL_COMPLETE);
9397}
9398
9399static int
9400ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9401{
9402	struct scsi_vpd_mode_page_policy *mpp_ptr;
9403	struct ctl_lun *lun;
9404	int data_len;
9405
9406	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9407
9408	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9409	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9410
9411	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9412	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9413	ctsio->kern_sg_entries = 0;
9414
9415	if (data_len < alloc_len) {
9416		ctsio->residual = alloc_len - data_len;
9417		ctsio->kern_data_len = data_len;
9418		ctsio->kern_total_len = data_len;
9419	} else {
9420		ctsio->residual = 0;
9421		ctsio->kern_data_len = alloc_len;
9422		ctsio->kern_total_len = alloc_len;
9423	}
9424	ctsio->kern_data_resid = 0;
9425	ctsio->kern_rel_offset = 0;
9426	ctsio->kern_sg_entries = 0;
9427
9428	/*
9429	 * The control device is always connected.  The disk device, on the
9430	 * other hand, may not be online all the time.
9431	 */
9432	if (lun != NULL)
9433		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9434				     lun->be_lun->lun_type;
9435	else
9436		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9437	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9438	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9439	mpp_ptr->descr[0].page_code = 0x3f;
9440	mpp_ptr->descr[0].subpage_code = 0xff;
9441	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9442
9443	ctl_set_success(ctsio);
9444	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9445	ctsio->be_move_done = ctl_config_move_done;
9446	ctl_datamove((union ctl_io *)ctsio);
9447	return (CTL_RETVAL_COMPLETE);
9448}
9449
9450/*
9451 * SCSI VPD page 0x83, the Device Identification page.
9452 */
9453static int
9454ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9455{
9456	struct scsi_vpd_device_id *devid_ptr;
9457	struct scsi_vpd_id_descriptor *desc;
9458	struct ctl_softc *softc;
9459	struct ctl_lun *lun;
9460	struct ctl_port *port;
9461	int data_len;
9462	uint8_t proto;
9463
9464	softc = control_softc;
9465
9466	port = softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9467	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9468
9469	data_len = sizeof(struct scsi_vpd_device_id) +
9470	    sizeof(struct scsi_vpd_id_descriptor) +
9471		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9472	    sizeof(struct scsi_vpd_id_descriptor) +
9473		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9474	if (lun && lun->lun_devid)
9475		data_len += lun->lun_devid->len;
9476	if (port->port_devid)
9477		data_len += port->port_devid->len;
9478	if (port->target_devid)
9479		data_len += port->target_devid->len;
9480
9481	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9482	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9483	ctsio->kern_sg_entries = 0;
9484
9485	if (data_len < alloc_len) {
9486		ctsio->residual = alloc_len - data_len;
9487		ctsio->kern_data_len = data_len;
9488		ctsio->kern_total_len = data_len;
9489	} else {
9490		ctsio->residual = 0;
9491		ctsio->kern_data_len = alloc_len;
9492		ctsio->kern_total_len = alloc_len;
9493	}
9494	ctsio->kern_data_resid = 0;
9495	ctsio->kern_rel_offset = 0;
9496	ctsio->kern_sg_entries = 0;
9497
9498	/*
9499	 * The control device is always connected.  The disk device, on the
9500	 * other hand, may not be online all the time.
9501	 */
9502	if (lun != NULL)
9503		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9504				     lun->be_lun->lun_type;
9505	else
9506		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9507	devid_ptr->page_code = SVPD_DEVICE_ID;
9508	scsi_ulto2b(data_len - 4, devid_ptr->length);
9509
9510	if (port->port_type == CTL_PORT_FC)
9511		proto = SCSI_PROTO_FC << 4;
9512	else if (port->port_type == CTL_PORT_ISCSI)
9513		proto = SCSI_PROTO_ISCSI << 4;
9514	else
9515		proto = SCSI_PROTO_SPI << 4;
9516	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9517
9518	/*
9519	 * We're using a LUN association here.  i.e., this device ID is a
9520	 * per-LUN identifier.
9521	 */
9522	if (lun && lun->lun_devid) {
9523		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9524		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9525		    lun->lun_devid->len);
9526	}
9527
9528	/*
9529	 * This is for the WWPN which is a port association.
9530	 */
9531	if (port->port_devid) {
9532		memcpy(desc, port->port_devid->data, port->port_devid->len);
9533		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9534		    port->port_devid->len);
9535	}
9536
9537	/*
9538	 * This is for the Relative Target Port(type 4h) identifier
9539	 */
9540	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9541	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9542	    SVPD_ID_TYPE_RELTARG;
9543	desc->length = 4;
9544	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9545	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9546	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9547
9548	/*
9549	 * This is for the Target Port Group(type 5h) identifier
9550	 */
9551	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9552	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9553	    SVPD_ID_TYPE_TPORTGRP;
9554	desc->length = 4;
9555	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9556	    &desc->identifier[2]);
9557	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9558	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9559
9560	/*
9561	 * This is for the Target identifier
9562	 */
9563	if (port->target_devid) {
9564		memcpy(desc, port->target_devid->data, port->target_devid->len);
9565	}
9566
9567	ctl_set_success(ctsio);
9568	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9569	ctsio->be_move_done = ctl_config_move_done;
9570	ctl_datamove((union ctl_io *)ctsio);
9571	return (CTL_RETVAL_COMPLETE);
9572}
9573
9574static int
9575ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9576{
9577	struct ctl_softc *softc = control_softc;
9578	struct scsi_vpd_scsi_ports *sp;
9579	struct scsi_vpd_port_designation *pd;
9580	struct scsi_vpd_port_designation_cont *pdc;
9581	struct ctl_lun *lun;
9582	struct ctl_port *port;
9583	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
9584	int num_target_port_groups;
9585
9586	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9587
9588	if (softc->is_single)
9589		num_target_port_groups = 1;
9590	else
9591		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
9592	num_target_ports = 0;
9593	iid_len = 0;
9594	id_len = 0;
9595	mtx_lock(&softc->ctl_lock);
9596	STAILQ_FOREACH(port, &softc->port_list, links) {
9597		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9598			continue;
9599		if (lun != NULL &&
9600		    ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9601			continue;
9602		num_target_ports++;
9603		if (port->init_devid)
9604			iid_len += port->init_devid->len;
9605		if (port->port_devid)
9606			id_len += port->port_devid->len;
9607	}
9608	mtx_unlock(&softc->ctl_lock);
9609
9610	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
9611	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9612	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9613	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9614	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9615	ctsio->kern_sg_entries = 0;
9616
9617	if (data_len < alloc_len) {
9618		ctsio->residual = alloc_len - data_len;
9619		ctsio->kern_data_len = data_len;
9620		ctsio->kern_total_len = data_len;
9621	} else {
9622		ctsio->residual = 0;
9623		ctsio->kern_data_len = alloc_len;
9624		ctsio->kern_total_len = alloc_len;
9625	}
9626	ctsio->kern_data_resid = 0;
9627	ctsio->kern_rel_offset = 0;
9628	ctsio->kern_sg_entries = 0;
9629
9630	/*
9631	 * The control device is always connected.  The disk device, on the
9632	 * other hand, may not be online all the time.  Need to change this
9633	 * to figure out whether the disk device is actually online or not.
9634	 */
9635	if (lun != NULL)
9636		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9637				  lun->be_lun->lun_type;
9638	else
9639		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9640
9641	sp->page_code = SVPD_SCSI_PORTS;
9642	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9643	    sp->page_length);
9644	pd = &sp->design[0];
9645
9646	mtx_lock(&softc->ctl_lock);
9647	pg = softc->port_offset / CTL_MAX_PORTS;
9648	for (g = 0; g < num_target_port_groups; g++) {
9649		STAILQ_FOREACH(port, &softc->port_list, links) {
9650			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9651				continue;
9652			if (lun != NULL &&
9653			    ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9654				continue;
9655			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
9656			scsi_ulto2b(p, pd->relative_port_id);
9657			if (port->init_devid && g == pg) {
9658				iid_len = port->init_devid->len;
9659				memcpy(pd->initiator_transportid,
9660				    port->init_devid->data, port->init_devid->len);
9661			} else
9662				iid_len = 0;
9663			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9664			pdc = (struct scsi_vpd_port_designation_cont *)
9665			    (&pd->initiator_transportid[iid_len]);
9666			if (port->port_devid && g == pg) {
9667				id_len = port->port_devid->len;
9668				memcpy(pdc->target_port_descriptors,
9669				    port->port_devid->data, port->port_devid->len);
9670			} else
9671				id_len = 0;
9672			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9673			pd = (struct scsi_vpd_port_designation *)
9674			    ((uint8_t *)pdc->target_port_descriptors + id_len);
9675		}
9676	}
9677	mtx_unlock(&softc->ctl_lock);
9678
9679	ctl_set_success(ctsio);
9680	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9681	ctsio->be_move_done = ctl_config_move_done;
9682	ctl_datamove((union ctl_io *)ctsio);
9683	return (CTL_RETVAL_COMPLETE);
9684}
9685
9686static int
9687ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9688{
9689	struct scsi_vpd_block_limits *bl_ptr;
9690	struct ctl_lun *lun;
9691	int bs;
9692
9693	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9694
9695	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9696	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9697	ctsio->kern_sg_entries = 0;
9698
9699	if (sizeof(*bl_ptr) < alloc_len) {
9700		ctsio->residual = alloc_len - sizeof(*bl_ptr);
9701		ctsio->kern_data_len = sizeof(*bl_ptr);
9702		ctsio->kern_total_len = sizeof(*bl_ptr);
9703	} else {
9704		ctsio->residual = 0;
9705		ctsio->kern_data_len = alloc_len;
9706		ctsio->kern_total_len = alloc_len;
9707	}
9708	ctsio->kern_data_resid = 0;
9709	ctsio->kern_rel_offset = 0;
9710	ctsio->kern_sg_entries = 0;
9711
9712	/*
9713	 * The control device is always connected.  The disk device, on the
9714	 * other hand, may not be online all the time.  Need to change this
9715	 * to figure out whether the disk device is actually online or not.
9716	 */
9717	if (lun != NULL)
9718		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9719				  lun->be_lun->lun_type;
9720	else
9721		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9722
9723	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9724	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9725	bl_ptr->max_cmp_write_len = 0xff;
9726	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9727	if (lun != NULL) {
9728		bs = lun->be_lun->blocksize;
9729		scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9730		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9731			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9732			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9733			if (lun->be_lun->ublockexp != 0) {
9734				scsi_ulto4b((1 << lun->be_lun->ublockexp),
9735				    bl_ptr->opt_unmap_grain);
9736				scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9737				    bl_ptr->unmap_grain_align);
9738			}
9739		}
9740		scsi_ulto4b(lun->be_lun->atomicblock,
9741		    bl_ptr->max_atomic_transfer_length);
9742		scsi_ulto4b(0, bl_ptr->atomic_alignment);
9743		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9744	}
9745	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9746
9747	ctl_set_success(ctsio);
9748	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9749	ctsio->be_move_done = ctl_config_move_done;
9750	ctl_datamove((union ctl_io *)ctsio);
9751	return (CTL_RETVAL_COMPLETE);
9752}
9753
9754static int
9755ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9756{
9757	struct scsi_vpd_block_device_characteristics *bdc_ptr;
9758	struct ctl_lun *lun;
9759	const char *value;
9760	u_int i;
9761
9762	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9763
9764	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9765	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9766	ctsio->kern_sg_entries = 0;
9767
9768	if (sizeof(*bdc_ptr) < alloc_len) {
9769		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
9770		ctsio->kern_data_len = sizeof(*bdc_ptr);
9771		ctsio->kern_total_len = sizeof(*bdc_ptr);
9772	} else {
9773		ctsio->residual = 0;
9774		ctsio->kern_data_len = alloc_len;
9775		ctsio->kern_total_len = alloc_len;
9776	}
9777	ctsio->kern_data_resid = 0;
9778	ctsio->kern_rel_offset = 0;
9779	ctsio->kern_sg_entries = 0;
9780
9781	/*
9782	 * The control device is always connected.  The disk device, on the
9783	 * other hand, may not be online all the time.  Need to change this
9784	 * to figure out whether the disk device is actually online or not.
9785	 */
9786	if (lun != NULL)
9787		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9788				  lun->be_lun->lun_type;
9789	else
9790		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9791	bdc_ptr->page_code = SVPD_BDC;
9792	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9793	if (lun != NULL &&
9794	    (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9795		i = strtol(value, NULL, 0);
9796	else
9797		i = CTL_DEFAULT_ROTATION_RATE;
9798	scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9799	if (lun != NULL &&
9800	    (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9801		i = strtol(value, NULL, 0);
9802	else
9803		i = 0;
9804	bdc_ptr->wab_wac_ff = (i & 0x0f);
9805	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9806
9807	ctl_set_success(ctsio);
9808	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9809	ctsio->be_move_done = ctl_config_move_done;
9810	ctl_datamove((union ctl_io *)ctsio);
9811	return (CTL_RETVAL_COMPLETE);
9812}
9813
9814static int
9815ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9816{
9817	struct scsi_vpd_logical_block_prov *lbp_ptr;
9818	struct ctl_lun *lun;
9819
9820	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9821
9822	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9823	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9824	ctsio->kern_sg_entries = 0;
9825
9826	if (sizeof(*lbp_ptr) < alloc_len) {
9827		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9828		ctsio->kern_data_len = sizeof(*lbp_ptr);
9829		ctsio->kern_total_len = sizeof(*lbp_ptr);
9830	} else {
9831		ctsio->residual = 0;
9832		ctsio->kern_data_len = alloc_len;
9833		ctsio->kern_total_len = alloc_len;
9834	}
9835	ctsio->kern_data_resid = 0;
9836	ctsio->kern_rel_offset = 0;
9837	ctsio->kern_sg_entries = 0;
9838
9839	/*
9840	 * The control device is always connected.  The disk device, on the
9841	 * other hand, may not be online all the time.  Need to change this
9842	 * to figure out whether the disk device is actually online or not.
9843	 */
9844	if (lun != NULL)
9845		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9846				  lun->be_lun->lun_type;
9847	else
9848		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9849
9850	lbp_ptr->page_code = SVPD_LBP;
9851	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9852	lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
9853	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9854		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
9855		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
9856		lbp_ptr->prov_type = SVPD_LBP_THIN;
9857	}
9858
9859	ctl_set_success(ctsio);
9860	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9861	ctsio->be_move_done = ctl_config_move_done;
9862	ctl_datamove((union ctl_io *)ctsio);
9863	return (CTL_RETVAL_COMPLETE);
9864}
9865
9866/*
9867 * INQUIRY with the EVPD bit set.
9868 */
9869static int
9870ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9871{
9872	struct ctl_lun *lun;
9873	struct scsi_inquiry *cdb;
9874	int alloc_len, retval;
9875
9876	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9877	cdb = (struct scsi_inquiry *)ctsio->cdb;
9878	alloc_len = scsi_2btoul(cdb->length);
9879
9880	switch (cdb->page_code) {
9881	case SVPD_SUPPORTED_PAGES:
9882		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9883		break;
9884	case SVPD_UNIT_SERIAL_NUMBER:
9885		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9886		break;
9887	case SVPD_DEVICE_ID:
9888		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9889		break;
9890	case SVPD_EXTENDED_INQUIRY_DATA:
9891		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
9892		break;
9893	case SVPD_MODE_PAGE_POLICY:
9894		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
9895		break;
9896	case SVPD_SCSI_PORTS:
9897		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
9898		break;
9899	case SVPD_SCSI_TPC:
9900		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
9901		break;
9902	case SVPD_BLOCK_LIMITS:
9903		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9904			goto err;
9905		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9906		break;
9907	case SVPD_BDC:
9908		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9909			goto err;
9910		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
9911		break;
9912	case SVPD_LBP:
9913		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9914			goto err;
9915		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9916		break;
9917	default:
9918err:
9919		ctl_set_invalid_field(ctsio,
9920				      /*sks_valid*/ 1,
9921				      /*command*/ 1,
9922				      /*field*/ 2,
9923				      /*bit_valid*/ 0,
9924				      /*bit*/ 0);
9925		ctl_done((union ctl_io *)ctsio);
9926		retval = CTL_RETVAL_COMPLETE;
9927		break;
9928	}
9929
9930	return (retval);
9931}
9932
9933/*
9934 * Standard INQUIRY data.
9935 */
9936static int
9937ctl_inquiry_std(struct ctl_scsiio *ctsio)
9938{
9939	struct scsi_inquiry_data *inq_ptr;
9940	struct scsi_inquiry *cdb;
9941	struct ctl_softc *softc;
9942	struct ctl_lun *lun;
9943	char *val;
9944	uint32_t alloc_len, data_len;
9945	ctl_port_type port_type;
9946
9947	softc = control_softc;
9948
9949	/*
9950	 * Figure out whether we're talking to a Fibre Channel port or not.
9951	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9952	 * SCSI front ends.
9953	 */
9954	port_type = softc->ctl_ports[
9955	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
9956	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
9957		port_type = CTL_PORT_SCSI;
9958
9959	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9960	cdb = (struct scsi_inquiry *)ctsio->cdb;
9961	alloc_len = scsi_2btoul(cdb->length);
9962
9963	/*
9964	 * We malloc the full inquiry data size here and fill it
9965	 * in.  If the user only asks for less, we'll give him
9966	 * that much.
9967	 */
9968	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
9969	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9970	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9971	ctsio->kern_sg_entries = 0;
9972	ctsio->kern_data_resid = 0;
9973	ctsio->kern_rel_offset = 0;
9974
9975	if (data_len < alloc_len) {
9976		ctsio->residual = alloc_len - data_len;
9977		ctsio->kern_data_len = data_len;
9978		ctsio->kern_total_len = data_len;
9979	} else {
9980		ctsio->residual = 0;
9981		ctsio->kern_data_len = alloc_len;
9982		ctsio->kern_total_len = alloc_len;
9983	}
9984
9985	/*
9986	 * If we have a LUN configured, report it as connected.  Otherwise,
9987	 * report that it is offline or no device is supported, depending
9988	 * on the value of inquiry_pq_no_lun.
9989	 *
9990	 * According to the spec (SPC-4 r34), the peripheral qualifier
9991	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9992	 *
9993	 * "A peripheral device having the specified peripheral device type
9994	 * is not connected to this logical unit. However, the device
9995	 * server is capable of supporting the specified peripheral device
9996	 * type on this logical unit."
9997	 *
9998	 * According to the same spec, the peripheral qualifier
9999	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10000	 *
10001	 * "The device server is not capable of supporting a peripheral
10002	 * device on this logical unit. For this peripheral qualifier the
10003	 * peripheral device type shall be set to 1Fh. All other peripheral
10004	 * device type values are reserved for this peripheral qualifier."
10005	 *
10006	 * Given the text, it would seem that we probably want to report that
10007	 * the LUN is offline here.  There is no LUN connected, but we can
10008	 * support a LUN at the given LUN number.
10009	 *
10010	 * In the real world, though, it sounds like things are a little
10011	 * different:
10012	 *
10013	 * - Linux, when presented with a LUN with the offline peripheral
10014	 *   qualifier, will create an sg driver instance for it.  So when
10015	 *   you attach it to CTL, you wind up with a ton of sg driver
10016	 *   instances.  (One for every LUN that Linux bothered to probe.)
10017	 *   Linux does this despite the fact that it issues a REPORT LUNs
10018	 *   to LUN 0 to get the inventory of supported LUNs.
10019	 *
10020	 * - There is other anecdotal evidence (from Emulex folks) about
10021	 *   arrays that use the offline peripheral qualifier for LUNs that
10022	 *   are on the "passive" path in an active/passive array.
10023	 *
10024	 * So the solution is provide a hopefully reasonable default
10025	 * (return bad/no LUN) and allow the user to change the behavior
10026	 * with a tunable/sysctl variable.
10027	 */
10028	if (lun != NULL)
10029		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10030				  lun->be_lun->lun_type;
10031	else if (softc->inquiry_pq_no_lun == 0)
10032		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10033	else
10034		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10035
10036	/* RMB in byte 2 is 0 */
10037	inq_ptr->version = SCSI_REV_SPC4;
10038
10039	/*
10040	 * According to SAM-3, even if a device only supports a single
10041	 * level of LUN addressing, it should still set the HISUP bit:
10042	 *
10043	 * 4.9.1 Logical unit numbers overview
10044	 *
10045	 * All logical unit number formats described in this standard are
10046	 * hierarchical in structure even when only a single level in that
10047	 * hierarchy is used. The HISUP bit shall be set to one in the
10048	 * standard INQUIRY data (see SPC-2) when any logical unit number
10049	 * format described in this standard is used.  Non-hierarchical
10050	 * formats are outside the scope of this standard.
10051	 *
10052	 * Therefore we set the HiSup bit here.
10053	 *
10054	 * The reponse format is 2, per SPC-3.
10055	 */
10056	inq_ptr->response_format = SID_HiSup | 2;
10057
10058	inq_ptr->additional_length = data_len -
10059	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10060	CTL_DEBUG_PRINT(("additional_length = %d\n",
10061			 inq_ptr->additional_length));
10062
10063	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10064	/* 16 bit addressing */
10065	if (port_type == CTL_PORT_SCSI)
10066		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10067	/* XXX set the SID_MultiP bit here if we're actually going to
10068	   respond on multiple ports */
10069	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10070
10071	/* 16 bit data bus, synchronous transfers */
10072	if (port_type == CTL_PORT_SCSI)
10073		inq_ptr->flags = SID_WBus16 | SID_Sync;
10074	/*
10075	 * XXX KDM do we want to support tagged queueing on the control
10076	 * device at all?
10077	 */
10078	if ((lun == NULL)
10079	 || (lun->be_lun->lun_type != T_PROCESSOR))
10080		inq_ptr->flags |= SID_CmdQue;
10081	/*
10082	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10083	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10084	 * name and 4 bytes for the revision.
10085	 */
10086	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10087	    "vendor")) == NULL) {
10088		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10089	} else {
10090		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10091		strncpy(inq_ptr->vendor, val,
10092		    min(sizeof(inq_ptr->vendor), strlen(val)));
10093	}
10094	if (lun == NULL) {
10095		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10096		    sizeof(inq_ptr->product));
10097	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10098		switch (lun->be_lun->lun_type) {
10099		case T_DIRECT:
10100			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10101			    sizeof(inq_ptr->product));
10102			break;
10103		case T_PROCESSOR:
10104			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10105			    sizeof(inq_ptr->product));
10106			break;
10107		default:
10108			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10109			    sizeof(inq_ptr->product));
10110			break;
10111		}
10112	} else {
10113		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10114		strncpy(inq_ptr->product, val,
10115		    min(sizeof(inq_ptr->product), strlen(val)));
10116	}
10117
10118	/*
10119	 * XXX make this a macro somewhere so it automatically gets
10120	 * incremented when we make changes.
10121	 */
10122	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10123	    "revision")) == NULL) {
10124		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10125	} else {
10126		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10127		strncpy(inq_ptr->revision, val,
10128		    min(sizeof(inq_ptr->revision), strlen(val)));
10129	}
10130
10131	/*
10132	 * For parallel SCSI, we support double transition and single
10133	 * transition clocking.  We also support QAS (Quick Arbitration
10134	 * and Selection) and Information Unit transfers on both the
10135	 * control and array devices.
10136	 */
10137	if (port_type == CTL_PORT_SCSI)
10138		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10139				    SID_SPI_IUS;
10140
10141	/* SAM-5 (no version claimed) */
10142	scsi_ulto2b(0x00A0, inq_ptr->version1);
10143	/* SPC-4 (no version claimed) */
10144	scsi_ulto2b(0x0460, inq_ptr->version2);
10145	if (port_type == CTL_PORT_FC) {
10146		/* FCP-2 ANSI INCITS.350:2003 */
10147		scsi_ulto2b(0x0917, inq_ptr->version3);
10148	} else if (port_type == CTL_PORT_SCSI) {
10149		/* SPI-4 ANSI INCITS.362:200x */
10150		scsi_ulto2b(0x0B56, inq_ptr->version3);
10151	} else if (port_type == CTL_PORT_ISCSI) {
10152		/* iSCSI (no version claimed) */
10153		scsi_ulto2b(0x0960, inq_ptr->version3);
10154	} else if (port_type == CTL_PORT_SAS) {
10155		/* SAS (no version claimed) */
10156		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10157	}
10158
10159	if (lun == NULL) {
10160		/* SBC-4 (no version claimed) */
10161		scsi_ulto2b(0x0600, inq_ptr->version4);
10162	} else {
10163		switch (lun->be_lun->lun_type) {
10164		case T_DIRECT:
10165			/* SBC-4 (no version claimed) */
10166			scsi_ulto2b(0x0600, inq_ptr->version4);
10167			break;
10168		case T_PROCESSOR:
10169		default:
10170			break;
10171		}
10172	}
10173
10174	ctl_set_success(ctsio);
10175	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10176	ctsio->be_move_done = ctl_config_move_done;
10177	ctl_datamove((union ctl_io *)ctsio);
10178	return (CTL_RETVAL_COMPLETE);
10179}
10180
10181int
10182ctl_inquiry(struct ctl_scsiio *ctsio)
10183{
10184	struct scsi_inquiry *cdb;
10185	int retval;
10186
10187	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10188
10189	cdb = (struct scsi_inquiry *)ctsio->cdb;
10190	if (cdb->byte2 & SI_EVPD)
10191		retval = ctl_inquiry_evpd(ctsio);
10192	else if (cdb->page_code == 0)
10193		retval = ctl_inquiry_std(ctsio);
10194	else {
10195		ctl_set_invalid_field(ctsio,
10196				      /*sks_valid*/ 1,
10197				      /*command*/ 1,
10198				      /*field*/ 2,
10199				      /*bit_valid*/ 0,
10200				      /*bit*/ 0);
10201		ctl_done((union ctl_io *)ctsio);
10202		return (CTL_RETVAL_COMPLETE);
10203	}
10204
10205	return (retval);
10206}
10207
10208/*
10209 * For known CDB types, parse the LBA and length.
10210 */
10211static int
10212ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10213{
10214	if (io->io_hdr.io_type != CTL_IO_SCSI)
10215		return (1);
10216
10217	switch (io->scsiio.cdb[0]) {
10218	case COMPARE_AND_WRITE: {
10219		struct scsi_compare_and_write *cdb;
10220
10221		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10222
10223		*lba = scsi_8btou64(cdb->addr);
10224		*len = cdb->length;
10225		break;
10226	}
10227	case READ_6:
10228	case WRITE_6: {
10229		struct scsi_rw_6 *cdb;
10230
10231		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10232
10233		*lba = scsi_3btoul(cdb->addr);
10234		/* only 5 bits are valid in the most significant address byte */
10235		*lba &= 0x1fffff;
10236		*len = cdb->length;
10237		break;
10238	}
10239	case READ_10:
10240	case WRITE_10: {
10241		struct scsi_rw_10 *cdb;
10242
10243		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10244
10245		*lba = scsi_4btoul(cdb->addr);
10246		*len = scsi_2btoul(cdb->length);
10247		break;
10248	}
10249	case WRITE_VERIFY_10: {
10250		struct scsi_write_verify_10 *cdb;
10251
10252		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10253
10254		*lba = scsi_4btoul(cdb->addr);
10255		*len = scsi_2btoul(cdb->length);
10256		break;
10257	}
10258	case READ_12:
10259	case WRITE_12: {
10260		struct scsi_rw_12 *cdb;
10261
10262		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10263
10264		*lba = scsi_4btoul(cdb->addr);
10265		*len = scsi_4btoul(cdb->length);
10266		break;
10267	}
10268	case WRITE_VERIFY_12: {
10269		struct scsi_write_verify_12 *cdb;
10270
10271		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10272
10273		*lba = scsi_4btoul(cdb->addr);
10274		*len = scsi_4btoul(cdb->length);
10275		break;
10276	}
10277	case READ_16:
10278	case WRITE_16:
10279	case WRITE_ATOMIC_16: {
10280		struct scsi_rw_16 *cdb;
10281
10282		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10283
10284		*lba = scsi_8btou64(cdb->addr);
10285		*len = scsi_4btoul(cdb->length);
10286		break;
10287	}
10288	case WRITE_VERIFY_16: {
10289		struct scsi_write_verify_16 *cdb;
10290
10291		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10292
10293		*lba = scsi_8btou64(cdb->addr);
10294		*len = scsi_4btoul(cdb->length);
10295		break;
10296	}
10297	case WRITE_SAME_10: {
10298		struct scsi_write_same_10 *cdb;
10299
10300		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10301
10302		*lba = scsi_4btoul(cdb->addr);
10303		*len = scsi_2btoul(cdb->length);
10304		break;
10305	}
10306	case WRITE_SAME_16: {
10307		struct scsi_write_same_16 *cdb;
10308
10309		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10310
10311		*lba = scsi_8btou64(cdb->addr);
10312		*len = scsi_4btoul(cdb->length);
10313		break;
10314	}
10315	case VERIFY_10: {
10316		struct scsi_verify_10 *cdb;
10317
10318		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10319
10320		*lba = scsi_4btoul(cdb->addr);
10321		*len = scsi_2btoul(cdb->length);
10322		break;
10323	}
10324	case VERIFY_12: {
10325		struct scsi_verify_12 *cdb;
10326
10327		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10328
10329		*lba = scsi_4btoul(cdb->addr);
10330		*len = scsi_4btoul(cdb->length);
10331		break;
10332	}
10333	case VERIFY_16: {
10334		struct scsi_verify_16 *cdb;
10335
10336		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10337
10338		*lba = scsi_8btou64(cdb->addr);
10339		*len = scsi_4btoul(cdb->length);
10340		break;
10341	}
10342	case UNMAP: {
10343		*lba = 0;
10344		*len = UINT64_MAX;
10345		break;
10346	}
10347	case SERVICE_ACTION_IN: {	/* GET LBA STATUS */
10348		struct scsi_get_lba_status *cdb;
10349
10350		cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10351		*lba = scsi_8btou64(cdb->addr);
10352		*len = UINT32_MAX;
10353		break;
10354	}
10355	default:
10356		return (1);
10357		break; /* NOTREACHED */
10358	}
10359
10360	return (0);
10361}
10362
10363static ctl_action
10364ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10365    bool seq)
10366{
10367	uint64_t endlba1, endlba2;
10368
10369	endlba1 = lba1 + len1 - (seq ? 0 : 1);
10370	endlba2 = lba2 + len2 - 1;
10371
10372	if ((endlba1 < lba2) || (endlba2 < lba1))
10373		return (CTL_ACTION_PASS);
10374	else
10375		return (CTL_ACTION_BLOCK);
10376}
10377
10378static int
10379ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10380{
10381	struct ctl_ptr_len_flags *ptrlen;
10382	struct scsi_unmap_desc *buf, *end, *range;
10383	uint64_t lba;
10384	uint32_t len;
10385
10386	/* If not UNMAP -- go other way. */
10387	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10388	    io->scsiio.cdb[0] != UNMAP)
10389		return (CTL_ACTION_ERROR);
10390
10391	/* If UNMAP without data -- block and wait for data. */
10392	ptrlen = (struct ctl_ptr_len_flags *)
10393	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10394	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10395	    ptrlen->ptr == NULL)
10396		return (CTL_ACTION_BLOCK);
10397
10398	/* UNMAP with data -- check for collision. */
10399	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10400	end = buf + ptrlen->len / sizeof(*buf);
10401	for (range = buf; range < end; range++) {
10402		lba = scsi_8btou64(range->lba);
10403		len = scsi_4btoul(range->length);
10404		if ((lba < lba2 + len2) && (lba + len > lba2))
10405			return (CTL_ACTION_BLOCK);
10406	}
10407	return (CTL_ACTION_PASS);
10408}
10409
10410static ctl_action
10411ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10412{
10413	uint64_t lba1, lba2;
10414	uint64_t len1, len2;
10415	int retval;
10416
10417	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10418		return (CTL_ACTION_ERROR);
10419
10420	retval = ctl_extent_check_unmap(io1, lba2, len2);
10421	if (retval != CTL_ACTION_ERROR)
10422		return (retval);
10423
10424	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10425		return (CTL_ACTION_ERROR);
10426
10427	return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10428}
10429
10430static ctl_action
10431ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10432{
10433	uint64_t lba1, lba2;
10434	uint64_t len1, len2;
10435
10436	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10437		return (CTL_ACTION_ERROR);
10438	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10439		return (CTL_ACTION_ERROR);
10440
10441	if (lba1 + len1 == lba2)
10442		return (CTL_ACTION_BLOCK);
10443	return (CTL_ACTION_PASS);
10444}
10445
10446static ctl_action
10447ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10448    union ctl_io *ooa_io)
10449{
10450	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10451	ctl_serialize_action *serialize_row;
10452
10453	/*
10454	 * The initiator attempted multiple untagged commands at the same
10455	 * time.  Can't do that.
10456	 */
10457	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10458	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10459	 && ((pending_io->io_hdr.nexus.targ_port ==
10460	      ooa_io->io_hdr.nexus.targ_port)
10461	  && (pending_io->io_hdr.nexus.initid.id ==
10462	      ooa_io->io_hdr.nexus.initid.id))
10463	 && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10464	      CTL_FLAG_STATUS_SENT)) == 0))
10465		return (CTL_ACTION_OVERLAP);
10466
10467	/*
10468	 * The initiator attempted to send multiple tagged commands with
10469	 * the same ID.  (It's fine if different initiators have the same
10470	 * tag ID.)
10471	 *
10472	 * Even if all of those conditions are true, we don't kill the I/O
10473	 * if the command ahead of us has been aborted.  We won't end up
10474	 * sending it to the FETD, and it's perfectly legal to resend a
10475	 * command with the same tag number as long as the previous
10476	 * instance of this tag number has been aborted somehow.
10477	 */
10478	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10479	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10480	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10481	 && ((pending_io->io_hdr.nexus.targ_port ==
10482	      ooa_io->io_hdr.nexus.targ_port)
10483	  && (pending_io->io_hdr.nexus.initid.id ==
10484	      ooa_io->io_hdr.nexus.initid.id))
10485	 && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10486	      CTL_FLAG_STATUS_SENT)) == 0))
10487		return (CTL_ACTION_OVERLAP_TAG);
10488
10489	/*
10490	 * If we get a head of queue tag, SAM-3 says that we should
10491	 * immediately execute it.
10492	 *
10493	 * What happens if this command would normally block for some other
10494	 * reason?  e.g. a request sense with a head of queue tag
10495	 * immediately after a write.  Normally that would block, but this
10496	 * will result in its getting executed immediately...
10497	 *
10498	 * We currently return "pass" instead of "skip", so we'll end up
10499	 * going through the rest of the queue to check for overlapped tags.
10500	 *
10501	 * XXX KDM check for other types of blockage first??
10502	 */
10503	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10504		return (CTL_ACTION_PASS);
10505
10506	/*
10507	 * Ordered tags have to block until all items ahead of them
10508	 * have completed.  If we get called with an ordered tag, we always
10509	 * block, if something else is ahead of us in the queue.
10510	 */
10511	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10512		return (CTL_ACTION_BLOCK);
10513
10514	/*
10515	 * Simple tags get blocked until all head of queue and ordered tags
10516	 * ahead of them have completed.  I'm lumping untagged commands in
10517	 * with simple tags here.  XXX KDM is that the right thing to do?
10518	 */
10519	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10520	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10521	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10522	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10523		return (CTL_ACTION_BLOCK);
10524
10525	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10526	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10527
10528	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10529
10530	switch (serialize_row[pending_entry->seridx]) {
10531	case CTL_SER_BLOCK:
10532		return (CTL_ACTION_BLOCK);
10533	case CTL_SER_EXTENT:
10534		return (ctl_extent_check(ooa_io, pending_io,
10535		    (lun->serseq == CTL_LUN_SERSEQ_ON)));
10536	case CTL_SER_EXTENTOPT:
10537		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10538		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10539			return (ctl_extent_check(ooa_io, pending_io,
10540			    (lun->serseq == CTL_LUN_SERSEQ_ON)));
10541		return (CTL_ACTION_PASS);
10542	case CTL_SER_EXTENTSEQ:
10543		if (lun->serseq != CTL_LUN_SERSEQ_OFF)
10544			return (ctl_extent_check_seq(ooa_io, pending_io));
10545		return (CTL_ACTION_PASS);
10546	case CTL_SER_PASS:
10547		return (CTL_ACTION_PASS);
10548	case CTL_SER_BLOCKOPT:
10549		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10550		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10551			return (CTL_ACTION_BLOCK);
10552		return (CTL_ACTION_PASS);
10553	case CTL_SER_SKIP:
10554		return (CTL_ACTION_SKIP);
10555	default:
10556		panic("invalid serialization value %d",
10557		      serialize_row[pending_entry->seridx]);
10558	}
10559
10560	return (CTL_ACTION_ERROR);
10561}
10562
10563/*
10564 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10565 * Assumptions:
10566 * - pending_io is generally either incoming, or on the blocked queue
10567 * - starting I/O is the I/O we want to start the check with.
10568 */
10569static ctl_action
10570ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10571	      union ctl_io *starting_io)
10572{
10573	union ctl_io *ooa_io;
10574	ctl_action action;
10575
10576	mtx_assert(&lun->lun_lock, MA_OWNED);
10577
10578	/*
10579	 * Run back along the OOA queue, starting with the current
10580	 * blocked I/O and going through every I/O before it on the
10581	 * queue.  If starting_io is NULL, we'll just end up returning
10582	 * CTL_ACTION_PASS.
10583	 */
10584	for (ooa_io = starting_io; ooa_io != NULL;
10585	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10586	     ooa_links)){
10587
10588		/*
10589		 * This routine just checks to see whether
10590		 * cur_blocked is blocked by ooa_io, which is ahead
10591		 * of it in the queue.  It doesn't queue/dequeue
10592		 * cur_blocked.
10593		 */
10594		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10595		switch (action) {
10596		case CTL_ACTION_BLOCK:
10597		case CTL_ACTION_OVERLAP:
10598		case CTL_ACTION_OVERLAP_TAG:
10599		case CTL_ACTION_SKIP:
10600		case CTL_ACTION_ERROR:
10601			return (action);
10602			break; /* NOTREACHED */
10603		case CTL_ACTION_PASS:
10604			break;
10605		default:
10606			panic("invalid action %d", action);
10607			break;  /* NOTREACHED */
10608		}
10609	}
10610
10611	return (CTL_ACTION_PASS);
10612}
10613
10614/*
10615 * Assumptions:
10616 * - An I/O has just completed, and has been removed from the per-LUN OOA
10617 *   queue, so some items on the blocked queue may now be unblocked.
10618 */
10619static int
10620ctl_check_blocked(struct ctl_lun *lun)
10621{
10622	union ctl_io *cur_blocked, *next_blocked;
10623
10624	mtx_assert(&lun->lun_lock, MA_OWNED);
10625
10626	/*
10627	 * Run forward from the head of the blocked queue, checking each
10628	 * entry against the I/Os prior to it on the OOA queue to see if
10629	 * there is still any blockage.
10630	 *
10631	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10632	 * with our removing a variable on it while it is traversing the
10633	 * list.
10634	 */
10635	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10636	     cur_blocked != NULL; cur_blocked = next_blocked) {
10637		union ctl_io *prev_ooa;
10638		ctl_action action;
10639
10640		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10641							  blocked_links);
10642
10643		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10644						      ctl_ooaq, ooa_links);
10645
10646		/*
10647		 * If cur_blocked happens to be the first item in the OOA
10648		 * queue now, prev_ooa will be NULL, and the action
10649		 * returned will just be CTL_ACTION_PASS.
10650		 */
10651		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10652
10653		switch (action) {
10654		case CTL_ACTION_BLOCK:
10655			/* Nothing to do here, still blocked */
10656			break;
10657		case CTL_ACTION_OVERLAP:
10658		case CTL_ACTION_OVERLAP_TAG:
10659			/*
10660			 * This shouldn't happen!  In theory we've already
10661			 * checked this command for overlap...
10662			 */
10663			break;
10664		case CTL_ACTION_PASS:
10665		case CTL_ACTION_SKIP: {
10666			const struct ctl_cmd_entry *entry;
10667			int isc_retval;
10668
10669			/*
10670			 * The skip case shouldn't happen, this transaction
10671			 * should have never made it onto the blocked queue.
10672			 */
10673			/*
10674			 * This I/O is no longer blocked, we can remove it
10675			 * from the blocked queue.  Since this is a TAILQ
10676			 * (doubly linked list), we can do O(1) removals
10677			 * from any place on the list.
10678			 */
10679			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10680				     blocked_links);
10681			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10682
10683			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10684				/*
10685				 * Need to send IO back to original side to
10686				 * run
10687				 */
10688				union ctl_ha_msg msg_info;
10689
10690				msg_info.hdr.original_sc =
10691					cur_blocked->io_hdr.original_sc;
10692				msg_info.hdr.serializing_sc = cur_blocked;
10693				msg_info.hdr.msg_type = CTL_MSG_R2R;
10694				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10695				     &msg_info, sizeof(msg_info), 0)) >
10696				     CTL_HA_STATUS_SUCCESS) {
10697					printf("CTL:Check Blocked error from "
10698					       "ctl_ha_msg_send %d\n",
10699					       isc_retval);
10700				}
10701				break;
10702			}
10703			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
10704
10705			/*
10706			 * Check this I/O for LUN state changes that may
10707			 * have happened while this command was blocked.
10708			 * The LUN state may have been changed by a command
10709			 * ahead of us in the queue, so we need to re-check
10710			 * for any states that can be caused by SCSI
10711			 * commands.
10712			 */
10713			if (ctl_scsiio_lun_check(lun, entry,
10714						 &cur_blocked->scsiio) == 0) {
10715				cur_blocked->io_hdr.flags |=
10716				                      CTL_FLAG_IS_WAS_ON_RTR;
10717				ctl_enqueue_rtr(cur_blocked);
10718			} else
10719				ctl_done(cur_blocked);
10720			break;
10721		}
10722		default:
10723			/*
10724			 * This probably shouldn't happen -- we shouldn't
10725			 * get CTL_ACTION_ERROR, or anything else.
10726			 */
10727			break;
10728		}
10729	}
10730
10731	return (CTL_RETVAL_COMPLETE);
10732}
10733
10734/*
10735 * This routine (with one exception) checks LUN flags that can be set by
10736 * commands ahead of us in the OOA queue.  These flags have to be checked
10737 * when a command initially comes in, and when we pull a command off the
10738 * blocked queue and are preparing to execute it.  The reason we have to
10739 * check these flags for commands on the blocked queue is that the LUN
10740 * state may have been changed by a command ahead of us while we're on the
10741 * blocked queue.
10742 *
10743 * Ordering is somewhat important with these checks, so please pay
10744 * careful attention to the placement of any new checks.
10745 */
10746static int
10747ctl_scsiio_lun_check(struct ctl_lun *lun,
10748    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10749{
10750	struct ctl_softc *softc = lun->ctl_softc;
10751	int retval;
10752	uint32_t residx;
10753
10754	retval = 0;
10755
10756	mtx_assert(&lun->lun_lock, MA_OWNED);
10757
10758	/*
10759	 * If this shelf is a secondary shelf controller, we have to reject
10760	 * any media access commands.
10761	 */
10762	if ((softc->flags & CTL_FLAG_ACTIVE_SHELF) == 0 &&
10763	    (entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0) {
10764		ctl_set_lun_standby(ctsio);
10765		retval = 1;
10766		goto bailout;
10767	}
10768
10769	if (entry->pattern & CTL_LUN_PAT_WRITE) {
10770		if (lun->flags & CTL_LUN_READONLY) {
10771			ctl_set_sense(ctsio, /*current_error*/ 1,
10772			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
10773			    /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
10774			retval = 1;
10775			goto bailout;
10776		}
10777		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
10778		    .eca_and_aen & SCP_SWP) != 0) {
10779			ctl_set_sense(ctsio, /*current_error*/ 1,
10780			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
10781			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
10782			retval = 1;
10783			goto bailout;
10784		}
10785	}
10786
10787	/*
10788	 * Check for a reservation conflict.  If this command isn't allowed
10789	 * even on reserved LUNs, and if this initiator isn't the one who
10790	 * reserved us, reject the command with a reservation conflict.
10791	 */
10792	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10793	if ((lun->flags & CTL_LUN_RESERVED)
10794	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10795		if (lun->res_idx != residx) {
10796			ctl_set_reservation_conflict(ctsio);
10797			retval = 1;
10798			goto bailout;
10799		}
10800	}
10801
10802	if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
10803	    (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
10804		/* No reservation or command is allowed. */;
10805	} else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
10806	    (lun->res_type == SPR_TYPE_WR_EX ||
10807	     lun->res_type == SPR_TYPE_WR_EX_RO ||
10808	     lun->res_type == SPR_TYPE_WR_EX_AR)) {
10809		/* The command is allowed for Write Exclusive resv. */;
10810	} else {
10811		/*
10812		 * if we aren't registered or it's a res holder type
10813		 * reservation and this isn't the res holder then set a
10814		 * conflict.
10815		 */
10816		if (ctl_get_prkey(lun, residx) == 0
10817		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10818			ctl_set_reservation_conflict(ctsio);
10819			retval = 1;
10820			goto bailout;
10821		}
10822
10823	}
10824
10825	if ((lun->flags & CTL_LUN_OFFLINE)
10826	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10827		ctl_set_lun_not_ready(ctsio);
10828		retval = 1;
10829		goto bailout;
10830	}
10831
10832	/*
10833	 * If the LUN is stopped, see if this particular command is allowed
10834	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10835	 */
10836	if ((lun->flags & CTL_LUN_STOPPED)
10837	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10838		/* "Logical unit not ready, initializing cmd. required" */
10839		ctl_set_lun_stopped(ctsio);
10840		retval = 1;
10841		goto bailout;
10842	}
10843
10844	if ((lun->flags & CTL_LUN_INOPERABLE)
10845	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10846		/* "Medium format corrupted" */
10847		ctl_set_medium_format_corrupted(ctsio);
10848		retval = 1;
10849		goto bailout;
10850	}
10851
10852bailout:
10853	return (retval);
10854
10855}
10856
10857static void
10858ctl_failover_io(union ctl_io *io, int have_lock)
10859{
10860	ctl_set_busy(&io->scsiio);
10861	ctl_done(io);
10862}
10863
10864#ifdef notyet
10865static void
10866ctl_failover(void)
10867{
10868	struct ctl_lun *lun;
10869	struct ctl_softc *softc;
10870	union ctl_io *next_io, *pending_io;
10871	union ctl_io *io;
10872	int lun_idx;
10873
10874	softc = control_softc;
10875
10876	mtx_lock(&softc->ctl_lock);
10877	/*
10878	 * Remove any cmds from the other SC from the rtr queue.  These
10879	 * will obviously only be for LUNs for which we're the primary.
10880	 * We can't send status or get/send data for these commands.
10881	 * Since they haven't been executed yet, we can just remove them.
10882	 * We'll either abort them or delete them below, depending on
10883	 * which HA mode we're in.
10884	 */
10885#ifdef notyet
10886	mtx_lock(&softc->queue_lock);
10887	for (io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue);
10888	     io != NULL; io = next_io) {
10889		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10890		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10891			STAILQ_REMOVE(&softc->rtr_queue, &io->io_hdr,
10892				      ctl_io_hdr, links);
10893	}
10894	mtx_unlock(&softc->queue_lock);
10895#endif
10896
10897	for (lun_idx=0; lun_idx < softc->num_luns; lun_idx++) {
10898		lun = softc->ctl_luns[lun_idx];
10899		if (lun==NULL)
10900			continue;
10901
10902		/*
10903		 * Processor LUNs are primary on both sides.
10904		 * XXX will this always be true?
10905		 */
10906		if (lun->be_lun->lun_type == T_PROCESSOR)
10907			continue;
10908
10909		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10910		 && (softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10911			printf("FAILOVER: primary lun %d\n", lun_idx);
10912		        /*
10913			 * Remove all commands from the other SC. First from the
10914			 * blocked queue then from the ooa queue. Once we have
10915			 * removed them. Call ctl_check_blocked to see if there
10916			 * is anything that can run.
10917			 */
10918			for (io = (union ctl_io *)TAILQ_FIRST(
10919			     &lun->blocked_queue); io != NULL; io = next_io) {
10920
10921		        	next_io = (union ctl_io *)TAILQ_NEXT(
10922				    &io->io_hdr, blocked_links);
10923
10924				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10925					TAILQ_REMOVE(&lun->blocked_queue,
10926						     &io->io_hdr,blocked_links);
10927					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10928					TAILQ_REMOVE(&lun->ooa_queue,
10929						     &io->io_hdr, ooa_links);
10930
10931					ctl_free_io(io);
10932				}
10933			}
10934
10935			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10936	     		     io != NULL; io = next_io) {
10937
10938		        	next_io = (union ctl_io *)TAILQ_NEXT(
10939				    &io->io_hdr, ooa_links);
10940
10941				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10942
10943					TAILQ_REMOVE(&lun->ooa_queue,
10944						&io->io_hdr,
10945					     	ooa_links);
10946
10947					ctl_free_io(io);
10948				}
10949			}
10950			ctl_check_blocked(lun);
10951		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10952			&& (softc->ha_mode == CTL_HA_MODE_XFER)) {
10953
10954			printf("FAILOVER: primary lun %d\n", lun_idx);
10955			/*
10956			 * Abort all commands from the other SC.  We can't
10957			 * send status back for them now.  These should get
10958			 * cleaned up when they are completed or come out
10959			 * for a datamove operation.
10960			 */
10961			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10962	     		     io != NULL; io = next_io) {
10963		        	next_io = (union ctl_io *)TAILQ_NEXT(
10964					&io->io_hdr, ooa_links);
10965
10966				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10967					io->io_hdr.flags |= CTL_FLAG_ABORT;
10968			}
10969		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10970			&& (softc->ha_mode == CTL_HA_MODE_XFER)) {
10971
10972			printf("FAILOVER: secondary lun %d\n", lun_idx);
10973
10974			lun->flags |= CTL_LUN_PRIMARY_SC;
10975
10976			/*
10977			 * We send all I/O that was sent to this controller
10978			 * and redirected to the other side back with
10979			 * busy status, and have the initiator retry it.
10980			 * Figuring out how much data has been transferred,
10981			 * etc. and picking up where we left off would be
10982			 * very tricky.
10983			 *
10984			 * XXX KDM need to remove I/O from the blocked
10985			 * queue as well!
10986			 */
10987			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10988			     &lun->ooa_queue); pending_io != NULL;
10989			     pending_io = next_io) {
10990
10991				next_io =  (union ctl_io *)TAILQ_NEXT(
10992					&pending_io->io_hdr, ooa_links);
10993
10994				pending_io->io_hdr.flags &=
10995					~CTL_FLAG_SENT_2OTHER_SC;
10996
10997				if (pending_io->io_hdr.flags &
10998				    CTL_FLAG_IO_ACTIVE) {
10999					pending_io->io_hdr.flags |=
11000						CTL_FLAG_FAILOVER;
11001				} else {
11002					ctl_set_busy(&pending_io->scsiio);
11003					ctl_done(pending_io);
11004				}
11005			}
11006
11007			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
11008		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11009			&& (softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11010			printf("FAILOVER: secondary lun %d\n", lun_idx);
11011			/*
11012			 * if the first io on the OOA is not on the RtR queue
11013			 * add it.
11014			 */
11015			lun->flags |= CTL_LUN_PRIMARY_SC;
11016
11017			pending_io = (union ctl_io *)TAILQ_FIRST(
11018			    &lun->ooa_queue);
11019			if (pending_io==NULL) {
11020				printf("Nothing on OOA queue\n");
11021				continue;
11022			}
11023
11024			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11025			if ((pending_io->io_hdr.flags &
11026			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11027				pending_io->io_hdr.flags |=
11028				    CTL_FLAG_IS_WAS_ON_RTR;
11029				ctl_enqueue_rtr(pending_io);
11030			}
11031#if 0
11032			else
11033			{
11034				printf("Tag 0x%04x is running\n",
11035				      pending_io->scsiio.tag_num);
11036			}
11037#endif
11038
11039			next_io = (union ctl_io *)TAILQ_NEXT(
11040			    &pending_io->io_hdr, ooa_links);
11041			for (pending_io=next_io; pending_io != NULL;
11042			     pending_io = next_io) {
11043				pending_io->io_hdr.flags &=
11044				    ~CTL_FLAG_SENT_2OTHER_SC;
11045				next_io = (union ctl_io *)TAILQ_NEXT(
11046					&pending_io->io_hdr, ooa_links);
11047				if (pending_io->io_hdr.flags &
11048				    CTL_FLAG_IS_WAS_ON_RTR) {
11049#if 0
11050				        printf("Tag 0x%04x is running\n",
11051				      		pending_io->scsiio.tag_num);
11052#endif
11053					continue;
11054				}
11055
11056				switch (ctl_check_ooa(lun, pending_io,
11057			            (union ctl_io *)TAILQ_PREV(
11058				    &pending_io->io_hdr, ctl_ooaq,
11059				    ooa_links))) {
11060
11061				case CTL_ACTION_BLOCK:
11062					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11063							  &pending_io->io_hdr,
11064							  blocked_links);
11065					pending_io->io_hdr.flags |=
11066					    CTL_FLAG_BLOCKED;
11067					break;
11068				case CTL_ACTION_PASS:
11069				case CTL_ACTION_SKIP:
11070					pending_io->io_hdr.flags |=
11071					    CTL_FLAG_IS_WAS_ON_RTR;
11072					ctl_enqueue_rtr(pending_io);
11073					break;
11074				case CTL_ACTION_OVERLAP:
11075					ctl_set_overlapped_cmd(
11076					    (struct ctl_scsiio *)pending_io);
11077					ctl_done(pending_io);
11078					break;
11079				case CTL_ACTION_OVERLAP_TAG:
11080					ctl_set_overlapped_tag(
11081					    (struct ctl_scsiio *)pending_io,
11082					    pending_io->scsiio.tag_num & 0xff);
11083					ctl_done(pending_io);
11084					break;
11085				case CTL_ACTION_ERROR:
11086				default:
11087					ctl_set_internal_failure(
11088						(struct ctl_scsiio *)pending_io,
11089						0,  // sks_valid
11090						0); //retry count
11091					ctl_done(pending_io);
11092					break;
11093				}
11094			}
11095
11096			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
11097		} else {
11098			panic("Unhandled HA mode failover, LUN flags = %#x, "
11099			      "ha_mode = #%x", lun->flags, softc->ha_mode);
11100		}
11101	}
11102	ctl_pause_rtr = 0;
11103	mtx_unlock(&softc->ctl_lock);
11104}
11105#endif
11106
11107static int
11108ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11109{
11110	struct ctl_lun *lun;
11111	const struct ctl_cmd_entry *entry;
11112	uint32_t initidx, targ_lun;
11113	int retval;
11114
11115	retval = 0;
11116
11117	lun = NULL;
11118
11119	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11120	if ((targ_lun < CTL_MAX_LUNS)
11121	 && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11122		/*
11123		 * If the LUN is invalid, pretend that it doesn't exist.
11124		 * It will go away as soon as all pending I/O has been
11125		 * completed.
11126		 */
11127		mtx_lock(&lun->lun_lock);
11128		if (lun->flags & CTL_LUN_DISABLED) {
11129			mtx_unlock(&lun->lun_lock);
11130			lun = NULL;
11131			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11132			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11133		} else {
11134			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11135			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11136				lun->be_lun;
11137			if (lun->be_lun->lun_type == T_PROCESSOR) {
11138				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11139			}
11140
11141			/*
11142			 * Every I/O goes into the OOA queue for a
11143			 * particular LUN, and stays there until completion.
11144			 */
11145#ifdef CTL_TIME_IO
11146			if (TAILQ_EMPTY(&lun->ooa_queue)) {
11147				lun->idle_time += getsbinuptime() -
11148				    lun->last_busy;
11149			}
11150#endif
11151			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11152			    ooa_links);
11153		}
11154	} else {
11155		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11156		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11157	}
11158
11159	/* Get command entry and return error if it is unsuppotyed. */
11160	entry = ctl_validate_command(ctsio);
11161	if (entry == NULL) {
11162		if (lun)
11163			mtx_unlock(&lun->lun_lock);
11164		return (retval);
11165	}
11166
11167	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11168	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11169
11170	/*
11171	 * Check to see whether we can send this command to LUNs that don't
11172	 * exist.  This should pretty much only be the case for inquiry
11173	 * and request sense.  Further checks, below, really require having
11174	 * a LUN, so we can't really check the command anymore.  Just put
11175	 * it on the rtr queue.
11176	 */
11177	if (lun == NULL) {
11178		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11179			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11180			ctl_enqueue_rtr((union ctl_io *)ctsio);
11181			return (retval);
11182		}
11183
11184		ctl_set_unsupported_lun(ctsio);
11185		ctl_done((union ctl_io *)ctsio);
11186		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11187		return (retval);
11188	} else {
11189		/*
11190		 * Make sure we support this particular command on this LUN.
11191		 * e.g., we don't support writes to the control LUN.
11192		 */
11193		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11194			mtx_unlock(&lun->lun_lock);
11195			ctl_set_invalid_opcode(ctsio);
11196			ctl_done((union ctl_io *)ctsio);
11197			return (retval);
11198		}
11199	}
11200
11201	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11202
11203#ifdef CTL_WITH_CA
11204	/*
11205	 * If we've got a request sense, it'll clear the contingent
11206	 * allegiance condition.  Otherwise, if we have a CA condition for
11207	 * this initiator, clear it, because it sent down a command other
11208	 * than request sense.
11209	 */
11210	if ((ctsio->cdb[0] != REQUEST_SENSE)
11211	 && (ctl_is_set(lun->have_ca, initidx)))
11212		ctl_clear_mask(lun->have_ca, initidx);
11213#endif
11214
11215	/*
11216	 * If the command has this flag set, it handles its own unit
11217	 * attention reporting, we shouldn't do anything.  Otherwise we
11218	 * check for any pending unit attentions, and send them back to the
11219	 * initiator.  We only do this when a command initially comes in,
11220	 * not when we pull it off the blocked queue.
11221	 *
11222	 * According to SAM-3, section 5.3.2, the order that things get
11223	 * presented back to the host is basically unit attentions caused
11224	 * by some sort of reset event, busy status, reservation conflicts
11225	 * or task set full, and finally any other status.
11226	 *
11227	 * One issue here is that some of the unit attentions we report
11228	 * don't fall into the "reset" category (e.g. "reported luns data
11229	 * has changed").  So reporting it here, before the reservation
11230	 * check, may be technically wrong.  I guess the only thing to do
11231	 * would be to check for and report the reset events here, and then
11232	 * check for the other unit attention types after we check for a
11233	 * reservation conflict.
11234	 *
11235	 * XXX KDM need to fix this
11236	 */
11237	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11238		ctl_ua_type ua_type;
11239		scsi_sense_data_type sense_format;
11240
11241		if (lun->flags & CTL_LUN_SENSE_DESC)
11242			sense_format = SSD_TYPE_DESC;
11243		else
11244			sense_format = SSD_TYPE_FIXED;
11245
11246		ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11247		    sense_format);
11248		if (ua_type != CTL_UA_NONE) {
11249			mtx_unlock(&lun->lun_lock);
11250			ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11251			ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11252			ctsio->sense_len = SSD_FULL_SIZE;
11253			ctl_done((union ctl_io *)ctsio);
11254			return (retval);
11255		}
11256	}
11257
11258
11259	if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11260		mtx_unlock(&lun->lun_lock);
11261		ctl_done((union ctl_io *)ctsio);
11262		return (retval);
11263	}
11264
11265	/*
11266	 * XXX CHD this is where we want to send IO to other side if
11267	 * this LUN is secondary on this SC. We will need to make a copy
11268	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11269	 * the copy we send as FROM_OTHER.
11270	 * We also need to stuff the address of the original IO so we can
11271	 * find it easily. Something similar will need be done on the other
11272	 * side so when we are done we can find the copy.
11273	 */
11274	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11275		union ctl_ha_msg msg_info;
11276		int isc_retval;
11277
11278		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11279
11280		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11281		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11282#if 0
11283		printf("1. ctsio %p\n", ctsio);
11284#endif
11285		msg_info.hdr.serializing_sc = NULL;
11286		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11287		msg_info.scsi.tag_num = ctsio->tag_num;
11288		msg_info.scsi.tag_type = ctsio->tag_type;
11289		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11290
11291		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11292
11293		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11294		    (void *)&msg_info, sizeof(msg_info), 0)) >
11295		    CTL_HA_STATUS_SUCCESS) {
11296			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11297			       isc_retval);
11298			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11299		} else {
11300#if 0
11301			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11302#endif
11303		}
11304
11305		/*
11306		 * XXX KDM this I/O is off the incoming queue, but hasn't
11307		 * been inserted on any other queue.  We may need to come
11308		 * up with a holding queue while we wait for serialization
11309		 * so that we have an idea of what we're waiting for from
11310		 * the other side.
11311		 */
11312		mtx_unlock(&lun->lun_lock);
11313		return (retval);
11314	}
11315
11316	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11317			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11318			      ctl_ooaq, ooa_links))) {
11319	case CTL_ACTION_BLOCK:
11320		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11321		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11322				  blocked_links);
11323		mtx_unlock(&lun->lun_lock);
11324		return (retval);
11325	case CTL_ACTION_PASS:
11326	case CTL_ACTION_SKIP:
11327		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11328		mtx_unlock(&lun->lun_lock);
11329		ctl_enqueue_rtr((union ctl_io *)ctsio);
11330		break;
11331	case CTL_ACTION_OVERLAP:
11332		mtx_unlock(&lun->lun_lock);
11333		ctl_set_overlapped_cmd(ctsio);
11334		ctl_done((union ctl_io *)ctsio);
11335		break;
11336	case CTL_ACTION_OVERLAP_TAG:
11337		mtx_unlock(&lun->lun_lock);
11338		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11339		ctl_done((union ctl_io *)ctsio);
11340		break;
11341	case CTL_ACTION_ERROR:
11342	default:
11343		mtx_unlock(&lun->lun_lock);
11344		ctl_set_internal_failure(ctsio,
11345					 /*sks_valid*/ 0,
11346					 /*retry_count*/ 0);
11347		ctl_done((union ctl_io *)ctsio);
11348		break;
11349	}
11350	return (retval);
11351}
11352
11353const struct ctl_cmd_entry *
11354ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11355{
11356	const struct ctl_cmd_entry *entry;
11357	int service_action;
11358
11359	entry = &ctl_cmd_table[ctsio->cdb[0]];
11360	if (sa)
11361		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11362	if (entry->flags & CTL_CMD_FLAG_SA5) {
11363		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11364		entry = &((const struct ctl_cmd_entry *)
11365		    entry->execute)[service_action];
11366	}
11367	return (entry);
11368}
11369
11370const struct ctl_cmd_entry *
11371ctl_validate_command(struct ctl_scsiio *ctsio)
11372{
11373	const struct ctl_cmd_entry *entry;
11374	int i, sa;
11375	uint8_t diff;
11376
11377	entry = ctl_get_cmd_entry(ctsio, &sa);
11378	if (entry->execute == NULL) {
11379		if (sa)
11380			ctl_set_invalid_field(ctsio,
11381					      /*sks_valid*/ 1,
11382					      /*command*/ 1,
11383					      /*field*/ 1,
11384					      /*bit_valid*/ 1,
11385					      /*bit*/ 4);
11386		else
11387			ctl_set_invalid_opcode(ctsio);
11388		ctl_done((union ctl_io *)ctsio);
11389		return (NULL);
11390	}
11391	KASSERT(entry->length > 0,
11392	    ("Not defined length for command 0x%02x/0x%02x",
11393	     ctsio->cdb[0], ctsio->cdb[1]));
11394	for (i = 1; i < entry->length; i++) {
11395		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11396		if (diff == 0)
11397			continue;
11398		ctl_set_invalid_field(ctsio,
11399				      /*sks_valid*/ 1,
11400				      /*command*/ 1,
11401				      /*field*/ i,
11402				      /*bit_valid*/ 1,
11403				      /*bit*/ fls(diff) - 1);
11404		ctl_done((union ctl_io *)ctsio);
11405		return (NULL);
11406	}
11407	return (entry);
11408}
11409
11410static int
11411ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11412{
11413
11414	switch (lun_type) {
11415	case T_PROCESSOR:
11416		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11417		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11418			return (0);
11419		break;
11420	case T_DIRECT:
11421		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11422		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11423			return (0);
11424		break;
11425	default:
11426		return (0);
11427	}
11428	return (1);
11429}
11430
11431static int
11432ctl_scsiio(struct ctl_scsiio *ctsio)
11433{
11434	int retval;
11435	const struct ctl_cmd_entry *entry;
11436
11437	retval = CTL_RETVAL_COMPLETE;
11438
11439	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11440
11441	entry = ctl_get_cmd_entry(ctsio, NULL);
11442
11443	/*
11444	 * If this I/O has been aborted, just send it straight to
11445	 * ctl_done() without executing it.
11446	 */
11447	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11448		ctl_done((union ctl_io *)ctsio);
11449		goto bailout;
11450	}
11451
11452	/*
11453	 * All the checks should have been handled by ctl_scsiio_precheck().
11454	 * We should be clear now to just execute the I/O.
11455	 */
11456	retval = entry->execute(ctsio);
11457
11458bailout:
11459	return (retval);
11460}
11461
11462/*
11463 * Since we only implement one target right now, a bus reset simply resets
11464 * our single target.
11465 */
11466static int
11467ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11468{
11469	return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11470}
11471
11472static int
11473ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11474		 ctl_ua_type ua_type)
11475{
11476	struct ctl_lun *lun;
11477	int retval;
11478
11479	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11480		union ctl_ha_msg msg_info;
11481
11482		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11483		msg_info.hdr.nexus = io->io_hdr.nexus;
11484		if (ua_type==CTL_UA_TARG_RESET)
11485			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11486		else
11487			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11488		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11489		msg_info.hdr.original_sc = NULL;
11490		msg_info.hdr.serializing_sc = NULL;
11491		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11492		    (void *)&msg_info, sizeof(msg_info), 0)) {
11493		}
11494	}
11495	retval = 0;
11496
11497	mtx_lock(&softc->ctl_lock);
11498	STAILQ_FOREACH(lun, &softc->lun_list, links)
11499		retval += ctl_lun_reset(lun, io, ua_type);
11500	mtx_unlock(&softc->ctl_lock);
11501
11502	return (retval);
11503}
11504
11505/*
11506 * The LUN should always be set.  The I/O is optional, and is used to
11507 * distinguish between I/Os sent by this initiator, and by other
11508 * initiators.  We set unit attention for initiators other than this one.
11509 * SAM-3 is vague on this point.  It does say that a unit attention should
11510 * be established for other initiators when a LUN is reset (see section
11511 * 5.7.3), but it doesn't specifically say that the unit attention should
11512 * be established for this particular initiator when a LUN is reset.  Here
11513 * is the relevant text, from SAM-3 rev 8:
11514 *
11515 * 5.7.2 When a SCSI initiator port aborts its own tasks
11516 *
11517 * When a SCSI initiator port causes its own task(s) to be aborted, no
11518 * notification that the task(s) have been aborted shall be returned to
11519 * the SCSI initiator port other than the completion response for the
11520 * command or task management function action that caused the task(s) to
11521 * be aborted and notification(s) associated with related effects of the
11522 * action (e.g., a reset unit attention condition).
11523 *
11524 * XXX KDM for now, we're setting unit attention for all initiators.
11525 */
11526static int
11527ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11528{
11529	union ctl_io *xio;
11530#if 0
11531	uint32_t initidx;
11532#endif
11533#ifdef CTL_WITH_CA
11534	int i;
11535#endif
11536
11537	mtx_lock(&lun->lun_lock);
11538	/*
11539	 * Run through the OOA queue and abort each I/O.
11540	 */
11541	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11542	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11543		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11544	}
11545
11546	/*
11547	 * This version sets unit attention for every
11548	 */
11549#if 0
11550	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11551	ctl_est_ua_all(lun, initidx, ua_type);
11552#else
11553	ctl_est_ua_all(lun, -1, ua_type);
11554#endif
11555
11556	/*
11557	 * A reset (any kind, really) clears reservations established with
11558	 * RESERVE/RELEASE.  It does not clear reservations established
11559	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11560	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11561	 * reservations made with the RESERVE/RELEASE commands, because
11562	 * those commands are obsolete in SPC-3.
11563	 */
11564	lun->flags &= ~CTL_LUN_RESERVED;
11565
11566#ifdef CTL_WITH_CA
11567	for (i = 0; i < CTL_MAX_INITIATORS; i++)
11568		ctl_clear_mask(lun->have_ca, i);
11569#endif
11570	mtx_unlock(&lun->lun_lock);
11571
11572	return (0);
11573}
11574
11575static void
11576ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11577    int other_sc)
11578{
11579	union ctl_io *xio;
11580
11581	mtx_assert(&lun->lun_lock, MA_OWNED);
11582
11583	/*
11584	 * Run through the OOA queue and attempt to find the given I/O.
11585	 * The target port, initiator ID, tag type and tag number have to
11586	 * match the values that we got from the initiator.  If we have an
11587	 * untagged command to abort, simply abort the first untagged command
11588	 * we come to.  We only allow one untagged command at a time of course.
11589	 */
11590	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11591	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11592
11593		if ((targ_port == UINT32_MAX ||
11594		     targ_port == xio->io_hdr.nexus.targ_port) &&
11595		    (init_id == UINT32_MAX ||
11596		     init_id == xio->io_hdr.nexus.initid.id)) {
11597			if (targ_port != xio->io_hdr.nexus.targ_port ||
11598			    init_id != xio->io_hdr.nexus.initid.id)
11599				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11600			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11601			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11602				union ctl_ha_msg msg_info;
11603
11604				msg_info.hdr.nexus = xio->io_hdr.nexus;
11605				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11606				msg_info.task.tag_num = xio->scsiio.tag_num;
11607				msg_info.task.tag_type = xio->scsiio.tag_type;
11608				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11609				msg_info.hdr.original_sc = NULL;
11610				msg_info.hdr.serializing_sc = NULL;
11611				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11612				    (void *)&msg_info, sizeof(msg_info), 0);
11613			}
11614		}
11615	}
11616}
11617
11618static int
11619ctl_abort_task_set(union ctl_io *io)
11620{
11621	struct ctl_softc *softc = control_softc;
11622	struct ctl_lun *lun;
11623	uint32_t targ_lun;
11624
11625	/*
11626	 * Look up the LUN.
11627	 */
11628	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11629	mtx_lock(&softc->ctl_lock);
11630	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
11631		lun = softc->ctl_luns[targ_lun];
11632	else {
11633		mtx_unlock(&softc->ctl_lock);
11634		return (1);
11635	}
11636
11637	mtx_lock(&lun->lun_lock);
11638	mtx_unlock(&softc->ctl_lock);
11639	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11640		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11641		    io->io_hdr.nexus.initid.id,
11642		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11643	} else { /* CTL_TASK_CLEAR_TASK_SET */
11644		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11645		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11646	}
11647	mtx_unlock(&lun->lun_lock);
11648	return (0);
11649}
11650
11651static int
11652ctl_i_t_nexus_reset(union ctl_io *io)
11653{
11654	struct ctl_softc *softc = control_softc;
11655	struct ctl_lun *lun;
11656	uint32_t initidx, residx;
11657
11658	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11659	residx = ctl_get_resindex(&io->io_hdr.nexus);
11660	mtx_lock(&softc->ctl_lock);
11661	STAILQ_FOREACH(lun, &softc->lun_list, links) {
11662		mtx_lock(&lun->lun_lock);
11663		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11664		    io->io_hdr.nexus.initid.id,
11665		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11666#ifdef CTL_WITH_CA
11667		ctl_clear_mask(lun->have_ca, initidx);
11668#endif
11669		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
11670			lun->flags &= ~CTL_LUN_RESERVED;
11671		ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
11672		mtx_unlock(&lun->lun_lock);
11673	}
11674	mtx_unlock(&softc->ctl_lock);
11675	return (0);
11676}
11677
11678static int
11679ctl_abort_task(union ctl_io *io)
11680{
11681	union ctl_io *xio;
11682	struct ctl_lun *lun;
11683	struct ctl_softc *softc;
11684#if 0
11685	struct sbuf sb;
11686	char printbuf[128];
11687#endif
11688	int found;
11689	uint32_t targ_lun;
11690
11691	softc = control_softc;
11692	found = 0;
11693
11694	/*
11695	 * Look up the LUN.
11696	 */
11697	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11698	mtx_lock(&softc->ctl_lock);
11699	if ((targ_lun < CTL_MAX_LUNS)
11700	 && (softc->ctl_luns[targ_lun] != NULL))
11701		lun = softc->ctl_luns[targ_lun];
11702	else {
11703		mtx_unlock(&softc->ctl_lock);
11704		return (1);
11705	}
11706
11707#if 0
11708	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11709	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11710#endif
11711
11712	mtx_lock(&lun->lun_lock);
11713	mtx_unlock(&softc->ctl_lock);
11714	/*
11715	 * Run through the OOA queue and attempt to find the given I/O.
11716	 * The target port, initiator ID, tag type and tag number have to
11717	 * match the values that we got from the initiator.  If we have an
11718	 * untagged command to abort, simply abort the first untagged command
11719	 * we come to.  We only allow one untagged command at a time of course.
11720	 */
11721	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11722	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11723#if 0
11724		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11725
11726		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11727			    lun->lun, xio->scsiio.tag_num,
11728			    xio->scsiio.tag_type,
11729			    (xio->io_hdr.blocked_links.tqe_prev
11730			    == NULL) ? "" : " BLOCKED",
11731			    (xio->io_hdr.flags &
11732			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11733			    (xio->io_hdr.flags &
11734			    CTL_FLAG_ABORT) ? " ABORT" : "",
11735			    (xio->io_hdr.flags &
11736			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11737		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11738		sbuf_finish(&sb);
11739		printf("%s\n", sbuf_data(&sb));
11740#endif
11741
11742		if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11743		 || (xio->io_hdr.nexus.initid.id != io->io_hdr.nexus.initid.id)
11744		 || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11745			continue;
11746
11747		/*
11748		 * If the abort says that the task is untagged, the
11749		 * task in the queue must be untagged.  Otherwise,
11750		 * we just check to see whether the tag numbers
11751		 * match.  This is because the QLogic firmware
11752		 * doesn't pass back the tag type in an abort
11753		 * request.
11754		 */
11755#if 0
11756		if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11757		  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11758		 || (xio->scsiio.tag_num == io->taskio.tag_num))
11759#endif
11760		/*
11761		 * XXX KDM we've got problems with FC, because it
11762		 * doesn't send down a tag type with aborts.  So we
11763		 * can only really go by the tag number...
11764		 * This may cause problems with parallel SCSI.
11765		 * Need to figure that out!!
11766		 */
11767		if (xio->scsiio.tag_num == io->taskio.tag_num) {
11768			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11769			found = 1;
11770			if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
11771			    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11772				union ctl_ha_msg msg_info;
11773
11774				io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11775				msg_info.hdr.nexus = io->io_hdr.nexus;
11776				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11777				msg_info.task.tag_num = io->taskio.tag_num;
11778				msg_info.task.tag_type = io->taskio.tag_type;
11779				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11780				msg_info.hdr.original_sc = NULL;
11781				msg_info.hdr.serializing_sc = NULL;
11782#if 0
11783				printf("Sent Abort to other side\n");
11784#endif
11785				if (ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11786				    (void *)&msg_info, sizeof(msg_info), 0) !=
11787				    CTL_HA_STATUS_SUCCESS) {
11788				}
11789			}
11790#if 0
11791			printf("ctl_abort_task: found I/O to abort\n");
11792#endif
11793		}
11794	}
11795	mtx_unlock(&lun->lun_lock);
11796
11797	if (found == 0) {
11798		/*
11799		 * This isn't really an error.  It's entirely possible for
11800		 * the abort and command completion to cross on the wire.
11801		 * This is more of an informative/diagnostic error.
11802		 */
11803#if 0
11804		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11805		       "%d:%d:%d:%d tag %d type %d\n",
11806		       io->io_hdr.nexus.initid.id,
11807		       io->io_hdr.nexus.targ_port,
11808		       io->io_hdr.nexus.targ_target.id,
11809		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11810		       io->taskio.tag_type);
11811#endif
11812	}
11813	return (0);
11814}
11815
11816static void
11817ctl_run_task(union ctl_io *io)
11818{
11819	struct ctl_softc *softc = control_softc;
11820	int retval = 1;
11821	const char *task_desc;
11822
11823	CTL_DEBUG_PRINT(("ctl_run_task\n"));
11824
11825	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
11826	    ("ctl_run_task: Unextected io_type %d\n",
11827	     io->io_hdr.io_type));
11828
11829	task_desc = ctl_scsi_task_string(&io->taskio);
11830	if (task_desc != NULL) {
11831#ifdef NEEDTOPORT
11832		csevent_log(CSC_CTL | CSC_SHELF_SW |
11833			    CTL_TASK_REPORT,
11834			    csevent_LogType_Trace,
11835			    csevent_Severity_Information,
11836			    csevent_AlertLevel_Green,
11837			    csevent_FRU_Firmware,
11838			    csevent_FRU_Unknown,
11839			    "CTL: received task: %s",task_desc);
11840#endif
11841	} else {
11842#ifdef NEEDTOPORT
11843		csevent_log(CSC_CTL | CSC_SHELF_SW |
11844			    CTL_TASK_REPORT,
11845			    csevent_LogType_Trace,
11846			    csevent_Severity_Information,
11847			    csevent_AlertLevel_Green,
11848			    csevent_FRU_Firmware,
11849			    csevent_FRU_Unknown,
11850			    "CTL: received unknown task "
11851			    "type: %d (%#x)",
11852			    io->taskio.task_action,
11853			    io->taskio.task_action);
11854#endif
11855	}
11856	switch (io->taskio.task_action) {
11857	case CTL_TASK_ABORT_TASK:
11858		retval = ctl_abort_task(io);
11859		break;
11860	case CTL_TASK_ABORT_TASK_SET:
11861	case CTL_TASK_CLEAR_TASK_SET:
11862		retval = ctl_abort_task_set(io);
11863		break;
11864	case CTL_TASK_CLEAR_ACA:
11865		break;
11866	case CTL_TASK_I_T_NEXUS_RESET:
11867		retval = ctl_i_t_nexus_reset(io);
11868		break;
11869	case CTL_TASK_LUN_RESET: {
11870		struct ctl_lun *lun;
11871		uint32_t targ_lun;
11872
11873		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11874		mtx_lock(&softc->ctl_lock);
11875		if ((targ_lun < CTL_MAX_LUNS)
11876		 && (softc->ctl_luns[targ_lun] != NULL))
11877			lun = softc->ctl_luns[targ_lun];
11878		else {
11879			mtx_unlock(&softc->ctl_lock);
11880			retval = 1;
11881			break;
11882		}
11883
11884		if (!(io->io_hdr.flags &
11885		    CTL_FLAG_FROM_OTHER_SC)) {
11886			union ctl_ha_msg msg_info;
11887
11888			io->io_hdr.flags |=
11889				CTL_FLAG_SENT_2OTHER_SC;
11890			msg_info.hdr.msg_type =
11891				CTL_MSG_MANAGE_TASKS;
11892			msg_info.hdr.nexus = io->io_hdr.nexus;
11893			msg_info.task.task_action =
11894				CTL_TASK_LUN_RESET;
11895			msg_info.hdr.original_sc = NULL;
11896			msg_info.hdr.serializing_sc = NULL;
11897			if (CTL_HA_STATUS_SUCCESS !=
11898			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11899			    (void *)&msg_info,
11900			    sizeof(msg_info), 0)) {
11901			}
11902		}
11903
11904		retval = ctl_lun_reset(lun, io,
11905				       CTL_UA_LUN_RESET);
11906		mtx_unlock(&softc->ctl_lock);
11907		break;
11908	}
11909	case CTL_TASK_TARGET_RESET:
11910		retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
11911		break;
11912	case CTL_TASK_BUS_RESET:
11913		retval = ctl_bus_reset(softc, io);
11914		break;
11915	case CTL_TASK_PORT_LOGIN:
11916		break;
11917	case CTL_TASK_PORT_LOGOUT:
11918		break;
11919	default:
11920		printf("ctl_run_task: got unknown task management event %d\n",
11921		       io->taskio.task_action);
11922		break;
11923	}
11924	if (retval == 0)
11925		io->io_hdr.status = CTL_SUCCESS;
11926	else
11927		io->io_hdr.status = CTL_ERROR;
11928	ctl_done(io);
11929}
11930
11931/*
11932 * For HA operation.  Handle commands that come in from the other
11933 * controller.
11934 */
11935static void
11936ctl_handle_isc(union ctl_io *io)
11937{
11938	int free_io;
11939	struct ctl_lun *lun;
11940	struct ctl_softc *softc;
11941	uint32_t targ_lun;
11942
11943	softc = control_softc;
11944
11945	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11946	lun = softc->ctl_luns[targ_lun];
11947
11948	switch (io->io_hdr.msg_type) {
11949	case CTL_MSG_SERIALIZE:
11950		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
11951		break;
11952	case CTL_MSG_R2R: {
11953		const struct ctl_cmd_entry *entry;
11954
11955		/*
11956		 * This is only used in SER_ONLY mode.
11957		 */
11958		free_io = 0;
11959		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
11960		mtx_lock(&lun->lun_lock);
11961		if (ctl_scsiio_lun_check(lun,
11962		    entry, (struct ctl_scsiio *)io) != 0) {
11963			mtx_unlock(&lun->lun_lock);
11964			ctl_done(io);
11965			break;
11966		}
11967		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11968		mtx_unlock(&lun->lun_lock);
11969		ctl_enqueue_rtr(io);
11970		break;
11971	}
11972	case CTL_MSG_FINISH_IO:
11973		if (softc->ha_mode == CTL_HA_MODE_XFER) {
11974			free_io = 0;
11975			ctl_done(io);
11976		} else {
11977			free_io = 1;
11978			mtx_lock(&lun->lun_lock);
11979			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11980				     ooa_links);
11981			ctl_check_blocked(lun);
11982			mtx_unlock(&lun->lun_lock);
11983		}
11984		break;
11985	case CTL_MSG_PERS_ACTION:
11986		ctl_hndl_per_res_out_on_other_sc(
11987			(union ctl_ha_msg *)&io->presio.pr_msg);
11988		free_io = 1;
11989		break;
11990	case CTL_MSG_BAD_JUJU:
11991		free_io = 0;
11992		ctl_done(io);
11993		break;
11994	case CTL_MSG_DATAMOVE:
11995		/* Only used in XFER mode */
11996		free_io = 0;
11997		ctl_datamove_remote(io);
11998		break;
11999	case CTL_MSG_DATAMOVE_DONE:
12000		/* Only used in XFER mode */
12001		free_io = 0;
12002		io->scsiio.be_move_done(io);
12003		break;
12004	default:
12005		free_io = 1;
12006		printf("%s: Invalid message type %d\n",
12007		       __func__, io->io_hdr.msg_type);
12008		break;
12009	}
12010	if (free_io)
12011		ctl_free_io(io);
12012
12013}
12014
12015
12016/*
12017 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12018 * there is no match.
12019 */
12020static ctl_lun_error_pattern
12021ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12022{
12023	const struct ctl_cmd_entry *entry;
12024	ctl_lun_error_pattern filtered_pattern, pattern;
12025
12026	pattern = desc->error_pattern;
12027
12028	/*
12029	 * XXX KDM we need more data passed into this function to match a
12030	 * custom pattern, and we actually need to implement custom pattern
12031	 * matching.
12032	 */
12033	if (pattern & CTL_LUN_PAT_CMD)
12034		return (CTL_LUN_PAT_CMD);
12035
12036	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12037		return (CTL_LUN_PAT_ANY);
12038
12039	entry = ctl_get_cmd_entry(ctsio, NULL);
12040
12041	filtered_pattern = entry->pattern & pattern;
12042
12043	/*
12044	 * If the user requested specific flags in the pattern (e.g.
12045	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12046	 * flags.
12047	 *
12048	 * If the user did not specify any flags, it doesn't matter whether
12049	 * or not the command supports the flags.
12050	 */
12051	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12052	     (pattern & ~CTL_LUN_PAT_MASK))
12053		return (CTL_LUN_PAT_NONE);
12054
12055	/*
12056	 * If the user asked for a range check, see if the requested LBA
12057	 * range overlaps with this command's LBA range.
12058	 */
12059	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12060		uint64_t lba1;
12061		uint64_t len1;
12062		ctl_action action;
12063		int retval;
12064
12065		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12066		if (retval != 0)
12067			return (CTL_LUN_PAT_NONE);
12068
12069		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12070					      desc->lba_range.len, FALSE);
12071		/*
12072		 * A "pass" means that the LBA ranges don't overlap, so
12073		 * this doesn't match the user's range criteria.
12074		 */
12075		if (action == CTL_ACTION_PASS)
12076			return (CTL_LUN_PAT_NONE);
12077	}
12078
12079	return (filtered_pattern);
12080}
12081
12082static void
12083ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12084{
12085	struct ctl_error_desc *desc, *desc2;
12086
12087	mtx_assert(&lun->lun_lock, MA_OWNED);
12088
12089	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12090		ctl_lun_error_pattern pattern;
12091		/*
12092		 * Check to see whether this particular command matches
12093		 * the pattern in the descriptor.
12094		 */
12095		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12096		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12097			continue;
12098
12099		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12100		case CTL_LUN_INJ_ABORTED:
12101			ctl_set_aborted(&io->scsiio);
12102			break;
12103		case CTL_LUN_INJ_MEDIUM_ERR:
12104			ctl_set_medium_error(&io->scsiio);
12105			break;
12106		case CTL_LUN_INJ_UA:
12107			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12108			 * OCCURRED */
12109			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12110			break;
12111		case CTL_LUN_INJ_CUSTOM:
12112			/*
12113			 * We're assuming the user knows what he is doing.
12114			 * Just copy the sense information without doing
12115			 * checks.
12116			 */
12117			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12118			      MIN(sizeof(desc->custom_sense),
12119				  sizeof(io->scsiio.sense_data)));
12120			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12121			io->scsiio.sense_len = SSD_FULL_SIZE;
12122			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12123			break;
12124		case CTL_LUN_INJ_NONE:
12125		default:
12126			/*
12127			 * If this is an error injection type we don't know
12128			 * about, clear the continuous flag (if it is set)
12129			 * so it will get deleted below.
12130			 */
12131			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12132			break;
12133		}
12134		/*
12135		 * By default, each error injection action is a one-shot
12136		 */
12137		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12138			continue;
12139
12140		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12141
12142		free(desc, M_CTL);
12143	}
12144}
12145
12146#ifdef CTL_IO_DELAY
12147static void
12148ctl_datamove_timer_wakeup(void *arg)
12149{
12150	union ctl_io *io;
12151
12152	io = (union ctl_io *)arg;
12153
12154	ctl_datamove(io);
12155}
12156#endif /* CTL_IO_DELAY */
12157
12158void
12159ctl_datamove(union ctl_io *io)
12160{
12161	void (*fe_datamove)(union ctl_io *io);
12162
12163	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12164
12165	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12166
12167#ifdef CTL_TIME_IO
12168	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12169		char str[256];
12170		char path_str[64];
12171		struct sbuf sb;
12172
12173		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12174		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12175
12176		sbuf_cat(&sb, path_str);
12177		switch (io->io_hdr.io_type) {
12178		case CTL_IO_SCSI:
12179			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12180			sbuf_printf(&sb, "\n");
12181			sbuf_cat(&sb, path_str);
12182			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12183				    io->scsiio.tag_num, io->scsiio.tag_type);
12184			break;
12185		case CTL_IO_TASK:
12186			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12187				    "Tag Type: %d\n", io->taskio.task_action,
12188				    io->taskio.tag_num, io->taskio.tag_type);
12189			break;
12190		default:
12191			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12192			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12193			break;
12194		}
12195		sbuf_cat(&sb, path_str);
12196		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12197			    (intmax_t)time_uptime - io->io_hdr.start_time);
12198		sbuf_finish(&sb);
12199		printf("%s", sbuf_data(&sb));
12200	}
12201#endif /* CTL_TIME_IO */
12202
12203#ifdef CTL_IO_DELAY
12204	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12205		struct ctl_lun *lun;
12206
12207		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12208
12209		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12210	} else {
12211		struct ctl_lun *lun;
12212
12213		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12214		if ((lun != NULL)
12215		 && (lun->delay_info.datamove_delay > 0)) {
12216			struct callout *callout;
12217
12218			callout = (struct callout *)&io->io_hdr.timer_bytes;
12219			callout_init(callout, /*mpsafe*/ 1);
12220			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12221			callout_reset(callout,
12222				      lun->delay_info.datamove_delay * hz,
12223				      ctl_datamove_timer_wakeup, io);
12224			if (lun->delay_info.datamove_type ==
12225			    CTL_DELAY_TYPE_ONESHOT)
12226				lun->delay_info.datamove_delay = 0;
12227			return;
12228		}
12229	}
12230#endif
12231
12232	/*
12233	 * This command has been aborted.  Set the port status, so we fail
12234	 * the data move.
12235	 */
12236	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12237		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12238		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12239		       io->io_hdr.nexus.targ_port,
12240		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12241		       io->io_hdr.nexus.targ_lun);
12242		io->io_hdr.port_status = 31337;
12243		/*
12244		 * Note that the backend, in this case, will get the
12245		 * callback in its context.  In other cases it may get
12246		 * called in the frontend's interrupt thread context.
12247		 */
12248		io->scsiio.be_move_done(io);
12249		return;
12250	}
12251
12252	/* Don't confuse frontend with zero length data move. */
12253	if (io->scsiio.kern_data_len == 0) {
12254		io->scsiio.be_move_done(io);
12255		return;
12256	}
12257
12258	/*
12259	 * If we're in XFER mode and this I/O is from the other shelf
12260	 * controller, we need to send the DMA to the other side to
12261	 * actually transfer the data to/from the host.  In serialize only
12262	 * mode the transfer happens below CTL and ctl_datamove() is only
12263	 * called on the machine that originally received the I/O.
12264	 */
12265	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12266	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12267		union ctl_ha_msg msg;
12268		uint32_t sg_entries_sent;
12269		int do_sg_copy;
12270		int i;
12271
12272		memset(&msg, 0, sizeof(msg));
12273		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12274		msg.hdr.original_sc = io->io_hdr.original_sc;
12275		msg.hdr.serializing_sc = io;
12276		msg.hdr.nexus = io->io_hdr.nexus;
12277		msg.dt.flags = io->io_hdr.flags;
12278		/*
12279		 * We convert everything into a S/G list here.  We can't
12280		 * pass by reference, only by value between controllers.
12281		 * So we can't pass a pointer to the S/G list, only as many
12282		 * S/G entries as we can fit in here.  If it's possible for
12283		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12284		 * then we need to break this up into multiple transfers.
12285		 */
12286		if (io->scsiio.kern_sg_entries == 0) {
12287			msg.dt.kern_sg_entries = 1;
12288			/*
12289			 * If this is in cached memory, flush the cache
12290			 * before we send the DMA request to the other
12291			 * controller.  We want to do this in either the
12292			 * read or the write case.  The read case is
12293			 * straightforward.  In the write case, we want to
12294			 * make sure nothing is in the local cache that
12295			 * could overwrite the DMAed data.
12296			 */
12297			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12298				/*
12299				 * XXX KDM use bus_dmamap_sync() here.
12300				 */
12301			}
12302
12303			/*
12304			 * Convert to a physical address if this is a
12305			 * virtual address.
12306			 */
12307			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12308				msg.dt.sg_list[0].addr =
12309					io->scsiio.kern_data_ptr;
12310			} else {
12311				/*
12312				 * XXX KDM use busdma here!
12313				 */
12314#if 0
12315				msg.dt.sg_list[0].addr = (void *)
12316					vtophys(io->scsiio.kern_data_ptr);
12317#endif
12318			}
12319
12320			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12321			do_sg_copy = 0;
12322		} else {
12323			struct ctl_sg_entry *sgl;
12324
12325			do_sg_copy = 1;
12326			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12327			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12328			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12329				/*
12330				 * XXX KDM use bus_dmamap_sync() here.
12331				 */
12332			}
12333		}
12334
12335		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12336		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12337		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12338		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12339		msg.dt.sg_sequence = 0;
12340
12341		/*
12342		 * Loop until we've sent all of the S/G entries.  On the
12343		 * other end, we'll recompose these S/G entries into one
12344		 * contiguous list before passing it to the
12345		 */
12346		for (sg_entries_sent = 0; sg_entries_sent <
12347		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12348			msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list)/
12349				sizeof(msg.dt.sg_list[0])),
12350				msg.dt.kern_sg_entries - sg_entries_sent);
12351
12352			if (do_sg_copy != 0) {
12353				struct ctl_sg_entry *sgl;
12354				int j;
12355
12356				sgl = (struct ctl_sg_entry *)
12357					io->scsiio.kern_data_ptr;
12358				/*
12359				 * If this is in cached memory, flush the cache
12360				 * before we send the DMA request to the other
12361				 * controller.  We want to do this in either
12362				 * the * read or the write case.  The read
12363				 * case is straightforward.  In the write
12364				 * case, we want to make sure nothing is
12365				 * in the local cache that could overwrite
12366				 * the DMAed data.
12367				 */
12368
12369				for (i = sg_entries_sent, j = 0;
12370				     i < msg.dt.cur_sg_entries; i++, j++) {
12371					if ((io->io_hdr.flags &
12372					     CTL_FLAG_NO_DATASYNC) == 0) {
12373						/*
12374						 * XXX KDM use bus_dmamap_sync()
12375						 */
12376					}
12377					if ((io->io_hdr.flags &
12378					     CTL_FLAG_BUS_ADDR) == 0) {
12379						/*
12380						 * XXX KDM use busdma.
12381						 */
12382#if 0
12383						msg.dt.sg_list[j].addr =(void *)
12384						       vtophys(sgl[i].addr);
12385#endif
12386					} else {
12387						msg.dt.sg_list[j].addr =
12388							sgl[i].addr;
12389					}
12390					msg.dt.sg_list[j].len = sgl[i].len;
12391				}
12392			}
12393
12394			sg_entries_sent += msg.dt.cur_sg_entries;
12395			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12396				msg.dt.sg_last = 1;
12397			else
12398				msg.dt.sg_last = 0;
12399
12400			/*
12401			 * XXX KDM drop and reacquire the lock here?
12402			 */
12403			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12404			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12405				/*
12406				 * XXX do something here.
12407				 */
12408			}
12409
12410			msg.dt.sent_sg_entries = sg_entries_sent;
12411		}
12412		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12413		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12414			ctl_failover_io(io, /*have_lock*/ 0);
12415
12416	} else {
12417
12418		/*
12419		 * Lookup the fe_datamove() function for this particular
12420		 * front end.
12421		 */
12422		fe_datamove =
12423		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12424
12425		fe_datamove(io);
12426	}
12427}
12428
12429static void
12430ctl_send_datamove_done(union ctl_io *io, int have_lock)
12431{
12432	union ctl_ha_msg msg;
12433	int isc_status;
12434
12435	memset(&msg, 0, sizeof(msg));
12436
12437	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12438	msg.hdr.original_sc = io;
12439	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12440	msg.hdr.nexus = io->io_hdr.nexus;
12441	msg.hdr.status = io->io_hdr.status;
12442	msg.scsi.tag_num = io->scsiio.tag_num;
12443	msg.scsi.tag_type = io->scsiio.tag_type;
12444	msg.scsi.scsi_status = io->scsiio.scsi_status;
12445	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12446	       sizeof(io->scsiio.sense_data));
12447	msg.scsi.sense_len = io->scsiio.sense_len;
12448	msg.scsi.sense_residual = io->scsiio.sense_residual;
12449	msg.scsi.fetd_status = io->io_hdr.port_status;
12450	msg.scsi.residual = io->scsiio.residual;
12451	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12452
12453	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12454		ctl_failover_io(io, /*have_lock*/ have_lock);
12455		return;
12456	}
12457
12458	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12459	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12460		/* XXX do something if this fails */
12461	}
12462
12463}
12464
12465/*
12466 * The DMA to the remote side is done, now we need to tell the other side
12467 * we're done so it can continue with its data movement.
12468 */
12469static void
12470ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12471{
12472	union ctl_io *io;
12473
12474	io = rq->context;
12475
12476	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12477		printf("%s: ISC DMA write failed with error %d", __func__,
12478		       rq->ret);
12479		ctl_set_internal_failure(&io->scsiio,
12480					 /*sks_valid*/ 1,
12481					 /*retry_count*/ rq->ret);
12482	}
12483
12484	ctl_dt_req_free(rq);
12485
12486	/*
12487	 * In this case, we had to malloc the memory locally.  Free it.
12488	 */
12489	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12490		int i;
12491		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12492			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12493	}
12494	/*
12495	 * The data is in local and remote memory, so now we need to send
12496	 * status (good or back) back to the other side.
12497	 */
12498	ctl_send_datamove_done(io, /*have_lock*/ 0);
12499}
12500
12501/*
12502 * We've moved the data from the host/controller into local memory.  Now we
12503 * need to push it over to the remote controller's memory.
12504 */
12505static int
12506ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12507{
12508	int retval;
12509
12510	retval = 0;
12511
12512	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12513					  ctl_datamove_remote_write_cb);
12514
12515	return (retval);
12516}
12517
12518static void
12519ctl_datamove_remote_write(union ctl_io *io)
12520{
12521	int retval;
12522	void (*fe_datamove)(union ctl_io *io);
12523
12524	/*
12525	 * - Get the data from the host/HBA into local memory.
12526	 * - DMA memory from the local controller to the remote controller.
12527	 * - Send status back to the remote controller.
12528	 */
12529
12530	retval = ctl_datamove_remote_sgl_setup(io);
12531	if (retval != 0)
12532		return;
12533
12534	/* Switch the pointer over so the FETD knows what to do */
12535	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12536
12537	/*
12538	 * Use a custom move done callback, since we need to send completion
12539	 * back to the other controller, not to the backend on this side.
12540	 */
12541	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12542
12543	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12544
12545	fe_datamove(io);
12546
12547	return;
12548
12549}
12550
12551static int
12552ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12553{
12554#if 0
12555	char str[256];
12556	char path_str[64];
12557	struct sbuf sb;
12558#endif
12559
12560	/*
12561	 * In this case, we had to malloc the memory locally.  Free it.
12562	 */
12563	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12564		int i;
12565		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12566			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12567	}
12568
12569#if 0
12570	scsi_path_string(io, path_str, sizeof(path_str));
12571	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12572	sbuf_cat(&sb, path_str);
12573	scsi_command_string(&io->scsiio, NULL, &sb);
12574	sbuf_printf(&sb, "\n");
12575	sbuf_cat(&sb, path_str);
12576	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12577		    io->scsiio.tag_num, io->scsiio.tag_type);
12578	sbuf_cat(&sb, path_str);
12579	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12580		    io->io_hdr.flags, io->io_hdr.status);
12581	sbuf_finish(&sb);
12582	printk("%s", sbuf_data(&sb));
12583#endif
12584
12585
12586	/*
12587	 * The read is done, now we need to send status (good or bad) back
12588	 * to the other side.
12589	 */
12590	ctl_send_datamove_done(io, /*have_lock*/ 0);
12591
12592	return (0);
12593}
12594
12595static void
12596ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12597{
12598	union ctl_io *io;
12599	void (*fe_datamove)(union ctl_io *io);
12600
12601	io = rq->context;
12602
12603	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12604		printf("%s: ISC DMA read failed with error %d", __func__,
12605		       rq->ret);
12606		ctl_set_internal_failure(&io->scsiio,
12607					 /*sks_valid*/ 1,
12608					 /*retry_count*/ rq->ret);
12609	}
12610
12611	ctl_dt_req_free(rq);
12612
12613	/* Switch the pointer over so the FETD knows what to do */
12614	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12615
12616	/*
12617	 * Use a custom move done callback, since we need to send completion
12618	 * back to the other controller, not to the backend on this side.
12619	 */
12620	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12621
12622	/* XXX KDM add checks like the ones in ctl_datamove? */
12623
12624	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12625
12626	fe_datamove(io);
12627}
12628
12629static int
12630ctl_datamove_remote_sgl_setup(union ctl_io *io)
12631{
12632	struct ctl_sg_entry *local_sglist, *remote_sglist;
12633	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12634	struct ctl_softc *softc;
12635	int retval;
12636	int i;
12637
12638	retval = 0;
12639	softc = control_softc;
12640
12641	local_sglist = io->io_hdr.local_sglist;
12642	local_dma_sglist = io->io_hdr.local_dma_sglist;
12643	remote_sglist = io->io_hdr.remote_sglist;
12644	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12645
12646	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12647		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12648			local_sglist[i].len = remote_sglist[i].len;
12649
12650			/*
12651			 * XXX Detect the situation where the RS-level I/O
12652			 * redirector on the other side has already read the
12653			 * data off of the AOR RS on this side, and
12654			 * transferred it to remote (mirror) memory on the
12655			 * other side.  Since we already have the data in
12656			 * memory here, we just need to use it.
12657			 *
12658			 * XXX KDM this can probably be removed once we
12659			 * get the cache device code in and take the
12660			 * current AOR implementation out.
12661			 */
12662#ifdef NEEDTOPORT
12663			if ((remote_sglist[i].addr >=
12664			     (void *)vtophys(softc->mirr->addr))
12665			 && (remote_sglist[i].addr <
12666			     ((void *)vtophys(softc->mirr->addr) +
12667			     CacheMirrorOffset))) {
12668				local_sglist[i].addr = remote_sglist[i].addr -
12669					CacheMirrorOffset;
12670				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12671				     CTL_FLAG_DATA_IN)
12672					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12673			} else {
12674				local_sglist[i].addr = remote_sglist[i].addr +
12675					CacheMirrorOffset;
12676			}
12677#endif
12678#if 0
12679			printf("%s: local %p, remote %p, len %d\n",
12680			       __func__, local_sglist[i].addr,
12681			       remote_sglist[i].addr, local_sglist[i].len);
12682#endif
12683		}
12684	} else {
12685		uint32_t len_to_go;
12686
12687		/*
12688		 * In this case, we don't have automatically allocated
12689		 * memory for this I/O on this controller.  This typically
12690		 * happens with internal CTL I/O -- e.g. inquiry, mode
12691		 * sense, etc.  Anything coming from RAIDCore will have
12692		 * a mirror area available.
12693		 */
12694		len_to_go = io->scsiio.kern_data_len;
12695
12696		/*
12697		 * Clear the no datasync flag, we have to use malloced
12698		 * buffers.
12699		 */
12700		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12701
12702		/*
12703		 * The difficult thing here is that the size of the various
12704		 * S/G segments may be different than the size from the
12705		 * remote controller.  That'll make it harder when DMAing
12706		 * the data back to the other side.
12707		 */
12708		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12709		     sizeof(io->io_hdr.remote_sglist[0])) &&
12710		     (len_to_go > 0); i++) {
12711			local_sglist[i].len = MIN(len_to_go, 131072);
12712			CTL_SIZE_8B(local_dma_sglist[i].len,
12713				    local_sglist[i].len);
12714			local_sglist[i].addr =
12715				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
12716
12717			local_dma_sglist[i].addr = local_sglist[i].addr;
12718
12719			if (local_sglist[i].addr == NULL) {
12720				int j;
12721
12722				printf("malloc failed for %zd bytes!",
12723				       local_dma_sglist[i].len);
12724				for (j = 0; j < i; j++) {
12725					free(local_sglist[j].addr, M_CTL);
12726				}
12727				ctl_set_internal_failure(&io->scsiio,
12728							 /*sks_valid*/ 1,
12729							 /*retry_count*/ 4857);
12730				retval = 1;
12731				goto bailout_error;
12732
12733			}
12734			/* XXX KDM do we need a sync here? */
12735
12736			len_to_go -= local_sglist[i].len;
12737		}
12738		/*
12739		 * Reset the number of S/G entries accordingly.  The
12740		 * original number of S/G entries is available in
12741		 * rem_sg_entries.
12742		 */
12743		io->scsiio.kern_sg_entries = i;
12744
12745#if 0
12746		printf("%s: kern_sg_entries = %d\n", __func__,
12747		       io->scsiio.kern_sg_entries);
12748		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12749			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
12750			       local_sglist[i].addr, local_sglist[i].len,
12751			       local_dma_sglist[i].len);
12752#endif
12753	}
12754
12755
12756	return (retval);
12757
12758bailout_error:
12759
12760	ctl_send_datamove_done(io, /*have_lock*/ 0);
12761
12762	return (retval);
12763}
12764
12765static int
12766ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12767			 ctl_ha_dt_cb callback)
12768{
12769	struct ctl_ha_dt_req *rq;
12770	struct ctl_sg_entry *remote_sglist, *local_sglist;
12771	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
12772	uint32_t local_used, remote_used, total_used;
12773	int retval;
12774	int i, j;
12775
12776	retval = 0;
12777
12778	rq = ctl_dt_req_alloc();
12779
12780	/*
12781	 * If we failed to allocate the request, and if the DMA didn't fail
12782	 * anyway, set busy status.  This is just a resource allocation
12783	 * failure.
12784	 */
12785	if ((rq == NULL)
12786	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12787		ctl_set_busy(&io->scsiio);
12788
12789	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12790
12791		if (rq != NULL)
12792			ctl_dt_req_free(rq);
12793
12794		/*
12795		 * The data move failed.  We need to return status back
12796		 * to the other controller.  No point in trying to DMA
12797		 * data to the remote controller.
12798		 */
12799
12800		ctl_send_datamove_done(io, /*have_lock*/ 0);
12801
12802		retval = 1;
12803
12804		goto bailout;
12805	}
12806
12807	local_sglist = io->io_hdr.local_sglist;
12808	local_dma_sglist = io->io_hdr.local_dma_sglist;
12809	remote_sglist = io->io_hdr.remote_sglist;
12810	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12811	local_used = 0;
12812	remote_used = 0;
12813	total_used = 0;
12814
12815	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12816		rq->ret = CTL_HA_STATUS_SUCCESS;
12817		rq->context = io;
12818		callback(rq);
12819		goto bailout;
12820	}
12821
12822	/*
12823	 * Pull/push the data over the wire from/to the other controller.
12824	 * This takes into account the possibility that the local and
12825	 * remote sglists may not be identical in terms of the size of
12826	 * the elements and the number of elements.
12827	 *
12828	 * One fundamental assumption here is that the length allocated for
12829	 * both the local and remote sglists is identical.  Otherwise, we've
12830	 * essentially got a coding error of some sort.
12831	 */
12832	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12833		int isc_ret;
12834		uint32_t cur_len, dma_length;
12835		uint8_t *tmp_ptr;
12836
12837		rq->id = CTL_HA_DATA_CTL;
12838		rq->command = command;
12839		rq->context = io;
12840
12841		/*
12842		 * Both pointers should be aligned.  But it is possible
12843		 * that the allocation length is not.  They should both
12844		 * also have enough slack left over at the end, though,
12845		 * to round up to the next 8 byte boundary.
12846		 */
12847		cur_len = MIN(local_sglist[i].len - local_used,
12848			      remote_sglist[j].len - remote_used);
12849
12850		/*
12851		 * In this case, we have a size issue and need to decrease
12852		 * the size, except in the case where we actually have less
12853		 * than 8 bytes left.  In that case, we need to increase
12854		 * the DMA length to get the last bit.
12855		 */
12856		if ((cur_len & 0x7) != 0) {
12857			if (cur_len > 0x7) {
12858				cur_len = cur_len - (cur_len & 0x7);
12859				dma_length = cur_len;
12860			} else {
12861				CTL_SIZE_8B(dma_length, cur_len);
12862			}
12863
12864		} else
12865			dma_length = cur_len;
12866
12867		/*
12868		 * If we had to allocate memory for this I/O, instead of using
12869		 * the non-cached mirror memory, we'll need to flush the cache
12870		 * before trying to DMA to the other controller.
12871		 *
12872		 * We could end up doing this multiple times for the same
12873		 * segment if we have a larger local segment than remote
12874		 * segment.  That shouldn't be an issue.
12875		 */
12876		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12877			/*
12878			 * XXX KDM use bus_dmamap_sync() here.
12879			 */
12880		}
12881
12882		rq->size = dma_length;
12883
12884		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12885		tmp_ptr += local_used;
12886
12887		/* Use physical addresses when talking to ISC hardware */
12888		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12889			/* XXX KDM use busdma */
12890#if 0
12891			rq->local = vtophys(tmp_ptr);
12892#endif
12893		} else
12894			rq->local = tmp_ptr;
12895
12896		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12897		tmp_ptr += remote_used;
12898		rq->remote = tmp_ptr;
12899
12900		rq->callback = NULL;
12901
12902		local_used += cur_len;
12903		if (local_used >= local_sglist[i].len) {
12904			i++;
12905			local_used = 0;
12906		}
12907
12908		remote_used += cur_len;
12909		if (remote_used >= remote_sglist[j].len) {
12910			j++;
12911			remote_used = 0;
12912		}
12913		total_used += cur_len;
12914
12915		if (total_used >= io->scsiio.kern_data_len)
12916			rq->callback = callback;
12917
12918		if ((rq->size & 0x7) != 0) {
12919			printf("%s: warning: size %d is not on 8b boundary\n",
12920			       __func__, rq->size);
12921		}
12922		if (((uintptr_t)rq->local & 0x7) != 0) {
12923			printf("%s: warning: local %p not on 8b boundary\n",
12924			       __func__, rq->local);
12925		}
12926		if (((uintptr_t)rq->remote & 0x7) != 0) {
12927			printf("%s: warning: remote %p not on 8b boundary\n",
12928			       __func__, rq->local);
12929		}
12930#if 0
12931		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12932		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12933		       rq->local, rq->remote, rq->size);
12934#endif
12935
12936		isc_ret = ctl_dt_single(rq);
12937		if (isc_ret == CTL_HA_STATUS_WAIT)
12938			continue;
12939
12940		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12941			rq->ret = CTL_HA_STATUS_SUCCESS;
12942		} else {
12943			rq->ret = isc_ret;
12944		}
12945		callback(rq);
12946		goto bailout;
12947	}
12948
12949bailout:
12950	return (retval);
12951
12952}
12953
12954static void
12955ctl_datamove_remote_read(union ctl_io *io)
12956{
12957	int retval;
12958	int i;
12959
12960	/*
12961	 * This will send an error to the other controller in the case of a
12962	 * failure.
12963	 */
12964	retval = ctl_datamove_remote_sgl_setup(io);
12965	if (retval != 0)
12966		return;
12967
12968	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12969					  ctl_datamove_remote_read_cb);
12970	if ((retval != 0)
12971	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12972		/*
12973		 * Make sure we free memory if there was an error..  The
12974		 * ctl_datamove_remote_xfer() function will send the
12975		 * datamove done message, or call the callback with an
12976		 * error if there is a problem.
12977		 */
12978		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12979			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12980	}
12981
12982	return;
12983}
12984
12985/*
12986 * Process a datamove request from the other controller.  This is used for
12987 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12988 * first.  Once that is complete, the data gets DMAed into the remote
12989 * controller's memory.  For reads, we DMA from the remote controller's
12990 * memory into our memory first, and then move it out to the FETD.
12991 */
12992static void
12993ctl_datamove_remote(union ctl_io *io)
12994{
12995	struct ctl_softc *softc;
12996
12997	softc = control_softc;
12998
12999	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13000
13001	/*
13002	 * Note that we look for an aborted I/O here, but don't do some of
13003	 * the other checks that ctl_datamove() normally does.
13004	 * We don't need to run the datamove delay code, since that should
13005	 * have been done if need be on the other controller.
13006	 */
13007	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13008		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13009		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13010		       io->io_hdr.nexus.targ_port,
13011		       io->io_hdr.nexus.targ_target.id,
13012		       io->io_hdr.nexus.targ_lun);
13013		io->io_hdr.port_status = 31338;
13014		ctl_send_datamove_done(io, /*have_lock*/ 0);
13015		return;
13016	}
13017
13018	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13019		ctl_datamove_remote_write(io);
13020	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13021		ctl_datamove_remote_read(io);
13022	} else {
13023		union ctl_ha_msg msg;
13024		struct scsi_sense_data *sense;
13025		uint8_t sks[3];
13026		int retry_count;
13027
13028		memset(&msg, 0, sizeof(msg));
13029
13030		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13031		msg.hdr.status = CTL_SCSI_ERROR;
13032		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13033
13034		retry_count = 4243;
13035
13036		sense = &msg.scsi.sense_data;
13037		sks[0] = SSD_SCS_VALID;
13038		sks[1] = (retry_count >> 8) & 0xff;
13039		sks[2] = retry_count & 0xff;
13040
13041		/* "Internal target failure" */
13042		scsi_set_sense_data(sense,
13043				    /*sense_format*/ SSD_TYPE_NONE,
13044				    /*current_error*/ 1,
13045				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13046				    /*asc*/ 0x44,
13047				    /*ascq*/ 0x00,
13048				    /*type*/ SSD_ELEM_SKS,
13049				    /*size*/ sizeof(sks),
13050				    /*data*/ sks,
13051				    SSD_ELEM_NONE);
13052
13053		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13054		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13055			ctl_failover_io(io, /*have_lock*/ 1);
13056			return;
13057		}
13058
13059		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13060		    CTL_HA_STATUS_SUCCESS) {
13061			/* XXX KDM what to do if this fails? */
13062		}
13063		return;
13064	}
13065
13066}
13067
13068static int
13069ctl_process_done(union ctl_io *io)
13070{
13071	struct ctl_lun *lun;
13072	struct ctl_softc *softc = control_softc;
13073	void (*fe_done)(union ctl_io *io);
13074	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13075
13076	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13077
13078	fe_done = softc->ctl_ports[targ_port]->fe_done;
13079
13080#ifdef CTL_TIME_IO
13081	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13082		char str[256];
13083		char path_str[64];
13084		struct sbuf sb;
13085
13086		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13087		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13088
13089		sbuf_cat(&sb, path_str);
13090		switch (io->io_hdr.io_type) {
13091		case CTL_IO_SCSI:
13092			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13093			sbuf_printf(&sb, "\n");
13094			sbuf_cat(&sb, path_str);
13095			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13096				    io->scsiio.tag_num, io->scsiio.tag_type);
13097			break;
13098		case CTL_IO_TASK:
13099			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13100				    "Tag Type: %d\n", io->taskio.task_action,
13101				    io->taskio.tag_num, io->taskio.tag_type);
13102			break;
13103		default:
13104			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13105			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13106			break;
13107		}
13108		sbuf_cat(&sb, path_str);
13109		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13110			    (intmax_t)time_uptime - io->io_hdr.start_time);
13111		sbuf_finish(&sb);
13112		printf("%s", sbuf_data(&sb));
13113	}
13114#endif /* CTL_TIME_IO */
13115
13116	switch (io->io_hdr.io_type) {
13117	case CTL_IO_SCSI:
13118		break;
13119	case CTL_IO_TASK:
13120		if (ctl_debug & CTL_DEBUG_INFO)
13121			ctl_io_error_print(io, NULL);
13122		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13123			ctl_free_io(io);
13124		else
13125			fe_done(io);
13126		return (CTL_RETVAL_COMPLETE);
13127	default:
13128		panic("ctl_process_done: invalid io type %d\n",
13129		      io->io_hdr.io_type);
13130		break; /* NOTREACHED */
13131	}
13132
13133	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13134	if (lun == NULL) {
13135		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13136				 io->io_hdr.nexus.targ_mapped_lun));
13137		goto bailout;
13138	}
13139
13140	mtx_lock(&lun->lun_lock);
13141
13142	/*
13143	 * Check to see if we have any errors to inject here.  We only
13144	 * inject errors for commands that don't already have errors set.
13145	 */
13146	if ((STAILQ_FIRST(&lun->error_list) != NULL) &&
13147	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13148	    ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13149		ctl_inject_error(lun, io);
13150
13151	/*
13152	 * XXX KDM how do we treat commands that aren't completed
13153	 * successfully?
13154	 *
13155	 * XXX KDM should we also track I/O latency?
13156	 */
13157	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13158	    io->io_hdr.io_type == CTL_IO_SCSI) {
13159#ifdef CTL_TIME_IO
13160		struct bintime cur_bt;
13161#endif
13162		int type;
13163
13164		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13165		    CTL_FLAG_DATA_IN)
13166			type = CTL_STATS_READ;
13167		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13168		    CTL_FLAG_DATA_OUT)
13169			type = CTL_STATS_WRITE;
13170		else
13171			type = CTL_STATS_NO_IO;
13172
13173		lun->stats.ports[targ_port].bytes[type] +=
13174		    io->scsiio.kern_total_len;
13175		lun->stats.ports[targ_port].operations[type]++;
13176#ifdef CTL_TIME_IO
13177		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13178		   &io->io_hdr.dma_bt);
13179		lun->stats.ports[targ_port].num_dmas[type] +=
13180		    io->io_hdr.num_dmas;
13181		getbintime(&cur_bt);
13182		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13183		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13184#endif
13185	}
13186
13187	/*
13188	 * Remove this from the OOA queue.
13189	 */
13190	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13191#ifdef CTL_TIME_IO
13192	if (TAILQ_EMPTY(&lun->ooa_queue))
13193		lun->last_busy = getsbinuptime();
13194#endif
13195
13196	/*
13197	 * Run through the blocked queue on this LUN and see if anything
13198	 * has become unblocked, now that this transaction is done.
13199	 */
13200	ctl_check_blocked(lun);
13201
13202	/*
13203	 * If the LUN has been invalidated, free it if there is nothing
13204	 * left on its OOA queue.
13205	 */
13206	if ((lun->flags & CTL_LUN_INVALID)
13207	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13208		mtx_unlock(&lun->lun_lock);
13209		mtx_lock(&softc->ctl_lock);
13210		ctl_free_lun(lun);
13211		mtx_unlock(&softc->ctl_lock);
13212	} else
13213		mtx_unlock(&lun->lun_lock);
13214
13215bailout:
13216
13217	/*
13218	 * If this command has been aborted, make sure we set the status
13219	 * properly.  The FETD is responsible for freeing the I/O and doing
13220	 * whatever it needs to do to clean up its state.
13221	 */
13222	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13223		ctl_set_task_aborted(&io->scsiio);
13224
13225	/*
13226	 * If enabled, print command error status.
13227	 */
13228	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13229	    (ctl_debug & CTL_DEBUG_INFO) != 0)
13230		ctl_io_error_print(io, NULL);
13231
13232	/*
13233	 * Tell the FETD or the other shelf controller we're done with this
13234	 * command.  Note that only SCSI commands get to this point.  Task
13235	 * management commands are completed above.
13236	 *
13237	 * We only send status to the other controller if we're in XFER
13238	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13239	 * received the I/O (from CTL's perspective), and so the status is
13240	 * generated there.
13241	 *
13242	 * XXX KDM if we hold the lock here, we could cause a deadlock
13243	 * if the frontend comes back in in this context to queue
13244	 * something.
13245	 */
13246	if ((softc->ha_mode == CTL_HA_MODE_XFER)
13247	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13248		union ctl_ha_msg msg;
13249
13250		memset(&msg, 0, sizeof(msg));
13251		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13252		msg.hdr.original_sc = io->io_hdr.original_sc;
13253		msg.hdr.nexus = io->io_hdr.nexus;
13254		msg.hdr.status = io->io_hdr.status;
13255		msg.scsi.scsi_status = io->scsiio.scsi_status;
13256		msg.scsi.tag_num = io->scsiio.tag_num;
13257		msg.scsi.tag_type = io->scsiio.tag_type;
13258		msg.scsi.sense_len = io->scsiio.sense_len;
13259		msg.scsi.sense_residual = io->scsiio.sense_residual;
13260		msg.scsi.residual = io->scsiio.residual;
13261		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13262		       sizeof(io->scsiio.sense_data));
13263		/*
13264		 * We copy this whether or not this is an I/O-related
13265		 * command.  Otherwise, we'd have to go and check to see
13266		 * whether it's a read/write command, and it really isn't
13267		 * worth it.
13268		 */
13269		memcpy(&msg.scsi.lbalen,
13270		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13271		       sizeof(msg.scsi.lbalen));
13272
13273		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13274				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13275			/* XXX do something here */
13276		}
13277
13278		ctl_free_io(io);
13279	} else
13280		fe_done(io);
13281
13282	return (CTL_RETVAL_COMPLETE);
13283}
13284
13285#ifdef CTL_WITH_CA
13286/*
13287 * Front end should call this if it doesn't do autosense.  When the request
13288 * sense comes back in from the initiator, we'll dequeue this and send it.
13289 */
13290int
13291ctl_queue_sense(union ctl_io *io)
13292{
13293	struct ctl_lun *lun;
13294	struct ctl_port *port;
13295	struct ctl_softc *softc;
13296	uint32_t initidx, targ_lun;
13297
13298	softc = control_softc;
13299
13300	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13301
13302	/*
13303	 * LUN lookup will likely move to the ctl_work_thread() once we
13304	 * have our new queueing infrastructure (that doesn't put things on
13305	 * a per-LUN queue initially).  That is so that we can handle
13306	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13307	 * can't deal with that right now.
13308	 */
13309	mtx_lock(&softc->ctl_lock);
13310
13311	/*
13312	 * If we don't have a LUN for this, just toss the sense
13313	 * information.
13314	 */
13315	port = ctl_io_port(&ctsio->io_hdr);
13316	targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13317	if ((targ_lun < CTL_MAX_LUNS)
13318	 && (softc->ctl_luns[targ_lun] != NULL))
13319		lun = softc->ctl_luns[targ_lun];
13320	else
13321		goto bailout;
13322
13323	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13324
13325	mtx_lock(&lun->lun_lock);
13326	/*
13327	 * Already have CA set for this LUN...toss the sense information.
13328	 */
13329	if (ctl_is_set(lun->have_ca, initidx)) {
13330		mtx_unlock(&lun->lun_lock);
13331		goto bailout;
13332	}
13333
13334	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13335	       MIN(sizeof(lun->pending_sense[initidx]),
13336	       sizeof(io->scsiio.sense_data)));
13337	ctl_set_mask(lun->have_ca, initidx);
13338	mtx_unlock(&lun->lun_lock);
13339
13340bailout:
13341	mtx_unlock(&softc->ctl_lock);
13342
13343	ctl_free_io(io);
13344
13345	return (CTL_RETVAL_COMPLETE);
13346}
13347#endif
13348
13349/*
13350 * Primary command inlet from frontend ports.  All SCSI and task I/O
13351 * requests must go through this function.
13352 */
13353int
13354ctl_queue(union ctl_io *io)
13355{
13356	struct ctl_port *port;
13357
13358	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13359
13360#ifdef CTL_TIME_IO
13361	io->io_hdr.start_time = time_uptime;
13362	getbintime(&io->io_hdr.start_bt);
13363#endif /* CTL_TIME_IO */
13364
13365	/* Map FE-specific LUN ID into global one. */
13366	port = ctl_io_port(&io->io_hdr);
13367	io->io_hdr.nexus.targ_mapped_lun =
13368	    ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13369
13370	switch (io->io_hdr.io_type) {
13371	case CTL_IO_SCSI:
13372	case CTL_IO_TASK:
13373		if (ctl_debug & CTL_DEBUG_CDB)
13374			ctl_io_print(io);
13375		ctl_enqueue_incoming(io);
13376		break;
13377	default:
13378		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13379		return (EINVAL);
13380	}
13381
13382	return (CTL_RETVAL_COMPLETE);
13383}
13384
13385#ifdef CTL_IO_DELAY
13386static void
13387ctl_done_timer_wakeup(void *arg)
13388{
13389	union ctl_io *io;
13390
13391	io = (union ctl_io *)arg;
13392	ctl_done(io);
13393}
13394#endif /* CTL_IO_DELAY */
13395
13396void
13397ctl_done(union ctl_io *io)
13398{
13399
13400	/*
13401	 * Enable this to catch duplicate completion issues.
13402	 */
13403#if 0
13404	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13405		printf("%s: type %d msg %d cdb %x iptl: "
13406		       "%d:%d:%d:%d tag 0x%04x "
13407		       "flag %#x status %x\n",
13408			__func__,
13409			io->io_hdr.io_type,
13410			io->io_hdr.msg_type,
13411			io->scsiio.cdb[0],
13412			io->io_hdr.nexus.initid.id,
13413			io->io_hdr.nexus.targ_port,
13414			io->io_hdr.nexus.targ_target.id,
13415			io->io_hdr.nexus.targ_lun,
13416			(io->io_hdr.io_type ==
13417			CTL_IO_TASK) ?
13418			io->taskio.tag_num :
13419			io->scsiio.tag_num,
13420		        io->io_hdr.flags,
13421			io->io_hdr.status);
13422	} else
13423		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13424#endif
13425
13426	/*
13427	 * This is an internal copy of an I/O, and should not go through
13428	 * the normal done processing logic.
13429	 */
13430	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13431		return;
13432
13433	/*
13434	 * We need to send a msg to the serializing shelf to finish the IO
13435	 * as well.  We don't send a finish message to the other shelf if
13436	 * this is a task management command.  Task management commands
13437	 * aren't serialized in the OOA queue, but rather just executed on
13438	 * both shelf controllers for commands that originated on that
13439	 * controller.
13440	 */
13441	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13442	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13443		union ctl_ha_msg msg_io;
13444
13445		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13446		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13447		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13448		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13449		}
13450		/* continue on to finish IO */
13451	}
13452#ifdef CTL_IO_DELAY
13453	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13454		struct ctl_lun *lun;
13455
13456		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13457
13458		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13459	} else {
13460		struct ctl_lun *lun;
13461
13462		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13463
13464		if ((lun != NULL)
13465		 && (lun->delay_info.done_delay > 0)) {
13466			struct callout *callout;
13467
13468			callout = (struct callout *)&io->io_hdr.timer_bytes;
13469			callout_init(callout, /*mpsafe*/ 1);
13470			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13471			callout_reset(callout,
13472				      lun->delay_info.done_delay * hz,
13473				      ctl_done_timer_wakeup, io);
13474			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13475				lun->delay_info.done_delay = 0;
13476			return;
13477		}
13478	}
13479#endif /* CTL_IO_DELAY */
13480
13481	ctl_enqueue_done(io);
13482}
13483
13484int
13485ctl_isc(struct ctl_scsiio *ctsio)
13486{
13487	struct ctl_lun *lun;
13488	int retval;
13489
13490	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13491
13492	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13493
13494	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13495
13496	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13497
13498	return (retval);
13499}
13500
13501
13502static void
13503ctl_work_thread(void *arg)
13504{
13505	struct ctl_thread *thr = (struct ctl_thread *)arg;
13506	struct ctl_softc *softc = thr->ctl_softc;
13507	union ctl_io *io;
13508	int retval;
13509
13510	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13511
13512	for (;;) {
13513		retval = 0;
13514
13515		/*
13516		 * We handle the queues in this order:
13517		 * - ISC
13518		 * - done queue (to free up resources, unblock other commands)
13519		 * - RtR queue
13520		 * - incoming queue
13521		 *
13522		 * If those queues are empty, we break out of the loop and
13523		 * go to sleep.
13524		 */
13525		mtx_lock(&thr->queue_lock);
13526		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13527		if (io != NULL) {
13528			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13529			mtx_unlock(&thr->queue_lock);
13530			ctl_handle_isc(io);
13531			continue;
13532		}
13533		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13534		if (io != NULL) {
13535			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13536			/* clear any blocked commands, call fe_done */
13537			mtx_unlock(&thr->queue_lock);
13538			retval = ctl_process_done(io);
13539			continue;
13540		}
13541		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13542		if (io != NULL) {
13543			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13544			mtx_unlock(&thr->queue_lock);
13545			if (io->io_hdr.io_type == CTL_IO_TASK)
13546				ctl_run_task(io);
13547			else
13548				ctl_scsiio_precheck(softc, &io->scsiio);
13549			continue;
13550		}
13551		if (!ctl_pause_rtr) {
13552			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13553			if (io != NULL) {
13554				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13555				mtx_unlock(&thr->queue_lock);
13556				retval = ctl_scsiio(&io->scsiio);
13557				if (retval != CTL_RETVAL_COMPLETE)
13558					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13559				continue;
13560			}
13561		}
13562
13563		/* Sleep until we have something to do. */
13564		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13565	}
13566}
13567
13568static void
13569ctl_lun_thread(void *arg)
13570{
13571	struct ctl_softc *softc = (struct ctl_softc *)arg;
13572	struct ctl_be_lun *be_lun;
13573	int retval;
13574
13575	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13576
13577	for (;;) {
13578		retval = 0;
13579		mtx_lock(&softc->ctl_lock);
13580		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13581		if (be_lun != NULL) {
13582			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13583			mtx_unlock(&softc->ctl_lock);
13584			ctl_create_lun(be_lun);
13585			continue;
13586		}
13587
13588		/* Sleep until we have something to do. */
13589		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13590		    PDROP | PRIBIO, "-", 0);
13591	}
13592}
13593
13594static void
13595ctl_thresh_thread(void *arg)
13596{
13597	struct ctl_softc *softc = (struct ctl_softc *)arg;
13598	struct ctl_lun *lun;
13599	struct ctl_be_lun *be_lun;
13600	struct scsi_da_rw_recovery_page *rwpage;
13601	struct ctl_logical_block_provisioning_page *page;
13602	const char *attr;
13603	uint64_t thres, val;
13604	int i, e;
13605
13606	CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13607
13608	for (;;) {
13609		mtx_lock(&softc->ctl_lock);
13610		STAILQ_FOREACH(lun, &softc->lun_list, links) {
13611			be_lun = lun->be_lun;
13612			if ((lun->flags & CTL_LUN_DISABLED) ||
13613			    (lun->flags & CTL_LUN_OFFLINE) ||
13614			    lun->backend->lun_attr == NULL)
13615				continue;
13616			rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
13617			if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
13618				continue;
13619			e = 0;
13620			page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
13621			for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13622				if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13623					continue;
13624				thres = scsi_4btoul(page->descr[i].count);
13625				thres <<= CTL_LBP_EXPONENT;
13626				switch (page->descr[i].resource) {
13627				case 0x01:
13628					attr = "blocksavail";
13629					break;
13630				case 0x02:
13631					attr = "blocksused";
13632					break;
13633				case 0xf1:
13634					attr = "poolblocksavail";
13635					break;
13636				case 0xf2:
13637					attr = "poolblocksused";
13638					break;
13639				default:
13640					continue;
13641				}
13642				mtx_unlock(&softc->ctl_lock); // XXX
13643				val = lun->backend->lun_attr(
13644				    lun->be_lun->be_lun, attr);
13645				mtx_lock(&softc->ctl_lock);
13646				if (val == UINT64_MAX)
13647					continue;
13648				if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13649				    == SLBPPD_ARMING_INC)
13650					e |= (val >= thres);
13651				else
13652					e |= (val <= thres);
13653			}
13654			mtx_lock(&lun->lun_lock);
13655			if (e) {
13656				if (lun->lasttpt == 0 ||
13657				    time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13658					lun->lasttpt = time_uptime;
13659					ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13660				}
13661			} else {
13662				lun->lasttpt = 0;
13663				ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13664			}
13665			mtx_unlock(&lun->lun_lock);
13666		}
13667		mtx_unlock(&softc->ctl_lock);
13668		pause("-", CTL_LBP_PERIOD * hz);
13669	}
13670}
13671
13672static void
13673ctl_enqueue_incoming(union ctl_io *io)
13674{
13675	struct ctl_softc *softc = control_softc;
13676	struct ctl_thread *thr;
13677	u_int idx;
13678
13679	idx = (io->io_hdr.nexus.targ_port * 127 +
13680	       io->io_hdr.nexus.initid.id) % worker_threads;
13681	thr = &softc->threads[idx];
13682	mtx_lock(&thr->queue_lock);
13683	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13684	mtx_unlock(&thr->queue_lock);
13685	wakeup(thr);
13686}
13687
13688static void
13689ctl_enqueue_rtr(union ctl_io *io)
13690{
13691	struct ctl_softc *softc = control_softc;
13692	struct ctl_thread *thr;
13693
13694	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13695	mtx_lock(&thr->queue_lock);
13696	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13697	mtx_unlock(&thr->queue_lock);
13698	wakeup(thr);
13699}
13700
13701static void
13702ctl_enqueue_done(union ctl_io *io)
13703{
13704	struct ctl_softc *softc = control_softc;
13705	struct ctl_thread *thr;
13706
13707	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13708	mtx_lock(&thr->queue_lock);
13709	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13710	mtx_unlock(&thr->queue_lock);
13711	wakeup(thr);
13712}
13713
13714#ifdef notyet
13715static void
13716ctl_enqueue_isc(union ctl_io *io)
13717{
13718	struct ctl_softc *softc = control_softc;
13719	struct ctl_thread *thr;
13720
13721	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13722	mtx_lock(&thr->queue_lock);
13723	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13724	mtx_unlock(&thr->queue_lock);
13725	wakeup(thr);
13726}
13727
13728/* Initialization and failover */
13729
13730void
13731ctl_init_isc_msg(void)
13732{
13733	printf("CTL: Still calling this thing\n");
13734}
13735
13736/*
13737 * Init component
13738 * 	Initializes component into configuration defined by bootMode
13739 *	(see hasc-sv.c)
13740 *  	returns hasc_Status:
13741 * 		OK
13742 *		ERROR - fatal error
13743 */
13744static ctl_ha_comp_status
13745ctl_isc_init(struct ctl_ha_component *c)
13746{
13747	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13748
13749	c->status = ret;
13750	return ret;
13751}
13752
13753/* Start component
13754 * 	Starts component in state requested. If component starts successfully,
13755 *	it must set its own state to the requestrd state
13756 *	When requested state is HASC_STATE_HA, the component may refine it
13757 * 	by adding _SLAVE or _MASTER flags.
13758 *	Currently allowed state transitions are:
13759 *	UNKNOWN->HA		- initial startup
13760 *	UNKNOWN->SINGLE - initial startup when no parter detected
13761 *	HA->SINGLE		- failover
13762 * returns ctl_ha_comp_status:
13763 * 		OK	- component successfully started in requested state
13764 *		FAILED  - could not start the requested state, failover may
13765 * 			  be possible
13766 *		ERROR	- fatal error detected, no future startup possible
13767 */
13768static ctl_ha_comp_status
13769ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13770{
13771	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13772
13773	printf("%s: go\n", __func__);
13774
13775	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13776	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13777		control_softc->is_single = 0;
13778		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13779		    != CTL_HA_STATUS_SUCCESS) {
13780			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13781			ret = CTL_HA_COMP_STATUS_ERROR;
13782		}
13783	} else if (CTL_HA_STATE_IS_HA(c->state)
13784		&& CTL_HA_STATE_IS_SINGLE(state)){
13785		// HA->SINGLE transition
13786	        ctl_failover();
13787		control_softc->is_single = 1;
13788	} else {
13789		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13790		       c->state, state);
13791		ret = CTL_HA_COMP_STATUS_ERROR;
13792	}
13793	if (CTL_HA_STATE_IS_SINGLE(state))
13794		control_softc->is_single = 1;
13795
13796	c->state = state;
13797	c->status = ret;
13798	return ret;
13799}
13800
13801/*
13802 * Quiesce component
13803 * The component must clear any error conditions (set status to OK) and
13804 * prepare itself to another Start call
13805 * returns ctl_ha_comp_status:
13806 * 	OK
13807 *	ERROR
13808 */
13809static ctl_ha_comp_status
13810ctl_isc_quiesce(struct ctl_ha_component *c)
13811{
13812	int ret = CTL_HA_COMP_STATUS_OK;
13813
13814	ctl_pause_rtr = 1;
13815	c->status = ret;
13816	return ret;
13817}
13818
13819struct ctl_ha_component ctl_ha_component_ctlisc =
13820{
13821	.name = "CTL ISC",
13822	.state = CTL_HA_STATE_UNKNOWN,
13823	.init = ctl_isc_init,
13824	.start = ctl_isc_start,
13825	.quiesce = ctl_isc_quiesce
13826};
13827#endif
13828
13829/*
13830 *  vim: ts=8
13831 */
13832