ctl.c revision 288727
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 288727 2015-10-05 08:51:20Z 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;
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	lun->ctl_softc = ctl_softc;
4173#ifdef CTL_TIME_IO
4174	lun->last_busy = getsbinuptime();
4175#endif
4176	TAILQ_INIT(&lun->ooa_queue);
4177	TAILQ_INIT(&lun->blocked_queue);
4178	STAILQ_INIT(&lun->error_list);
4179	ctl_tpc_lun_init(lun);
4180
4181	/*
4182	 * Initialize the mode and log page index.
4183	 */
4184	ctl_init_page_index(lun);
4185	ctl_init_log_page_index(lun);
4186
4187	/*
4188	 * Now, before we insert this lun on the lun list, set the lun
4189	 * inventory changed UA for all other luns.
4190	 */
4191	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4192		mtx_lock(&nlun->lun_lock);
4193		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4194		mtx_unlock(&nlun->lun_lock);
4195	}
4196
4197	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4198
4199	ctl_softc->ctl_luns[lun_number] = lun;
4200
4201	ctl_softc->num_luns++;
4202
4203	/* Setup statistics gathering */
4204	lun->stats.device_type = be_lun->lun_type;
4205	lun->stats.lun_number = lun_number;
4206	if (lun->stats.device_type == T_DIRECT)
4207		lun->stats.blocksize = be_lun->blocksize;
4208	else
4209		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4210	for (i = 0;i < CTL_MAX_PORTS;i++)
4211		lun->stats.ports[i].targ_port = i;
4212
4213	mtx_unlock(&ctl_softc->ctl_lock);
4214
4215	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4216	return (0);
4217}
4218
4219/*
4220 * Delete a LUN.
4221 * Assumptions:
4222 * - LUN has already been marked invalid and any pending I/O has been taken
4223 *   care of.
4224 */
4225static int
4226ctl_free_lun(struct ctl_lun *lun)
4227{
4228	struct ctl_softc *softc;
4229	struct ctl_lun *nlun;
4230	int i;
4231
4232	softc = lun->ctl_softc;
4233
4234	mtx_assert(&softc->ctl_lock, MA_OWNED);
4235
4236	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4237
4238	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4239
4240	softc->ctl_luns[lun->lun] = NULL;
4241
4242	if (!TAILQ_EMPTY(&lun->ooa_queue))
4243		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4244
4245	softc->num_luns--;
4246
4247	/*
4248	 * Tell the backend to free resources, if this LUN has a backend.
4249	 */
4250	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4251	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4252
4253	ctl_tpc_lun_shutdown(lun);
4254	mtx_destroy(&lun->lun_lock);
4255	free(lun->lun_devid, M_CTL);
4256	for (i = 0; i < CTL_MAX_PORTS; i++)
4257		free(lun->pending_ua[i], M_CTL);
4258	for (i = 0; i < 2 * CTL_MAX_PORTS; i++)
4259		free(lun->pr_keys[i], M_CTL);
4260	free(lun->write_buffer, M_CTL);
4261	if (lun->flags & CTL_LUN_MALLOCED)
4262		free(lun, M_CTL);
4263
4264	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4265		mtx_lock(&nlun->lun_lock);
4266		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4267		mtx_unlock(&nlun->lun_lock);
4268	}
4269
4270	return (0);
4271}
4272
4273static void
4274ctl_create_lun(struct ctl_be_lun *be_lun)
4275{
4276	struct ctl_softc *softc;
4277
4278	softc = control_softc;
4279
4280	/*
4281	 * ctl_alloc_lun() should handle all potential failure cases.
4282	 */
4283	ctl_alloc_lun(softc, NULL, be_lun);
4284}
4285
4286int
4287ctl_add_lun(struct ctl_be_lun *be_lun)
4288{
4289	struct ctl_softc *softc = control_softc;
4290
4291	mtx_lock(&softc->ctl_lock);
4292	STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4293	mtx_unlock(&softc->ctl_lock);
4294	wakeup(&softc->pending_lun_queue);
4295
4296	return (0);
4297}
4298
4299int
4300ctl_enable_lun(struct ctl_be_lun *be_lun)
4301{
4302	struct ctl_softc *softc;
4303	struct ctl_port *port, *nport;
4304	struct ctl_lun *lun;
4305	int retval;
4306
4307	lun = (struct ctl_lun *)be_lun->ctl_lun;
4308	softc = lun->ctl_softc;
4309
4310	mtx_lock(&softc->ctl_lock);
4311	mtx_lock(&lun->lun_lock);
4312	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4313		/*
4314		 * eh?  Why did we get called if the LUN is already
4315		 * enabled?
4316		 */
4317		mtx_unlock(&lun->lun_lock);
4318		mtx_unlock(&softc->ctl_lock);
4319		return (0);
4320	}
4321	lun->flags &= ~CTL_LUN_DISABLED;
4322	mtx_unlock(&lun->lun_lock);
4323
4324	for (port = STAILQ_FIRST(&softc->port_list); port != NULL; port = nport) {
4325		nport = STAILQ_NEXT(port, links);
4326		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4327		    port->lun_map != NULL || port->lun_enable == NULL)
4328			continue;
4329
4330		/*
4331		 * Drop the lock while we call the FETD's enable routine.
4332		 * This can lead to a callback into CTL (at least in the
4333		 * case of the internal initiator frontend.
4334		 */
4335		mtx_unlock(&softc->ctl_lock);
4336		retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4337		mtx_lock(&softc->ctl_lock);
4338		if (retval != 0) {
4339			printf("%s: FETD %s port %d returned error "
4340			       "%d for lun_enable on lun %jd\n",
4341			       __func__, port->port_name, port->targ_port,
4342			       retval, (intmax_t)lun->lun);
4343		}
4344	}
4345
4346	mtx_unlock(&softc->ctl_lock);
4347
4348	return (0);
4349}
4350
4351int
4352ctl_disable_lun(struct ctl_be_lun *be_lun)
4353{
4354	struct ctl_softc *softc;
4355	struct ctl_port *port;
4356	struct ctl_lun *lun;
4357	int retval;
4358
4359	lun = (struct ctl_lun *)be_lun->ctl_lun;
4360	softc = lun->ctl_softc;
4361
4362	mtx_lock(&softc->ctl_lock);
4363	mtx_lock(&lun->lun_lock);
4364	if (lun->flags & CTL_LUN_DISABLED) {
4365		mtx_unlock(&lun->lun_lock);
4366		mtx_unlock(&softc->ctl_lock);
4367		return (0);
4368	}
4369	lun->flags |= CTL_LUN_DISABLED;
4370	mtx_unlock(&lun->lun_lock);
4371
4372	STAILQ_FOREACH(port, &softc->port_list, links) {
4373		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4374		    port->lun_map != NULL || port->lun_disable == NULL)
4375			continue;
4376
4377		/*
4378		 * Drop the lock before we call the frontend's disable
4379		 * routine, to avoid lock order reversals.
4380		 *
4381		 * XXX KDM what happens if the frontend list changes while
4382		 * we're traversing it?  It's unlikely, but should be handled.
4383		 */
4384		mtx_unlock(&softc->ctl_lock);
4385		retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4386		mtx_lock(&softc->ctl_lock);
4387		if (retval != 0) {
4388			printf("%s: FETD %s port %d returned error "
4389			       "%d for lun_disable on lun %jd\n",
4390			       __func__, port->port_name, port->targ_port,
4391			       retval, (intmax_t)lun->lun);
4392		}
4393	}
4394
4395	mtx_unlock(&softc->ctl_lock);
4396
4397	return (0);
4398}
4399
4400int
4401ctl_start_lun(struct ctl_be_lun *be_lun)
4402{
4403	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4404
4405	mtx_lock(&lun->lun_lock);
4406	lun->flags &= ~CTL_LUN_STOPPED;
4407	mtx_unlock(&lun->lun_lock);
4408	return (0);
4409}
4410
4411int
4412ctl_stop_lun(struct ctl_be_lun *be_lun)
4413{
4414	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4415
4416	mtx_lock(&lun->lun_lock);
4417	lun->flags |= CTL_LUN_STOPPED;
4418	mtx_unlock(&lun->lun_lock);
4419	return (0);
4420}
4421
4422int
4423ctl_lun_offline(struct ctl_be_lun *be_lun)
4424{
4425	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4426
4427	mtx_lock(&lun->lun_lock);
4428	lun->flags |= CTL_LUN_OFFLINE;
4429	mtx_unlock(&lun->lun_lock);
4430	return (0);
4431}
4432
4433int
4434ctl_lun_online(struct ctl_be_lun *be_lun)
4435{
4436	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4437
4438	mtx_lock(&lun->lun_lock);
4439	lun->flags &= ~CTL_LUN_OFFLINE;
4440	mtx_unlock(&lun->lun_lock);
4441	return (0);
4442}
4443
4444int
4445ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4446{
4447	struct ctl_softc *softc;
4448	struct ctl_lun *lun;
4449
4450	lun = (struct ctl_lun *)be_lun->ctl_lun;
4451	softc = lun->ctl_softc;
4452
4453	mtx_lock(&lun->lun_lock);
4454
4455	/*
4456	 * The LUN needs to be disabled before it can be marked invalid.
4457	 */
4458	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4459		mtx_unlock(&lun->lun_lock);
4460		return (-1);
4461	}
4462	/*
4463	 * Mark the LUN invalid.
4464	 */
4465	lun->flags |= CTL_LUN_INVALID;
4466
4467	/*
4468	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4469	 * If we have something in the OOA queue, we'll free it when the
4470	 * last I/O completes.
4471	 */
4472	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4473		mtx_unlock(&lun->lun_lock);
4474		mtx_lock(&softc->ctl_lock);
4475		ctl_free_lun(lun);
4476		mtx_unlock(&softc->ctl_lock);
4477	} else
4478		mtx_unlock(&lun->lun_lock);
4479
4480	return (0);
4481}
4482
4483int
4484ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4485{
4486	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4487
4488	mtx_lock(&lun->lun_lock);
4489	lun->flags |= CTL_LUN_INOPERABLE;
4490	mtx_unlock(&lun->lun_lock);
4491	return (0);
4492}
4493
4494int
4495ctl_lun_operable(struct ctl_be_lun *be_lun)
4496{
4497	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4498
4499	mtx_lock(&lun->lun_lock);
4500	lun->flags &= ~CTL_LUN_INOPERABLE;
4501	mtx_unlock(&lun->lun_lock);
4502	return (0);
4503}
4504
4505void
4506ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4507{
4508	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4509
4510	mtx_lock(&lun->lun_lock);
4511	ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGED);
4512	mtx_unlock(&lun->lun_lock);
4513}
4514
4515/*
4516 * Backend "memory move is complete" callback for requests that never
4517 * make it down to say RAIDCore's configuration code.
4518 */
4519int
4520ctl_config_move_done(union ctl_io *io)
4521{
4522	int retval;
4523
4524	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4525	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4526	    ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
4527
4528	if ((io->io_hdr.port_status != 0) &&
4529	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4530	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4531		/*
4532		 * For hardware error sense keys, the sense key
4533		 * specific value is defined to be a retry count,
4534		 * but we use it to pass back an internal FETD
4535		 * error code.  XXX KDM  Hopefully the FETD is only
4536		 * using 16 bits for an error code, since that's
4537		 * all the space we have in the sks field.
4538		 */
4539		ctl_set_internal_failure(&io->scsiio,
4540					 /*sks_valid*/ 1,
4541					 /*retry_count*/
4542					 io->io_hdr.port_status);
4543	}
4544
4545	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
4546	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4547	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
4548	    ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4549		/*
4550		 * XXX KDM just assuming a single pointer here, and not a
4551		 * S/G list.  If we start using S/G lists for config data,
4552		 * we'll need to know how to clean them up here as well.
4553		 */
4554		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4555			free(io->scsiio.kern_data_ptr, M_CTL);
4556		ctl_done(io);
4557		retval = CTL_RETVAL_COMPLETE;
4558	} else {
4559		/*
4560		 * XXX KDM now we need to continue data movement.  Some
4561		 * options:
4562		 * - call ctl_scsiio() again?  We don't do this for data
4563		 *   writes, because for those at least we know ahead of
4564		 *   time where the write will go and how long it is.  For
4565		 *   config writes, though, that information is largely
4566		 *   contained within the write itself, thus we need to
4567		 *   parse out the data again.
4568		 *
4569		 * - Call some other function once the data is in?
4570		 */
4571		if (ctl_debug & CTL_DEBUG_CDB_DATA)
4572			ctl_data_print(io);
4573
4574		/*
4575		 * XXX KDM call ctl_scsiio() again for now, and check flag
4576		 * bits to see whether we're allocated or not.
4577		 */
4578		retval = ctl_scsiio(&io->scsiio);
4579	}
4580	return (retval);
4581}
4582
4583/*
4584 * This gets called by a backend driver when it is done with a
4585 * data_submit method.
4586 */
4587void
4588ctl_data_submit_done(union ctl_io *io)
4589{
4590	/*
4591	 * If the IO_CONT flag is set, we need to call the supplied
4592	 * function to continue processing the I/O, instead of completing
4593	 * the I/O just yet.
4594	 *
4595	 * If there is an error, though, we don't want to keep processing.
4596	 * Instead, just send status back to the initiator.
4597	 */
4598	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4599	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4600	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4601	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4602		io->scsiio.io_cont(io);
4603		return;
4604	}
4605	ctl_done(io);
4606}
4607
4608/*
4609 * This gets called by a backend driver when it is done with a
4610 * configuration write.
4611 */
4612void
4613ctl_config_write_done(union ctl_io *io)
4614{
4615	uint8_t *buf;
4616
4617	/*
4618	 * If the IO_CONT flag is set, we need to call the supplied
4619	 * function to continue processing the I/O, instead of completing
4620	 * the I/O just yet.
4621	 *
4622	 * If there is an error, though, we don't want to keep processing.
4623	 * Instead, just send status back to the initiator.
4624	 */
4625	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4626	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4627	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4628	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4629		io->scsiio.io_cont(io);
4630		return;
4631	}
4632	/*
4633	 * Since a configuration write can be done for commands that actually
4634	 * have data allocated, like write buffer, and commands that have
4635	 * no data, like start/stop unit, we need to check here.
4636	 */
4637	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4638		buf = io->scsiio.kern_data_ptr;
4639	else
4640		buf = NULL;
4641	ctl_done(io);
4642	if (buf)
4643		free(buf, M_CTL);
4644}
4645
4646void
4647ctl_config_read_done(union ctl_io *io)
4648{
4649	uint8_t *buf;
4650
4651	/*
4652	 * If there is some error -- we are done, skip data transfer.
4653	 */
4654	if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
4655	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4656	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
4657		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4658			buf = io->scsiio.kern_data_ptr;
4659		else
4660			buf = NULL;
4661		ctl_done(io);
4662		if (buf)
4663			free(buf, M_CTL);
4664		return;
4665	}
4666
4667	/*
4668	 * If the IO_CONT flag is set, we need to call the supplied
4669	 * function to continue processing the I/O, instead of completing
4670	 * the I/O just yet.
4671	 */
4672	if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
4673		io->scsiio.io_cont(io);
4674		return;
4675	}
4676
4677	ctl_datamove(io);
4678}
4679
4680/*
4681 * SCSI release command.
4682 */
4683int
4684ctl_scsi_release(struct ctl_scsiio *ctsio)
4685{
4686	int length, longid, thirdparty_id, resv_id;
4687	struct ctl_lun *lun;
4688	uint32_t residx;
4689
4690	length = 0;
4691	resv_id = 0;
4692
4693	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
4694
4695	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
4696	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4697
4698	switch (ctsio->cdb[0]) {
4699	case RELEASE_10: {
4700		struct scsi_release_10 *cdb;
4701
4702		cdb = (struct scsi_release_10 *)ctsio->cdb;
4703
4704		if (cdb->byte2 & SR10_LONGID)
4705			longid = 1;
4706		else
4707			thirdparty_id = cdb->thirdparty_id;
4708
4709		resv_id = cdb->resv_id;
4710		length = scsi_2btoul(cdb->length);
4711		break;
4712	}
4713	}
4714
4715
4716	/*
4717	 * XXX KDM right now, we only support LUN reservation.  We don't
4718	 * support 3rd party reservations, or extent reservations, which
4719	 * might actually need the parameter list.  If we've gotten this
4720	 * far, we've got a LUN reservation.  Anything else got kicked out
4721	 * above.  So, according to SPC, ignore the length.
4722	 */
4723	length = 0;
4724
4725	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
4726	 && (length > 0)) {
4727		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
4728		ctsio->kern_data_len = length;
4729		ctsio->kern_total_len = length;
4730		ctsio->kern_data_resid = 0;
4731		ctsio->kern_rel_offset = 0;
4732		ctsio->kern_sg_entries = 0;
4733		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
4734		ctsio->be_move_done = ctl_config_move_done;
4735		ctl_datamove((union ctl_io *)ctsio);
4736
4737		return (CTL_RETVAL_COMPLETE);
4738	}
4739
4740	if (length > 0)
4741		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
4742
4743	mtx_lock(&lun->lun_lock);
4744
4745	/*
4746	 * According to SPC, it is not an error for an intiator to attempt
4747	 * to release a reservation on a LUN that isn't reserved, or that
4748	 * is reserved by another initiator.  The reservation can only be
4749	 * released, though, by the initiator who made it or by one of
4750	 * several reset type events.
4751	 */
4752	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
4753			lun->flags &= ~CTL_LUN_RESERVED;
4754
4755	mtx_unlock(&lun->lun_lock);
4756
4757	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
4758		free(ctsio->kern_data_ptr, M_CTL);
4759		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
4760	}
4761
4762	ctl_set_success(ctsio);
4763	ctl_done((union ctl_io *)ctsio);
4764	return (CTL_RETVAL_COMPLETE);
4765}
4766
4767int
4768ctl_scsi_reserve(struct ctl_scsiio *ctsio)
4769{
4770	int extent, thirdparty, longid;
4771	int resv_id, length;
4772	uint64_t thirdparty_id;
4773	struct ctl_lun *lun;
4774	uint32_t residx;
4775
4776	extent = 0;
4777	thirdparty = 0;
4778	longid = 0;
4779	resv_id = 0;
4780	length = 0;
4781	thirdparty_id = 0;
4782
4783	CTL_DEBUG_PRINT(("ctl_reserve\n"));
4784
4785	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
4786	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4787
4788	switch (ctsio->cdb[0]) {
4789	case RESERVE_10: {
4790		struct scsi_reserve_10 *cdb;
4791
4792		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
4793
4794		if (cdb->byte2 & SR10_LONGID)
4795			longid = 1;
4796		else
4797			thirdparty_id = cdb->thirdparty_id;
4798
4799		resv_id = cdb->resv_id;
4800		length = scsi_2btoul(cdb->length);
4801		break;
4802	}
4803	}
4804
4805	/*
4806	 * XXX KDM right now, we only support LUN reservation.  We don't
4807	 * support 3rd party reservations, or extent reservations, which
4808	 * might actually need the parameter list.  If we've gotten this
4809	 * far, we've got a LUN reservation.  Anything else got kicked out
4810	 * above.  So, according to SPC, ignore the length.
4811	 */
4812	length = 0;
4813
4814	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
4815	 && (length > 0)) {
4816		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
4817		ctsio->kern_data_len = length;
4818		ctsio->kern_total_len = length;
4819		ctsio->kern_data_resid = 0;
4820		ctsio->kern_rel_offset = 0;
4821		ctsio->kern_sg_entries = 0;
4822		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
4823		ctsio->be_move_done = ctl_config_move_done;
4824		ctl_datamove((union ctl_io *)ctsio);
4825
4826		return (CTL_RETVAL_COMPLETE);
4827	}
4828
4829	if (length > 0)
4830		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
4831
4832	mtx_lock(&lun->lun_lock);
4833	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
4834		ctl_set_reservation_conflict(ctsio);
4835		goto bailout;
4836	}
4837
4838	lun->flags |= CTL_LUN_RESERVED;
4839	lun->res_idx = residx;
4840
4841	ctl_set_success(ctsio);
4842
4843bailout:
4844	mtx_unlock(&lun->lun_lock);
4845
4846	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
4847		free(ctsio->kern_data_ptr, M_CTL);
4848		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
4849	}
4850
4851	ctl_done((union ctl_io *)ctsio);
4852	return (CTL_RETVAL_COMPLETE);
4853}
4854
4855int
4856ctl_start_stop(struct ctl_scsiio *ctsio)
4857{
4858	struct scsi_start_stop_unit *cdb;
4859	struct ctl_lun *lun;
4860	int retval;
4861
4862	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
4863
4864	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4865	retval = 0;
4866
4867	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
4868
4869	/*
4870	 * XXX KDM
4871	 * We don't support the immediate bit on a stop unit.  In order to
4872	 * do that, we would need to code up a way to know that a stop is
4873	 * pending, and hold off any new commands until it completes, one
4874	 * way or another.  Then we could accept or reject those commands
4875	 * depending on its status.  We would almost need to do the reverse
4876	 * of what we do below for an immediate start -- return the copy of
4877	 * the ctl_io to the FETD with status to send to the host (and to
4878	 * free the copy!) and then free the original I/O once the stop
4879	 * actually completes.  That way, the OOA queue mechanism can work
4880	 * to block commands that shouldn't proceed.  Another alternative
4881	 * would be to put the copy in the queue in place of the original,
4882	 * and return the original back to the caller.  That could be
4883	 * slightly safer..
4884	 */
4885	if ((cdb->byte2 & SSS_IMMED)
4886	 && ((cdb->how & SSS_START) == 0)) {
4887		ctl_set_invalid_field(ctsio,
4888				      /*sks_valid*/ 1,
4889				      /*command*/ 1,
4890				      /*field*/ 1,
4891				      /*bit_valid*/ 1,
4892				      /*bit*/ 0);
4893		ctl_done((union ctl_io *)ctsio);
4894		return (CTL_RETVAL_COMPLETE);
4895	}
4896
4897	if ((lun->flags & CTL_LUN_PR_RESERVED)
4898	 && ((cdb->how & SSS_START)==0)) {
4899		uint32_t residx;
4900
4901		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
4902		if (ctl_get_prkey(lun, residx) == 0
4903		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
4904
4905			ctl_set_reservation_conflict(ctsio);
4906			ctl_done((union ctl_io *)ctsio);
4907			return (CTL_RETVAL_COMPLETE);
4908		}
4909	}
4910
4911	/*
4912	 * If there is no backend on this device, we can't start or stop
4913	 * it.  In theory we shouldn't get any start/stop commands in the
4914	 * first place at this level if the LUN doesn't have a backend.
4915	 * That should get stopped by the command decode code.
4916	 */
4917	if (lun->backend == NULL) {
4918		ctl_set_invalid_opcode(ctsio);
4919		ctl_done((union ctl_io *)ctsio);
4920		return (CTL_RETVAL_COMPLETE);
4921	}
4922
4923	/*
4924	 * XXX KDM Copan-specific offline behavior.
4925	 * Figure out a reasonable way to port this?
4926	 */
4927#ifdef NEEDTOPORT
4928	mtx_lock(&lun->lun_lock);
4929
4930	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
4931	 && (lun->flags & CTL_LUN_OFFLINE)) {
4932		/*
4933		 * If the LUN is offline, and the on/offline bit isn't set,
4934		 * reject the start or stop.  Otherwise, let it through.
4935		 */
4936		mtx_unlock(&lun->lun_lock);
4937		ctl_set_lun_not_ready(ctsio);
4938		ctl_done((union ctl_io *)ctsio);
4939	} else {
4940		mtx_unlock(&lun->lun_lock);
4941#endif /* NEEDTOPORT */
4942		/*
4943		 * This could be a start or a stop when we're online,
4944		 * or a stop/offline or start/online.  A start or stop when
4945		 * we're offline is covered in the case above.
4946		 */
4947		/*
4948		 * In the non-immediate case, we send the request to
4949		 * the backend and return status to the user when
4950		 * it is done.
4951		 *
4952		 * In the immediate case, we allocate a new ctl_io
4953		 * to hold a copy of the request, and send that to
4954		 * the backend.  We then set good status on the
4955		 * user's request and return it immediately.
4956		 */
4957		if (cdb->byte2 & SSS_IMMED) {
4958			union ctl_io *new_io;
4959
4960			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
4961			ctl_copy_io((union ctl_io *)ctsio, new_io);
4962			retval = lun->backend->config_write(new_io);
4963			ctl_set_success(ctsio);
4964			ctl_done((union ctl_io *)ctsio);
4965		} else {
4966			retval = lun->backend->config_write(
4967				(union ctl_io *)ctsio);
4968		}
4969#ifdef NEEDTOPORT
4970	}
4971#endif
4972	return (retval);
4973}
4974
4975/*
4976 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
4977 * we don't really do anything with the LBA and length fields if the user
4978 * passes them in.  Instead we'll just flush out the cache for the entire
4979 * LUN.
4980 */
4981int
4982ctl_sync_cache(struct ctl_scsiio *ctsio)
4983{
4984	struct ctl_lun *lun;
4985	struct ctl_softc *softc;
4986	struct ctl_lba_len_flags *lbalen;
4987	uint64_t starting_lba;
4988	uint32_t block_count;
4989	int retval;
4990	uint8_t byte2;
4991
4992	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
4993
4994	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4995	softc = lun->ctl_softc;
4996	retval = 0;
4997
4998	switch (ctsio->cdb[0]) {
4999	case SYNCHRONIZE_CACHE: {
5000		struct scsi_sync_cache *cdb;
5001		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5002
5003		starting_lba = scsi_4btoul(cdb->begin_lba);
5004		block_count = scsi_2btoul(cdb->lb_count);
5005		byte2 = cdb->byte2;
5006		break;
5007	}
5008	case SYNCHRONIZE_CACHE_16: {
5009		struct scsi_sync_cache_16 *cdb;
5010		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5011
5012		starting_lba = scsi_8btou64(cdb->begin_lba);
5013		block_count = scsi_4btoul(cdb->lb_count);
5014		byte2 = cdb->byte2;
5015		break;
5016	}
5017	default:
5018		ctl_set_invalid_opcode(ctsio);
5019		ctl_done((union ctl_io *)ctsio);
5020		goto bailout;
5021		break; /* NOTREACHED */
5022	}
5023
5024	/*
5025	 * We check the LBA and length, but don't do anything with them.
5026	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5027	 * get flushed.  This check will just help satisfy anyone who wants
5028	 * to see an error for an out of range LBA.
5029	 */
5030	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5031		ctl_set_lba_out_of_range(ctsio);
5032		ctl_done((union ctl_io *)ctsio);
5033		goto bailout;
5034	}
5035
5036	/*
5037	 * If this LUN has no backend, we can't flush the cache anyway.
5038	 */
5039	if (lun->backend == NULL) {
5040		ctl_set_invalid_opcode(ctsio);
5041		ctl_done((union ctl_io *)ctsio);
5042		goto bailout;
5043	}
5044
5045	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5046	lbalen->lba = starting_lba;
5047	lbalen->len = block_count;
5048	lbalen->flags = byte2;
5049
5050	/*
5051	 * Check to see whether we're configured to send the SYNCHRONIZE
5052	 * CACHE command directly to the back end.
5053	 */
5054	mtx_lock(&lun->lun_lock);
5055	if ((softc->flags & CTL_FLAG_REAL_SYNC)
5056	 && (++(lun->sync_count) >= lun->sync_interval)) {
5057		lun->sync_count = 0;
5058		mtx_unlock(&lun->lun_lock);
5059		retval = lun->backend->config_write((union ctl_io *)ctsio);
5060	} else {
5061		mtx_unlock(&lun->lun_lock);
5062		ctl_set_success(ctsio);
5063		ctl_done((union ctl_io *)ctsio);
5064	}
5065
5066bailout:
5067
5068	return (retval);
5069}
5070
5071int
5072ctl_format(struct ctl_scsiio *ctsio)
5073{
5074	struct scsi_format *cdb;
5075	struct ctl_lun *lun;
5076	int length, defect_list_len;
5077
5078	CTL_DEBUG_PRINT(("ctl_format\n"));
5079
5080	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5081
5082	cdb = (struct scsi_format *)ctsio->cdb;
5083
5084	length = 0;
5085	if (cdb->byte2 & SF_FMTDATA) {
5086		if (cdb->byte2 & SF_LONGLIST)
5087			length = sizeof(struct scsi_format_header_long);
5088		else
5089			length = sizeof(struct scsi_format_header_short);
5090	}
5091
5092	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5093	 && (length > 0)) {
5094		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5095		ctsio->kern_data_len = length;
5096		ctsio->kern_total_len = length;
5097		ctsio->kern_data_resid = 0;
5098		ctsio->kern_rel_offset = 0;
5099		ctsio->kern_sg_entries = 0;
5100		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5101		ctsio->be_move_done = ctl_config_move_done;
5102		ctl_datamove((union ctl_io *)ctsio);
5103
5104		return (CTL_RETVAL_COMPLETE);
5105	}
5106
5107	defect_list_len = 0;
5108
5109	if (cdb->byte2 & SF_FMTDATA) {
5110		if (cdb->byte2 & SF_LONGLIST) {
5111			struct scsi_format_header_long *header;
5112
5113			header = (struct scsi_format_header_long *)
5114				ctsio->kern_data_ptr;
5115
5116			defect_list_len = scsi_4btoul(header->defect_list_len);
5117			if (defect_list_len != 0) {
5118				ctl_set_invalid_field(ctsio,
5119						      /*sks_valid*/ 1,
5120						      /*command*/ 0,
5121						      /*field*/ 2,
5122						      /*bit_valid*/ 0,
5123						      /*bit*/ 0);
5124				goto bailout;
5125			}
5126		} else {
5127			struct scsi_format_header_short *header;
5128
5129			header = (struct scsi_format_header_short *)
5130				ctsio->kern_data_ptr;
5131
5132			defect_list_len = scsi_2btoul(header->defect_list_len);
5133			if (defect_list_len != 0) {
5134				ctl_set_invalid_field(ctsio,
5135						      /*sks_valid*/ 1,
5136						      /*command*/ 0,
5137						      /*field*/ 2,
5138						      /*bit_valid*/ 0,
5139						      /*bit*/ 0);
5140				goto bailout;
5141			}
5142		}
5143	}
5144
5145	/*
5146	 * The format command will clear out the "Medium format corrupted"
5147	 * status if set by the configuration code.  That status is really
5148	 * just a way to notify the host that we have lost the media, and
5149	 * get them to issue a command that will basically make them think
5150	 * they're blowing away the media.
5151	 */
5152	mtx_lock(&lun->lun_lock);
5153	lun->flags &= ~CTL_LUN_INOPERABLE;
5154	mtx_unlock(&lun->lun_lock);
5155
5156	ctl_set_success(ctsio);
5157bailout:
5158
5159	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5160		free(ctsio->kern_data_ptr, M_CTL);
5161		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5162	}
5163
5164	ctl_done((union ctl_io *)ctsio);
5165	return (CTL_RETVAL_COMPLETE);
5166}
5167
5168int
5169ctl_read_buffer(struct ctl_scsiio *ctsio)
5170{
5171	struct scsi_read_buffer *cdb;
5172	struct ctl_lun *lun;
5173	int buffer_offset, len;
5174	static uint8_t descr[4];
5175	static uint8_t echo_descr[4] = { 0 };
5176
5177	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5178
5179	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5180	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5181
5182	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5183	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5184	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5185		ctl_set_invalid_field(ctsio,
5186				      /*sks_valid*/ 1,
5187				      /*command*/ 1,
5188				      /*field*/ 1,
5189				      /*bit_valid*/ 1,
5190				      /*bit*/ 4);
5191		ctl_done((union ctl_io *)ctsio);
5192		return (CTL_RETVAL_COMPLETE);
5193	}
5194
5195	len = scsi_3btoul(cdb->length);
5196	buffer_offset = scsi_3btoul(cdb->offset);
5197
5198	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5199		ctl_set_invalid_field(ctsio,
5200				      /*sks_valid*/ 1,
5201				      /*command*/ 1,
5202				      /*field*/ 6,
5203				      /*bit_valid*/ 0,
5204				      /*bit*/ 0);
5205		ctl_done((union ctl_io *)ctsio);
5206		return (CTL_RETVAL_COMPLETE);
5207	}
5208
5209	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5210		descr[0] = 0;
5211		scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5212		ctsio->kern_data_ptr = descr;
5213		len = min(len, sizeof(descr));
5214	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5215		ctsio->kern_data_ptr = echo_descr;
5216		len = min(len, sizeof(echo_descr));
5217	} else {
5218		if (lun->write_buffer == NULL) {
5219			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5220			    M_CTL, M_WAITOK);
5221		}
5222		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5223	}
5224	ctsio->kern_data_len = len;
5225	ctsio->kern_total_len = len;
5226	ctsio->kern_data_resid = 0;
5227	ctsio->kern_rel_offset = 0;
5228	ctsio->kern_sg_entries = 0;
5229	ctl_set_success(ctsio);
5230	ctsio->be_move_done = ctl_config_move_done;
5231	ctl_datamove((union ctl_io *)ctsio);
5232	return (CTL_RETVAL_COMPLETE);
5233}
5234
5235int
5236ctl_write_buffer(struct ctl_scsiio *ctsio)
5237{
5238	struct scsi_write_buffer *cdb;
5239	struct ctl_lun *lun;
5240	int buffer_offset, len;
5241
5242	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5243
5244	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5245	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5246
5247	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5248		ctl_set_invalid_field(ctsio,
5249				      /*sks_valid*/ 1,
5250				      /*command*/ 1,
5251				      /*field*/ 1,
5252				      /*bit_valid*/ 1,
5253				      /*bit*/ 4);
5254		ctl_done((union ctl_io *)ctsio);
5255		return (CTL_RETVAL_COMPLETE);
5256	}
5257
5258	len = scsi_3btoul(cdb->length);
5259	buffer_offset = scsi_3btoul(cdb->offset);
5260
5261	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5262		ctl_set_invalid_field(ctsio,
5263				      /*sks_valid*/ 1,
5264				      /*command*/ 1,
5265				      /*field*/ 6,
5266				      /*bit_valid*/ 0,
5267				      /*bit*/ 0);
5268		ctl_done((union ctl_io *)ctsio);
5269		return (CTL_RETVAL_COMPLETE);
5270	}
5271
5272	/*
5273	 * If we've got a kernel request that hasn't been malloced yet,
5274	 * malloc it and tell the caller the data buffer is here.
5275	 */
5276	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5277		if (lun->write_buffer == NULL) {
5278			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5279			    M_CTL, M_WAITOK);
5280		}
5281		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5282		ctsio->kern_data_len = len;
5283		ctsio->kern_total_len = len;
5284		ctsio->kern_data_resid = 0;
5285		ctsio->kern_rel_offset = 0;
5286		ctsio->kern_sg_entries = 0;
5287		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5288		ctsio->be_move_done = ctl_config_move_done;
5289		ctl_datamove((union ctl_io *)ctsio);
5290
5291		return (CTL_RETVAL_COMPLETE);
5292	}
5293
5294	ctl_set_success(ctsio);
5295	ctl_done((union ctl_io *)ctsio);
5296	return (CTL_RETVAL_COMPLETE);
5297}
5298
5299int
5300ctl_write_same(struct ctl_scsiio *ctsio)
5301{
5302	struct ctl_lun *lun;
5303	struct ctl_lba_len_flags *lbalen;
5304	uint64_t lba;
5305	uint32_t num_blocks;
5306	int len, retval;
5307	uint8_t byte2;
5308
5309	retval = CTL_RETVAL_COMPLETE;
5310
5311	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5312
5313	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5314
5315	switch (ctsio->cdb[0]) {
5316	case WRITE_SAME_10: {
5317		struct scsi_write_same_10 *cdb;
5318
5319		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5320
5321		lba = scsi_4btoul(cdb->addr);
5322		num_blocks = scsi_2btoul(cdb->length);
5323		byte2 = cdb->byte2;
5324		break;
5325	}
5326	case WRITE_SAME_16: {
5327		struct scsi_write_same_16 *cdb;
5328
5329		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5330
5331		lba = scsi_8btou64(cdb->addr);
5332		num_blocks = scsi_4btoul(cdb->length);
5333		byte2 = cdb->byte2;
5334		break;
5335	}
5336	default:
5337		/*
5338		 * We got a command we don't support.  This shouldn't
5339		 * happen, commands should be filtered out above us.
5340		 */
5341		ctl_set_invalid_opcode(ctsio);
5342		ctl_done((union ctl_io *)ctsio);
5343
5344		return (CTL_RETVAL_COMPLETE);
5345		break; /* NOTREACHED */
5346	}
5347
5348	/* NDOB and ANCHOR flags can be used only together with UNMAP */
5349	if ((byte2 & SWS_UNMAP) == 0 &&
5350	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5351		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5352		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5353		ctl_done((union ctl_io *)ctsio);
5354		return (CTL_RETVAL_COMPLETE);
5355	}
5356
5357	/*
5358	 * The first check is to make sure we're in bounds, the second
5359	 * check is to catch wrap-around problems.  If the lba + num blocks
5360	 * is less than the lba, then we've wrapped around and the block
5361	 * range is invalid anyway.
5362	 */
5363	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5364	 || ((lba + num_blocks) < lba)) {
5365		ctl_set_lba_out_of_range(ctsio);
5366		ctl_done((union ctl_io *)ctsio);
5367		return (CTL_RETVAL_COMPLETE);
5368	}
5369
5370	/* Zero number of blocks means "to the last logical block" */
5371	if (num_blocks == 0) {
5372		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5373			ctl_set_invalid_field(ctsio,
5374					      /*sks_valid*/ 0,
5375					      /*command*/ 1,
5376					      /*field*/ 0,
5377					      /*bit_valid*/ 0,
5378					      /*bit*/ 0);
5379			ctl_done((union ctl_io *)ctsio);
5380			return (CTL_RETVAL_COMPLETE);
5381		}
5382		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5383	}
5384
5385	len = lun->be_lun->blocksize;
5386
5387	/*
5388	 * If we've got a kernel request that hasn't been malloced yet,
5389	 * malloc it and tell the caller the data buffer is here.
5390	 */
5391	if ((byte2 & SWS_NDOB) == 0 &&
5392	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5393		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5394		ctsio->kern_data_len = len;
5395		ctsio->kern_total_len = len;
5396		ctsio->kern_data_resid = 0;
5397		ctsio->kern_rel_offset = 0;
5398		ctsio->kern_sg_entries = 0;
5399		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5400		ctsio->be_move_done = ctl_config_move_done;
5401		ctl_datamove((union ctl_io *)ctsio);
5402
5403		return (CTL_RETVAL_COMPLETE);
5404	}
5405
5406	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5407	lbalen->lba = lba;
5408	lbalen->len = num_blocks;
5409	lbalen->flags = byte2;
5410	retval = lun->backend->config_write((union ctl_io *)ctsio);
5411
5412	return (retval);
5413}
5414
5415int
5416ctl_unmap(struct ctl_scsiio *ctsio)
5417{
5418	struct ctl_lun *lun;
5419	struct scsi_unmap *cdb;
5420	struct ctl_ptr_len_flags *ptrlen;
5421	struct scsi_unmap_header *hdr;
5422	struct scsi_unmap_desc *buf, *end, *endnz, *range;
5423	uint64_t lba;
5424	uint32_t num_blocks;
5425	int len, retval;
5426	uint8_t byte2;
5427
5428	retval = CTL_RETVAL_COMPLETE;
5429
5430	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5431
5432	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5433	cdb = (struct scsi_unmap *)ctsio->cdb;
5434
5435	len = scsi_2btoul(cdb->length);
5436	byte2 = cdb->byte2;
5437
5438	/*
5439	 * If we've got a kernel request that hasn't been malloced yet,
5440	 * malloc it and tell the caller the data buffer is here.
5441	 */
5442	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5443		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5444		ctsio->kern_data_len = len;
5445		ctsio->kern_total_len = len;
5446		ctsio->kern_data_resid = 0;
5447		ctsio->kern_rel_offset = 0;
5448		ctsio->kern_sg_entries = 0;
5449		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5450		ctsio->be_move_done = ctl_config_move_done;
5451		ctl_datamove((union ctl_io *)ctsio);
5452
5453		return (CTL_RETVAL_COMPLETE);
5454	}
5455
5456	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5457	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5458	if (len < sizeof (*hdr) ||
5459	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5460	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5461	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5462		ctl_set_invalid_field(ctsio,
5463				      /*sks_valid*/ 0,
5464				      /*command*/ 0,
5465				      /*field*/ 0,
5466				      /*bit_valid*/ 0,
5467				      /*bit*/ 0);
5468		goto done;
5469	}
5470	len = scsi_2btoul(hdr->desc_length);
5471	buf = (struct scsi_unmap_desc *)(hdr + 1);
5472	end = buf + len / sizeof(*buf);
5473
5474	endnz = buf;
5475	for (range = buf; range < end; range++) {
5476		lba = scsi_8btou64(range->lba);
5477		num_blocks = scsi_4btoul(range->length);
5478		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5479		 || ((lba + num_blocks) < lba)) {
5480			ctl_set_lba_out_of_range(ctsio);
5481			ctl_done((union ctl_io *)ctsio);
5482			return (CTL_RETVAL_COMPLETE);
5483		}
5484		if (num_blocks != 0)
5485			endnz = range + 1;
5486	}
5487
5488	/*
5489	 * Block backend can not handle zero last range.
5490	 * Filter it out and return if there is nothing left.
5491	 */
5492	len = (uint8_t *)endnz - (uint8_t *)buf;
5493	if (len == 0) {
5494		ctl_set_success(ctsio);
5495		goto done;
5496	}
5497
5498	mtx_lock(&lun->lun_lock);
5499	ptrlen = (struct ctl_ptr_len_flags *)
5500	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5501	ptrlen->ptr = (void *)buf;
5502	ptrlen->len = len;
5503	ptrlen->flags = byte2;
5504	ctl_check_blocked(lun);
5505	mtx_unlock(&lun->lun_lock);
5506
5507	retval = lun->backend->config_write((union ctl_io *)ctsio);
5508	return (retval);
5509
5510done:
5511	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5512		free(ctsio->kern_data_ptr, M_CTL);
5513		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5514	}
5515	ctl_done((union ctl_io *)ctsio);
5516	return (CTL_RETVAL_COMPLETE);
5517}
5518
5519/*
5520 * Note that this function currently doesn't actually do anything inside
5521 * CTL to enforce things if the DQue bit is turned on.
5522 *
5523 * Also note that this function can't be used in the default case, because
5524 * the DQue bit isn't set in the changeable mask for the control mode page
5525 * anyway.  This is just here as an example for how to implement a page
5526 * handler, and a placeholder in case we want to allow the user to turn
5527 * tagged queueing on and off.
5528 *
5529 * The D_SENSE bit handling is functional, however, and will turn
5530 * descriptor sense on and off for a given LUN.
5531 */
5532int
5533ctl_control_page_handler(struct ctl_scsiio *ctsio,
5534			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5535{
5536	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5537	struct ctl_lun *lun;
5538	int set_ua;
5539	uint32_t initidx;
5540
5541	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5542	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5543	set_ua = 0;
5544
5545	user_cp = (struct scsi_control_page *)page_ptr;
5546	current_cp = (struct scsi_control_page *)
5547		(page_index->page_data + (page_index->page_len *
5548		CTL_PAGE_CURRENT));
5549	saved_cp = (struct scsi_control_page *)
5550		(page_index->page_data + (page_index->page_len *
5551		CTL_PAGE_SAVED));
5552
5553	mtx_lock(&lun->lun_lock);
5554	if (((current_cp->rlec & SCP_DSENSE) == 0)
5555	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5556		/*
5557		 * Descriptor sense is currently turned off and the user
5558		 * wants to turn it on.
5559		 */
5560		current_cp->rlec |= SCP_DSENSE;
5561		saved_cp->rlec |= SCP_DSENSE;
5562		lun->flags |= CTL_LUN_SENSE_DESC;
5563		set_ua = 1;
5564	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
5565		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
5566		/*
5567		 * Descriptor sense is currently turned on, and the user
5568		 * wants to turn it off.
5569		 */
5570		current_cp->rlec &= ~SCP_DSENSE;
5571		saved_cp->rlec &= ~SCP_DSENSE;
5572		lun->flags &= ~CTL_LUN_SENSE_DESC;
5573		set_ua = 1;
5574	}
5575	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
5576	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
5577		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5578		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5579		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5580		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5581		set_ua = 1;
5582	}
5583	if ((current_cp->eca_and_aen & SCP_SWP) !=
5584	    (user_cp->eca_and_aen & SCP_SWP)) {
5585		current_cp->eca_and_aen &= ~SCP_SWP;
5586		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5587		saved_cp->eca_and_aen &= ~SCP_SWP;
5588		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5589		set_ua = 1;
5590	}
5591	if (set_ua != 0)
5592		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5593	mtx_unlock(&lun->lun_lock);
5594
5595	return (0);
5596}
5597
5598int
5599ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
5600		     struct ctl_page_index *page_index, uint8_t *page_ptr)
5601{
5602	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
5603	struct ctl_lun *lun;
5604	int set_ua;
5605	uint32_t initidx;
5606
5607	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5608	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5609	set_ua = 0;
5610
5611	user_cp = (struct scsi_caching_page *)page_ptr;
5612	current_cp = (struct scsi_caching_page *)
5613		(page_index->page_data + (page_index->page_len *
5614		CTL_PAGE_CURRENT));
5615	saved_cp = (struct scsi_caching_page *)
5616		(page_index->page_data + (page_index->page_len *
5617		CTL_PAGE_SAVED));
5618
5619	mtx_lock(&lun->lun_lock);
5620	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
5621	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
5622		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
5623		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
5624		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
5625		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
5626		set_ua = 1;
5627	}
5628	if (set_ua != 0)
5629		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5630	mtx_unlock(&lun->lun_lock);
5631
5632	return (0);
5633}
5634
5635int
5636ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
5637				struct ctl_page_index *page_index,
5638				uint8_t *page_ptr)
5639{
5640	uint8_t *c;
5641	int i;
5642
5643	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
5644	ctl_time_io_secs =
5645		(c[0] << 8) |
5646		(c[1] << 0) |
5647		0;
5648	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
5649	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
5650	printf("page data:");
5651	for (i=0; i<8; i++)
5652		printf(" %.2x",page_ptr[i]);
5653	printf("\n");
5654	return (0);
5655}
5656
5657int
5658ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
5659			       struct ctl_page_index *page_index,
5660			       int pc)
5661{
5662	struct copan_debugconf_subpage *page;
5663
5664	page = (struct copan_debugconf_subpage *)page_index->page_data +
5665		(page_index->page_len * pc);
5666
5667	switch (pc) {
5668	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5669	case SMS_PAGE_CTRL_DEFAULT >> 6:
5670	case SMS_PAGE_CTRL_SAVED >> 6:
5671		/*
5672		 * We don't update the changable or default bits for this page.
5673		 */
5674		break;
5675	case SMS_PAGE_CTRL_CURRENT >> 6:
5676		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
5677		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
5678		break;
5679	default:
5680#ifdef NEEDTOPORT
5681		EPRINT(0, "Invalid PC %d!!", pc);
5682#endif /* NEEDTOPORT */
5683		break;
5684	}
5685	return (0);
5686}
5687
5688
5689static int
5690ctl_do_mode_select(union ctl_io *io)
5691{
5692	struct scsi_mode_page_header *page_header;
5693	struct ctl_page_index *page_index;
5694	struct ctl_scsiio *ctsio;
5695	int control_dev, page_len;
5696	int page_len_offset, page_len_size;
5697	union ctl_modepage_info *modepage_info;
5698	struct ctl_lun *lun;
5699	int *len_left, *len_used;
5700	int retval, i;
5701
5702	ctsio = &io->scsiio;
5703	page_index = NULL;
5704	page_len = 0;
5705	retval = CTL_RETVAL_COMPLETE;
5706
5707	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5708
5709	if (lun->be_lun->lun_type != T_DIRECT)
5710		control_dev = 1;
5711	else
5712		control_dev = 0;
5713
5714	modepage_info = (union ctl_modepage_info *)
5715		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5716	len_left = &modepage_info->header.len_left;
5717	len_used = &modepage_info->header.len_used;
5718
5719do_next_page:
5720
5721	page_header = (struct scsi_mode_page_header *)
5722		(ctsio->kern_data_ptr + *len_used);
5723
5724	if (*len_left == 0) {
5725		free(ctsio->kern_data_ptr, M_CTL);
5726		ctl_set_success(ctsio);
5727		ctl_done((union ctl_io *)ctsio);
5728		return (CTL_RETVAL_COMPLETE);
5729	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
5730
5731		free(ctsio->kern_data_ptr, M_CTL);
5732		ctl_set_param_len_error(ctsio);
5733		ctl_done((union ctl_io *)ctsio);
5734		return (CTL_RETVAL_COMPLETE);
5735
5736	} else if ((page_header->page_code & SMPH_SPF)
5737		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
5738
5739		free(ctsio->kern_data_ptr, M_CTL);
5740		ctl_set_param_len_error(ctsio);
5741		ctl_done((union ctl_io *)ctsio);
5742		return (CTL_RETVAL_COMPLETE);
5743	}
5744
5745
5746	/*
5747	 * XXX KDM should we do something with the block descriptor?
5748	 */
5749	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5750
5751		if ((control_dev != 0)
5752		 && (lun->mode_pages.index[i].page_flags &
5753		     CTL_PAGE_FLAG_DISK_ONLY))
5754			continue;
5755
5756		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5757		    (page_header->page_code & SMPH_PC_MASK))
5758			continue;
5759
5760		/*
5761		 * If neither page has a subpage code, then we've got a
5762		 * match.
5763		 */
5764		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
5765		 && ((page_header->page_code & SMPH_SPF) == 0)) {
5766			page_index = &lun->mode_pages.index[i];
5767			page_len = page_header->page_length;
5768			break;
5769		}
5770
5771		/*
5772		 * If both pages have subpages, then the subpage numbers
5773		 * have to match.
5774		 */
5775		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
5776		  && (page_header->page_code & SMPH_SPF)) {
5777			struct scsi_mode_page_header_sp *sph;
5778
5779			sph = (struct scsi_mode_page_header_sp *)page_header;
5780
5781			if (lun->mode_pages.index[i].subpage ==
5782			    sph->subpage) {
5783				page_index = &lun->mode_pages.index[i];
5784				page_len = scsi_2btoul(sph->page_length);
5785				break;
5786			}
5787		}
5788	}
5789
5790	/*
5791	 * If we couldn't find the page, or if we don't have a mode select
5792	 * handler for it, send back an error to the user.
5793	 */
5794	if ((page_index == NULL)
5795	 || (page_index->select_handler == NULL)) {
5796		ctl_set_invalid_field(ctsio,
5797				      /*sks_valid*/ 1,
5798				      /*command*/ 0,
5799				      /*field*/ *len_used,
5800				      /*bit_valid*/ 0,
5801				      /*bit*/ 0);
5802		free(ctsio->kern_data_ptr, M_CTL);
5803		ctl_done((union ctl_io *)ctsio);
5804		return (CTL_RETVAL_COMPLETE);
5805	}
5806
5807	if (page_index->page_code & SMPH_SPF) {
5808		page_len_offset = 2;
5809		page_len_size = 2;
5810	} else {
5811		page_len_size = 1;
5812		page_len_offset = 1;
5813	}
5814
5815	/*
5816	 * If the length the initiator gives us isn't the one we specify in
5817	 * the mode page header, or if they didn't specify enough data in
5818	 * the CDB to avoid truncating this page, kick out the request.
5819	 */
5820	if ((page_len != (page_index->page_len - page_len_offset -
5821			  page_len_size))
5822	 || (*len_left < page_index->page_len)) {
5823
5824
5825		ctl_set_invalid_field(ctsio,
5826				      /*sks_valid*/ 1,
5827				      /*command*/ 0,
5828				      /*field*/ *len_used + page_len_offset,
5829				      /*bit_valid*/ 0,
5830				      /*bit*/ 0);
5831		free(ctsio->kern_data_ptr, M_CTL);
5832		ctl_done((union ctl_io *)ctsio);
5833		return (CTL_RETVAL_COMPLETE);
5834	}
5835
5836	/*
5837	 * Run through the mode page, checking to make sure that the bits
5838	 * the user changed are actually legal for him to change.
5839	 */
5840	for (i = 0; i < page_index->page_len; i++) {
5841		uint8_t *user_byte, *change_mask, *current_byte;
5842		int bad_bit;
5843		int j;
5844
5845		user_byte = (uint8_t *)page_header + i;
5846		change_mask = page_index->page_data +
5847			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
5848		current_byte = page_index->page_data +
5849			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
5850
5851		/*
5852		 * Check to see whether the user set any bits in this byte
5853		 * that he is not allowed to set.
5854		 */
5855		if ((*user_byte & ~(*change_mask)) ==
5856		    (*current_byte & ~(*change_mask)))
5857			continue;
5858
5859		/*
5860		 * Go through bit by bit to determine which one is illegal.
5861		 */
5862		bad_bit = 0;
5863		for (j = 7; j >= 0; j--) {
5864			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
5865			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
5866				bad_bit = i;
5867				break;
5868			}
5869		}
5870		ctl_set_invalid_field(ctsio,
5871				      /*sks_valid*/ 1,
5872				      /*command*/ 0,
5873				      /*field*/ *len_used + i,
5874				      /*bit_valid*/ 1,
5875				      /*bit*/ bad_bit);
5876		free(ctsio->kern_data_ptr, M_CTL);
5877		ctl_done((union ctl_io *)ctsio);
5878		return (CTL_RETVAL_COMPLETE);
5879	}
5880
5881	/*
5882	 * Decrement these before we call the page handler, since we may
5883	 * end up getting called back one way or another before the handler
5884	 * returns to this context.
5885	 */
5886	*len_left -= page_index->page_len;
5887	*len_used += page_index->page_len;
5888
5889	retval = page_index->select_handler(ctsio, page_index,
5890					    (uint8_t *)page_header);
5891
5892	/*
5893	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
5894	 * wait until this queued command completes to finish processing
5895	 * the mode page.  If it returns anything other than
5896	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
5897	 * already set the sense information, freed the data pointer, and
5898	 * completed the io for us.
5899	 */
5900	if (retval != CTL_RETVAL_COMPLETE)
5901		goto bailout_no_done;
5902
5903	/*
5904	 * If the initiator sent us more than one page, parse the next one.
5905	 */
5906	if (*len_left > 0)
5907		goto do_next_page;
5908
5909	ctl_set_success(ctsio);
5910	free(ctsio->kern_data_ptr, M_CTL);
5911	ctl_done((union ctl_io *)ctsio);
5912
5913bailout_no_done:
5914
5915	return (CTL_RETVAL_COMPLETE);
5916
5917}
5918
5919int
5920ctl_mode_select(struct ctl_scsiio *ctsio)
5921{
5922	int param_len, pf, sp;
5923	int header_size, bd_len;
5924	int len_left, len_used;
5925	struct ctl_page_index *page_index;
5926	struct ctl_lun *lun;
5927	int control_dev, page_len;
5928	union ctl_modepage_info *modepage_info;
5929	int retval;
5930
5931	pf = 0;
5932	sp = 0;
5933	page_len = 0;
5934	len_used = 0;
5935	len_left = 0;
5936	retval = 0;
5937	bd_len = 0;
5938	page_index = NULL;
5939
5940	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5941
5942	if (lun->be_lun->lun_type != T_DIRECT)
5943		control_dev = 1;
5944	else
5945		control_dev = 0;
5946
5947	switch (ctsio->cdb[0]) {
5948	case MODE_SELECT_6: {
5949		struct scsi_mode_select_6 *cdb;
5950
5951		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
5952
5953		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
5954		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
5955
5956		param_len = cdb->length;
5957		header_size = sizeof(struct scsi_mode_header_6);
5958		break;
5959	}
5960	case MODE_SELECT_10: {
5961		struct scsi_mode_select_10 *cdb;
5962
5963		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
5964
5965		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
5966		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
5967
5968		param_len = scsi_2btoul(cdb->length);
5969		header_size = sizeof(struct scsi_mode_header_10);
5970		break;
5971	}
5972	default:
5973		ctl_set_invalid_opcode(ctsio);
5974		ctl_done((union ctl_io *)ctsio);
5975		return (CTL_RETVAL_COMPLETE);
5976		break; /* NOTREACHED */
5977	}
5978
5979	/*
5980	 * From SPC-3:
5981	 * "A parameter list length of zero indicates that the Data-Out Buffer
5982	 * shall be empty. This condition shall not be considered as an error."
5983	 */
5984	if (param_len == 0) {
5985		ctl_set_success(ctsio);
5986		ctl_done((union ctl_io *)ctsio);
5987		return (CTL_RETVAL_COMPLETE);
5988	}
5989
5990	/*
5991	 * Since we'll hit this the first time through, prior to
5992	 * allocation, we don't need to free a data buffer here.
5993	 */
5994	if (param_len < header_size) {
5995		ctl_set_param_len_error(ctsio);
5996		ctl_done((union ctl_io *)ctsio);
5997		return (CTL_RETVAL_COMPLETE);
5998	}
5999
6000	/*
6001	 * Allocate the data buffer and grab the user's data.  In theory,
6002	 * we shouldn't have to sanity check the parameter list length here
6003	 * because the maximum size is 64K.  We should be able to malloc
6004	 * that much without too many problems.
6005	 */
6006	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6007		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6008		ctsio->kern_data_len = param_len;
6009		ctsio->kern_total_len = param_len;
6010		ctsio->kern_data_resid = 0;
6011		ctsio->kern_rel_offset = 0;
6012		ctsio->kern_sg_entries = 0;
6013		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6014		ctsio->be_move_done = ctl_config_move_done;
6015		ctl_datamove((union ctl_io *)ctsio);
6016
6017		return (CTL_RETVAL_COMPLETE);
6018	}
6019
6020	switch (ctsio->cdb[0]) {
6021	case MODE_SELECT_6: {
6022		struct scsi_mode_header_6 *mh6;
6023
6024		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6025		bd_len = mh6->blk_desc_len;
6026		break;
6027	}
6028	case MODE_SELECT_10: {
6029		struct scsi_mode_header_10 *mh10;
6030
6031		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6032		bd_len = scsi_2btoul(mh10->blk_desc_len);
6033		break;
6034	}
6035	default:
6036		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6037		break;
6038	}
6039
6040	if (param_len < (header_size + bd_len)) {
6041		free(ctsio->kern_data_ptr, M_CTL);
6042		ctl_set_param_len_error(ctsio);
6043		ctl_done((union ctl_io *)ctsio);
6044		return (CTL_RETVAL_COMPLETE);
6045	}
6046
6047	/*
6048	 * Set the IO_CONT flag, so that if this I/O gets passed to
6049	 * ctl_config_write_done(), it'll get passed back to
6050	 * ctl_do_mode_select() for further processing, or completion if
6051	 * we're all done.
6052	 */
6053	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6054	ctsio->io_cont = ctl_do_mode_select;
6055
6056	modepage_info = (union ctl_modepage_info *)
6057		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6058
6059	memset(modepage_info, 0, sizeof(*modepage_info));
6060
6061	len_left = param_len - header_size - bd_len;
6062	len_used = header_size + bd_len;
6063
6064	modepage_info->header.len_left = len_left;
6065	modepage_info->header.len_used = len_used;
6066
6067	return (ctl_do_mode_select((union ctl_io *)ctsio));
6068}
6069
6070int
6071ctl_mode_sense(struct ctl_scsiio *ctsio)
6072{
6073	struct ctl_lun *lun;
6074	int pc, page_code, dbd, llba, subpage;
6075	int alloc_len, page_len, header_len, total_len;
6076	struct scsi_mode_block_descr *block_desc;
6077	struct ctl_page_index *page_index;
6078	int control_dev;
6079
6080	dbd = 0;
6081	llba = 0;
6082	block_desc = NULL;
6083	page_index = NULL;
6084
6085	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6086
6087	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6088
6089	if (lun->be_lun->lun_type != T_DIRECT)
6090		control_dev = 1;
6091	else
6092		control_dev = 0;
6093
6094	switch (ctsio->cdb[0]) {
6095	case MODE_SENSE_6: {
6096		struct scsi_mode_sense_6 *cdb;
6097
6098		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6099
6100		header_len = sizeof(struct scsi_mode_hdr_6);
6101		if (cdb->byte2 & SMS_DBD)
6102			dbd = 1;
6103		else
6104			header_len += sizeof(struct scsi_mode_block_descr);
6105
6106		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6107		page_code = cdb->page & SMS_PAGE_CODE;
6108		subpage = cdb->subpage;
6109		alloc_len = cdb->length;
6110		break;
6111	}
6112	case MODE_SENSE_10: {
6113		struct scsi_mode_sense_10 *cdb;
6114
6115		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6116
6117		header_len = sizeof(struct scsi_mode_hdr_10);
6118
6119		if (cdb->byte2 & SMS_DBD)
6120			dbd = 1;
6121		else
6122			header_len += sizeof(struct scsi_mode_block_descr);
6123		if (cdb->byte2 & SMS10_LLBAA)
6124			llba = 1;
6125		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6126		page_code = cdb->page & SMS_PAGE_CODE;
6127		subpage = cdb->subpage;
6128		alloc_len = scsi_2btoul(cdb->length);
6129		break;
6130	}
6131	default:
6132		ctl_set_invalid_opcode(ctsio);
6133		ctl_done((union ctl_io *)ctsio);
6134		return (CTL_RETVAL_COMPLETE);
6135		break; /* NOTREACHED */
6136	}
6137
6138	/*
6139	 * We have to make a first pass through to calculate the size of
6140	 * the pages that match the user's query.  Then we allocate enough
6141	 * memory to hold it, and actually copy the data into the buffer.
6142	 */
6143	switch (page_code) {
6144	case SMS_ALL_PAGES_PAGE: {
6145		int i;
6146
6147		page_len = 0;
6148
6149		/*
6150		 * At the moment, values other than 0 and 0xff here are
6151		 * reserved according to SPC-3.
6152		 */
6153		if ((subpage != SMS_SUBPAGE_PAGE_0)
6154		 && (subpage != SMS_SUBPAGE_ALL)) {
6155			ctl_set_invalid_field(ctsio,
6156					      /*sks_valid*/ 1,
6157					      /*command*/ 1,
6158					      /*field*/ 3,
6159					      /*bit_valid*/ 0,
6160					      /*bit*/ 0);
6161			ctl_done((union ctl_io *)ctsio);
6162			return (CTL_RETVAL_COMPLETE);
6163		}
6164
6165		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6166			if ((control_dev != 0)
6167			 && (lun->mode_pages.index[i].page_flags &
6168			     CTL_PAGE_FLAG_DISK_ONLY))
6169				continue;
6170
6171			/*
6172			 * We don't use this subpage if the user didn't
6173			 * request all subpages.
6174			 */
6175			if ((lun->mode_pages.index[i].subpage != 0)
6176			 && (subpage == SMS_SUBPAGE_PAGE_0))
6177				continue;
6178
6179#if 0
6180			printf("found page %#x len %d\n",
6181			       lun->mode_pages.index[i].page_code &
6182			       SMPH_PC_MASK,
6183			       lun->mode_pages.index[i].page_len);
6184#endif
6185			page_len += lun->mode_pages.index[i].page_len;
6186		}
6187		break;
6188	}
6189	default: {
6190		int i;
6191
6192		page_len = 0;
6193
6194		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6195			/* Look for the right page code */
6196			if ((lun->mode_pages.index[i].page_code &
6197			     SMPH_PC_MASK) != page_code)
6198				continue;
6199
6200			/* Look for the right subpage or the subpage wildcard*/
6201			if ((lun->mode_pages.index[i].subpage != subpage)
6202			 && (subpage != SMS_SUBPAGE_ALL))
6203				continue;
6204
6205			/* Make sure the page is supported for this dev type */
6206			if ((control_dev != 0)
6207			 && (lun->mode_pages.index[i].page_flags &
6208			     CTL_PAGE_FLAG_DISK_ONLY))
6209				continue;
6210
6211#if 0
6212			printf("found page %#x len %d\n",
6213			       lun->mode_pages.index[i].page_code &
6214			       SMPH_PC_MASK,
6215			       lun->mode_pages.index[i].page_len);
6216#endif
6217
6218			page_len += lun->mode_pages.index[i].page_len;
6219		}
6220
6221		if (page_len == 0) {
6222			ctl_set_invalid_field(ctsio,
6223					      /*sks_valid*/ 1,
6224					      /*command*/ 1,
6225					      /*field*/ 2,
6226					      /*bit_valid*/ 1,
6227					      /*bit*/ 5);
6228			ctl_done((union ctl_io *)ctsio);
6229			return (CTL_RETVAL_COMPLETE);
6230		}
6231		break;
6232	}
6233	}
6234
6235	total_len = header_len + page_len;
6236#if 0
6237	printf("header_len = %d, page_len = %d, total_len = %d\n",
6238	       header_len, page_len, total_len);
6239#endif
6240
6241	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6242	ctsio->kern_sg_entries = 0;
6243	ctsio->kern_data_resid = 0;
6244	ctsio->kern_rel_offset = 0;
6245	if (total_len < alloc_len) {
6246		ctsio->residual = alloc_len - total_len;
6247		ctsio->kern_data_len = total_len;
6248		ctsio->kern_total_len = total_len;
6249	} else {
6250		ctsio->residual = 0;
6251		ctsio->kern_data_len = alloc_len;
6252		ctsio->kern_total_len = alloc_len;
6253	}
6254
6255	switch (ctsio->cdb[0]) {
6256	case MODE_SENSE_6: {
6257		struct scsi_mode_hdr_6 *header;
6258
6259		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6260
6261		header->datalen = MIN(total_len - 1, 254);
6262		if (control_dev == 0) {
6263			header->dev_specific = 0x10; /* DPOFUA */
6264			if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6265			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6266			    .eca_and_aen & SCP_SWP) != 0)
6267				    header->dev_specific |= 0x80; /* WP */
6268		}
6269		if (dbd)
6270			header->block_descr_len = 0;
6271		else
6272			header->block_descr_len =
6273				sizeof(struct scsi_mode_block_descr);
6274		block_desc = (struct scsi_mode_block_descr *)&header[1];
6275		break;
6276	}
6277	case MODE_SENSE_10: {
6278		struct scsi_mode_hdr_10 *header;
6279		int datalen;
6280
6281		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6282
6283		datalen = MIN(total_len - 2, 65533);
6284		scsi_ulto2b(datalen, header->datalen);
6285		if (control_dev == 0) {
6286			header->dev_specific = 0x10; /* DPOFUA */
6287			if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6288			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6289			    .eca_and_aen & SCP_SWP) != 0)
6290				    header->dev_specific |= 0x80; /* WP */
6291		}
6292		if (dbd)
6293			scsi_ulto2b(0, header->block_descr_len);
6294		else
6295			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6296				    header->block_descr_len);
6297		block_desc = (struct scsi_mode_block_descr *)&header[1];
6298		break;
6299	}
6300	default:
6301		panic("invalid CDB type %#x", ctsio->cdb[0]);
6302		break; /* NOTREACHED */
6303	}
6304
6305	/*
6306	 * If we've got a disk, use its blocksize in the block
6307	 * descriptor.  Otherwise, just set it to 0.
6308	 */
6309	if (dbd == 0) {
6310		if (control_dev == 0)
6311			scsi_ulto3b(lun->be_lun->blocksize,
6312				    block_desc->block_len);
6313		else
6314			scsi_ulto3b(0, block_desc->block_len);
6315	}
6316
6317	switch (page_code) {
6318	case SMS_ALL_PAGES_PAGE: {
6319		int i, data_used;
6320
6321		data_used = header_len;
6322		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6323			struct ctl_page_index *page_index;
6324
6325			page_index = &lun->mode_pages.index[i];
6326
6327			if ((control_dev != 0)
6328			 && (page_index->page_flags &
6329			    CTL_PAGE_FLAG_DISK_ONLY))
6330				continue;
6331
6332			/*
6333			 * We don't use this subpage if the user didn't
6334			 * request all subpages.  We already checked (above)
6335			 * to make sure the user only specified a subpage
6336			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6337			 */
6338			if ((page_index->subpage != 0)
6339			 && (subpage == SMS_SUBPAGE_PAGE_0))
6340				continue;
6341
6342			/*
6343			 * Call the handler, if it exists, to update the
6344			 * page to the latest values.
6345			 */
6346			if (page_index->sense_handler != NULL)
6347				page_index->sense_handler(ctsio, page_index,pc);
6348
6349			memcpy(ctsio->kern_data_ptr + data_used,
6350			       page_index->page_data +
6351			       (page_index->page_len * pc),
6352			       page_index->page_len);
6353			data_used += page_index->page_len;
6354		}
6355		break;
6356	}
6357	default: {
6358		int i, data_used;
6359
6360		data_used = header_len;
6361
6362		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6363			struct ctl_page_index *page_index;
6364
6365			page_index = &lun->mode_pages.index[i];
6366
6367			/* Look for the right page code */
6368			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6369				continue;
6370
6371			/* Look for the right subpage or the subpage wildcard*/
6372			if ((page_index->subpage != subpage)
6373			 && (subpage != SMS_SUBPAGE_ALL))
6374				continue;
6375
6376			/* Make sure the page is supported for this dev type */
6377			if ((control_dev != 0)
6378			 && (page_index->page_flags &
6379			     CTL_PAGE_FLAG_DISK_ONLY))
6380				continue;
6381
6382			/*
6383			 * Call the handler, if it exists, to update the
6384			 * page to the latest values.
6385			 */
6386			if (page_index->sense_handler != NULL)
6387				page_index->sense_handler(ctsio, page_index,pc);
6388
6389			memcpy(ctsio->kern_data_ptr + data_used,
6390			       page_index->page_data +
6391			       (page_index->page_len * pc),
6392			       page_index->page_len);
6393			data_used += page_index->page_len;
6394		}
6395		break;
6396	}
6397	}
6398
6399	ctl_set_success(ctsio);
6400	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6401	ctsio->be_move_done = ctl_config_move_done;
6402	ctl_datamove((union ctl_io *)ctsio);
6403	return (CTL_RETVAL_COMPLETE);
6404}
6405
6406int
6407ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6408			       struct ctl_page_index *page_index,
6409			       int pc)
6410{
6411	struct ctl_lun *lun;
6412	struct scsi_log_param_header *phdr;
6413	uint8_t *data;
6414	uint64_t val;
6415
6416	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6417	data = page_index->page_data;
6418
6419	if (lun->backend->lun_attr != NULL &&
6420	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6421	     != UINT64_MAX) {
6422		phdr = (struct scsi_log_param_header *)data;
6423		scsi_ulto2b(0x0001, phdr->param_code);
6424		phdr->param_control = SLP_LBIN | SLP_LP;
6425		phdr->param_len = 8;
6426		data = (uint8_t *)(phdr + 1);
6427		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6428		data[4] = 0x02; /* per-pool */
6429		data += phdr->param_len;
6430	}
6431
6432	if (lun->backend->lun_attr != NULL &&
6433	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6434	     != UINT64_MAX) {
6435		phdr = (struct scsi_log_param_header *)data;
6436		scsi_ulto2b(0x0002, phdr->param_code);
6437		phdr->param_control = SLP_LBIN | SLP_LP;
6438		phdr->param_len = 8;
6439		data = (uint8_t *)(phdr + 1);
6440		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6441		data[4] = 0x01; /* per-LUN */
6442		data += phdr->param_len;
6443	}
6444
6445	if (lun->backend->lun_attr != NULL &&
6446	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6447	     != UINT64_MAX) {
6448		phdr = (struct scsi_log_param_header *)data;
6449		scsi_ulto2b(0x00f1, phdr->param_code);
6450		phdr->param_control = SLP_LBIN | SLP_LP;
6451		phdr->param_len = 8;
6452		data = (uint8_t *)(phdr + 1);
6453		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6454		data[4] = 0x02; /* per-pool */
6455		data += phdr->param_len;
6456	}
6457
6458	if (lun->backend->lun_attr != NULL &&
6459	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6460	     != UINT64_MAX) {
6461		phdr = (struct scsi_log_param_header *)data;
6462		scsi_ulto2b(0x00f2, phdr->param_code);
6463		phdr->param_control = SLP_LBIN | SLP_LP;
6464		phdr->param_len = 8;
6465		data = (uint8_t *)(phdr + 1);
6466		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6467		data[4] = 0x02; /* per-pool */
6468		data += phdr->param_len;
6469	}
6470
6471	page_index->page_len = data - page_index->page_data;
6472	return (0);
6473}
6474
6475int
6476ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6477			       struct ctl_page_index *page_index,
6478			       int pc)
6479{
6480	struct ctl_lun *lun;
6481	struct stat_page *data;
6482	uint64_t rn, wn, rb, wb;
6483	struct bintime rt, wt;
6484	int i;
6485
6486	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6487	data = (struct stat_page *)page_index->page_data;
6488
6489	scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6490	data->sap.hdr.param_control = SLP_LBIN;
6491	data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6492	    sizeof(struct scsi_log_param_header);
6493	rn = wn = rb = wb = 0;
6494	bintime_clear(&rt);
6495	bintime_clear(&wt);
6496	for (i = 0; i < CTL_MAX_PORTS; i++) {
6497		rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6498		wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6499		rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6500		wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6501		bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6502		bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6503	}
6504	scsi_u64to8b(rn, data->sap.read_num);
6505	scsi_u64to8b(wn, data->sap.write_num);
6506	if (lun->stats.blocksize > 0) {
6507		scsi_u64to8b(wb / lun->stats.blocksize,
6508		    data->sap.recvieved_lba);
6509		scsi_u64to8b(rb / lun->stats.blocksize,
6510		    data->sap.transmitted_lba);
6511	}
6512	scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6513	    data->sap.read_int);
6514	scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6515	    data->sap.write_int);
6516	scsi_u64to8b(0, data->sap.weighted_num);
6517	scsi_u64to8b(0, data->sap.weighted_int);
6518	scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6519	data->it.hdr.param_control = SLP_LBIN;
6520	data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6521	    sizeof(struct scsi_log_param_header);
6522#ifdef CTL_TIME_IO
6523	scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6524#endif
6525	scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6526	data->it.hdr.param_control = SLP_LBIN;
6527	data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6528	    sizeof(struct scsi_log_param_header);
6529	scsi_ulto4b(3, data->ti.exponent);
6530	scsi_ulto4b(1, data->ti.integer);
6531
6532	page_index->page_len = sizeof(*data);
6533	return (0);
6534}
6535
6536int
6537ctl_log_sense(struct ctl_scsiio *ctsio)
6538{
6539	struct ctl_lun *lun;
6540	int i, pc, page_code, subpage;
6541	int alloc_len, total_len;
6542	struct ctl_page_index *page_index;
6543	struct scsi_log_sense *cdb;
6544	struct scsi_log_header *header;
6545
6546	CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6547
6548	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6549	cdb = (struct scsi_log_sense *)ctsio->cdb;
6550	pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6551	page_code = cdb->page & SLS_PAGE_CODE;
6552	subpage = cdb->subpage;
6553	alloc_len = scsi_2btoul(cdb->length);
6554
6555	page_index = NULL;
6556	for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6557		page_index = &lun->log_pages.index[i];
6558
6559		/* Look for the right page code */
6560		if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6561			continue;
6562
6563		/* Look for the right subpage or the subpage wildcard*/
6564		if (page_index->subpage != subpage)
6565			continue;
6566
6567		break;
6568	}
6569	if (i >= CTL_NUM_LOG_PAGES) {
6570		ctl_set_invalid_field(ctsio,
6571				      /*sks_valid*/ 1,
6572				      /*command*/ 1,
6573				      /*field*/ 2,
6574				      /*bit_valid*/ 0,
6575				      /*bit*/ 0);
6576		ctl_done((union ctl_io *)ctsio);
6577		return (CTL_RETVAL_COMPLETE);
6578	}
6579
6580	total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6581
6582	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6583	ctsio->kern_sg_entries = 0;
6584	ctsio->kern_data_resid = 0;
6585	ctsio->kern_rel_offset = 0;
6586	if (total_len < alloc_len) {
6587		ctsio->residual = alloc_len - total_len;
6588		ctsio->kern_data_len = total_len;
6589		ctsio->kern_total_len = total_len;
6590	} else {
6591		ctsio->residual = 0;
6592		ctsio->kern_data_len = alloc_len;
6593		ctsio->kern_total_len = alloc_len;
6594	}
6595
6596	header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6597	header->page = page_index->page_code;
6598	if (page_index->subpage) {
6599		header->page |= SL_SPF;
6600		header->subpage = page_index->subpage;
6601	}
6602	scsi_ulto2b(page_index->page_len, header->datalen);
6603
6604	/*
6605	 * Call the handler, if it exists, to update the
6606	 * page to the latest values.
6607	 */
6608	if (page_index->sense_handler != NULL)
6609		page_index->sense_handler(ctsio, page_index, pc);
6610
6611	memcpy(header + 1, page_index->page_data, page_index->page_len);
6612
6613	ctl_set_success(ctsio);
6614	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6615	ctsio->be_move_done = ctl_config_move_done;
6616	ctl_datamove((union ctl_io *)ctsio);
6617	return (CTL_RETVAL_COMPLETE);
6618}
6619
6620int
6621ctl_read_capacity(struct ctl_scsiio *ctsio)
6622{
6623	struct scsi_read_capacity *cdb;
6624	struct scsi_read_capacity_data *data;
6625	struct ctl_lun *lun;
6626	uint32_t lba;
6627
6628	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6629
6630	cdb = (struct scsi_read_capacity *)ctsio->cdb;
6631
6632	lba = scsi_4btoul(cdb->addr);
6633	if (((cdb->pmi & SRC_PMI) == 0)
6634	 && (lba != 0)) {
6635		ctl_set_invalid_field(/*ctsio*/ ctsio,
6636				      /*sks_valid*/ 1,
6637				      /*command*/ 1,
6638				      /*field*/ 2,
6639				      /*bit_valid*/ 0,
6640				      /*bit*/ 0);
6641		ctl_done((union ctl_io *)ctsio);
6642		return (CTL_RETVAL_COMPLETE);
6643	}
6644
6645	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6646
6647	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6648	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6649	ctsio->residual = 0;
6650	ctsio->kern_data_len = sizeof(*data);
6651	ctsio->kern_total_len = sizeof(*data);
6652	ctsio->kern_data_resid = 0;
6653	ctsio->kern_rel_offset = 0;
6654	ctsio->kern_sg_entries = 0;
6655
6656	/*
6657	 * If the maximum LBA is greater than 0xfffffffe, the user must
6658	 * issue a SERVICE ACTION IN (16) command, with the read capacity
6659	 * serivce action set.
6660	 */
6661	if (lun->be_lun->maxlba > 0xfffffffe)
6662		scsi_ulto4b(0xffffffff, data->addr);
6663	else
6664		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6665
6666	/*
6667	 * XXX KDM this may not be 512 bytes...
6668	 */
6669	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6670
6671	ctl_set_success(ctsio);
6672	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6673	ctsio->be_move_done = ctl_config_move_done;
6674	ctl_datamove((union ctl_io *)ctsio);
6675	return (CTL_RETVAL_COMPLETE);
6676}
6677
6678int
6679ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6680{
6681	struct scsi_read_capacity_16 *cdb;
6682	struct scsi_read_capacity_data_long *data;
6683	struct ctl_lun *lun;
6684	uint64_t lba;
6685	uint32_t alloc_len;
6686
6687	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6688
6689	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6690
6691	alloc_len = scsi_4btoul(cdb->alloc_len);
6692	lba = scsi_8btou64(cdb->addr);
6693
6694	if ((cdb->reladr & SRC16_PMI)
6695	 && (lba != 0)) {
6696		ctl_set_invalid_field(/*ctsio*/ ctsio,
6697				      /*sks_valid*/ 1,
6698				      /*command*/ 1,
6699				      /*field*/ 2,
6700				      /*bit_valid*/ 0,
6701				      /*bit*/ 0);
6702		ctl_done((union ctl_io *)ctsio);
6703		return (CTL_RETVAL_COMPLETE);
6704	}
6705
6706	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6707
6708	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6709	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6710
6711	if (sizeof(*data) < alloc_len) {
6712		ctsio->residual = alloc_len - sizeof(*data);
6713		ctsio->kern_data_len = sizeof(*data);
6714		ctsio->kern_total_len = sizeof(*data);
6715	} else {
6716		ctsio->residual = 0;
6717		ctsio->kern_data_len = alloc_len;
6718		ctsio->kern_total_len = alloc_len;
6719	}
6720	ctsio->kern_data_resid = 0;
6721	ctsio->kern_rel_offset = 0;
6722	ctsio->kern_sg_entries = 0;
6723
6724	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6725	/* XXX KDM this may not be 512 bytes... */
6726	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6727	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6728	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6729	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6730		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6731
6732	ctl_set_success(ctsio);
6733	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6734	ctsio->be_move_done = ctl_config_move_done;
6735	ctl_datamove((union ctl_io *)ctsio);
6736	return (CTL_RETVAL_COMPLETE);
6737}
6738
6739int
6740ctl_get_lba_status(struct ctl_scsiio *ctsio)
6741{
6742	struct scsi_get_lba_status *cdb;
6743	struct scsi_get_lba_status_data *data;
6744	struct ctl_lun *lun;
6745	struct ctl_lba_len_flags *lbalen;
6746	uint64_t lba;
6747	uint32_t alloc_len, total_len;
6748	int retval;
6749
6750	CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
6751
6752	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6753	cdb = (struct scsi_get_lba_status *)ctsio->cdb;
6754	lba = scsi_8btou64(cdb->addr);
6755	alloc_len = scsi_4btoul(cdb->alloc_len);
6756
6757	if (lba > lun->be_lun->maxlba) {
6758		ctl_set_lba_out_of_range(ctsio);
6759		ctl_done((union ctl_io *)ctsio);
6760		return (CTL_RETVAL_COMPLETE);
6761	}
6762
6763	total_len = sizeof(*data) + sizeof(data->descr[0]);
6764	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6765	data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
6766
6767	if (total_len < alloc_len) {
6768		ctsio->residual = alloc_len - total_len;
6769		ctsio->kern_data_len = total_len;
6770		ctsio->kern_total_len = total_len;
6771	} else {
6772		ctsio->residual = 0;
6773		ctsio->kern_data_len = alloc_len;
6774		ctsio->kern_total_len = alloc_len;
6775	}
6776	ctsio->kern_data_resid = 0;
6777	ctsio->kern_rel_offset = 0;
6778	ctsio->kern_sg_entries = 0;
6779
6780	/* Fill dummy data in case backend can't tell anything. */
6781	scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
6782	scsi_u64to8b(lba, data->descr[0].addr);
6783	scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
6784	    data->descr[0].length);
6785	data->descr[0].status = 0; /* Mapped or unknown. */
6786
6787	ctl_set_success(ctsio);
6788	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6789	ctsio->be_move_done = ctl_config_move_done;
6790
6791	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6792	lbalen->lba = lba;
6793	lbalen->len = total_len;
6794	lbalen->flags = 0;
6795	retval = lun->backend->config_read((union ctl_io *)ctsio);
6796	return (CTL_RETVAL_COMPLETE);
6797}
6798
6799int
6800ctl_read_defect(struct ctl_scsiio *ctsio)
6801{
6802	struct scsi_read_defect_data_10 *ccb10;
6803	struct scsi_read_defect_data_12 *ccb12;
6804	struct scsi_read_defect_data_hdr_10 *data10;
6805	struct scsi_read_defect_data_hdr_12 *data12;
6806	uint32_t alloc_len, data_len;
6807	uint8_t format;
6808
6809	CTL_DEBUG_PRINT(("ctl_read_defect\n"));
6810
6811	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
6812		ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
6813		format = ccb10->format;
6814		alloc_len = scsi_2btoul(ccb10->alloc_length);
6815		data_len = sizeof(*data10);
6816	} else {
6817		ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
6818		format = ccb12->format;
6819		alloc_len = scsi_4btoul(ccb12->alloc_length);
6820		data_len = sizeof(*data12);
6821	}
6822	if (alloc_len == 0) {
6823		ctl_set_success(ctsio);
6824		ctl_done((union ctl_io *)ctsio);
6825		return (CTL_RETVAL_COMPLETE);
6826	}
6827
6828	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
6829	if (data_len < alloc_len) {
6830		ctsio->residual = alloc_len - data_len;
6831		ctsio->kern_data_len = data_len;
6832		ctsio->kern_total_len = data_len;
6833	} else {
6834		ctsio->residual = 0;
6835		ctsio->kern_data_len = alloc_len;
6836		ctsio->kern_total_len = alloc_len;
6837	}
6838	ctsio->kern_data_resid = 0;
6839	ctsio->kern_rel_offset = 0;
6840	ctsio->kern_sg_entries = 0;
6841
6842	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
6843		data10 = (struct scsi_read_defect_data_hdr_10 *)
6844		    ctsio->kern_data_ptr;
6845		data10->format = format;
6846		scsi_ulto2b(0, data10->length);
6847	} else {
6848		data12 = (struct scsi_read_defect_data_hdr_12 *)
6849		    ctsio->kern_data_ptr;
6850		data12->format = format;
6851		scsi_ulto2b(0, data12->generation);
6852		scsi_ulto4b(0, data12->length);
6853	}
6854
6855	ctl_set_success(ctsio);
6856	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6857	ctsio->be_move_done = ctl_config_move_done;
6858	ctl_datamove((union ctl_io *)ctsio);
6859	return (CTL_RETVAL_COMPLETE);
6860}
6861
6862int
6863ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
6864{
6865	struct scsi_maintenance_in *cdb;
6866	int retval;
6867	int alloc_len, ext, total_len = 0, g, p, pc, pg, gs, os;
6868	int num_target_port_groups, num_target_ports;
6869	struct ctl_lun *lun;
6870	struct ctl_softc *softc;
6871	struct ctl_port *port;
6872	struct scsi_target_group_data *rtg_ptr;
6873	struct scsi_target_group_data_extended *rtg_ext_ptr;
6874	struct scsi_target_port_group_descriptor *tpg_desc;
6875
6876	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
6877
6878	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
6879	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6880	softc = lun->ctl_softc;
6881
6882	retval = CTL_RETVAL_COMPLETE;
6883
6884	switch (cdb->byte2 & STG_PDF_MASK) {
6885	case STG_PDF_LENGTH:
6886		ext = 0;
6887		break;
6888	case STG_PDF_EXTENDED:
6889		ext = 1;
6890		break;
6891	default:
6892		ctl_set_invalid_field(/*ctsio*/ ctsio,
6893				      /*sks_valid*/ 1,
6894				      /*command*/ 1,
6895				      /*field*/ 2,
6896				      /*bit_valid*/ 1,
6897				      /*bit*/ 5);
6898		ctl_done((union ctl_io *)ctsio);
6899		return(retval);
6900	}
6901
6902	if (softc->is_single)
6903		num_target_port_groups = 1;
6904	else
6905		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
6906	num_target_ports = 0;
6907	mtx_lock(&softc->ctl_lock);
6908	STAILQ_FOREACH(port, &softc->port_list, links) {
6909		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
6910			continue;
6911		if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
6912			continue;
6913		num_target_ports++;
6914	}
6915	mtx_unlock(&softc->ctl_lock);
6916
6917	if (ext)
6918		total_len = sizeof(struct scsi_target_group_data_extended);
6919	else
6920		total_len = sizeof(struct scsi_target_group_data);
6921	total_len += sizeof(struct scsi_target_port_group_descriptor) *
6922		num_target_port_groups +
6923	    sizeof(struct scsi_target_port_descriptor) *
6924		num_target_ports * num_target_port_groups;
6925
6926	alloc_len = scsi_4btoul(cdb->length);
6927
6928	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6929
6930	ctsio->kern_sg_entries = 0;
6931
6932	if (total_len < alloc_len) {
6933		ctsio->residual = alloc_len - total_len;
6934		ctsio->kern_data_len = total_len;
6935		ctsio->kern_total_len = total_len;
6936	} else {
6937		ctsio->residual = 0;
6938		ctsio->kern_data_len = alloc_len;
6939		ctsio->kern_total_len = alloc_len;
6940	}
6941	ctsio->kern_data_resid = 0;
6942	ctsio->kern_rel_offset = 0;
6943
6944	if (ext) {
6945		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
6946		    ctsio->kern_data_ptr;
6947		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
6948		rtg_ext_ptr->format_type = 0x10;
6949		rtg_ext_ptr->implicit_transition_time = 0;
6950		tpg_desc = &rtg_ext_ptr->groups[0];
6951	} else {
6952		rtg_ptr = (struct scsi_target_group_data *)
6953		    ctsio->kern_data_ptr;
6954		scsi_ulto4b(total_len - 4, rtg_ptr->length);
6955		tpg_desc = &rtg_ptr->groups[0];
6956	}
6957
6958	mtx_lock(&softc->ctl_lock);
6959	pg = softc->port_offset / CTL_MAX_PORTS;
6960	if (softc->flags & CTL_FLAG_ACTIVE_SHELF) {
6961		if (softc->ha_mode == CTL_HA_MODE_ACT_STBY) {
6962			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
6963			os = TPG_ASYMMETRIC_ACCESS_STANDBY;
6964		} else if (lun->flags & CTL_LUN_PRIMARY_SC) {
6965			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
6966			os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6967		} else {
6968			gs = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
6969			os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
6970		}
6971	} else {
6972		gs = TPG_ASYMMETRIC_ACCESS_STANDBY;
6973		os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
6974	}
6975	for (g = 0; g < num_target_port_groups; g++) {
6976		tpg_desc->pref_state = (g == pg) ? gs : os;
6977		tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP;
6978		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
6979		tpg_desc->status = TPG_IMPLICIT;
6980		pc = 0;
6981		STAILQ_FOREACH(port, &softc->port_list, links) {
6982			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
6983				continue;
6984			if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
6985				continue;
6986			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
6987			scsi_ulto2b(p, tpg_desc->descriptors[pc].
6988			    relative_target_port_identifier);
6989			pc++;
6990		}
6991		tpg_desc->target_port_count = pc;
6992		tpg_desc = (struct scsi_target_port_group_descriptor *)
6993		    &tpg_desc->descriptors[pc];
6994	}
6995	mtx_unlock(&softc->ctl_lock);
6996
6997	ctl_set_success(ctsio);
6998	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6999	ctsio->be_move_done = ctl_config_move_done;
7000	ctl_datamove((union ctl_io *)ctsio);
7001	return(retval);
7002}
7003
7004int
7005ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7006{
7007	struct ctl_lun *lun;
7008	struct scsi_report_supported_opcodes *cdb;
7009	const struct ctl_cmd_entry *entry, *sentry;
7010	struct scsi_report_supported_opcodes_all *all;
7011	struct scsi_report_supported_opcodes_descr *descr;
7012	struct scsi_report_supported_opcodes_one *one;
7013	int retval;
7014	int alloc_len, total_len;
7015	int opcode, service_action, i, j, num;
7016
7017	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7018
7019	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7020	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7021
7022	retval = CTL_RETVAL_COMPLETE;
7023
7024	opcode = cdb->requested_opcode;
7025	service_action = scsi_2btoul(cdb->requested_service_action);
7026	switch (cdb->options & RSO_OPTIONS_MASK) {
7027	case RSO_OPTIONS_ALL:
7028		num = 0;
7029		for (i = 0; i < 256; i++) {
7030			entry = &ctl_cmd_table[i];
7031			if (entry->flags & CTL_CMD_FLAG_SA5) {
7032				for (j = 0; j < 32; j++) {
7033					sentry = &((const struct ctl_cmd_entry *)
7034					    entry->execute)[j];
7035					if (ctl_cmd_applicable(
7036					    lun->be_lun->lun_type, sentry))
7037						num++;
7038				}
7039			} else {
7040				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7041				    entry))
7042					num++;
7043			}
7044		}
7045		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7046		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7047		break;
7048	case RSO_OPTIONS_OC:
7049		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7050			ctl_set_invalid_field(/*ctsio*/ ctsio,
7051					      /*sks_valid*/ 1,
7052					      /*command*/ 1,
7053					      /*field*/ 2,
7054					      /*bit_valid*/ 1,
7055					      /*bit*/ 2);
7056			ctl_done((union ctl_io *)ctsio);
7057			return (CTL_RETVAL_COMPLETE);
7058		}
7059		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7060		break;
7061	case RSO_OPTIONS_OC_SA:
7062		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7063		    service_action >= 32) {
7064			ctl_set_invalid_field(/*ctsio*/ ctsio,
7065					      /*sks_valid*/ 1,
7066					      /*command*/ 1,
7067					      /*field*/ 2,
7068					      /*bit_valid*/ 1,
7069					      /*bit*/ 2);
7070			ctl_done((union ctl_io *)ctsio);
7071			return (CTL_RETVAL_COMPLETE);
7072		}
7073		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7074		break;
7075	default:
7076		ctl_set_invalid_field(/*ctsio*/ ctsio,
7077				      /*sks_valid*/ 1,
7078				      /*command*/ 1,
7079				      /*field*/ 2,
7080				      /*bit_valid*/ 1,
7081				      /*bit*/ 2);
7082		ctl_done((union ctl_io *)ctsio);
7083		return (CTL_RETVAL_COMPLETE);
7084	}
7085
7086	alloc_len = scsi_4btoul(cdb->length);
7087
7088	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7089
7090	ctsio->kern_sg_entries = 0;
7091
7092	if (total_len < alloc_len) {
7093		ctsio->residual = alloc_len - total_len;
7094		ctsio->kern_data_len = total_len;
7095		ctsio->kern_total_len = total_len;
7096	} else {
7097		ctsio->residual = 0;
7098		ctsio->kern_data_len = alloc_len;
7099		ctsio->kern_total_len = alloc_len;
7100	}
7101	ctsio->kern_data_resid = 0;
7102	ctsio->kern_rel_offset = 0;
7103
7104	switch (cdb->options & RSO_OPTIONS_MASK) {
7105	case RSO_OPTIONS_ALL:
7106		all = (struct scsi_report_supported_opcodes_all *)
7107		    ctsio->kern_data_ptr;
7108		num = 0;
7109		for (i = 0; i < 256; i++) {
7110			entry = &ctl_cmd_table[i];
7111			if (entry->flags & CTL_CMD_FLAG_SA5) {
7112				for (j = 0; j < 32; j++) {
7113					sentry = &((const struct ctl_cmd_entry *)
7114					    entry->execute)[j];
7115					if (!ctl_cmd_applicable(
7116					    lun->be_lun->lun_type, sentry))
7117						continue;
7118					descr = &all->descr[num++];
7119					descr->opcode = i;
7120					scsi_ulto2b(j, descr->service_action);
7121					descr->flags = RSO_SERVACTV;
7122					scsi_ulto2b(sentry->length,
7123					    descr->cdb_length);
7124				}
7125			} else {
7126				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7127				    entry))
7128					continue;
7129				descr = &all->descr[num++];
7130				descr->opcode = i;
7131				scsi_ulto2b(0, descr->service_action);
7132				descr->flags = 0;
7133				scsi_ulto2b(entry->length, descr->cdb_length);
7134			}
7135		}
7136		scsi_ulto4b(
7137		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7138		    all->length);
7139		break;
7140	case RSO_OPTIONS_OC:
7141		one = (struct scsi_report_supported_opcodes_one *)
7142		    ctsio->kern_data_ptr;
7143		entry = &ctl_cmd_table[opcode];
7144		goto fill_one;
7145	case RSO_OPTIONS_OC_SA:
7146		one = (struct scsi_report_supported_opcodes_one *)
7147		    ctsio->kern_data_ptr;
7148		entry = &ctl_cmd_table[opcode];
7149		entry = &((const struct ctl_cmd_entry *)
7150		    entry->execute)[service_action];
7151fill_one:
7152		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7153			one->support = 3;
7154			scsi_ulto2b(entry->length, one->cdb_length);
7155			one->cdb_usage[0] = opcode;
7156			memcpy(&one->cdb_usage[1], entry->usage,
7157			    entry->length - 1);
7158		} else
7159			one->support = 1;
7160		break;
7161	}
7162
7163	ctl_set_success(ctsio);
7164	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7165	ctsio->be_move_done = ctl_config_move_done;
7166	ctl_datamove((union ctl_io *)ctsio);
7167	return(retval);
7168}
7169
7170int
7171ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7172{
7173	struct scsi_report_supported_tmf *cdb;
7174	struct scsi_report_supported_tmf_data *data;
7175	int retval;
7176	int alloc_len, total_len;
7177
7178	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7179
7180	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7181
7182	retval = CTL_RETVAL_COMPLETE;
7183
7184	total_len = sizeof(struct scsi_report_supported_tmf_data);
7185	alloc_len = scsi_4btoul(cdb->length);
7186
7187	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7188
7189	ctsio->kern_sg_entries = 0;
7190
7191	if (total_len < alloc_len) {
7192		ctsio->residual = alloc_len - total_len;
7193		ctsio->kern_data_len = total_len;
7194		ctsio->kern_total_len = total_len;
7195	} else {
7196		ctsio->residual = 0;
7197		ctsio->kern_data_len = alloc_len;
7198		ctsio->kern_total_len = alloc_len;
7199	}
7200	ctsio->kern_data_resid = 0;
7201	ctsio->kern_rel_offset = 0;
7202
7203	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7204	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7205	data->byte2 |= RST_ITNRS;
7206
7207	ctl_set_success(ctsio);
7208	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7209	ctsio->be_move_done = ctl_config_move_done;
7210	ctl_datamove((union ctl_io *)ctsio);
7211	return (retval);
7212}
7213
7214int
7215ctl_report_timestamp(struct ctl_scsiio *ctsio)
7216{
7217	struct scsi_report_timestamp *cdb;
7218	struct scsi_report_timestamp_data *data;
7219	struct timeval tv;
7220	int64_t timestamp;
7221	int retval;
7222	int alloc_len, total_len;
7223
7224	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7225
7226	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7227
7228	retval = CTL_RETVAL_COMPLETE;
7229
7230	total_len = sizeof(struct scsi_report_timestamp_data);
7231	alloc_len = scsi_4btoul(cdb->length);
7232
7233	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7234
7235	ctsio->kern_sg_entries = 0;
7236
7237	if (total_len < alloc_len) {
7238		ctsio->residual = alloc_len - total_len;
7239		ctsio->kern_data_len = total_len;
7240		ctsio->kern_total_len = total_len;
7241	} else {
7242		ctsio->residual = 0;
7243		ctsio->kern_data_len = alloc_len;
7244		ctsio->kern_total_len = alloc_len;
7245	}
7246	ctsio->kern_data_resid = 0;
7247	ctsio->kern_rel_offset = 0;
7248
7249	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7250	scsi_ulto2b(sizeof(*data) - 2, data->length);
7251	data->origin = RTS_ORIG_OUTSIDE;
7252	getmicrotime(&tv);
7253	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7254	scsi_ulto4b(timestamp >> 16, data->timestamp);
7255	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7256
7257	ctl_set_success(ctsio);
7258	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7259	ctsio->be_move_done = ctl_config_move_done;
7260	ctl_datamove((union ctl_io *)ctsio);
7261	return (retval);
7262}
7263
7264int
7265ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7266{
7267	struct scsi_per_res_in *cdb;
7268	int alloc_len, total_len = 0;
7269	/* struct scsi_per_res_in_rsrv in_data; */
7270	struct ctl_lun *lun;
7271	struct ctl_softc *softc;
7272	uint64_t key;
7273
7274	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7275
7276	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7277
7278	alloc_len = scsi_2btoul(cdb->length);
7279
7280	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7281	softc = lun->ctl_softc;
7282
7283retry:
7284	mtx_lock(&lun->lun_lock);
7285	switch (cdb->action) {
7286	case SPRI_RK: /* read keys */
7287		total_len = sizeof(struct scsi_per_res_in_keys) +
7288			lun->pr_key_count *
7289			sizeof(struct scsi_per_res_key);
7290		break;
7291	case SPRI_RR: /* read reservation */
7292		if (lun->flags & CTL_LUN_PR_RESERVED)
7293			total_len = sizeof(struct scsi_per_res_in_rsrv);
7294		else
7295			total_len = sizeof(struct scsi_per_res_in_header);
7296		break;
7297	case SPRI_RC: /* report capabilities */
7298		total_len = sizeof(struct scsi_per_res_cap);
7299		break;
7300	case SPRI_RS: /* read full status */
7301		total_len = sizeof(struct scsi_per_res_in_header) +
7302		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7303		    lun->pr_key_count;
7304		break;
7305	default:
7306		panic("Invalid PR type %x", cdb->action);
7307	}
7308	mtx_unlock(&lun->lun_lock);
7309
7310	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7311
7312	if (total_len < alloc_len) {
7313		ctsio->residual = alloc_len - total_len;
7314		ctsio->kern_data_len = total_len;
7315		ctsio->kern_total_len = total_len;
7316	} else {
7317		ctsio->residual = 0;
7318		ctsio->kern_data_len = alloc_len;
7319		ctsio->kern_total_len = alloc_len;
7320	}
7321
7322	ctsio->kern_data_resid = 0;
7323	ctsio->kern_rel_offset = 0;
7324	ctsio->kern_sg_entries = 0;
7325
7326	mtx_lock(&lun->lun_lock);
7327	switch (cdb->action) {
7328	case SPRI_RK: { // read keys
7329        struct scsi_per_res_in_keys *res_keys;
7330		int i, key_count;
7331
7332		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7333
7334		/*
7335		 * We had to drop the lock to allocate our buffer, which
7336		 * leaves time for someone to come in with another
7337		 * persistent reservation.  (That is unlikely, though,
7338		 * since this should be the only persistent reservation
7339		 * command active right now.)
7340		 */
7341		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7342		    (lun->pr_key_count *
7343		     sizeof(struct scsi_per_res_key)))){
7344			mtx_unlock(&lun->lun_lock);
7345			free(ctsio->kern_data_ptr, M_CTL);
7346			printf("%s: reservation length changed, retrying\n",
7347			       __func__);
7348			goto retry;
7349		}
7350
7351		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7352
7353		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7354			     lun->pr_key_count, res_keys->header.length);
7355
7356		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7357			if ((key = ctl_get_prkey(lun, i)) == 0)
7358				continue;
7359
7360			/*
7361			 * We used lun->pr_key_count to calculate the
7362			 * size to allocate.  If it turns out the number of
7363			 * initiators with the registered flag set is
7364			 * larger than that (i.e. they haven't been kept in
7365			 * sync), we've got a problem.
7366			 */
7367			if (key_count >= lun->pr_key_count) {
7368#ifdef NEEDTOPORT
7369				csevent_log(CSC_CTL | CSC_SHELF_SW |
7370					    CTL_PR_ERROR,
7371					    csevent_LogType_Fault,
7372					    csevent_AlertLevel_Yellow,
7373					    csevent_FRU_ShelfController,
7374					    csevent_FRU_Firmware,
7375				        csevent_FRU_Unknown,
7376					    "registered keys %d >= key "
7377					    "count %d", key_count,
7378					    lun->pr_key_count);
7379#endif
7380				key_count++;
7381				continue;
7382			}
7383			scsi_u64to8b(key, res_keys->keys[key_count].key);
7384			key_count++;
7385		}
7386		break;
7387	}
7388	case SPRI_RR: { // read reservation
7389		struct scsi_per_res_in_rsrv *res;
7390		int tmp_len, header_only;
7391
7392		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7393
7394		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7395
7396		if (lun->flags & CTL_LUN_PR_RESERVED)
7397		{
7398			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7399			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7400				    res->header.length);
7401			header_only = 0;
7402		} else {
7403			tmp_len = sizeof(struct scsi_per_res_in_header);
7404			scsi_ulto4b(0, res->header.length);
7405			header_only = 1;
7406		}
7407
7408		/*
7409		 * We had to drop the lock to allocate our buffer, which
7410		 * leaves time for someone to come in with another
7411		 * persistent reservation.  (That is unlikely, though,
7412		 * since this should be the only persistent reservation
7413		 * command active right now.)
7414		 */
7415		if (tmp_len != total_len) {
7416			mtx_unlock(&lun->lun_lock);
7417			free(ctsio->kern_data_ptr, M_CTL);
7418			printf("%s: reservation status changed, retrying\n",
7419			       __func__);
7420			goto retry;
7421		}
7422
7423		/*
7424		 * No reservation held, so we're done.
7425		 */
7426		if (header_only != 0)
7427			break;
7428
7429		/*
7430		 * If the registration is an All Registrants type, the key
7431		 * is 0, since it doesn't really matter.
7432		 */
7433		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7434			scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7435			    res->data.reservation);
7436		}
7437		res->data.scopetype = lun->res_type;
7438		break;
7439	}
7440	case SPRI_RC:     //report capabilities
7441	{
7442		struct scsi_per_res_cap *res_cap;
7443		uint16_t type_mask;
7444
7445		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7446		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7447		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7448		type_mask = SPRI_TM_WR_EX_AR |
7449			    SPRI_TM_EX_AC_RO |
7450			    SPRI_TM_WR_EX_RO |
7451			    SPRI_TM_EX_AC |
7452			    SPRI_TM_WR_EX |
7453			    SPRI_TM_EX_AC_AR;
7454		scsi_ulto2b(type_mask, res_cap->type_mask);
7455		break;
7456	}
7457	case SPRI_RS: { // read full status
7458		struct scsi_per_res_in_full *res_status;
7459		struct scsi_per_res_in_full_desc *res_desc;
7460		struct ctl_port *port;
7461		int i, len;
7462
7463		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7464
7465		/*
7466		 * We had to drop the lock to allocate our buffer, which
7467		 * leaves time for someone to come in with another
7468		 * persistent reservation.  (That is unlikely, though,
7469		 * since this should be the only persistent reservation
7470		 * command active right now.)
7471		 */
7472		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7473		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7474		     lun->pr_key_count)){
7475			mtx_unlock(&lun->lun_lock);
7476			free(ctsio->kern_data_ptr, M_CTL);
7477			printf("%s: reservation length changed, retrying\n",
7478			       __func__);
7479			goto retry;
7480		}
7481
7482		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7483
7484		res_desc = &res_status->desc[0];
7485		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7486			if ((key = ctl_get_prkey(lun, i)) == 0)
7487				continue;
7488
7489			scsi_u64to8b(key, res_desc->res_key.key);
7490			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7491			    (lun->pr_res_idx == i ||
7492			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7493				res_desc->flags = SPRI_FULL_R_HOLDER;
7494				res_desc->scopetype = lun->res_type;
7495			}
7496			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7497			    res_desc->rel_trgt_port_id);
7498			len = 0;
7499			port = softc->ctl_ports[
7500			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7501			if (port != NULL)
7502				len = ctl_create_iid(port,
7503				    i % CTL_MAX_INIT_PER_PORT,
7504				    res_desc->transport_id);
7505			scsi_ulto4b(len, res_desc->additional_length);
7506			res_desc = (struct scsi_per_res_in_full_desc *)
7507			    &res_desc->transport_id[len];
7508		}
7509		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7510		    res_status->header.length);
7511		break;
7512	}
7513	default:
7514		/*
7515		 * This is a bug, because we just checked for this above,
7516		 * and should have returned an error.
7517		 */
7518		panic("Invalid PR type %x", cdb->action);
7519		break; /* NOTREACHED */
7520	}
7521	mtx_unlock(&lun->lun_lock);
7522
7523	ctl_set_success(ctsio);
7524	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7525	ctsio->be_move_done = ctl_config_move_done;
7526	ctl_datamove((union ctl_io *)ctsio);
7527	return (CTL_RETVAL_COMPLETE);
7528}
7529
7530static void
7531ctl_est_res_ua(struct ctl_lun *lun, uint32_t residx, ctl_ua_type ua)
7532{
7533	int off = lun->ctl_softc->persis_offset;
7534
7535	if (residx >= off && residx < off + CTL_MAX_INITIATORS)
7536		ctl_est_ua(lun, residx - off, ua);
7537}
7538
7539/*
7540 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7541 * it should return.
7542 */
7543static int
7544ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7545		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7546		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7547		struct scsi_per_res_out_parms* param)
7548{
7549	union ctl_ha_msg persis_io;
7550	int retval, i;
7551	int isc_retval;
7552
7553	retval = 0;
7554
7555	mtx_lock(&lun->lun_lock);
7556	if (sa_res_key == 0) {
7557		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7558			/* validate scope and type */
7559			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7560			     SPR_LU_SCOPE) {
7561				mtx_unlock(&lun->lun_lock);
7562				ctl_set_invalid_field(/*ctsio*/ ctsio,
7563						      /*sks_valid*/ 1,
7564						      /*command*/ 1,
7565						      /*field*/ 2,
7566						      /*bit_valid*/ 1,
7567						      /*bit*/ 4);
7568				ctl_done((union ctl_io *)ctsio);
7569				return (1);
7570			}
7571
7572		        if (type>8 || type==2 || type==4 || type==0) {
7573				mtx_unlock(&lun->lun_lock);
7574				ctl_set_invalid_field(/*ctsio*/ ctsio,
7575       	           				      /*sks_valid*/ 1,
7576						      /*command*/ 1,
7577						      /*field*/ 2,
7578						      /*bit_valid*/ 1,
7579						      /*bit*/ 0);
7580				ctl_done((union ctl_io *)ctsio);
7581				return (1);
7582		        }
7583
7584			/*
7585			 * Unregister everybody else and build UA for
7586			 * them
7587			 */
7588			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7589				if (i == residx || ctl_get_prkey(lun, i) == 0)
7590					continue;
7591
7592				ctl_clr_prkey(lun, i);
7593				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7594			}
7595			lun->pr_key_count = 1;
7596			lun->res_type = type;
7597			if (lun->res_type != SPR_TYPE_WR_EX_AR
7598			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7599				lun->pr_res_idx = residx;
7600
7601			/* send msg to other side */
7602			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7603			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7604			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7605			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7606			persis_io.pr.pr_info.res_type = type;
7607			memcpy(persis_io.pr.pr_info.sa_res_key,
7608			       param->serv_act_res_key,
7609			       sizeof(param->serv_act_res_key));
7610			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7611			     &persis_io, sizeof(persis_io), 0)) >
7612			     CTL_HA_STATUS_SUCCESS) {
7613				printf("CTL:Persis Out error returned "
7614				       "from ctl_ha_msg_send %d\n",
7615				       isc_retval);
7616			}
7617		} else {
7618			/* not all registrants */
7619			mtx_unlock(&lun->lun_lock);
7620			free(ctsio->kern_data_ptr, M_CTL);
7621			ctl_set_invalid_field(ctsio,
7622					      /*sks_valid*/ 1,
7623					      /*command*/ 0,
7624					      /*field*/ 8,
7625					      /*bit_valid*/ 0,
7626					      /*bit*/ 0);
7627			ctl_done((union ctl_io *)ctsio);
7628			return (1);
7629		}
7630	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7631		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7632		int found = 0;
7633
7634		if (res_key == sa_res_key) {
7635			/* special case */
7636			/*
7637			 * The spec implies this is not good but doesn't
7638			 * say what to do. There are two choices either
7639			 * generate a res conflict or check condition
7640			 * with illegal field in parameter data. Since
7641			 * that is what is done when the sa_res_key is
7642			 * zero I'll take that approach since this has
7643			 * to do with the sa_res_key.
7644			 */
7645			mtx_unlock(&lun->lun_lock);
7646			free(ctsio->kern_data_ptr, M_CTL);
7647			ctl_set_invalid_field(ctsio,
7648					      /*sks_valid*/ 1,
7649					      /*command*/ 0,
7650					      /*field*/ 8,
7651					      /*bit_valid*/ 0,
7652					      /*bit*/ 0);
7653			ctl_done((union ctl_io *)ctsio);
7654			return (1);
7655		}
7656
7657		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7658			if (ctl_get_prkey(lun, i) != sa_res_key)
7659				continue;
7660
7661			found = 1;
7662			ctl_clr_prkey(lun, i);
7663			lun->pr_key_count--;
7664			ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7665		}
7666		if (!found) {
7667			mtx_unlock(&lun->lun_lock);
7668			free(ctsio->kern_data_ptr, M_CTL);
7669			ctl_set_reservation_conflict(ctsio);
7670			ctl_done((union ctl_io *)ctsio);
7671			return (CTL_RETVAL_COMPLETE);
7672		}
7673		/* send msg to other side */
7674		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7675		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7676		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7677		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7678		persis_io.pr.pr_info.res_type = type;
7679		memcpy(persis_io.pr.pr_info.sa_res_key,
7680		       param->serv_act_res_key,
7681		       sizeof(param->serv_act_res_key));
7682		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7683		     &persis_io, sizeof(persis_io), 0)) >
7684		     CTL_HA_STATUS_SUCCESS) {
7685			printf("CTL:Persis Out error returned from "
7686			       "ctl_ha_msg_send %d\n", isc_retval);
7687		}
7688	} else {
7689		/* Reserved but not all registrants */
7690		/* sa_res_key is res holder */
7691		if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7692			/* validate scope and type */
7693			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7694			     SPR_LU_SCOPE) {
7695				mtx_unlock(&lun->lun_lock);
7696				ctl_set_invalid_field(/*ctsio*/ ctsio,
7697						      /*sks_valid*/ 1,
7698						      /*command*/ 1,
7699						      /*field*/ 2,
7700						      /*bit_valid*/ 1,
7701						      /*bit*/ 4);
7702				ctl_done((union ctl_io *)ctsio);
7703				return (1);
7704			}
7705
7706			if (type>8 || type==2 || type==4 || type==0) {
7707				mtx_unlock(&lun->lun_lock);
7708				ctl_set_invalid_field(/*ctsio*/ ctsio,
7709						      /*sks_valid*/ 1,
7710						      /*command*/ 1,
7711						      /*field*/ 2,
7712						      /*bit_valid*/ 1,
7713						      /*bit*/ 0);
7714				ctl_done((union ctl_io *)ctsio);
7715				return (1);
7716			}
7717
7718			/*
7719			 * Do the following:
7720			 * if sa_res_key != res_key remove all
7721			 * registrants w/sa_res_key and generate UA
7722			 * for these registrants(Registrations
7723			 * Preempted) if it wasn't an exclusive
7724			 * reservation generate UA(Reservations
7725			 * Preempted) for all other registered nexuses
7726			 * if the type has changed. Establish the new
7727			 * reservation and holder. If res_key and
7728			 * sa_res_key are the same do the above
7729			 * except don't unregister the res holder.
7730			 */
7731
7732			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7733				if (i == residx || ctl_get_prkey(lun, i) == 0)
7734					continue;
7735
7736				if (sa_res_key == ctl_get_prkey(lun, i)) {
7737					ctl_clr_prkey(lun, i);
7738					lun->pr_key_count--;
7739					ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7740				} else if (type != lun->res_type
7741					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7742					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7743					ctl_est_res_ua(lun, i, CTL_UA_RES_RELEASE);
7744				}
7745			}
7746			lun->res_type = type;
7747			if (lun->res_type != SPR_TYPE_WR_EX_AR
7748			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7749				lun->pr_res_idx = residx;
7750			else
7751				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7752
7753			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7754			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7755			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7756			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7757			persis_io.pr.pr_info.res_type = type;
7758			memcpy(persis_io.pr.pr_info.sa_res_key,
7759			       param->serv_act_res_key,
7760			       sizeof(param->serv_act_res_key));
7761			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7762			     &persis_io, sizeof(persis_io), 0)) >
7763			     CTL_HA_STATUS_SUCCESS) {
7764				printf("CTL:Persis Out error returned "
7765				       "from ctl_ha_msg_send %d\n",
7766				       isc_retval);
7767			}
7768		} else {
7769			/*
7770			 * sa_res_key is not the res holder just
7771			 * remove registrants
7772			 */
7773			int found=0;
7774
7775			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7776				if (sa_res_key != ctl_get_prkey(lun, i))
7777					continue;
7778
7779				found = 1;
7780				ctl_clr_prkey(lun, i);
7781				lun->pr_key_count--;
7782				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7783			}
7784
7785			if (!found) {
7786				mtx_unlock(&lun->lun_lock);
7787				free(ctsio->kern_data_ptr, M_CTL);
7788				ctl_set_reservation_conflict(ctsio);
7789				ctl_done((union ctl_io *)ctsio);
7790		        	return (1);
7791			}
7792			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7793			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7794			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7795			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7796			persis_io.pr.pr_info.res_type = type;
7797			memcpy(persis_io.pr.pr_info.sa_res_key,
7798			       param->serv_act_res_key,
7799			       sizeof(param->serv_act_res_key));
7800			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7801			     &persis_io, sizeof(persis_io), 0)) >
7802			     CTL_HA_STATUS_SUCCESS) {
7803				printf("CTL:Persis Out error returned "
7804				       "from ctl_ha_msg_send %d\n",
7805				isc_retval);
7806			}
7807		}
7808	}
7809
7810	lun->PRGeneration++;
7811	mtx_unlock(&lun->lun_lock);
7812
7813	return (retval);
7814}
7815
7816static void
7817ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7818{
7819	uint64_t sa_res_key;
7820	int i;
7821
7822	sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7823
7824	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7825	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7826	 || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
7827		if (sa_res_key == 0) {
7828			/*
7829			 * Unregister everybody else and build UA for
7830			 * them
7831			 */
7832			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7833				if (i == msg->pr.pr_info.residx ||
7834				    ctl_get_prkey(lun, i) == 0)
7835					continue;
7836
7837				ctl_clr_prkey(lun, i);
7838				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7839			}
7840
7841			lun->pr_key_count = 1;
7842			lun->res_type = msg->pr.pr_info.res_type;
7843			if (lun->res_type != SPR_TYPE_WR_EX_AR
7844			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7845				lun->pr_res_idx = msg->pr.pr_info.residx;
7846		} else {
7847		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7848				if (sa_res_key == ctl_get_prkey(lun, i))
7849					continue;
7850
7851				ctl_clr_prkey(lun, i);
7852				lun->pr_key_count--;
7853				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7854			}
7855		}
7856	} else {
7857		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7858			if (i == msg->pr.pr_info.residx ||
7859			    ctl_get_prkey(lun, i) == 0)
7860				continue;
7861
7862			if (sa_res_key == ctl_get_prkey(lun, i)) {
7863				ctl_clr_prkey(lun, i);
7864				lun->pr_key_count--;
7865				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7866			} else if (msg->pr.pr_info.res_type != lun->res_type
7867				&& (lun->res_type == SPR_TYPE_WR_EX_RO
7868				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7869				ctl_est_res_ua(lun, i, CTL_UA_RES_RELEASE);
7870			}
7871		}
7872		lun->res_type = msg->pr.pr_info.res_type;
7873		if (lun->res_type != SPR_TYPE_WR_EX_AR
7874		 && lun->res_type != SPR_TYPE_EX_AC_AR)
7875			lun->pr_res_idx = msg->pr.pr_info.residx;
7876		else
7877			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7878	}
7879	lun->PRGeneration++;
7880
7881}
7882
7883
7884int
7885ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
7886{
7887	int retval;
7888	int isc_retval;
7889	u_int32_t param_len;
7890	struct scsi_per_res_out *cdb;
7891	struct ctl_lun *lun;
7892	struct scsi_per_res_out_parms* param;
7893	struct ctl_softc *softc;
7894	uint32_t residx;
7895	uint64_t res_key, sa_res_key, key;
7896	uint8_t type;
7897	union ctl_ha_msg persis_io;
7898	int    i;
7899
7900	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
7901
7902	retval = CTL_RETVAL_COMPLETE;
7903
7904	cdb = (struct scsi_per_res_out *)ctsio->cdb;
7905	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7906	softc = lun->ctl_softc;
7907
7908	/*
7909	 * We only support whole-LUN scope.  The scope & type are ignored for
7910	 * register, register and ignore existing key and clear.
7911	 * We sometimes ignore scope and type on preempts too!!
7912	 * Verify reservation type here as well.
7913	 */
7914	type = cdb->scope_type & SPR_TYPE_MASK;
7915	if ((cdb->action == SPRO_RESERVE)
7916	 || (cdb->action == SPRO_RELEASE)) {
7917		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
7918			ctl_set_invalid_field(/*ctsio*/ ctsio,
7919					      /*sks_valid*/ 1,
7920					      /*command*/ 1,
7921					      /*field*/ 2,
7922					      /*bit_valid*/ 1,
7923					      /*bit*/ 4);
7924			ctl_done((union ctl_io *)ctsio);
7925			return (CTL_RETVAL_COMPLETE);
7926		}
7927
7928		if (type>8 || type==2 || type==4 || type==0) {
7929			ctl_set_invalid_field(/*ctsio*/ ctsio,
7930					      /*sks_valid*/ 1,
7931					      /*command*/ 1,
7932					      /*field*/ 2,
7933					      /*bit_valid*/ 1,
7934					      /*bit*/ 0);
7935			ctl_done((union ctl_io *)ctsio);
7936			return (CTL_RETVAL_COMPLETE);
7937		}
7938	}
7939
7940	param_len = scsi_4btoul(cdb->length);
7941
7942	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
7943		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
7944		ctsio->kern_data_len = param_len;
7945		ctsio->kern_total_len = param_len;
7946		ctsio->kern_data_resid = 0;
7947		ctsio->kern_rel_offset = 0;
7948		ctsio->kern_sg_entries = 0;
7949		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7950		ctsio->be_move_done = ctl_config_move_done;
7951		ctl_datamove((union ctl_io *)ctsio);
7952
7953		return (CTL_RETVAL_COMPLETE);
7954	}
7955
7956	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
7957
7958	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
7959	res_key = scsi_8btou64(param->res_key.key);
7960	sa_res_key = scsi_8btou64(param->serv_act_res_key);
7961
7962	/*
7963	 * Validate the reservation key here except for SPRO_REG_IGNO
7964	 * This must be done for all other service actions
7965	 */
7966	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
7967		mtx_lock(&lun->lun_lock);
7968		if ((key = ctl_get_prkey(lun, residx)) != 0) {
7969			if (res_key != key) {
7970				/*
7971				 * The current key passed in doesn't match
7972				 * the one the initiator previously
7973				 * registered.
7974				 */
7975				mtx_unlock(&lun->lun_lock);
7976				free(ctsio->kern_data_ptr, M_CTL);
7977				ctl_set_reservation_conflict(ctsio);
7978				ctl_done((union ctl_io *)ctsio);
7979				return (CTL_RETVAL_COMPLETE);
7980			}
7981		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
7982			/*
7983			 * We are not registered
7984			 */
7985			mtx_unlock(&lun->lun_lock);
7986			free(ctsio->kern_data_ptr, M_CTL);
7987			ctl_set_reservation_conflict(ctsio);
7988			ctl_done((union ctl_io *)ctsio);
7989			return (CTL_RETVAL_COMPLETE);
7990		} else if (res_key != 0) {
7991			/*
7992			 * We are not registered and trying to register but
7993			 * the register key isn't zero.
7994			 */
7995			mtx_unlock(&lun->lun_lock);
7996			free(ctsio->kern_data_ptr, M_CTL);
7997			ctl_set_reservation_conflict(ctsio);
7998			ctl_done((union ctl_io *)ctsio);
7999			return (CTL_RETVAL_COMPLETE);
8000		}
8001		mtx_unlock(&lun->lun_lock);
8002	}
8003
8004	switch (cdb->action & SPRO_ACTION_MASK) {
8005	case SPRO_REGISTER:
8006	case SPRO_REG_IGNO: {
8007
8008#if 0
8009		printf("Registration received\n");
8010#endif
8011
8012		/*
8013		 * We don't support any of these options, as we report in
8014		 * the read capabilities request (see
8015		 * ctl_persistent_reserve_in(), above).
8016		 */
8017		if ((param->flags & SPR_SPEC_I_PT)
8018		 || (param->flags & SPR_ALL_TG_PT)
8019		 || (param->flags & SPR_APTPL)) {
8020			int bit_ptr;
8021
8022			if (param->flags & SPR_APTPL)
8023				bit_ptr = 0;
8024			else if (param->flags & SPR_ALL_TG_PT)
8025				bit_ptr = 2;
8026			else /* SPR_SPEC_I_PT */
8027				bit_ptr = 3;
8028
8029			free(ctsio->kern_data_ptr, M_CTL);
8030			ctl_set_invalid_field(ctsio,
8031					      /*sks_valid*/ 1,
8032					      /*command*/ 0,
8033					      /*field*/ 20,
8034					      /*bit_valid*/ 1,
8035					      /*bit*/ bit_ptr);
8036			ctl_done((union ctl_io *)ctsio);
8037			return (CTL_RETVAL_COMPLETE);
8038		}
8039
8040		mtx_lock(&lun->lun_lock);
8041
8042		/*
8043		 * The initiator wants to clear the
8044		 * key/unregister.
8045		 */
8046		if (sa_res_key == 0) {
8047			if ((res_key == 0
8048			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8049			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8050			  && ctl_get_prkey(lun, residx) == 0)) {
8051				mtx_unlock(&lun->lun_lock);
8052				goto done;
8053			}
8054
8055			ctl_clr_prkey(lun, residx);
8056			lun->pr_key_count--;
8057
8058			if (residx == lun->pr_res_idx) {
8059				lun->flags &= ~CTL_LUN_PR_RESERVED;
8060				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8061
8062				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8063				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8064				 && lun->pr_key_count) {
8065					/*
8066					 * If the reservation is a registrants
8067					 * only type we need to generate a UA
8068					 * for other registered inits.  The
8069					 * sense code should be RESERVATIONS
8070					 * RELEASED
8071					 */
8072
8073					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8074						if (ctl_get_prkey(lun, i +
8075						    softc->persis_offset) == 0)
8076							continue;
8077						ctl_est_ua(lun, i,
8078						    CTL_UA_RES_RELEASE);
8079					}
8080				}
8081				lun->res_type = 0;
8082			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8083				if (lun->pr_key_count==0) {
8084					lun->flags &= ~CTL_LUN_PR_RESERVED;
8085					lun->res_type = 0;
8086					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8087				}
8088			}
8089			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8090			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8091			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8092			persis_io.pr.pr_info.residx = residx;
8093			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8094			     &persis_io, sizeof(persis_io), 0 )) >
8095			     CTL_HA_STATUS_SUCCESS) {
8096				printf("CTL:Persis Out error returned from "
8097				       "ctl_ha_msg_send %d\n", isc_retval);
8098			}
8099		} else /* sa_res_key != 0 */ {
8100
8101			/*
8102			 * If we aren't registered currently then increment
8103			 * the key count and set the registered flag.
8104			 */
8105			ctl_alloc_prkey(lun, residx);
8106			if (ctl_get_prkey(lun, residx) == 0)
8107				lun->pr_key_count++;
8108			ctl_set_prkey(lun, residx, sa_res_key);
8109
8110			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8111			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8112			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8113			persis_io.pr.pr_info.residx = residx;
8114			memcpy(persis_io.pr.pr_info.sa_res_key,
8115			       param->serv_act_res_key,
8116			       sizeof(param->serv_act_res_key));
8117			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8118			     &persis_io, sizeof(persis_io), 0)) >
8119			     CTL_HA_STATUS_SUCCESS) {
8120				printf("CTL:Persis Out error returned from "
8121				       "ctl_ha_msg_send %d\n", isc_retval);
8122			}
8123		}
8124		lun->PRGeneration++;
8125		mtx_unlock(&lun->lun_lock);
8126
8127		break;
8128	}
8129	case SPRO_RESERVE:
8130#if 0
8131                printf("Reserve executed type %d\n", type);
8132#endif
8133		mtx_lock(&lun->lun_lock);
8134		if (lun->flags & CTL_LUN_PR_RESERVED) {
8135			/*
8136			 * if this isn't the reservation holder and it's
8137			 * not a "all registrants" type or if the type is
8138			 * different then we have a conflict
8139			 */
8140			if ((lun->pr_res_idx != residx
8141			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8142			 || lun->res_type != type) {
8143				mtx_unlock(&lun->lun_lock);
8144				free(ctsio->kern_data_ptr, M_CTL);
8145				ctl_set_reservation_conflict(ctsio);
8146				ctl_done((union ctl_io *)ctsio);
8147				return (CTL_RETVAL_COMPLETE);
8148			}
8149			mtx_unlock(&lun->lun_lock);
8150		} else /* create a reservation */ {
8151			/*
8152			 * If it's not an "all registrants" type record
8153			 * reservation holder
8154			 */
8155			if (type != SPR_TYPE_WR_EX_AR
8156			 && type != SPR_TYPE_EX_AC_AR)
8157				lun->pr_res_idx = residx; /* Res holder */
8158			else
8159				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8160
8161			lun->flags |= CTL_LUN_PR_RESERVED;
8162			lun->res_type = type;
8163
8164			mtx_unlock(&lun->lun_lock);
8165
8166			/* send msg to other side */
8167			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8168			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8169			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8170			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8171			persis_io.pr.pr_info.res_type = type;
8172			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8173			     &persis_io, sizeof(persis_io), 0)) >
8174			     CTL_HA_STATUS_SUCCESS) {
8175				printf("CTL:Persis Out error returned from "
8176				       "ctl_ha_msg_send %d\n", isc_retval);
8177			}
8178		}
8179		break;
8180
8181	case SPRO_RELEASE:
8182		mtx_lock(&lun->lun_lock);
8183		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8184			/* No reservation exists return good status */
8185			mtx_unlock(&lun->lun_lock);
8186			goto done;
8187		}
8188		/*
8189		 * Is this nexus a reservation holder?
8190		 */
8191		if (lun->pr_res_idx != residx
8192		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8193			/*
8194			 * not a res holder return good status but
8195			 * do nothing
8196			 */
8197			mtx_unlock(&lun->lun_lock);
8198			goto done;
8199		}
8200
8201		if (lun->res_type != type) {
8202			mtx_unlock(&lun->lun_lock);
8203			free(ctsio->kern_data_ptr, M_CTL);
8204			ctl_set_illegal_pr_release(ctsio);
8205			ctl_done((union ctl_io *)ctsio);
8206			return (CTL_RETVAL_COMPLETE);
8207		}
8208
8209		/* okay to release */
8210		lun->flags &= ~CTL_LUN_PR_RESERVED;
8211		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8212		lun->res_type = 0;
8213
8214		/*
8215		 * if this isn't an exclusive access
8216		 * res generate UA for all other
8217		 * registrants.
8218		 */
8219		if (type != SPR_TYPE_EX_AC
8220		 && type != SPR_TYPE_WR_EX) {
8221			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8222				if (i == residx ||
8223				    ctl_get_prkey(lun,
8224				     i + softc->persis_offset) == 0)
8225					continue;
8226				ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8227			}
8228		}
8229		mtx_unlock(&lun->lun_lock);
8230		/* Send msg to other side */
8231		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8232		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8233		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8234		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8235		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8236			printf("CTL:Persis Out error returned from "
8237			       "ctl_ha_msg_send %d\n", isc_retval);
8238		}
8239		break;
8240
8241	case SPRO_CLEAR:
8242		/* send msg to other side */
8243
8244		mtx_lock(&lun->lun_lock);
8245		lun->flags &= ~CTL_LUN_PR_RESERVED;
8246		lun->res_type = 0;
8247		lun->pr_key_count = 0;
8248		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8249
8250		ctl_clr_prkey(lun, residx);
8251		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8252			if (ctl_get_prkey(lun, i) != 0) {
8253				ctl_clr_prkey(lun, i);
8254				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8255			}
8256		lun->PRGeneration++;
8257		mtx_unlock(&lun->lun_lock);
8258		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8259		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8260		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8261		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8262		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8263			printf("CTL:Persis Out error returned from "
8264			       "ctl_ha_msg_send %d\n", isc_retval);
8265		}
8266		break;
8267
8268	case SPRO_PREEMPT:
8269	case SPRO_PRE_ABO: {
8270		int nretval;
8271
8272		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8273					  residx, ctsio, cdb, param);
8274		if (nretval != 0)
8275			return (CTL_RETVAL_COMPLETE);
8276		break;
8277	}
8278	default:
8279		panic("Invalid PR type %x", cdb->action);
8280	}
8281
8282done:
8283	free(ctsio->kern_data_ptr, M_CTL);
8284	ctl_set_success(ctsio);
8285	ctl_done((union ctl_io *)ctsio);
8286
8287	return (retval);
8288}
8289
8290/*
8291 * This routine is for handling a message from the other SC pertaining to
8292 * persistent reserve out. All the error checking will have been done
8293 * so only perorming the action need be done here to keep the two
8294 * in sync.
8295 */
8296static void
8297ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8298{
8299	struct ctl_lun *lun;
8300	struct ctl_softc *softc;
8301	int i;
8302	uint32_t targ_lun;
8303
8304	softc = control_softc;
8305
8306	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8307	lun = softc->ctl_luns[targ_lun];
8308	mtx_lock(&lun->lun_lock);
8309	switch(msg->pr.pr_info.action) {
8310	case CTL_PR_REG_KEY:
8311		ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8312		if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8313			lun->pr_key_count++;
8314		ctl_set_prkey(lun, msg->pr.pr_info.residx,
8315		    scsi_8btou64(msg->pr.pr_info.sa_res_key));
8316		lun->PRGeneration++;
8317		break;
8318
8319	case CTL_PR_UNREG_KEY:
8320		ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8321		lun->pr_key_count--;
8322
8323		/* XXX Need to see if the reservation has been released */
8324		/* if so do we need to generate UA? */
8325		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8326			lun->flags &= ~CTL_LUN_PR_RESERVED;
8327			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8328
8329			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8330			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8331			 && lun->pr_key_count) {
8332				/*
8333				 * If the reservation is a registrants
8334				 * only type we need to generate a UA
8335				 * for other registered inits.  The
8336				 * sense code should be RESERVATIONS
8337				 * RELEASED
8338				 */
8339
8340				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8341					if (ctl_get_prkey(lun, i +
8342					    softc->persis_offset) == 0)
8343						continue;
8344
8345					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8346				}
8347			}
8348			lun->res_type = 0;
8349		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8350			if (lun->pr_key_count==0) {
8351				lun->flags &= ~CTL_LUN_PR_RESERVED;
8352				lun->res_type = 0;
8353				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8354			}
8355		}
8356		lun->PRGeneration++;
8357		break;
8358
8359	case CTL_PR_RESERVE:
8360		lun->flags |= CTL_LUN_PR_RESERVED;
8361		lun->res_type = msg->pr.pr_info.res_type;
8362		lun->pr_res_idx = msg->pr.pr_info.residx;
8363
8364		break;
8365
8366	case CTL_PR_RELEASE:
8367		/*
8368		 * if this isn't an exclusive access res generate UA for all
8369		 * other registrants.
8370		 */
8371		if (lun->res_type != SPR_TYPE_EX_AC
8372		 && lun->res_type != SPR_TYPE_WR_EX) {
8373			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8374				if (ctl_get_prkey(lun, i + softc->persis_offset) != 0)
8375					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8376		}
8377
8378		lun->flags &= ~CTL_LUN_PR_RESERVED;
8379		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8380		lun->res_type = 0;
8381		break;
8382
8383	case CTL_PR_PREEMPT:
8384		ctl_pro_preempt_other(lun, msg);
8385		break;
8386	case CTL_PR_CLEAR:
8387		lun->flags &= ~CTL_LUN_PR_RESERVED;
8388		lun->res_type = 0;
8389		lun->pr_key_count = 0;
8390		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8391
8392		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8393			if (ctl_get_prkey(lun, i) == 0)
8394				continue;
8395			ctl_clr_prkey(lun, i);
8396			ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8397		}
8398		lun->PRGeneration++;
8399		break;
8400	}
8401
8402	mtx_unlock(&lun->lun_lock);
8403}
8404
8405int
8406ctl_read_write(struct ctl_scsiio *ctsio)
8407{
8408	struct ctl_lun *lun;
8409	struct ctl_lba_len_flags *lbalen;
8410	uint64_t lba;
8411	uint32_t num_blocks;
8412	int flags, retval;
8413	int isread;
8414
8415	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8416
8417	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8418
8419	flags = 0;
8420	retval = CTL_RETVAL_COMPLETE;
8421
8422	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8423	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8424	switch (ctsio->cdb[0]) {
8425	case READ_6:
8426	case WRITE_6: {
8427		struct scsi_rw_6 *cdb;
8428
8429		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8430
8431		lba = scsi_3btoul(cdb->addr);
8432		/* only 5 bits are valid in the most significant address byte */
8433		lba &= 0x1fffff;
8434		num_blocks = cdb->length;
8435		/*
8436		 * This is correct according to SBC-2.
8437		 */
8438		if (num_blocks == 0)
8439			num_blocks = 256;
8440		break;
8441	}
8442	case READ_10:
8443	case WRITE_10: {
8444		struct scsi_rw_10 *cdb;
8445
8446		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8447		if (cdb->byte2 & SRW10_FUA)
8448			flags |= CTL_LLF_FUA;
8449		if (cdb->byte2 & SRW10_DPO)
8450			flags |= CTL_LLF_DPO;
8451		lba = scsi_4btoul(cdb->addr);
8452		num_blocks = scsi_2btoul(cdb->length);
8453		break;
8454	}
8455	case WRITE_VERIFY_10: {
8456		struct scsi_write_verify_10 *cdb;
8457
8458		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8459		flags |= CTL_LLF_FUA;
8460		if (cdb->byte2 & SWV_DPO)
8461			flags |= CTL_LLF_DPO;
8462		lba = scsi_4btoul(cdb->addr);
8463		num_blocks = scsi_2btoul(cdb->length);
8464		break;
8465	}
8466	case READ_12:
8467	case WRITE_12: {
8468		struct scsi_rw_12 *cdb;
8469
8470		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8471		if (cdb->byte2 & SRW12_FUA)
8472			flags |= CTL_LLF_FUA;
8473		if (cdb->byte2 & SRW12_DPO)
8474			flags |= CTL_LLF_DPO;
8475		lba = scsi_4btoul(cdb->addr);
8476		num_blocks = scsi_4btoul(cdb->length);
8477		break;
8478	}
8479	case WRITE_VERIFY_12: {
8480		struct scsi_write_verify_12 *cdb;
8481
8482		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8483		flags |= CTL_LLF_FUA;
8484		if (cdb->byte2 & SWV_DPO)
8485			flags |= CTL_LLF_DPO;
8486		lba = scsi_4btoul(cdb->addr);
8487		num_blocks = scsi_4btoul(cdb->length);
8488		break;
8489	}
8490	case READ_16:
8491	case WRITE_16: {
8492		struct scsi_rw_16 *cdb;
8493
8494		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8495		if (cdb->byte2 & SRW12_FUA)
8496			flags |= CTL_LLF_FUA;
8497		if (cdb->byte2 & SRW12_DPO)
8498			flags |= CTL_LLF_DPO;
8499		lba = scsi_8btou64(cdb->addr);
8500		num_blocks = scsi_4btoul(cdb->length);
8501		break;
8502	}
8503	case WRITE_ATOMIC_16: {
8504		struct scsi_rw_16 *cdb;
8505
8506		if (lun->be_lun->atomicblock == 0) {
8507			ctl_set_invalid_opcode(ctsio);
8508			ctl_done((union ctl_io *)ctsio);
8509			return (CTL_RETVAL_COMPLETE);
8510		}
8511
8512		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8513		if (cdb->byte2 & SRW12_FUA)
8514			flags |= CTL_LLF_FUA;
8515		if (cdb->byte2 & SRW12_DPO)
8516			flags |= CTL_LLF_DPO;
8517		lba = scsi_8btou64(cdb->addr);
8518		num_blocks = scsi_4btoul(cdb->length);
8519		if (num_blocks > lun->be_lun->atomicblock) {
8520			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8521			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8522			    /*bit*/ 0);
8523			ctl_done((union ctl_io *)ctsio);
8524			return (CTL_RETVAL_COMPLETE);
8525		}
8526		break;
8527	}
8528	case WRITE_VERIFY_16: {
8529		struct scsi_write_verify_16 *cdb;
8530
8531		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8532		flags |= CTL_LLF_FUA;
8533		if (cdb->byte2 & SWV_DPO)
8534			flags |= CTL_LLF_DPO;
8535		lba = scsi_8btou64(cdb->addr);
8536		num_blocks = scsi_4btoul(cdb->length);
8537		break;
8538	}
8539	default:
8540		/*
8541		 * We got a command we don't support.  This shouldn't
8542		 * happen, commands should be filtered out above us.
8543		 */
8544		ctl_set_invalid_opcode(ctsio);
8545		ctl_done((union ctl_io *)ctsio);
8546
8547		return (CTL_RETVAL_COMPLETE);
8548		break; /* NOTREACHED */
8549	}
8550
8551	/*
8552	 * The first check is to make sure we're in bounds, the second
8553	 * check is to catch wrap-around problems.  If the lba + num blocks
8554	 * is less than the lba, then we've wrapped around and the block
8555	 * range is invalid anyway.
8556	 */
8557	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8558	 || ((lba + num_blocks) < lba)) {
8559		ctl_set_lba_out_of_range(ctsio);
8560		ctl_done((union ctl_io *)ctsio);
8561		return (CTL_RETVAL_COMPLETE);
8562	}
8563
8564	/*
8565	 * According to SBC-3, a transfer length of 0 is not an error.
8566	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8567	 * translates to 256 blocks for those commands.
8568	 */
8569	if (num_blocks == 0) {
8570		ctl_set_success(ctsio);
8571		ctl_done((union ctl_io *)ctsio);
8572		return (CTL_RETVAL_COMPLETE);
8573	}
8574
8575	/* Set FUA and/or DPO if caches are disabled. */
8576	if (isread) {
8577		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8578		    SCP_RCD) != 0)
8579			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8580	} else {
8581		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8582		    SCP_WCE) == 0)
8583			flags |= CTL_LLF_FUA;
8584	}
8585
8586	lbalen = (struct ctl_lba_len_flags *)
8587	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8588	lbalen->lba = lba;
8589	lbalen->len = num_blocks;
8590	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8591
8592	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8593	ctsio->kern_rel_offset = 0;
8594
8595	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8596
8597	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8598
8599	return (retval);
8600}
8601
8602static int
8603ctl_cnw_cont(union ctl_io *io)
8604{
8605	struct ctl_scsiio *ctsio;
8606	struct ctl_lun *lun;
8607	struct ctl_lba_len_flags *lbalen;
8608	int retval;
8609
8610	ctsio = &io->scsiio;
8611	ctsio->io_hdr.status = CTL_STATUS_NONE;
8612	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8613	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8614	lbalen = (struct ctl_lba_len_flags *)
8615	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8616	lbalen->flags &= ~CTL_LLF_COMPARE;
8617	lbalen->flags |= CTL_LLF_WRITE;
8618
8619	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8620	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8621	return (retval);
8622}
8623
8624int
8625ctl_cnw(struct ctl_scsiio *ctsio)
8626{
8627	struct ctl_lun *lun;
8628	struct ctl_lba_len_flags *lbalen;
8629	uint64_t lba;
8630	uint32_t num_blocks;
8631	int flags, retval;
8632
8633	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8634
8635	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8636
8637	flags = 0;
8638	retval = CTL_RETVAL_COMPLETE;
8639
8640	switch (ctsio->cdb[0]) {
8641	case COMPARE_AND_WRITE: {
8642		struct scsi_compare_and_write *cdb;
8643
8644		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8645		if (cdb->byte2 & SRW10_FUA)
8646			flags |= CTL_LLF_FUA;
8647		if (cdb->byte2 & SRW10_DPO)
8648			flags |= CTL_LLF_DPO;
8649		lba = scsi_8btou64(cdb->addr);
8650		num_blocks = cdb->length;
8651		break;
8652	}
8653	default:
8654		/*
8655		 * We got a command we don't support.  This shouldn't
8656		 * happen, commands should be filtered out above us.
8657		 */
8658		ctl_set_invalid_opcode(ctsio);
8659		ctl_done((union ctl_io *)ctsio);
8660
8661		return (CTL_RETVAL_COMPLETE);
8662		break; /* NOTREACHED */
8663	}
8664
8665	/*
8666	 * The first check is to make sure we're in bounds, the second
8667	 * check is to catch wrap-around problems.  If the lba + num blocks
8668	 * is less than the lba, then we've wrapped around and the block
8669	 * range is invalid anyway.
8670	 */
8671	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8672	 || ((lba + num_blocks) < lba)) {
8673		ctl_set_lba_out_of_range(ctsio);
8674		ctl_done((union ctl_io *)ctsio);
8675		return (CTL_RETVAL_COMPLETE);
8676	}
8677
8678	/*
8679	 * According to SBC-3, a transfer length of 0 is not an error.
8680	 */
8681	if (num_blocks == 0) {
8682		ctl_set_success(ctsio);
8683		ctl_done((union ctl_io *)ctsio);
8684		return (CTL_RETVAL_COMPLETE);
8685	}
8686
8687	/* Set FUA if write cache is disabled. */
8688	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8689	    SCP_WCE) == 0)
8690		flags |= CTL_LLF_FUA;
8691
8692	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8693	ctsio->kern_rel_offset = 0;
8694
8695	/*
8696	 * Set the IO_CONT flag, so that if this I/O gets passed to
8697	 * ctl_data_submit_done(), it'll get passed back to
8698	 * ctl_ctl_cnw_cont() for further processing.
8699	 */
8700	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8701	ctsio->io_cont = ctl_cnw_cont;
8702
8703	lbalen = (struct ctl_lba_len_flags *)
8704	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8705	lbalen->lba = lba;
8706	lbalen->len = num_blocks;
8707	lbalen->flags = CTL_LLF_COMPARE | flags;
8708
8709	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8710	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8711	return (retval);
8712}
8713
8714int
8715ctl_verify(struct ctl_scsiio *ctsio)
8716{
8717	struct ctl_lun *lun;
8718	struct ctl_lba_len_flags *lbalen;
8719	uint64_t lba;
8720	uint32_t num_blocks;
8721	int bytchk, flags;
8722	int retval;
8723
8724	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8725
8726	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8727
8728	bytchk = 0;
8729	flags = CTL_LLF_FUA;
8730	retval = CTL_RETVAL_COMPLETE;
8731
8732	switch (ctsio->cdb[0]) {
8733	case VERIFY_10: {
8734		struct scsi_verify_10 *cdb;
8735
8736		cdb = (struct scsi_verify_10 *)ctsio->cdb;
8737		if (cdb->byte2 & SVFY_BYTCHK)
8738			bytchk = 1;
8739		if (cdb->byte2 & SVFY_DPO)
8740			flags |= CTL_LLF_DPO;
8741		lba = scsi_4btoul(cdb->addr);
8742		num_blocks = scsi_2btoul(cdb->length);
8743		break;
8744	}
8745	case VERIFY_12: {
8746		struct scsi_verify_12 *cdb;
8747
8748		cdb = (struct scsi_verify_12 *)ctsio->cdb;
8749		if (cdb->byte2 & SVFY_BYTCHK)
8750			bytchk = 1;
8751		if (cdb->byte2 & SVFY_DPO)
8752			flags |= CTL_LLF_DPO;
8753		lba = scsi_4btoul(cdb->addr);
8754		num_blocks = scsi_4btoul(cdb->length);
8755		break;
8756	}
8757	case VERIFY_16: {
8758		struct scsi_rw_16 *cdb;
8759
8760		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8761		if (cdb->byte2 & SVFY_BYTCHK)
8762			bytchk = 1;
8763		if (cdb->byte2 & SVFY_DPO)
8764			flags |= CTL_LLF_DPO;
8765		lba = scsi_8btou64(cdb->addr);
8766		num_blocks = scsi_4btoul(cdb->length);
8767		break;
8768	}
8769	default:
8770		/*
8771		 * We got a command we don't support.  This shouldn't
8772		 * happen, commands should be filtered out above us.
8773		 */
8774		ctl_set_invalid_opcode(ctsio);
8775		ctl_done((union ctl_io *)ctsio);
8776		return (CTL_RETVAL_COMPLETE);
8777	}
8778
8779	/*
8780	 * The first check is to make sure we're in bounds, the second
8781	 * check is to catch wrap-around problems.  If the lba + num blocks
8782	 * is less than the lba, then we've wrapped around and the block
8783	 * range is invalid anyway.
8784	 */
8785	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8786	 || ((lba + num_blocks) < lba)) {
8787		ctl_set_lba_out_of_range(ctsio);
8788		ctl_done((union ctl_io *)ctsio);
8789		return (CTL_RETVAL_COMPLETE);
8790	}
8791
8792	/*
8793	 * According to SBC-3, a transfer length of 0 is not an error.
8794	 */
8795	if (num_blocks == 0) {
8796		ctl_set_success(ctsio);
8797		ctl_done((union ctl_io *)ctsio);
8798		return (CTL_RETVAL_COMPLETE);
8799	}
8800
8801	lbalen = (struct ctl_lba_len_flags *)
8802	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8803	lbalen->lba = lba;
8804	lbalen->len = num_blocks;
8805	if (bytchk) {
8806		lbalen->flags = CTL_LLF_COMPARE | flags;
8807		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8808	} else {
8809		lbalen->flags = CTL_LLF_VERIFY | flags;
8810		ctsio->kern_total_len = 0;
8811	}
8812	ctsio->kern_rel_offset = 0;
8813
8814	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
8815	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8816	return (retval);
8817}
8818
8819int
8820ctl_report_luns(struct ctl_scsiio *ctsio)
8821{
8822	struct ctl_softc *softc = control_softc;
8823	struct scsi_report_luns *cdb;
8824	struct scsi_report_luns_data *lun_data;
8825	struct ctl_lun *lun, *request_lun;
8826	struct ctl_port *port;
8827	int num_luns, retval;
8828	uint32_t alloc_len, lun_datalen;
8829	int num_filled, well_known;
8830	uint32_t initidx, targ_lun_id, lun_id;
8831
8832	retval = CTL_RETVAL_COMPLETE;
8833	well_known = 0;
8834
8835	cdb = (struct scsi_report_luns *)ctsio->cdb;
8836	port = ctl_io_port(&ctsio->io_hdr);
8837
8838	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
8839
8840	mtx_lock(&softc->ctl_lock);
8841	num_luns = 0;
8842	for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
8843		if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
8844			num_luns++;
8845	}
8846	mtx_unlock(&softc->ctl_lock);
8847
8848	switch (cdb->select_report) {
8849	case RPL_REPORT_DEFAULT:
8850	case RPL_REPORT_ALL:
8851		break;
8852	case RPL_REPORT_WELLKNOWN:
8853		well_known = 1;
8854		num_luns = 0;
8855		break;
8856	default:
8857		ctl_set_invalid_field(ctsio,
8858				      /*sks_valid*/ 1,
8859				      /*command*/ 1,
8860				      /*field*/ 2,
8861				      /*bit_valid*/ 0,
8862				      /*bit*/ 0);
8863		ctl_done((union ctl_io *)ctsio);
8864		return (retval);
8865		break; /* NOTREACHED */
8866	}
8867
8868	alloc_len = scsi_4btoul(cdb->length);
8869	/*
8870	 * The initiator has to allocate at least 16 bytes for this request,
8871	 * so he can at least get the header and the first LUN.  Otherwise
8872	 * we reject the request (per SPC-3 rev 14, section 6.21).
8873	 */
8874	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
8875	    sizeof(struct scsi_report_luns_lundata))) {
8876		ctl_set_invalid_field(ctsio,
8877				      /*sks_valid*/ 1,
8878				      /*command*/ 1,
8879				      /*field*/ 6,
8880				      /*bit_valid*/ 0,
8881				      /*bit*/ 0);
8882		ctl_done((union ctl_io *)ctsio);
8883		return (retval);
8884	}
8885
8886	request_lun = (struct ctl_lun *)
8887		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8888
8889	lun_datalen = sizeof(*lun_data) +
8890		(num_luns * sizeof(struct scsi_report_luns_lundata));
8891
8892	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
8893	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
8894	ctsio->kern_sg_entries = 0;
8895
8896	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8897
8898	mtx_lock(&softc->ctl_lock);
8899	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
8900		lun_id = ctl_lun_map_from_port(port, targ_lun_id);
8901		if (lun_id >= CTL_MAX_LUNS)
8902			continue;
8903		lun = softc->ctl_luns[lun_id];
8904		if (lun == NULL)
8905			continue;
8906
8907		if (targ_lun_id <= 0xff) {
8908			/*
8909			 * Peripheral addressing method, bus number 0.
8910			 */
8911			lun_data->luns[num_filled].lundata[0] =
8912				RPL_LUNDATA_ATYP_PERIPH;
8913			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
8914			num_filled++;
8915		} else if (targ_lun_id <= 0x3fff) {
8916			/*
8917			 * Flat addressing method.
8918			 */
8919			lun_data->luns[num_filled].lundata[0] =
8920				RPL_LUNDATA_ATYP_FLAT | (targ_lun_id >> 8);
8921			lun_data->luns[num_filled].lundata[1] =
8922				(targ_lun_id & 0xff);
8923			num_filled++;
8924		} else if (targ_lun_id <= 0xffffff) {
8925			/*
8926			 * Extended flat addressing method.
8927			 */
8928			lun_data->luns[num_filled].lundata[0] =
8929			    RPL_LUNDATA_ATYP_EXTLUN | 0x12;
8930			scsi_ulto3b(targ_lun_id,
8931			    &lun_data->luns[num_filled].lundata[1]);
8932			num_filled++;
8933		} else {
8934			printf("ctl_report_luns: bogus LUN number %jd, "
8935			       "skipping\n", (intmax_t)targ_lun_id);
8936		}
8937		/*
8938		 * According to SPC-3, rev 14 section 6.21:
8939		 *
8940		 * "The execution of a REPORT LUNS command to any valid and
8941		 * installed logical unit shall clear the REPORTED LUNS DATA
8942		 * HAS CHANGED unit attention condition for all logical
8943		 * units of that target with respect to the requesting
8944		 * initiator. A valid and installed logical unit is one
8945		 * having a PERIPHERAL QUALIFIER of 000b in the standard
8946		 * INQUIRY data (see 6.4.2)."
8947		 *
8948		 * If request_lun is NULL, the LUN this report luns command
8949		 * was issued to is either disabled or doesn't exist. In that
8950		 * case, we shouldn't clear any pending lun change unit
8951		 * attention.
8952		 */
8953		if (request_lun != NULL) {
8954			mtx_lock(&lun->lun_lock);
8955			ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
8956			mtx_unlock(&lun->lun_lock);
8957		}
8958	}
8959	mtx_unlock(&softc->ctl_lock);
8960
8961	/*
8962	 * It's quite possible that we've returned fewer LUNs than we allocated
8963	 * space for.  Trim it.
8964	 */
8965	lun_datalen = sizeof(*lun_data) +
8966		(num_filled * sizeof(struct scsi_report_luns_lundata));
8967
8968	if (lun_datalen < alloc_len) {
8969		ctsio->residual = alloc_len - lun_datalen;
8970		ctsio->kern_data_len = lun_datalen;
8971		ctsio->kern_total_len = lun_datalen;
8972	} else {
8973		ctsio->residual = 0;
8974		ctsio->kern_data_len = alloc_len;
8975		ctsio->kern_total_len = alloc_len;
8976	}
8977	ctsio->kern_data_resid = 0;
8978	ctsio->kern_rel_offset = 0;
8979	ctsio->kern_sg_entries = 0;
8980
8981	/*
8982	 * We set this to the actual data length, regardless of how much
8983	 * space we actually have to return results.  If the user looks at
8984	 * this value, he'll know whether or not he allocated enough space
8985	 * and reissue the command if necessary.  We don't support well
8986	 * known logical units, so if the user asks for that, return none.
8987	 */
8988	scsi_ulto4b(lun_datalen - 8, lun_data->length);
8989
8990	/*
8991	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
8992	 * this request.
8993	 */
8994	ctl_set_success(ctsio);
8995	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8996	ctsio->be_move_done = ctl_config_move_done;
8997	ctl_datamove((union ctl_io *)ctsio);
8998	return (retval);
8999}
9000
9001int
9002ctl_request_sense(struct ctl_scsiio *ctsio)
9003{
9004	struct scsi_request_sense *cdb;
9005	struct scsi_sense_data *sense_ptr;
9006	struct ctl_softc *ctl_softc;
9007	struct ctl_lun *lun;
9008	uint32_t initidx;
9009	int have_error;
9010	scsi_sense_data_type sense_format;
9011	ctl_ua_type ua_type;
9012
9013	cdb = (struct scsi_request_sense *)ctsio->cdb;
9014
9015	ctl_softc = control_softc;
9016	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9017
9018	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9019
9020	/*
9021	 * Determine which sense format the user wants.
9022	 */
9023	if (cdb->byte2 & SRS_DESC)
9024		sense_format = SSD_TYPE_DESC;
9025	else
9026		sense_format = SSD_TYPE_FIXED;
9027
9028	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9029	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9030	ctsio->kern_sg_entries = 0;
9031
9032	/*
9033	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9034	 * larger than the largest allowed value for the length field in the
9035	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9036	 */
9037	ctsio->residual = 0;
9038	ctsio->kern_data_len = cdb->length;
9039	ctsio->kern_total_len = cdb->length;
9040
9041	ctsio->kern_data_resid = 0;
9042	ctsio->kern_rel_offset = 0;
9043	ctsio->kern_sg_entries = 0;
9044
9045	/*
9046	 * If we don't have a LUN, we don't have any pending sense.
9047	 */
9048	if (lun == NULL)
9049		goto no_sense;
9050
9051	have_error = 0;
9052	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9053	/*
9054	 * Check for pending sense, and then for pending unit attentions.
9055	 * Pending sense gets returned first, then pending unit attentions.
9056	 */
9057	mtx_lock(&lun->lun_lock);
9058#ifdef CTL_WITH_CA
9059	if (ctl_is_set(lun->have_ca, initidx)) {
9060		scsi_sense_data_type stored_format;
9061
9062		/*
9063		 * Check to see which sense format was used for the stored
9064		 * sense data.
9065		 */
9066		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9067
9068		/*
9069		 * If the user requested a different sense format than the
9070		 * one we stored, then we need to convert it to the other
9071		 * format.  If we're going from descriptor to fixed format
9072		 * sense data, we may lose things in translation, depending
9073		 * on what options were used.
9074		 *
9075		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9076		 * for some reason we'll just copy it out as-is.
9077		 */
9078		if ((stored_format == SSD_TYPE_FIXED)
9079		 && (sense_format == SSD_TYPE_DESC))
9080			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9081			    &lun->pending_sense[initidx],
9082			    (struct scsi_sense_data_desc *)sense_ptr);
9083		else if ((stored_format == SSD_TYPE_DESC)
9084		      && (sense_format == SSD_TYPE_FIXED))
9085			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9086			    &lun->pending_sense[initidx],
9087			    (struct scsi_sense_data_fixed *)sense_ptr);
9088		else
9089			memcpy(sense_ptr, &lun->pending_sense[initidx],
9090			       MIN(sizeof(*sense_ptr),
9091			       sizeof(lun->pending_sense[initidx])));
9092
9093		ctl_clear_mask(lun->have_ca, initidx);
9094		have_error = 1;
9095	} else
9096#endif
9097	{
9098		ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9099		if (ua_type != CTL_UA_NONE)
9100			have_error = 1;
9101		if (ua_type == CTL_UA_LUN_CHANGE) {
9102			mtx_unlock(&lun->lun_lock);
9103			mtx_lock(&ctl_softc->ctl_lock);
9104			ctl_clr_ua_allluns(ctl_softc, initidx, ua_type);
9105			mtx_unlock(&ctl_softc->ctl_lock);
9106			mtx_lock(&lun->lun_lock);
9107		}
9108
9109	}
9110	mtx_unlock(&lun->lun_lock);
9111
9112	/*
9113	 * We already have a pending error, return it.
9114	 */
9115	if (have_error != 0) {
9116		/*
9117		 * We report the SCSI status as OK, since the status of the
9118		 * request sense command itself is OK.
9119		 * We report 0 for the sense length, because we aren't doing
9120		 * autosense in this case.  We're reporting sense as
9121		 * parameter data.
9122		 */
9123		ctl_set_success(ctsio);
9124		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9125		ctsio->be_move_done = ctl_config_move_done;
9126		ctl_datamove((union ctl_io *)ctsio);
9127		return (CTL_RETVAL_COMPLETE);
9128	}
9129
9130no_sense:
9131
9132	/*
9133	 * No sense information to report, so we report that everything is
9134	 * okay.
9135	 */
9136	ctl_set_sense_data(sense_ptr,
9137			   lun,
9138			   sense_format,
9139			   /*current_error*/ 1,
9140			   /*sense_key*/ SSD_KEY_NO_SENSE,
9141			   /*asc*/ 0x00,
9142			   /*ascq*/ 0x00,
9143			   SSD_ELEM_NONE);
9144
9145	/*
9146	 * We report 0 for the sense length, because we aren't doing
9147	 * autosense in this case.  We're reporting sense as parameter data.
9148	 */
9149	ctl_set_success(ctsio);
9150	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9151	ctsio->be_move_done = ctl_config_move_done;
9152	ctl_datamove((union ctl_io *)ctsio);
9153	return (CTL_RETVAL_COMPLETE);
9154}
9155
9156int
9157ctl_tur(struct ctl_scsiio *ctsio)
9158{
9159
9160	CTL_DEBUG_PRINT(("ctl_tur\n"));
9161
9162	ctl_set_success(ctsio);
9163	ctl_done((union ctl_io *)ctsio);
9164
9165	return (CTL_RETVAL_COMPLETE);
9166}
9167
9168#ifdef notyet
9169static int
9170ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9171{
9172
9173}
9174#endif
9175
9176/*
9177 * SCSI VPD page 0x00, the Supported VPD Pages page.
9178 */
9179static int
9180ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9181{
9182	struct scsi_vpd_supported_pages *pages;
9183	int sup_page_size;
9184	struct ctl_lun *lun;
9185	int p;
9186
9187	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9188
9189	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9190	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9191	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9192	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9193	ctsio->kern_sg_entries = 0;
9194
9195	if (sup_page_size < alloc_len) {
9196		ctsio->residual = alloc_len - sup_page_size;
9197		ctsio->kern_data_len = sup_page_size;
9198		ctsio->kern_total_len = sup_page_size;
9199	} else {
9200		ctsio->residual = 0;
9201		ctsio->kern_data_len = alloc_len;
9202		ctsio->kern_total_len = alloc_len;
9203	}
9204	ctsio->kern_data_resid = 0;
9205	ctsio->kern_rel_offset = 0;
9206	ctsio->kern_sg_entries = 0;
9207
9208	/*
9209	 * The control device is always connected.  The disk device, on the
9210	 * other hand, may not be online all the time.  Need to change this
9211	 * to figure out whether the disk device is actually online or not.
9212	 */
9213	if (lun != NULL)
9214		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9215				lun->be_lun->lun_type;
9216	else
9217		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9218
9219	p = 0;
9220	/* Supported VPD pages */
9221	pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9222	/* Serial Number */
9223	pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9224	/* Device Identification */
9225	pages->page_list[p++] = SVPD_DEVICE_ID;
9226	/* Extended INQUIRY Data */
9227	pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9228	/* Mode Page Policy */
9229	pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9230	/* SCSI Ports */
9231	pages->page_list[p++] = SVPD_SCSI_PORTS;
9232	/* Third-party Copy */
9233	pages->page_list[p++] = SVPD_SCSI_TPC;
9234	if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9235		/* Block limits */
9236		pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9237		/* Block Device Characteristics */
9238		pages->page_list[p++] = SVPD_BDC;
9239		/* Logical Block Provisioning */
9240		pages->page_list[p++] = SVPD_LBP;
9241	}
9242	pages->length = p;
9243
9244	ctl_set_success(ctsio);
9245	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9246	ctsio->be_move_done = ctl_config_move_done;
9247	ctl_datamove((union ctl_io *)ctsio);
9248	return (CTL_RETVAL_COMPLETE);
9249}
9250
9251/*
9252 * SCSI VPD page 0x80, the Unit Serial Number page.
9253 */
9254static int
9255ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9256{
9257	struct scsi_vpd_unit_serial_number *sn_ptr;
9258	struct ctl_lun *lun;
9259	int data_len;
9260
9261	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9262
9263	data_len = 4 + CTL_SN_LEN;
9264	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9265	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9266	if (data_len < alloc_len) {
9267		ctsio->residual = alloc_len - data_len;
9268		ctsio->kern_data_len = data_len;
9269		ctsio->kern_total_len = data_len;
9270	} else {
9271		ctsio->residual = 0;
9272		ctsio->kern_data_len = alloc_len;
9273		ctsio->kern_total_len = alloc_len;
9274	}
9275	ctsio->kern_data_resid = 0;
9276	ctsio->kern_rel_offset = 0;
9277	ctsio->kern_sg_entries = 0;
9278
9279	/*
9280	 * The control device is always connected.  The disk device, on the
9281	 * other hand, may not be online all the time.  Need to change this
9282	 * to figure out whether the disk device is actually online or not.
9283	 */
9284	if (lun != NULL)
9285		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9286				  lun->be_lun->lun_type;
9287	else
9288		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9289
9290	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9291	sn_ptr->length = CTL_SN_LEN;
9292	/*
9293	 * If we don't have a LUN, we just leave the serial number as
9294	 * all spaces.
9295	 */
9296	if (lun != NULL) {
9297		strncpy((char *)sn_ptr->serial_num,
9298			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9299	} else
9300		memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9301
9302	ctl_set_success(ctsio);
9303	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9304	ctsio->be_move_done = ctl_config_move_done;
9305	ctl_datamove((union ctl_io *)ctsio);
9306	return (CTL_RETVAL_COMPLETE);
9307}
9308
9309
9310/*
9311 * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9312 */
9313static int
9314ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9315{
9316	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9317	struct ctl_lun *lun;
9318	int data_len;
9319
9320	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9321
9322	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9323	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9324	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9325	ctsio->kern_sg_entries = 0;
9326
9327	if (data_len < alloc_len) {
9328		ctsio->residual = alloc_len - data_len;
9329		ctsio->kern_data_len = data_len;
9330		ctsio->kern_total_len = data_len;
9331	} else {
9332		ctsio->residual = 0;
9333		ctsio->kern_data_len = alloc_len;
9334		ctsio->kern_total_len = alloc_len;
9335	}
9336	ctsio->kern_data_resid = 0;
9337	ctsio->kern_rel_offset = 0;
9338	ctsio->kern_sg_entries = 0;
9339
9340	/*
9341	 * The control device is always connected.  The disk device, on the
9342	 * other hand, may not be online all the time.
9343	 */
9344	if (lun != NULL)
9345		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9346				     lun->be_lun->lun_type;
9347	else
9348		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9349	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9350	scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9351	/*
9352	 * We support head of queue, ordered and simple tags.
9353	 */
9354	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9355	/*
9356	 * Volatile cache supported.
9357	 */
9358	eid_ptr->flags3 = SVPD_EID_V_SUP;
9359
9360	/*
9361	 * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9362	 * attention for a particular IT nexus on all LUNs once we report
9363	 * it to that nexus once.  This bit is required as of SPC-4.
9364	 */
9365	eid_ptr->flags4 = SVPD_EID_LUICLT;
9366
9367	/*
9368	 * XXX KDM in order to correctly answer this, we would need
9369	 * information from the SIM to determine how much sense data it
9370	 * can send.  So this would really be a path inquiry field, most
9371	 * likely.  This can be set to a maximum of 252 according to SPC-4,
9372	 * but the hardware may or may not be able to support that much.
9373	 * 0 just means that the maximum sense data length is not reported.
9374	 */
9375	eid_ptr->max_sense_length = 0;
9376
9377	ctl_set_success(ctsio);
9378	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9379	ctsio->be_move_done = ctl_config_move_done;
9380	ctl_datamove((union ctl_io *)ctsio);
9381	return (CTL_RETVAL_COMPLETE);
9382}
9383
9384static int
9385ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9386{
9387	struct scsi_vpd_mode_page_policy *mpp_ptr;
9388	struct ctl_lun *lun;
9389	int data_len;
9390
9391	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9392
9393	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9394	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9395
9396	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9397	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9398	ctsio->kern_sg_entries = 0;
9399
9400	if (data_len < alloc_len) {
9401		ctsio->residual = alloc_len - data_len;
9402		ctsio->kern_data_len = data_len;
9403		ctsio->kern_total_len = data_len;
9404	} else {
9405		ctsio->residual = 0;
9406		ctsio->kern_data_len = alloc_len;
9407		ctsio->kern_total_len = alloc_len;
9408	}
9409	ctsio->kern_data_resid = 0;
9410	ctsio->kern_rel_offset = 0;
9411	ctsio->kern_sg_entries = 0;
9412
9413	/*
9414	 * The control device is always connected.  The disk device, on the
9415	 * other hand, may not be online all the time.
9416	 */
9417	if (lun != NULL)
9418		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9419				     lun->be_lun->lun_type;
9420	else
9421		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9422	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9423	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9424	mpp_ptr->descr[0].page_code = 0x3f;
9425	mpp_ptr->descr[0].subpage_code = 0xff;
9426	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9427
9428	ctl_set_success(ctsio);
9429	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9430	ctsio->be_move_done = ctl_config_move_done;
9431	ctl_datamove((union ctl_io *)ctsio);
9432	return (CTL_RETVAL_COMPLETE);
9433}
9434
9435/*
9436 * SCSI VPD page 0x83, the Device Identification page.
9437 */
9438static int
9439ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9440{
9441	struct scsi_vpd_device_id *devid_ptr;
9442	struct scsi_vpd_id_descriptor *desc;
9443	struct ctl_softc *softc;
9444	struct ctl_lun *lun;
9445	struct ctl_port *port;
9446	int data_len;
9447	uint8_t proto;
9448
9449	softc = control_softc;
9450
9451	port = softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9452	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9453
9454	data_len = sizeof(struct scsi_vpd_device_id) +
9455	    sizeof(struct scsi_vpd_id_descriptor) +
9456		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9457	    sizeof(struct scsi_vpd_id_descriptor) +
9458		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9459	if (lun && lun->lun_devid)
9460		data_len += lun->lun_devid->len;
9461	if (port->port_devid)
9462		data_len += port->port_devid->len;
9463	if (port->target_devid)
9464		data_len += port->target_devid->len;
9465
9466	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9467	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9468	ctsio->kern_sg_entries = 0;
9469
9470	if (data_len < alloc_len) {
9471		ctsio->residual = alloc_len - data_len;
9472		ctsio->kern_data_len = data_len;
9473		ctsio->kern_total_len = data_len;
9474	} else {
9475		ctsio->residual = 0;
9476		ctsio->kern_data_len = alloc_len;
9477		ctsio->kern_total_len = alloc_len;
9478	}
9479	ctsio->kern_data_resid = 0;
9480	ctsio->kern_rel_offset = 0;
9481	ctsio->kern_sg_entries = 0;
9482
9483	/*
9484	 * The control device is always connected.  The disk device, on the
9485	 * other hand, may not be online all the time.
9486	 */
9487	if (lun != NULL)
9488		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9489				     lun->be_lun->lun_type;
9490	else
9491		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9492	devid_ptr->page_code = SVPD_DEVICE_ID;
9493	scsi_ulto2b(data_len - 4, devid_ptr->length);
9494
9495	if (port->port_type == CTL_PORT_FC)
9496		proto = SCSI_PROTO_FC << 4;
9497	else if (port->port_type == CTL_PORT_ISCSI)
9498		proto = SCSI_PROTO_ISCSI << 4;
9499	else
9500		proto = SCSI_PROTO_SPI << 4;
9501	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9502
9503	/*
9504	 * We're using a LUN association here.  i.e., this device ID is a
9505	 * per-LUN identifier.
9506	 */
9507	if (lun && lun->lun_devid) {
9508		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9509		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9510		    lun->lun_devid->len);
9511	}
9512
9513	/*
9514	 * This is for the WWPN which is a port association.
9515	 */
9516	if (port->port_devid) {
9517		memcpy(desc, port->port_devid->data, port->port_devid->len);
9518		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9519		    port->port_devid->len);
9520	}
9521
9522	/*
9523	 * This is for the Relative Target Port(type 4h) identifier
9524	 */
9525	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9526	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9527	    SVPD_ID_TYPE_RELTARG;
9528	desc->length = 4;
9529	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9530	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9531	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9532
9533	/*
9534	 * This is for the Target Port Group(type 5h) identifier
9535	 */
9536	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9537	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9538	    SVPD_ID_TYPE_TPORTGRP;
9539	desc->length = 4;
9540	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9541	    &desc->identifier[2]);
9542	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9543	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9544
9545	/*
9546	 * This is for the Target identifier
9547	 */
9548	if (port->target_devid) {
9549		memcpy(desc, port->target_devid->data, port->target_devid->len);
9550	}
9551
9552	ctl_set_success(ctsio);
9553	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9554	ctsio->be_move_done = ctl_config_move_done;
9555	ctl_datamove((union ctl_io *)ctsio);
9556	return (CTL_RETVAL_COMPLETE);
9557}
9558
9559static int
9560ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9561{
9562	struct ctl_softc *softc = control_softc;
9563	struct scsi_vpd_scsi_ports *sp;
9564	struct scsi_vpd_port_designation *pd;
9565	struct scsi_vpd_port_designation_cont *pdc;
9566	struct ctl_lun *lun;
9567	struct ctl_port *port;
9568	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
9569	int num_target_port_groups;
9570
9571	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9572
9573	if (softc->is_single)
9574		num_target_port_groups = 1;
9575	else
9576		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
9577	num_target_ports = 0;
9578	iid_len = 0;
9579	id_len = 0;
9580	mtx_lock(&softc->ctl_lock);
9581	STAILQ_FOREACH(port, &softc->port_list, links) {
9582		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9583			continue;
9584		if (lun != NULL &&
9585		    ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9586			continue;
9587		num_target_ports++;
9588		if (port->init_devid)
9589			iid_len += port->init_devid->len;
9590		if (port->port_devid)
9591			id_len += port->port_devid->len;
9592	}
9593	mtx_unlock(&softc->ctl_lock);
9594
9595	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
9596	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9597	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9598	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9599	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9600	ctsio->kern_sg_entries = 0;
9601
9602	if (data_len < alloc_len) {
9603		ctsio->residual = alloc_len - data_len;
9604		ctsio->kern_data_len = data_len;
9605		ctsio->kern_total_len = data_len;
9606	} else {
9607		ctsio->residual = 0;
9608		ctsio->kern_data_len = alloc_len;
9609		ctsio->kern_total_len = alloc_len;
9610	}
9611	ctsio->kern_data_resid = 0;
9612	ctsio->kern_rel_offset = 0;
9613	ctsio->kern_sg_entries = 0;
9614
9615	/*
9616	 * The control device is always connected.  The disk device, on the
9617	 * other hand, may not be online all the time.  Need to change this
9618	 * to figure out whether the disk device is actually online or not.
9619	 */
9620	if (lun != NULL)
9621		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9622				  lun->be_lun->lun_type;
9623	else
9624		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9625
9626	sp->page_code = SVPD_SCSI_PORTS;
9627	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9628	    sp->page_length);
9629	pd = &sp->design[0];
9630
9631	mtx_lock(&softc->ctl_lock);
9632	pg = softc->port_offset / CTL_MAX_PORTS;
9633	for (g = 0; g < num_target_port_groups; g++) {
9634		STAILQ_FOREACH(port, &softc->port_list, links) {
9635			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9636				continue;
9637			if (lun != NULL &&
9638			    ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9639				continue;
9640			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
9641			scsi_ulto2b(p, pd->relative_port_id);
9642			if (port->init_devid && g == pg) {
9643				iid_len = port->init_devid->len;
9644				memcpy(pd->initiator_transportid,
9645				    port->init_devid->data, port->init_devid->len);
9646			} else
9647				iid_len = 0;
9648			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9649			pdc = (struct scsi_vpd_port_designation_cont *)
9650			    (&pd->initiator_transportid[iid_len]);
9651			if (port->port_devid && g == pg) {
9652				id_len = port->port_devid->len;
9653				memcpy(pdc->target_port_descriptors,
9654				    port->port_devid->data, port->port_devid->len);
9655			} else
9656				id_len = 0;
9657			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9658			pd = (struct scsi_vpd_port_designation *)
9659			    ((uint8_t *)pdc->target_port_descriptors + id_len);
9660		}
9661	}
9662	mtx_unlock(&softc->ctl_lock);
9663
9664	ctl_set_success(ctsio);
9665	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9666	ctsio->be_move_done = ctl_config_move_done;
9667	ctl_datamove((union ctl_io *)ctsio);
9668	return (CTL_RETVAL_COMPLETE);
9669}
9670
9671static int
9672ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9673{
9674	struct scsi_vpd_block_limits *bl_ptr;
9675	struct ctl_lun *lun;
9676	int bs;
9677
9678	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9679
9680	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9681	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9682	ctsio->kern_sg_entries = 0;
9683
9684	if (sizeof(*bl_ptr) < alloc_len) {
9685		ctsio->residual = alloc_len - sizeof(*bl_ptr);
9686		ctsio->kern_data_len = sizeof(*bl_ptr);
9687		ctsio->kern_total_len = sizeof(*bl_ptr);
9688	} else {
9689		ctsio->residual = 0;
9690		ctsio->kern_data_len = alloc_len;
9691		ctsio->kern_total_len = alloc_len;
9692	}
9693	ctsio->kern_data_resid = 0;
9694	ctsio->kern_rel_offset = 0;
9695	ctsio->kern_sg_entries = 0;
9696
9697	/*
9698	 * The control device is always connected.  The disk device, on the
9699	 * other hand, may not be online all the time.  Need to change this
9700	 * to figure out whether the disk device is actually online or not.
9701	 */
9702	if (lun != NULL)
9703		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9704				  lun->be_lun->lun_type;
9705	else
9706		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9707
9708	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9709	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9710	bl_ptr->max_cmp_write_len = 0xff;
9711	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9712	if (lun != NULL) {
9713		bs = lun->be_lun->blocksize;
9714		scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9715		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9716			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9717			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9718			if (lun->be_lun->ublockexp != 0) {
9719				scsi_ulto4b((1 << lun->be_lun->ublockexp),
9720				    bl_ptr->opt_unmap_grain);
9721				scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9722				    bl_ptr->unmap_grain_align);
9723			}
9724		}
9725		scsi_ulto4b(lun->be_lun->atomicblock,
9726		    bl_ptr->max_atomic_transfer_length);
9727		scsi_ulto4b(0, bl_ptr->atomic_alignment);
9728		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9729	}
9730	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9731
9732	ctl_set_success(ctsio);
9733	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9734	ctsio->be_move_done = ctl_config_move_done;
9735	ctl_datamove((union ctl_io *)ctsio);
9736	return (CTL_RETVAL_COMPLETE);
9737}
9738
9739static int
9740ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9741{
9742	struct scsi_vpd_block_device_characteristics *bdc_ptr;
9743	struct ctl_lun *lun;
9744	const char *value;
9745	u_int i;
9746
9747	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9748
9749	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9750	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9751	ctsio->kern_sg_entries = 0;
9752
9753	if (sizeof(*bdc_ptr) < alloc_len) {
9754		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
9755		ctsio->kern_data_len = sizeof(*bdc_ptr);
9756		ctsio->kern_total_len = sizeof(*bdc_ptr);
9757	} else {
9758		ctsio->residual = 0;
9759		ctsio->kern_data_len = alloc_len;
9760		ctsio->kern_total_len = alloc_len;
9761	}
9762	ctsio->kern_data_resid = 0;
9763	ctsio->kern_rel_offset = 0;
9764	ctsio->kern_sg_entries = 0;
9765
9766	/*
9767	 * The control device is always connected.  The disk device, on the
9768	 * other hand, may not be online all the time.  Need to change this
9769	 * to figure out whether the disk device is actually online or not.
9770	 */
9771	if (lun != NULL)
9772		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9773				  lun->be_lun->lun_type;
9774	else
9775		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9776	bdc_ptr->page_code = SVPD_BDC;
9777	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9778	if (lun != NULL &&
9779	    (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9780		i = strtol(value, NULL, 0);
9781	else
9782		i = CTL_DEFAULT_ROTATION_RATE;
9783	scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9784	if (lun != NULL &&
9785	    (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9786		i = strtol(value, NULL, 0);
9787	else
9788		i = 0;
9789	bdc_ptr->wab_wac_ff = (i & 0x0f);
9790	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9791
9792	ctl_set_success(ctsio);
9793	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9794	ctsio->be_move_done = ctl_config_move_done;
9795	ctl_datamove((union ctl_io *)ctsio);
9796	return (CTL_RETVAL_COMPLETE);
9797}
9798
9799static int
9800ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9801{
9802	struct scsi_vpd_logical_block_prov *lbp_ptr;
9803	struct ctl_lun *lun;
9804
9805	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9806
9807	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9808	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9809	ctsio->kern_sg_entries = 0;
9810
9811	if (sizeof(*lbp_ptr) < alloc_len) {
9812		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9813		ctsio->kern_data_len = sizeof(*lbp_ptr);
9814		ctsio->kern_total_len = sizeof(*lbp_ptr);
9815	} else {
9816		ctsio->residual = 0;
9817		ctsio->kern_data_len = alloc_len;
9818		ctsio->kern_total_len = alloc_len;
9819	}
9820	ctsio->kern_data_resid = 0;
9821	ctsio->kern_rel_offset = 0;
9822	ctsio->kern_sg_entries = 0;
9823
9824	/*
9825	 * The control device is always connected.  The disk device, on the
9826	 * other hand, may not be online all the time.  Need to change this
9827	 * to figure out whether the disk device is actually online or not.
9828	 */
9829	if (lun != NULL)
9830		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9831				  lun->be_lun->lun_type;
9832	else
9833		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9834
9835	lbp_ptr->page_code = SVPD_LBP;
9836	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9837	lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
9838	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9839		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
9840		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
9841		lbp_ptr->prov_type = SVPD_LBP_THIN;
9842	}
9843
9844	ctl_set_success(ctsio);
9845	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9846	ctsio->be_move_done = ctl_config_move_done;
9847	ctl_datamove((union ctl_io *)ctsio);
9848	return (CTL_RETVAL_COMPLETE);
9849}
9850
9851/*
9852 * INQUIRY with the EVPD bit set.
9853 */
9854static int
9855ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9856{
9857	struct ctl_lun *lun;
9858	struct scsi_inquiry *cdb;
9859	int alloc_len, retval;
9860
9861	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9862	cdb = (struct scsi_inquiry *)ctsio->cdb;
9863	alloc_len = scsi_2btoul(cdb->length);
9864
9865	switch (cdb->page_code) {
9866	case SVPD_SUPPORTED_PAGES:
9867		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9868		break;
9869	case SVPD_UNIT_SERIAL_NUMBER:
9870		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9871		break;
9872	case SVPD_DEVICE_ID:
9873		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9874		break;
9875	case SVPD_EXTENDED_INQUIRY_DATA:
9876		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
9877		break;
9878	case SVPD_MODE_PAGE_POLICY:
9879		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
9880		break;
9881	case SVPD_SCSI_PORTS:
9882		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
9883		break;
9884	case SVPD_SCSI_TPC:
9885		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
9886		break;
9887	case SVPD_BLOCK_LIMITS:
9888		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9889			goto err;
9890		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9891		break;
9892	case SVPD_BDC:
9893		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9894			goto err;
9895		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
9896		break;
9897	case SVPD_LBP:
9898		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9899			goto err;
9900		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9901		break;
9902	default:
9903err:
9904		ctl_set_invalid_field(ctsio,
9905				      /*sks_valid*/ 1,
9906				      /*command*/ 1,
9907				      /*field*/ 2,
9908				      /*bit_valid*/ 0,
9909				      /*bit*/ 0);
9910		ctl_done((union ctl_io *)ctsio);
9911		retval = CTL_RETVAL_COMPLETE;
9912		break;
9913	}
9914
9915	return (retval);
9916}
9917
9918/*
9919 * Standard INQUIRY data.
9920 */
9921static int
9922ctl_inquiry_std(struct ctl_scsiio *ctsio)
9923{
9924	struct scsi_inquiry_data *inq_ptr;
9925	struct scsi_inquiry *cdb;
9926	struct ctl_softc *softc;
9927	struct ctl_lun *lun;
9928	char *val;
9929	uint32_t alloc_len, data_len;
9930	ctl_port_type port_type;
9931
9932	softc = control_softc;
9933
9934	/*
9935	 * Figure out whether we're talking to a Fibre Channel port or not.
9936	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9937	 * SCSI front ends.
9938	 */
9939	port_type = softc->ctl_ports[
9940	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
9941	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
9942		port_type = CTL_PORT_SCSI;
9943
9944	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9945	cdb = (struct scsi_inquiry *)ctsio->cdb;
9946	alloc_len = scsi_2btoul(cdb->length);
9947
9948	/*
9949	 * We malloc the full inquiry data size here and fill it
9950	 * in.  If the user only asks for less, we'll give him
9951	 * that much.
9952	 */
9953	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
9954	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9955	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9956	ctsio->kern_sg_entries = 0;
9957	ctsio->kern_data_resid = 0;
9958	ctsio->kern_rel_offset = 0;
9959
9960	if (data_len < alloc_len) {
9961		ctsio->residual = alloc_len - data_len;
9962		ctsio->kern_data_len = data_len;
9963		ctsio->kern_total_len = data_len;
9964	} else {
9965		ctsio->residual = 0;
9966		ctsio->kern_data_len = alloc_len;
9967		ctsio->kern_total_len = alloc_len;
9968	}
9969
9970	/*
9971	 * If we have a LUN configured, report it as connected.  Otherwise,
9972	 * report that it is offline or no device is supported, depending
9973	 * on the value of inquiry_pq_no_lun.
9974	 *
9975	 * According to the spec (SPC-4 r34), the peripheral qualifier
9976	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9977	 *
9978	 * "A peripheral device having the specified peripheral device type
9979	 * is not connected to this logical unit. However, the device
9980	 * server is capable of supporting the specified peripheral device
9981	 * type on this logical unit."
9982	 *
9983	 * According to the same spec, the peripheral qualifier
9984	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
9985	 *
9986	 * "The device server is not capable of supporting a peripheral
9987	 * device on this logical unit. For this peripheral qualifier the
9988	 * peripheral device type shall be set to 1Fh. All other peripheral
9989	 * device type values are reserved for this peripheral qualifier."
9990	 *
9991	 * Given the text, it would seem that we probably want to report that
9992	 * the LUN is offline here.  There is no LUN connected, but we can
9993	 * support a LUN at the given LUN number.
9994	 *
9995	 * In the real world, though, it sounds like things are a little
9996	 * different:
9997	 *
9998	 * - Linux, when presented with a LUN with the offline peripheral
9999	 *   qualifier, will create an sg driver instance for it.  So when
10000	 *   you attach it to CTL, you wind up with a ton of sg driver
10001	 *   instances.  (One for every LUN that Linux bothered to probe.)
10002	 *   Linux does this despite the fact that it issues a REPORT LUNs
10003	 *   to LUN 0 to get the inventory of supported LUNs.
10004	 *
10005	 * - There is other anecdotal evidence (from Emulex folks) about
10006	 *   arrays that use the offline peripheral qualifier for LUNs that
10007	 *   are on the "passive" path in an active/passive array.
10008	 *
10009	 * So the solution is provide a hopefully reasonable default
10010	 * (return bad/no LUN) and allow the user to change the behavior
10011	 * with a tunable/sysctl variable.
10012	 */
10013	if (lun != NULL)
10014		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10015				  lun->be_lun->lun_type;
10016	else if (softc->inquiry_pq_no_lun == 0)
10017		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10018	else
10019		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10020
10021	/* RMB in byte 2 is 0 */
10022	inq_ptr->version = SCSI_REV_SPC4;
10023
10024	/*
10025	 * According to SAM-3, even if a device only supports a single
10026	 * level of LUN addressing, it should still set the HISUP bit:
10027	 *
10028	 * 4.9.1 Logical unit numbers overview
10029	 *
10030	 * All logical unit number formats described in this standard are
10031	 * hierarchical in structure even when only a single level in that
10032	 * hierarchy is used. The HISUP bit shall be set to one in the
10033	 * standard INQUIRY data (see SPC-2) when any logical unit number
10034	 * format described in this standard is used.  Non-hierarchical
10035	 * formats are outside the scope of this standard.
10036	 *
10037	 * Therefore we set the HiSup bit here.
10038	 *
10039	 * The reponse format is 2, per SPC-3.
10040	 */
10041	inq_ptr->response_format = SID_HiSup | 2;
10042
10043	inq_ptr->additional_length = data_len -
10044	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10045	CTL_DEBUG_PRINT(("additional_length = %d\n",
10046			 inq_ptr->additional_length));
10047
10048	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10049	/* 16 bit addressing */
10050	if (port_type == CTL_PORT_SCSI)
10051		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10052	/* XXX set the SID_MultiP bit here if we're actually going to
10053	   respond on multiple ports */
10054	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10055
10056	/* 16 bit data bus, synchronous transfers */
10057	if (port_type == CTL_PORT_SCSI)
10058		inq_ptr->flags = SID_WBus16 | SID_Sync;
10059	/*
10060	 * XXX KDM do we want to support tagged queueing on the control
10061	 * device at all?
10062	 */
10063	if ((lun == NULL)
10064	 || (lun->be_lun->lun_type != T_PROCESSOR))
10065		inq_ptr->flags |= SID_CmdQue;
10066	/*
10067	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10068	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10069	 * name and 4 bytes for the revision.
10070	 */
10071	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10072	    "vendor")) == NULL) {
10073		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10074	} else {
10075		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10076		strncpy(inq_ptr->vendor, val,
10077		    min(sizeof(inq_ptr->vendor), strlen(val)));
10078	}
10079	if (lun == NULL) {
10080		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10081		    sizeof(inq_ptr->product));
10082	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10083		switch (lun->be_lun->lun_type) {
10084		case T_DIRECT:
10085			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10086			    sizeof(inq_ptr->product));
10087			break;
10088		case T_PROCESSOR:
10089			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10090			    sizeof(inq_ptr->product));
10091			break;
10092		default:
10093			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10094			    sizeof(inq_ptr->product));
10095			break;
10096		}
10097	} else {
10098		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10099		strncpy(inq_ptr->product, val,
10100		    min(sizeof(inq_ptr->product), strlen(val)));
10101	}
10102
10103	/*
10104	 * XXX make this a macro somewhere so it automatically gets
10105	 * incremented when we make changes.
10106	 */
10107	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10108	    "revision")) == NULL) {
10109		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10110	} else {
10111		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10112		strncpy(inq_ptr->revision, val,
10113		    min(sizeof(inq_ptr->revision), strlen(val)));
10114	}
10115
10116	/*
10117	 * For parallel SCSI, we support double transition and single
10118	 * transition clocking.  We also support QAS (Quick Arbitration
10119	 * and Selection) and Information Unit transfers on both the
10120	 * control and array devices.
10121	 */
10122	if (port_type == CTL_PORT_SCSI)
10123		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10124				    SID_SPI_IUS;
10125
10126	/* SAM-5 (no version claimed) */
10127	scsi_ulto2b(0x00A0, inq_ptr->version1);
10128	/* SPC-4 (no version claimed) */
10129	scsi_ulto2b(0x0460, inq_ptr->version2);
10130	if (port_type == CTL_PORT_FC) {
10131		/* FCP-2 ANSI INCITS.350:2003 */
10132		scsi_ulto2b(0x0917, inq_ptr->version3);
10133	} else if (port_type == CTL_PORT_SCSI) {
10134		/* SPI-4 ANSI INCITS.362:200x */
10135		scsi_ulto2b(0x0B56, inq_ptr->version3);
10136	} else if (port_type == CTL_PORT_ISCSI) {
10137		/* iSCSI (no version claimed) */
10138		scsi_ulto2b(0x0960, inq_ptr->version3);
10139	} else if (port_type == CTL_PORT_SAS) {
10140		/* SAS (no version claimed) */
10141		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10142	}
10143
10144	if (lun == NULL) {
10145		/* SBC-4 (no version claimed) */
10146		scsi_ulto2b(0x0600, inq_ptr->version4);
10147	} else {
10148		switch (lun->be_lun->lun_type) {
10149		case T_DIRECT:
10150			/* SBC-4 (no version claimed) */
10151			scsi_ulto2b(0x0600, inq_ptr->version4);
10152			break;
10153		case T_PROCESSOR:
10154		default:
10155			break;
10156		}
10157	}
10158
10159	ctl_set_success(ctsio);
10160	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10161	ctsio->be_move_done = ctl_config_move_done;
10162	ctl_datamove((union ctl_io *)ctsio);
10163	return (CTL_RETVAL_COMPLETE);
10164}
10165
10166int
10167ctl_inquiry(struct ctl_scsiio *ctsio)
10168{
10169	struct scsi_inquiry *cdb;
10170	int retval;
10171
10172	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10173
10174	cdb = (struct scsi_inquiry *)ctsio->cdb;
10175	if (cdb->byte2 & SI_EVPD)
10176		retval = ctl_inquiry_evpd(ctsio);
10177	else if (cdb->page_code == 0)
10178		retval = ctl_inquiry_std(ctsio);
10179	else {
10180		ctl_set_invalid_field(ctsio,
10181				      /*sks_valid*/ 1,
10182				      /*command*/ 1,
10183				      /*field*/ 2,
10184				      /*bit_valid*/ 0,
10185				      /*bit*/ 0);
10186		ctl_done((union ctl_io *)ctsio);
10187		return (CTL_RETVAL_COMPLETE);
10188	}
10189
10190	return (retval);
10191}
10192
10193/*
10194 * For known CDB types, parse the LBA and length.
10195 */
10196static int
10197ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10198{
10199	if (io->io_hdr.io_type != CTL_IO_SCSI)
10200		return (1);
10201
10202	switch (io->scsiio.cdb[0]) {
10203	case COMPARE_AND_WRITE: {
10204		struct scsi_compare_and_write *cdb;
10205
10206		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10207
10208		*lba = scsi_8btou64(cdb->addr);
10209		*len = cdb->length;
10210		break;
10211	}
10212	case READ_6:
10213	case WRITE_6: {
10214		struct scsi_rw_6 *cdb;
10215
10216		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10217
10218		*lba = scsi_3btoul(cdb->addr);
10219		/* only 5 bits are valid in the most significant address byte */
10220		*lba &= 0x1fffff;
10221		*len = cdb->length;
10222		break;
10223	}
10224	case READ_10:
10225	case WRITE_10: {
10226		struct scsi_rw_10 *cdb;
10227
10228		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10229
10230		*lba = scsi_4btoul(cdb->addr);
10231		*len = scsi_2btoul(cdb->length);
10232		break;
10233	}
10234	case WRITE_VERIFY_10: {
10235		struct scsi_write_verify_10 *cdb;
10236
10237		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10238
10239		*lba = scsi_4btoul(cdb->addr);
10240		*len = scsi_2btoul(cdb->length);
10241		break;
10242	}
10243	case READ_12:
10244	case WRITE_12: {
10245		struct scsi_rw_12 *cdb;
10246
10247		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10248
10249		*lba = scsi_4btoul(cdb->addr);
10250		*len = scsi_4btoul(cdb->length);
10251		break;
10252	}
10253	case WRITE_VERIFY_12: {
10254		struct scsi_write_verify_12 *cdb;
10255
10256		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10257
10258		*lba = scsi_4btoul(cdb->addr);
10259		*len = scsi_4btoul(cdb->length);
10260		break;
10261	}
10262	case READ_16:
10263	case WRITE_16:
10264	case WRITE_ATOMIC_16: {
10265		struct scsi_rw_16 *cdb;
10266
10267		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10268
10269		*lba = scsi_8btou64(cdb->addr);
10270		*len = scsi_4btoul(cdb->length);
10271		break;
10272	}
10273	case WRITE_VERIFY_16: {
10274		struct scsi_write_verify_16 *cdb;
10275
10276		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10277
10278		*lba = scsi_8btou64(cdb->addr);
10279		*len = scsi_4btoul(cdb->length);
10280		break;
10281	}
10282	case WRITE_SAME_10: {
10283		struct scsi_write_same_10 *cdb;
10284
10285		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10286
10287		*lba = scsi_4btoul(cdb->addr);
10288		*len = scsi_2btoul(cdb->length);
10289		break;
10290	}
10291	case WRITE_SAME_16: {
10292		struct scsi_write_same_16 *cdb;
10293
10294		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10295
10296		*lba = scsi_8btou64(cdb->addr);
10297		*len = scsi_4btoul(cdb->length);
10298		break;
10299	}
10300	case VERIFY_10: {
10301		struct scsi_verify_10 *cdb;
10302
10303		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10304
10305		*lba = scsi_4btoul(cdb->addr);
10306		*len = scsi_2btoul(cdb->length);
10307		break;
10308	}
10309	case VERIFY_12: {
10310		struct scsi_verify_12 *cdb;
10311
10312		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10313
10314		*lba = scsi_4btoul(cdb->addr);
10315		*len = scsi_4btoul(cdb->length);
10316		break;
10317	}
10318	case VERIFY_16: {
10319		struct scsi_verify_16 *cdb;
10320
10321		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10322
10323		*lba = scsi_8btou64(cdb->addr);
10324		*len = scsi_4btoul(cdb->length);
10325		break;
10326	}
10327	case UNMAP: {
10328		*lba = 0;
10329		*len = UINT64_MAX;
10330		break;
10331	}
10332	case SERVICE_ACTION_IN: {	/* GET LBA STATUS */
10333		struct scsi_get_lba_status *cdb;
10334
10335		cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10336		*lba = scsi_8btou64(cdb->addr);
10337		*len = UINT32_MAX;
10338		break;
10339	}
10340	default:
10341		return (1);
10342		break; /* NOTREACHED */
10343	}
10344
10345	return (0);
10346}
10347
10348static ctl_action
10349ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10350    bool seq)
10351{
10352	uint64_t endlba1, endlba2;
10353
10354	endlba1 = lba1 + len1 - (seq ? 0 : 1);
10355	endlba2 = lba2 + len2 - 1;
10356
10357	if ((endlba1 < lba2) || (endlba2 < lba1))
10358		return (CTL_ACTION_PASS);
10359	else
10360		return (CTL_ACTION_BLOCK);
10361}
10362
10363static int
10364ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10365{
10366	struct ctl_ptr_len_flags *ptrlen;
10367	struct scsi_unmap_desc *buf, *end, *range;
10368	uint64_t lba;
10369	uint32_t len;
10370
10371	/* If not UNMAP -- go other way. */
10372	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10373	    io->scsiio.cdb[0] != UNMAP)
10374		return (CTL_ACTION_ERROR);
10375
10376	/* If UNMAP without data -- block and wait for data. */
10377	ptrlen = (struct ctl_ptr_len_flags *)
10378	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10379	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10380	    ptrlen->ptr == NULL)
10381		return (CTL_ACTION_BLOCK);
10382
10383	/* UNMAP with data -- check for collision. */
10384	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10385	end = buf + ptrlen->len / sizeof(*buf);
10386	for (range = buf; range < end; range++) {
10387		lba = scsi_8btou64(range->lba);
10388		len = scsi_4btoul(range->length);
10389		if ((lba < lba2 + len2) && (lba + len > lba2))
10390			return (CTL_ACTION_BLOCK);
10391	}
10392	return (CTL_ACTION_PASS);
10393}
10394
10395static ctl_action
10396ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10397{
10398	uint64_t lba1, lba2;
10399	uint64_t len1, len2;
10400	int retval;
10401
10402	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10403		return (CTL_ACTION_ERROR);
10404
10405	retval = ctl_extent_check_unmap(io1, lba2, len2);
10406	if (retval != CTL_ACTION_ERROR)
10407		return (retval);
10408
10409	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10410		return (CTL_ACTION_ERROR);
10411
10412	return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10413}
10414
10415static ctl_action
10416ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10417{
10418	uint64_t lba1, lba2;
10419	uint64_t len1, len2;
10420
10421	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10422		return (CTL_ACTION_ERROR);
10423	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10424		return (CTL_ACTION_ERROR);
10425
10426	if (lba1 + len1 == lba2)
10427		return (CTL_ACTION_BLOCK);
10428	return (CTL_ACTION_PASS);
10429}
10430
10431static ctl_action
10432ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10433    union ctl_io *ooa_io)
10434{
10435	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10436	ctl_serialize_action *serialize_row;
10437
10438	/*
10439	 * The initiator attempted multiple untagged commands at the same
10440	 * time.  Can't do that.
10441	 */
10442	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10443	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10444	 && ((pending_io->io_hdr.nexus.targ_port ==
10445	      ooa_io->io_hdr.nexus.targ_port)
10446	  && (pending_io->io_hdr.nexus.initid.id ==
10447	      ooa_io->io_hdr.nexus.initid.id))
10448	 && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10449	      CTL_FLAG_STATUS_SENT)) == 0))
10450		return (CTL_ACTION_OVERLAP);
10451
10452	/*
10453	 * The initiator attempted to send multiple tagged commands with
10454	 * the same ID.  (It's fine if different initiators have the same
10455	 * tag ID.)
10456	 *
10457	 * Even if all of those conditions are true, we don't kill the I/O
10458	 * if the command ahead of us has been aborted.  We won't end up
10459	 * sending it to the FETD, and it's perfectly legal to resend a
10460	 * command with the same tag number as long as the previous
10461	 * instance of this tag number has been aborted somehow.
10462	 */
10463	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10464	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10465	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10466	 && ((pending_io->io_hdr.nexus.targ_port ==
10467	      ooa_io->io_hdr.nexus.targ_port)
10468	  && (pending_io->io_hdr.nexus.initid.id ==
10469	      ooa_io->io_hdr.nexus.initid.id))
10470	 && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10471	      CTL_FLAG_STATUS_SENT)) == 0))
10472		return (CTL_ACTION_OVERLAP_TAG);
10473
10474	/*
10475	 * If we get a head of queue tag, SAM-3 says that we should
10476	 * immediately execute it.
10477	 *
10478	 * What happens if this command would normally block for some other
10479	 * reason?  e.g. a request sense with a head of queue tag
10480	 * immediately after a write.  Normally that would block, but this
10481	 * will result in its getting executed immediately...
10482	 *
10483	 * We currently return "pass" instead of "skip", so we'll end up
10484	 * going through the rest of the queue to check for overlapped tags.
10485	 *
10486	 * XXX KDM check for other types of blockage first??
10487	 */
10488	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10489		return (CTL_ACTION_PASS);
10490
10491	/*
10492	 * Ordered tags have to block until all items ahead of them
10493	 * have completed.  If we get called with an ordered tag, we always
10494	 * block, if something else is ahead of us in the queue.
10495	 */
10496	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10497		return (CTL_ACTION_BLOCK);
10498
10499	/*
10500	 * Simple tags get blocked until all head of queue and ordered tags
10501	 * ahead of them have completed.  I'm lumping untagged commands in
10502	 * with simple tags here.  XXX KDM is that the right thing to do?
10503	 */
10504	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10505	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10506	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10507	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10508		return (CTL_ACTION_BLOCK);
10509
10510	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10511	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10512
10513	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10514
10515	switch (serialize_row[pending_entry->seridx]) {
10516	case CTL_SER_BLOCK:
10517		return (CTL_ACTION_BLOCK);
10518	case CTL_SER_EXTENT:
10519		return (ctl_extent_check(ooa_io, pending_io,
10520		    (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10521	case CTL_SER_EXTENTOPT:
10522		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10523		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10524			return (ctl_extent_check(ooa_io, pending_io,
10525			    (lun->be_lun &&
10526			     lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10527		return (CTL_ACTION_PASS);
10528	case CTL_SER_EXTENTSEQ:
10529		if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10530			return (ctl_extent_check_seq(ooa_io, pending_io));
10531		return (CTL_ACTION_PASS);
10532	case CTL_SER_PASS:
10533		return (CTL_ACTION_PASS);
10534	case CTL_SER_BLOCKOPT:
10535		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10536		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10537			return (CTL_ACTION_BLOCK);
10538		return (CTL_ACTION_PASS);
10539	case CTL_SER_SKIP:
10540		return (CTL_ACTION_SKIP);
10541	default:
10542		panic("invalid serialization value %d",
10543		      serialize_row[pending_entry->seridx]);
10544	}
10545
10546	return (CTL_ACTION_ERROR);
10547}
10548
10549/*
10550 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10551 * Assumptions:
10552 * - pending_io is generally either incoming, or on the blocked queue
10553 * - starting I/O is the I/O we want to start the check with.
10554 */
10555static ctl_action
10556ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10557	      union ctl_io *starting_io)
10558{
10559	union ctl_io *ooa_io;
10560	ctl_action action;
10561
10562	mtx_assert(&lun->lun_lock, MA_OWNED);
10563
10564	/*
10565	 * Run back along the OOA queue, starting with the current
10566	 * blocked I/O and going through every I/O before it on the
10567	 * queue.  If starting_io is NULL, we'll just end up returning
10568	 * CTL_ACTION_PASS.
10569	 */
10570	for (ooa_io = starting_io; ooa_io != NULL;
10571	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10572	     ooa_links)){
10573
10574		/*
10575		 * This routine just checks to see whether
10576		 * cur_blocked is blocked by ooa_io, which is ahead
10577		 * of it in the queue.  It doesn't queue/dequeue
10578		 * cur_blocked.
10579		 */
10580		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10581		switch (action) {
10582		case CTL_ACTION_BLOCK:
10583		case CTL_ACTION_OVERLAP:
10584		case CTL_ACTION_OVERLAP_TAG:
10585		case CTL_ACTION_SKIP:
10586		case CTL_ACTION_ERROR:
10587			return (action);
10588			break; /* NOTREACHED */
10589		case CTL_ACTION_PASS:
10590			break;
10591		default:
10592			panic("invalid action %d", action);
10593			break;  /* NOTREACHED */
10594		}
10595	}
10596
10597	return (CTL_ACTION_PASS);
10598}
10599
10600/*
10601 * Assumptions:
10602 * - An I/O has just completed, and has been removed from the per-LUN OOA
10603 *   queue, so some items on the blocked queue may now be unblocked.
10604 */
10605static int
10606ctl_check_blocked(struct ctl_lun *lun)
10607{
10608	union ctl_io *cur_blocked, *next_blocked;
10609
10610	mtx_assert(&lun->lun_lock, MA_OWNED);
10611
10612	/*
10613	 * Run forward from the head of the blocked queue, checking each
10614	 * entry against the I/Os prior to it on the OOA queue to see if
10615	 * there is still any blockage.
10616	 *
10617	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10618	 * with our removing a variable on it while it is traversing the
10619	 * list.
10620	 */
10621	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10622	     cur_blocked != NULL; cur_blocked = next_blocked) {
10623		union ctl_io *prev_ooa;
10624		ctl_action action;
10625
10626		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10627							  blocked_links);
10628
10629		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10630						      ctl_ooaq, ooa_links);
10631
10632		/*
10633		 * If cur_blocked happens to be the first item in the OOA
10634		 * queue now, prev_ooa will be NULL, and the action
10635		 * returned will just be CTL_ACTION_PASS.
10636		 */
10637		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10638
10639		switch (action) {
10640		case CTL_ACTION_BLOCK:
10641			/* Nothing to do here, still blocked */
10642			break;
10643		case CTL_ACTION_OVERLAP:
10644		case CTL_ACTION_OVERLAP_TAG:
10645			/*
10646			 * This shouldn't happen!  In theory we've already
10647			 * checked this command for overlap...
10648			 */
10649			break;
10650		case CTL_ACTION_PASS:
10651		case CTL_ACTION_SKIP: {
10652			const struct ctl_cmd_entry *entry;
10653			int isc_retval;
10654
10655			/*
10656			 * The skip case shouldn't happen, this transaction
10657			 * should have never made it onto the blocked queue.
10658			 */
10659			/*
10660			 * This I/O is no longer blocked, we can remove it
10661			 * from the blocked queue.  Since this is a TAILQ
10662			 * (doubly linked list), we can do O(1) removals
10663			 * from any place on the list.
10664			 */
10665			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10666				     blocked_links);
10667			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10668
10669			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10670				/*
10671				 * Need to send IO back to original side to
10672				 * run
10673				 */
10674				union ctl_ha_msg msg_info;
10675
10676				msg_info.hdr.original_sc =
10677					cur_blocked->io_hdr.original_sc;
10678				msg_info.hdr.serializing_sc = cur_blocked;
10679				msg_info.hdr.msg_type = CTL_MSG_R2R;
10680				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10681				     &msg_info, sizeof(msg_info), 0)) >
10682				     CTL_HA_STATUS_SUCCESS) {
10683					printf("CTL:Check Blocked error from "
10684					       "ctl_ha_msg_send %d\n",
10685					       isc_retval);
10686				}
10687				break;
10688			}
10689			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
10690
10691			/*
10692			 * Check this I/O for LUN state changes that may
10693			 * have happened while this command was blocked.
10694			 * The LUN state may have been changed by a command
10695			 * ahead of us in the queue, so we need to re-check
10696			 * for any states that can be caused by SCSI
10697			 * commands.
10698			 */
10699			if (ctl_scsiio_lun_check(lun, entry,
10700						 &cur_blocked->scsiio) == 0) {
10701				cur_blocked->io_hdr.flags |=
10702				                      CTL_FLAG_IS_WAS_ON_RTR;
10703				ctl_enqueue_rtr(cur_blocked);
10704			} else
10705				ctl_done(cur_blocked);
10706			break;
10707		}
10708		default:
10709			/*
10710			 * This probably shouldn't happen -- we shouldn't
10711			 * get CTL_ACTION_ERROR, or anything else.
10712			 */
10713			break;
10714		}
10715	}
10716
10717	return (CTL_RETVAL_COMPLETE);
10718}
10719
10720/*
10721 * This routine (with one exception) checks LUN flags that can be set by
10722 * commands ahead of us in the OOA queue.  These flags have to be checked
10723 * when a command initially comes in, and when we pull a command off the
10724 * blocked queue and are preparing to execute it.  The reason we have to
10725 * check these flags for commands on the blocked queue is that the LUN
10726 * state may have been changed by a command ahead of us while we're on the
10727 * blocked queue.
10728 *
10729 * Ordering is somewhat important with these checks, so please pay
10730 * careful attention to the placement of any new checks.
10731 */
10732static int
10733ctl_scsiio_lun_check(struct ctl_lun *lun,
10734    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10735{
10736	struct ctl_softc *softc = lun->ctl_softc;
10737	int retval;
10738	uint32_t residx;
10739
10740	retval = 0;
10741
10742	mtx_assert(&lun->lun_lock, MA_OWNED);
10743
10744	/*
10745	 * If this shelf is a secondary shelf controller, we have to reject
10746	 * any media access commands.
10747	 */
10748	if ((softc->flags & CTL_FLAG_ACTIVE_SHELF) == 0 &&
10749	    (entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0) {
10750		ctl_set_lun_standby(ctsio);
10751		retval = 1;
10752		goto bailout;
10753	}
10754
10755	if (entry->pattern & CTL_LUN_PAT_WRITE) {
10756		if (lun->be_lun &&
10757		    lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
10758			ctl_set_sense(ctsio, /*current_error*/ 1,
10759			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
10760			    /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
10761			retval = 1;
10762			goto bailout;
10763		}
10764		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
10765		    .eca_and_aen & SCP_SWP) != 0) {
10766			ctl_set_sense(ctsio, /*current_error*/ 1,
10767			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
10768			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
10769			retval = 1;
10770			goto bailout;
10771		}
10772	}
10773
10774	/*
10775	 * Check for a reservation conflict.  If this command isn't allowed
10776	 * even on reserved LUNs, and if this initiator isn't the one who
10777	 * reserved us, reject the command with a reservation conflict.
10778	 */
10779	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10780	if ((lun->flags & CTL_LUN_RESERVED)
10781	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10782		if (lun->res_idx != residx) {
10783			ctl_set_reservation_conflict(ctsio);
10784			retval = 1;
10785			goto bailout;
10786		}
10787	}
10788
10789	if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
10790	    (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
10791		/* No reservation or command is allowed. */;
10792	} else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
10793	    (lun->res_type == SPR_TYPE_WR_EX ||
10794	     lun->res_type == SPR_TYPE_WR_EX_RO ||
10795	     lun->res_type == SPR_TYPE_WR_EX_AR)) {
10796		/* The command is allowed for Write Exclusive resv. */;
10797	} else {
10798		/*
10799		 * if we aren't registered or it's a res holder type
10800		 * reservation and this isn't the res holder then set a
10801		 * conflict.
10802		 */
10803		if (ctl_get_prkey(lun, residx) == 0
10804		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10805			ctl_set_reservation_conflict(ctsio);
10806			retval = 1;
10807			goto bailout;
10808		}
10809
10810	}
10811
10812	if ((lun->flags & CTL_LUN_OFFLINE)
10813	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10814		ctl_set_lun_not_ready(ctsio);
10815		retval = 1;
10816		goto bailout;
10817	}
10818
10819	/*
10820	 * If the LUN is stopped, see if this particular command is allowed
10821	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10822	 */
10823	if ((lun->flags & CTL_LUN_STOPPED)
10824	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10825		/* "Logical unit not ready, initializing cmd. required" */
10826		ctl_set_lun_stopped(ctsio);
10827		retval = 1;
10828		goto bailout;
10829	}
10830
10831	if ((lun->flags & CTL_LUN_INOPERABLE)
10832	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10833		/* "Medium format corrupted" */
10834		ctl_set_medium_format_corrupted(ctsio);
10835		retval = 1;
10836		goto bailout;
10837	}
10838
10839bailout:
10840	return (retval);
10841
10842}
10843
10844static void
10845ctl_failover_io(union ctl_io *io, int have_lock)
10846{
10847	ctl_set_busy(&io->scsiio);
10848	ctl_done(io);
10849}
10850
10851#ifdef notyet
10852static void
10853ctl_failover(void)
10854{
10855	struct ctl_lun *lun;
10856	struct ctl_softc *softc;
10857	union ctl_io *next_io, *pending_io;
10858	union ctl_io *io;
10859	int lun_idx;
10860
10861	softc = control_softc;
10862
10863	mtx_lock(&softc->ctl_lock);
10864	/*
10865	 * Remove any cmds from the other SC from the rtr queue.  These
10866	 * will obviously only be for LUNs for which we're the primary.
10867	 * We can't send status or get/send data for these commands.
10868	 * Since they haven't been executed yet, we can just remove them.
10869	 * We'll either abort them or delete them below, depending on
10870	 * which HA mode we're in.
10871	 */
10872#ifdef notyet
10873	mtx_lock(&softc->queue_lock);
10874	for (io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue);
10875	     io != NULL; io = next_io) {
10876		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10877		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10878			STAILQ_REMOVE(&softc->rtr_queue, &io->io_hdr,
10879				      ctl_io_hdr, links);
10880	}
10881	mtx_unlock(&softc->queue_lock);
10882#endif
10883
10884	for (lun_idx=0; lun_idx < softc->num_luns; lun_idx++) {
10885		lun = softc->ctl_luns[lun_idx];
10886		if (lun==NULL)
10887			continue;
10888
10889		/*
10890		 * Processor LUNs are primary on both sides.
10891		 * XXX will this always be true?
10892		 */
10893		if (lun->be_lun->lun_type == T_PROCESSOR)
10894			continue;
10895
10896		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10897		 && (softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10898			printf("FAILOVER: primary lun %d\n", lun_idx);
10899		        /*
10900			 * Remove all commands from the other SC. First from the
10901			 * blocked queue then from the ooa queue. Once we have
10902			 * removed them. Call ctl_check_blocked to see if there
10903			 * is anything that can run.
10904			 */
10905			for (io = (union ctl_io *)TAILQ_FIRST(
10906			     &lun->blocked_queue); io != NULL; io = next_io) {
10907
10908		        	next_io = (union ctl_io *)TAILQ_NEXT(
10909				    &io->io_hdr, blocked_links);
10910
10911				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10912					TAILQ_REMOVE(&lun->blocked_queue,
10913						     &io->io_hdr,blocked_links);
10914					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10915					TAILQ_REMOVE(&lun->ooa_queue,
10916						     &io->io_hdr, ooa_links);
10917
10918					ctl_free_io(io);
10919				}
10920			}
10921
10922			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10923	     		     io != NULL; io = next_io) {
10924
10925		        	next_io = (union ctl_io *)TAILQ_NEXT(
10926				    &io->io_hdr, ooa_links);
10927
10928				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10929
10930					TAILQ_REMOVE(&lun->ooa_queue,
10931						&io->io_hdr,
10932					     	ooa_links);
10933
10934					ctl_free_io(io);
10935				}
10936			}
10937			ctl_check_blocked(lun);
10938		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10939			&& (softc->ha_mode == CTL_HA_MODE_XFER)) {
10940
10941			printf("FAILOVER: primary lun %d\n", lun_idx);
10942			/*
10943			 * Abort all commands from the other SC.  We can't
10944			 * send status back for them now.  These should get
10945			 * cleaned up when they are completed or come out
10946			 * for a datamove operation.
10947			 */
10948			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10949	     		     io != NULL; io = next_io) {
10950		        	next_io = (union ctl_io *)TAILQ_NEXT(
10951					&io->io_hdr, ooa_links);
10952
10953				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10954					io->io_hdr.flags |= CTL_FLAG_ABORT;
10955			}
10956		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10957			&& (softc->ha_mode == CTL_HA_MODE_XFER)) {
10958
10959			printf("FAILOVER: secondary lun %d\n", lun_idx);
10960
10961			lun->flags |= CTL_LUN_PRIMARY_SC;
10962
10963			/*
10964			 * We send all I/O that was sent to this controller
10965			 * and redirected to the other side back with
10966			 * busy status, and have the initiator retry it.
10967			 * Figuring out how much data has been transferred,
10968			 * etc. and picking up where we left off would be
10969			 * very tricky.
10970			 *
10971			 * XXX KDM need to remove I/O from the blocked
10972			 * queue as well!
10973			 */
10974			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10975			     &lun->ooa_queue); pending_io != NULL;
10976			     pending_io = next_io) {
10977
10978				next_io =  (union ctl_io *)TAILQ_NEXT(
10979					&pending_io->io_hdr, ooa_links);
10980
10981				pending_io->io_hdr.flags &=
10982					~CTL_FLAG_SENT_2OTHER_SC;
10983
10984				if (pending_io->io_hdr.flags &
10985				    CTL_FLAG_IO_ACTIVE) {
10986					pending_io->io_hdr.flags |=
10987						CTL_FLAG_FAILOVER;
10988				} else {
10989					ctl_set_busy(&pending_io->scsiio);
10990					ctl_done(pending_io);
10991				}
10992			}
10993
10994			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
10995		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10996			&& (softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10997			printf("FAILOVER: secondary lun %d\n", lun_idx);
10998			/*
10999			 * if the first io on the OOA is not on the RtR queue
11000			 * add it.
11001			 */
11002			lun->flags |= CTL_LUN_PRIMARY_SC;
11003
11004			pending_io = (union ctl_io *)TAILQ_FIRST(
11005			    &lun->ooa_queue);
11006			if (pending_io==NULL) {
11007				printf("Nothing on OOA queue\n");
11008				continue;
11009			}
11010
11011			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11012			if ((pending_io->io_hdr.flags &
11013			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11014				pending_io->io_hdr.flags |=
11015				    CTL_FLAG_IS_WAS_ON_RTR;
11016				ctl_enqueue_rtr(pending_io);
11017			}
11018#if 0
11019			else
11020			{
11021				printf("Tag 0x%04x is running\n",
11022				      pending_io->scsiio.tag_num);
11023			}
11024#endif
11025
11026			next_io = (union ctl_io *)TAILQ_NEXT(
11027			    &pending_io->io_hdr, ooa_links);
11028			for (pending_io=next_io; pending_io != NULL;
11029			     pending_io = next_io) {
11030				pending_io->io_hdr.flags &=
11031				    ~CTL_FLAG_SENT_2OTHER_SC;
11032				next_io = (union ctl_io *)TAILQ_NEXT(
11033					&pending_io->io_hdr, ooa_links);
11034				if (pending_io->io_hdr.flags &
11035				    CTL_FLAG_IS_WAS_ON_RTR) {
11036#if 0
11037				        printf("Tag 0x%04x is running\n",
11038				      		pending_io->scsiio.tag_num);
11039#endif
11040					continue;
11041				}
11042
11043				switch (ctl_check_ooa(lun, pending_io,
11044			            (union ctl_io *)TAILQ_PREV(
11045				    &pending_io->io_hdr, ctl_ooaq,
11046				    ooa_links))) {
11047
11048				case CTL_ACTION_BLOCK:
11049					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11050							  &pending_io->io_hdr,
11051							  blocked_links);
11052					pending_io->io_hdr.flags |=
11053					    CTL_FLAG_BLOCKED;
11054					break;
11055				case CTL_ACTION_PASS:
11056				case CTL_ACTION_SKIP:
11057					pending_io->io_hdr.flags |=
11058					    CTL_FLAG_IS_WAS_ON_RTR;
11059					ctl_enqueue_rtr(pending_io);
11060					break;
11061				case CTL_ACTION_OVERLAP:
11062					ctl_set_overlapped_cmd(
11063					    (struct ctl_scsiio *)pending_io);
11064					ctl_done(pending_io);
11065					break;
11066				case CTL_ACTION_OVERLAP_TAG:
11067					ctl_set_overlapped_tag(
11068					    (struct ctl_scsiio *)pending_io,
11069					    pending_io->scsiio.tag_num & 0xff);
11070					ctl_done(pending_io);
11071					break;
11072				case CTL_ACTION_ERROR:
11073				default:
11074					ctl_set_internal_failure(
11075						(struct ctl_scsiio *)pending_io,
11076						0,  // sks_valid
11077						0); //retry count
11078					ctl_done(pending_io);
11079					break;
11080				}
11081			}
11082
11083			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
11084		} else {
11085			panic("Unhandled HA mode failover, LUN flags = %#x, "
11086			      "ha_mode = #%x", lun->flags, softc->ha_mode);
11087		}
11088	}
11089	ctl_pause_rtr = 0;
11090	mtx_unlock(&softc->ctl_lock);
11091}
11092#endif
11093
11094static int
11095ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11096{
11097	struct ctl_lun *lun;
11098	const struct ctl_cmd_entry *entry;
11099	uint32_t initidx, targ_lun;
11100	int retval;
11101
11102	retval = 0;
11103
11104	lun = NULL;
11105
11106	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11107	if ((targ_lun < CTL_MAX_LUNS)
11108	 && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11109		/*
11110		 * If the LUN is invalid, pretend that it doesn't exist.
11111		 * It will go away as soon as all pending I/O has been
11112		 * completed.
11113		 */
11114		mtx_lock(&lun->lun_lock);
11115		if (lun->flags & CTL_LUN_DISABLED) {
11116			mtx_unlock(&lun->lun_lock);
11117			lun = NULL;
11118			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11119			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11120		} else {
11121			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11122			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11123				lun->be_lun;
11124			if (lun->be_lun->lun_type == T_PROCESSOR) {
11125				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11126			}
11127
11128			/*
11129			 * Every I/O goes into the OOA queue for a
11130			 * particular LUN, and stays there until completion.
11131			 */
11132#ifdef CTL_TIME_IO
11133			if (TAILQ_EMPTY(&lun->ooa_queue)) {
11134				lun->idle_time += getsbinuptime() -
11135				    lun->last_busy;
11136			}
11137#endif
11138			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11139			    ooa_links);
11140		}
11141	} else {
11142		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11143		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11144	}
11145
11146	/* Get command entry and return error if it is unsuppotyed. */
11147	entry = ctl_validate_command(ctsio);
11148	if (entry == NULL) {
11149		if (lun)
11150			mtx_unlock(&lun->lun_lock);
11151		return (retval);
11152	}
11153
11154	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11155	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11156
11157	/*
11158	 * Check to see whether we can send this command to LUNs that don't
11159	 * exist.  This should pretty much only be the case for inquiry
11160	 * and request sense.  Further checks, below, really require having
11161	 * a LUN, so we can't really check the command anymore.  Just put
11162	 * it on the rtr queue.
11163	 */
11164	if (lun == NULL) {
11165		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11166			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11167			ctl_enqueue_rtr((union ctl_io *)ctsio);
11168			return (retval);
11169		}
11170
11171		ctl_set_unsupported_lun(ctsio);
11172		ctl_done((union ctl_io *)ctsio);
11173		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11174		return (retval);
11175	} else {
11176		/*
11177		 * Make sure we support this particular command on this LUN.
11178		 * e.g., we don't support writes to the control LUN.
11179		 */
11180		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11181			mtx_unlock(&lun->lun_lock);
11182			ctl_set_invalid_opcode(ctsio);
11183			ctl_done((union ctl_io *)ctsio);
11184			return (retval);
11185		}
11186	}
11187
11188	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11189
11190#ifdef CTL_WITH_CA
11191	/*
11192	 * If we've got a request sense, it'll clear the contingent
11193	 * allegiance condition.  Otherwise, if we have a CA condition for
11194	 * this initiator, clear it, because it sent down a command other
11195	 * than request sense.
11196	 */
11197	if ((ctsio->cdb[0] != REQUEST_SENSE)
11198	 && (ctl_is_set(lun->have_ca, initidx)))
11199		ctl_clear_mask(lun->have_ca, initidx);
11200#endif
11201
11202	/*
11203	 * If the command has this flag set, it handles its own unit
11204	 * attention reporting, we shouldn't do anything.  Otherwise we
11205	 * check for any pending unit attentions, and send them back to the
11206	 * initiator.  We only do this when a command initially comes in,
11207	 * not when we pull it off the blocked queue.
11208	 *
11209	 * According to SAM-3, section 5.3.2, the order that things get
11210	 * presented back to the host is basically unit attentions caused
11211	 * by some sort of reset event, busy status, reservation conflicts
11212	 * or task set full, and finally any other status.
11213	 *
11214	 * One issue here is that some of the unit attentions we report
11215	 * don't fall into the "reset" category (e.g. "reported luns data
11216	 * has changed").  So reporting it here, before the reservation
11217	 * check, may be technically wrong.  I guess the only thing to do
11218	 * would be to check for and report the reset events here, and then
11219	 * check for the other unit attention types after we check for a
11220	 * reservation conflict.
11221	 *
11222	 * XXX KDM need to fix this
11223	 */
11224	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11225		ctl_ua_type ua_type;
11226		scsi_sense_data_type sense_format;
11227
11228		if (lun->flags & CTL_LUN_SENSE_DESC)
11229			sense_format = SSD_TYPE_DESC;
11230		else
11231			sense_format = SSD_TYPE_FIXED;
11232
11233		ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11234		    sense_format);
11235		if (ua_type != CTL_UA_NONE) {
11236			mtx_unlock(&lun->lun_lock);
11237			ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11238			ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11239			ctsio->sense_len = SSD_FULL_SIZE;
11240			ctl_done((union ctl_io *)ctsio);
11241			return (retval);
11242		}
11243	}
11244
11245
11246	if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11247		mtx_unlock(&lun->lun_lock);
11248		ctl_done((union ctl_io *)ctsio);
11249		return (retval);
11250	}
11251
11252	/*
11253	 * XXX CHD this is where we want to send IO to other side if
11254	 * this LUN is secondary on this SC. We will need to make a copy
11255	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11256	 * the copy we send as FROM_OTHER.
11257	 * We also need to stuff the address of the original IO so we can
11258	 * find it easily. Something similar will need be done on the other
11259	 * side so when we are done we can find the copy.
11260	 */
11261	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11262		union ctl_ha_msg msg_info;
11263		int isc_retval;
11264
11265		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11266
11267		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11268		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11269#if 0
11270		printf("1. ctsio %p\n", ctsio);
11271#endif
11272		msg_info.hdr.serializing_sc = NULL;
11273		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11274		msg_info.scsi.tag_num = ctsio->tag_num;
11275		msg_info.scsi.tag_type = ctsio->tag_type;
11276		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11277
11278		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11279
11280		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11281		    (void *)&msg_info, sizeof(msg_info), 0)) >
11282		    CTL_HA_STATUS_SUCCESS) {
11283			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11284			       isc_retval);
11285			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11286		} else {
11287#if 0
11288			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11289#endif
11290		}
11291
11292		/*
11293		 * XXX KDM this I/O is off the incoming queue, but hasn't
11294		 * been inserted on any other queue.  We may need to come
11295		 * up with a holding queue while we wait for serialization
11296		 * so that we have an idea of what we're waiting for from
11297		 * the other side.
11298		 */
11299		mtx_unlock(&lun->lun_lock);
11300		return (retval);
11301	}
11302
11303	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11304			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11305			      ctl_ooaq, ooa_links))) {
11306	case CTL_ACTION_BLOCK:
11307		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11308		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11309				  blocked_links);
11310		mtx_unlock(&lun->lun_lock);
11311		return (retval);
11312	case CTL_ACTION_PASS:
11313	case CTL_ACTION_SKIP:
11314		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11315		mtx_unlock(&lun->lun_lock);
11316		ctl_enqueue_rtr((union ctl_io *)ctsio);
11317		break;
11318	case CTL_ACTION_OVERLAP:
11319		mtx_unlock(&lun->lun_lock);
11320		ctl_set_overlapped_cmd(ctsio);
11321		ctl_done((union ctl_io *)ctsio);
11322		break;
11323	case CTL_ACTION_OVERLAP_TAG:
11324		mtx_unlock(&lun->lun_lock);
11325		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11326		ctl_done((union ctl_io *)ctsio);
11327		break;
11328	case CTL_ACTION_ERROR:
11329	default:
11330		mtx_unlock(&lun->lun_lock);
11331		ctl_set_internal_failure(ctsio,
11332					 /*sks_valid*/ 0,
11333					 /*retry_count*/ 0);
11334		ctl_done((union ctl_io *)ctsio);
11335		break;
11336	}
11337	return (retval);
11338}
11339
11340const struct ctl_cmd_entry *
11341ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11342{
11343	const struct ctl_cmd_entry *entry;
11344	int service_action;
11345
11346	entry = &ctl_cmd_table[ctsio->cdb[0]];
11347	if (sa)
11348		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11349	if (entry->flags & CTL_CMD_FLAG_SA5) {
11350		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11351		entry = &((const struct ctl_cmd_entry *)
11352		    entry->execute)[service_action];
11353	}
11354	return (entry);
11355}
11356
11357const struct ctl_cmd_entry *
11358ctl_validate_command(struct ctl_scsiio *ctsio)
11359{
11360	const struct ctl_cmd_entry *entry;
11361	int i, sa;
11362	uint8_t diff;
11363
11364	entry = ctl_get_cmd_entry(ctsio, &sa);
11365	if (entry->execute == NULL) {
11366		if (sa)
11367			ctl_set_invalid_field(ctsio,
11368					      /*sks_valid*/ 1,
11369					      /*command*/ 1,
11370					      /*field*/ 1,
11371					      /*bit_valid*/ 1,
11372					      /*bit*/ 4);
11373		else
11374			ctl_set_invalid_opcode(ctsio);
11375		ctl_done((union ctl_io *)ctsio);
11376		return (NULL);
11377	}
11378	KASSERT(entry->length > 0,
11379	    ("Not defined length for command 0x%02x/0x%02x",
11380	     ctsio->cdb[0], ctsio->cdb[1]));
11381	for (i = 1; i < entry->length; i++) {
11382		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11383		if (diff == 0)
11384			continue;
11385		ctl_set_invalid_field(ctsio,
11386				      /*sks_valid*/ 1,
11387				      /*command*/ 1,
11388				      /*field*/ i,
11389				      /*bit_valid*/ 1,
11390				      /*bit*/ fls(diff) - 1);
11391		ctl_done((union ctl_io *)ctsio);
11392		return (NULL);
11393	}
11394	return (entry);
11395}
11396
11397static int
11398ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11399{
11400
11401	switch (lun_type) {
11402	case T_PROCESSOR:
11403		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11404		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11405			return (0);
11406		break;
11407	case T_DIRECT:
11408		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11409		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11410			return (0);
11411		break;
11412	default:
11413		return (0);
11414	}
11415	return (1);
11416}
11417
11418static int
11419ctl_scsiio(struct ctl_scsiio *ctsio)
11420{
11421	int retval;
11422	const struct ctl_cmd_entry *entry;
11423
11424	retval = CTL_RETVAL_COMPLETE;
11425
11426	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11427
11428	entry = ctl_get_cmd_entry(ctsio, NULL);
11429
11430	/*
11431	 * If this I/O has been aborted, just send it straight to
11432	 * ctl_done() without executing it.
11433	 */
11434	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11435		ctl_done((union ctl_io *)ctsio);
11436		goto bailout;
11437	}
11438
11439	/*
11440	 * All the checks should have been handled by ctl_scsiio_precheck().
11441	 * We should be clear now to just execute the I/O.
11442	 */
11443	retval = entry->execute(ctsio);
11444
11445bailout:
11446	return (retval);
11447}
11448
11449/*
11450 * Since we only implement one target right now, a bus reset simply resets
11451 * our single target.
11452 */
11453static int
11454ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11455{
11456	return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11457}
11458
11459static int
11460ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11461		 ctl_ua_type ua_type)
11462{
11463	struct ctl_lun *lun;
11464	int retval;
11465
11466	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11467		union ctl_ha_msg msg_info;
11468
11469		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11470		msg_info.hdr.nexus = io->io_hdr.nexus;
11471		if (ua_type==CTL_UA_TARG_RESET)
11472			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11473		else
11474			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11475		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11476		msg_info.hdr.original_sc = NULL;
11477		msg_info.hdr.serializing_sc = NULL;
11478		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11479		    (void *)&msg_info, sizeof(msg_info), 0)) {
11480		}
11481	}
11482	retval = 0;
11483
11484	mtx_lock(&softc->ctl_lock);
11485	STAILQ_FOREACH(lun, &softc->lun_list, links)
11486		retval += ctl_lun_reset(lun, io, ua_type);
11487	mtx_unlock(&softc->ctl_lock);
11488
11489	return (retval);
11490}
11491
11492/*
11493 * The LUN should always be set.  The I/O is optional, and is used to
11494 * distinguish between I/Os sent by this initiator, and by other
11495 * initiators.  We set unit attention for initiators other than this one.
11496 * SAM-3 is vague on this point.  It does say that a unit attention should
11497 * be established for other initiators when a LUN is reset (see section
11498 * 5.7.3), but it doesn't specifically say that the unit attention should
11499 * be established for this particular initiator when a LUN is reset.  Here
11500 * is the relevant text, from SAM-3 rev 8:
11501 *
11502 * 5.7.2 When a SCSI initiator port aborts its own tasks
11503 *
11504 * When a SCSI initiator port causes its own task(s) to be aborted, no
11505 * notification that the task(s) have been aborted shall be returned to
11506 * the SCSI initiator port other than the completion response for the
11507 * command or task management function action that caused the task(s) to
11508 * be aborted and notification(s) associated with related effects of the
11509 * action (e.g., a reset unit attention condition).
11510 *
11511 * XXX KDM for now, we're setting unit attention for all initiators.
11512 */
11513static int
11514ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11515{
11516	union ctl_io *xio;
11517#if 0
11518	uint32_t initidx;
11519#endif
11520#ifdef CTL_WITH_CA
11521	int i;
11522#endif
11523
11524	mtx_lock(&lun->lun_lock);
11525	/*
11526	 * Run through the OOA queue and abort each I/O.
11527	 */
11528	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11529	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11530		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11531	}
11532
11533	/*
11534	 * This version sets unit attention for every
11535	 */
11536#if 0
11537	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11538	ctl_est_ua_all(lun, initidx, ua_type);
11539#else
11540	ctl_est_ua_all(lun, -1, ua_type);
11541#endif
11542
11543	/*
11544	 * A reset (any kind, really) clears reservations established with
11545	 * RESERVE/RELEASE.  It does not clear reservations established
11546	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11547	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11548	 * reservations made with the RESERVE/RELEASE commands, because
11549	 * those commands are obsolete in SPC-3.
11550	 */
11551	lun->flags &= ~CTL_LUN_RESERVED;
11552
11553#ifdef CTL_WITH_CA
11554	for (i = 0; i < CTL_MAX_INITIATORS; i++)
11555		ctl_clear_mask(lun->have_ca, i);
11556#endif
11557	mtx_unlock(&lun->lun_lock);
11558
11559	return (0);
11560}
11561
11562static void
11563ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11564    int other_sc)
11565{
11566	union ctl_io *xio;
11567
11568	mtx_assert(&lun->lun_lock, MA_OWNED);
11569
11570	/*
11571	 * Run through the OOA queue and attempt to find the given I/O.
11572	 * The target port, initiator ID, tag type and tag number have to
11573	 * match the values that we got from the initiator.  If we have an
11574	 * untagged command to abort, simply abort the first untagged command
11575	 * we come to.  We only allow one untagged command at a time of course.
11576	 */
11577	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11578	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11579
11580		if ((targ_port == UINT32_MAX ||
11581		     targ_port == xio->io_hdr.nexus.targ_port) &&
11582		    (init_id == UINT32_MAX ||
11583		     init_id == xio->io_hdr.nexus.initid.id)) {
11584			if (targ_port != xio->io_hdr.nexus.targ_port ||
11585			    init_id != xio->io_hdr.nexus.initid.id)
11586				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11587			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11588			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11589				union ctl_ha_msg msg_info;
11590
11591				msg_info.hdr.nexus = xio->io_hdr.nexus;
11592				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11593				msg_info.task.tag_num = xio->scsiio.tag_num;
11594				msg_info.task.tag_type = xio->scsiio.tag_type;
11595				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11596				msg_info.hdr.original_sc = NULL;
11597				msg_info.hdr.serializing_sc = NULL;
11598				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11599				    (void *)&msg_info, sizeof(msg_info), 0);
11600			}
11601		}
11602	}
11603}
11604
11605static int
11606ctl_abort_task_set(union ctl_io *io)
11607{
11608	struct ctl_softc *softc = control_softc;
11609	struct ctl_lun *lun;
11610	uint32_t targ_lun;
11611
11612	/*
11613	 * Look up the LUN.
11614	 */
11615	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11616	mtx_lock(&softc->ctl_lock);
11617	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
11618		lun = softc->ctl_luns[targ_lun];
11619	else {
11620		mtx_unlock(&softc->ctl_lock);
11621		return (1);
11622	}
11623
11624	mtx_lock(&lun->lun_lock);
11625	mtx_unlock(&softc->ctl_lock);
11626	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11627		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11628		    io->io_hdr.nexus.initid.id,
11629		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11630	} else { /* CTL_TASK_CLEAR_TASK_SET */
11631		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11632		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11633	}
11634	mtx_unlock(&lun->lun_lock);
11635	return (0);
11636}
11637
11638static int
11639ctl_i_t_nexus_reset(union ctl_io *io)
11640{
11641	struct ctl_softc *softc = control_softc;
11642	struct ctl_lun *lun;
11643	uint32_t initidx, residx;
11644
11645	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11646	residx = ctl_get_resindex(&io->io_hdr.nexus);
11647	mtx_lock(&softc->ctl_lock);
11648	STAILQ_FOREACH(lun, &softc->lun_list, links) {
11649		mtx_lock(&lun->lun_lock);
11650		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11651		    io->io_hdr.nexus.initid.id,
11652		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11653#ifdef CTL_WITH_CA
11654		ctl_clear_mask(lun->have_ca, initidx);
11655#endif
11656		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
11657			lun->flags &= ~CTL_LUN_RESERVED;
11658		ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
11659		mtx_unlock(&lun->lun_lock);
11660	}
11661	mtx_unlock(&softc->ctl_lock);
11662	return (0);
11663}
11664
11665static int
11666ctl_abort_task(union ctl_io *io)
11667{
11668	union ctl_io *xio;
11669	struct ctl_lun *lun;
11670	struct ctl_softc *softc;
11671#if 0
11672	struct sbuf sb;
11673	char printbuf[128];
11674#endif
11675	int found;
11676	uint32_t targ_lun;
11677
11678	softc = control_softc;
11679	found = 0;
11680
11681	/*
11682	 * Look up the LUN.
11683	 */
11684	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11685	mtx_lock(&softc->ctl_lock);
11686	if ((targ_lun < CTL_MAX_LUNS)
11687	 && (softc->ctl_luns[targ_lun] != NULL))
11688		lun = softc->ctl_luns[targ_lun];
11689	else {
11690		mtx_unlock(&softc->ctl_lock);
11691		return (1);
11692	}
11693
11694#if 0
11695	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11696	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11697#endif
11698
11699	mtx_lock(&lun->lun_lock);
11700	mtx_unlock(&softc->ctl_lock);
11701	/*
11702	 * Run through the OOA queue and attempt to find the given I/O.
11703	 * The target port, initiator ID, tag type and tag number have to
11704	 * match the values that we got from the initiator.  If we have an
11705	 * untagged command to abort, simply abort the first untagged command
11706	 * we come to.  We only allow one untagged command at a time of course.
11707	 */
11708	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11709	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11710#if 0
11711		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11712
11713		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11714			    lun->lun, xio->scsiio.tag_num,
11715			    xio->scsiio.tag_type,
11716			    (xio->io_hdr.blocked_links.tqe_prev
11717			    == NULL) ? "" : " BLOCKED",
11718			    (xio->io_hdr.flags &
11719			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11720			    (xio->io_hdr.flags &
11721			    CTL_FLAG_ABORT) ? " ABORT" : "",
11722			    (xio->io_hdr.flags &
11723			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11724		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11725		sbuf_finish(&sb);
11726		printf("%s\n", sbuf_data(&sb));
11727#endif
11728
11729		if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11730		 || (xio->io_hdr.nexus.initid.id != io->io_hdr.nexus.initid.id)
11731		 || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11732			continue;
11733
11734		/*
11735		 * If the abort says that the task is untagged, the
11736		 * task in the queue must be untagged.  Otherwise,
11737		 * we just check to see whether the tag numbers
11738		 * match.  This is because the QLogic firmware
11739		 * doesn't pass back the tag type in an abort
11740		 * request.
11741		 */
11742#if 0
11743		if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11744		  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11745		 || (xio->scsiio.tag_num == io->taskio.tag_num))
11746#endif
11747		/*
11748		 * XXX KDM we've got problems with FC, because it
11749		 * doesn't send down a tag type with aborts.  So we
11750		 * can only really go by the tag number...
11751		 * This may cause problems with parallel SCSI.
11752		 * Need to figure that out!!
11753		 */
11754		if (xio->scsiio.tag_num == io->taskio.tag_num) {
11755			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11756			found = 1;
11757			if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
11758			    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11759				union ctl_ha_msg msg_info;
11760
11761				io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11762				msg_info.hdr.nexus = io->io_hdr.nexus;
11763				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11764				msg_info.task.tag_num = io->taskio.tag_num;
11765				msg_info.task.tag_type = io->taskio.tag_type;
11766				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11767				msg_info.hdr.original_sc = NULL;
11768				msg_info.hdr.serializing_sc = NULL;
11769#if 0
11770				printf("Sent Abort to other side\n");
11771#endif
11772				if (ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11773				    (void *)&msg_info, sizeof(msg_info), 0) !=
11774				    CTL_HA_STATUS_SUCCESS) {
11775				}
11776			}
11777#if 0
11778			printf("ctl_abort_task: found I/O to abort\n");
11779#endif
11780		}
11781	}
11782	mtx_unlock(&lun->lun_lock);
11783
11784	if (found == 0) {
11785		/*
11786		 * This isn't really an error.  It's entirely possible for
11787		 * the abort and command completion to cross on the wire.
11788		 * This is more of an informative/diagnostic error.
11789		 */
11790#if 0
11791		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11792		       "%d:%d:%d:%d tag %d type %d\n",
11793		       io->io_hdr.nexus.initid.id,
11794		       io->io_hdr.nexus.targ_port,
11795		       io->io_hdr.nexus.targ_target.id,
11796		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11797		       io->taskio.tag_type);
11798#endif
11799	}
11800	return (0);
11801}
11802
11803static void
11804ctl_run_task(union ctl_io *io)
11805{
11806	struct ctl_softc *softc = control_softc;
11807	int retval = 1;
11808	const char *task_desc;
11809
11810	CTL_DEBUG_PRINT(("ctl_run_task\n"));
11811
11812	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
11813	    ("ctl_run_task: Unextected io_type %d\n",
11814	     io->io_hdr.io_type));
11815
11816	task_desc = ctl_scsi_task_string(&io->taskio);
11817	if (task_desc != NULL) {
11818#ifdef NEEDTOPORT
11819		csevent_log(CSC_CTL | CSC_SHELF_SW |
11820			    CTL_TASK_REPORT,
11821			    csevent_LogType_Trace,
11822			    csevent_Severity_Information,
11823			    csevent_AlertLevel_Green,
11824			    csevent_FRU_Firmware,
11825			    csevent_FRU_Unknown,
11826			    "CTL: received task: %s",task_desc);
11827#endif
11828	} else {
11829#ifdef NEEDTOPORT
11830		csevent_log(CSC_CTL | CSC_SHELF_SW |
11831			    CTL_TASK_REPORT,
11832			    csevent_LogType_Trace,
11833			    csevent_Severity_Information,
11834			    csevent_AlertLevel_Green,
11835			    csevent_FRU_Firmware,
11836			    csevent_FRU_Unknown,
11837			    "CTL: received unknown task "
11838			    "type: %d (%#x)",
11839			    io->taskio.task_action,
11840			    io->taskio.task_action);
11841#endif
11842	}
11843	switch (io->taskio.task_action) {
11844	case CTL_TASK_ABORT_TASK:
11845		retval = ctl_abort_task(io);
11846		break;
11847	case CTL_TASK_ABORT_TASK_SET:
11848	case CTL_TASK_CLEAR_TASK_SET:
11849		retval = ctl_abort_task_set(io);
11850		break;
11851	case CTL_TASK_CLEAR_ACA:
11852		break;
11853	case CTL_TASK_I_T_NEXUS_RESET:
11854		retval = ctl_i_t_nexus_reset(io);
11855		break;
11856	case CTL_TASK_LUN_RESET: {
11857		struct ctl_lun *lun;
11858		uint32_t targ_lun;
11859
11860		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11861		mtx_lock(&softc->ctl_lock);
11862		if ((targ_lun < CTL_MAX_LUNS)
11863		 && (softc->ctl_luns[targ_lun] != NULL))
11864			lun = softc->ctl_luns[targ_lun];
11865		else {
11866			mtx_unlock(&softc->ctl_lock);
11867			retval = 1;
11868			break;
11869		}
11870
11871		if (!(io->io_hdr.flags &
11872		    CTL_FLAG_FROM_OTHER_SC)) {
11873			union ctl_ha_msg msg_info;
11874
11875			io->io_hdr.flags |=
11876				CTL_FLAG_SENT_2OTHER_SC;
11877			msg_info.hdr.msg_type =
11878				CTL_MSG_MANAGE_TASKS;
11879			msg_info.hdr.nexus = io->io_hdr.nexus;
11880			msg_info.task.task_action =
11881				CTL_TASK_LUN_RESET;
11882			msg_info.hdr.original_sc = NULL;
11883			msg_info.hdr.serializing_sc = NULL;
11884			if (CTL_HA_STATUS_SUCCESS !=
11885			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11886			    (void *)&msg_info,
11887			    sizeof(msg_info), 0)) {
11888			}
11889		}
11890
11891		retval = ctl_lun_reset(lun, io,
11892				       CTL_UA_LUN_RESET);
11893		mtx_unlock(&softc->ctl_lock);
11894		break;
11895	}
11896	case CTL_TASK_TARGET_RESET:
11897		retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
11898		break;
11899	case CTL_TASK_BUS_RESET:
11900		retval = ctl_bus_reset(softc, io);
11901		break;
11902	case CTL_TASK_PORT_LOGIN:
11903		break;
11904	case CTL_TASK_PORT_LOGOUT:
11905		break;
11906	default:
11907		printf("ctl_run_task: got unknown task management event %d\n",
11908		       io->taskio.task_action);
11909		break;
11910	}
11911	if (retval == 0)
11912		io->io_hdr.status = CTL_SUCCESS;
11913	else
11914		io->io_hdr.status = CTL_ERROR;
11915	ctl_done(io);
11916}
11917
11918/*
11919 * For HA operation.  Handle commands that come in from the other
11920 * controller.
11921 */
11922static void
11923ctl_handle_isc(union ctl_io *io)
11924{
11925	int free_io;
11926	struct ctl_lun *lun;
11927	struct ctl_softc *softc;
11928	uint32_t targ_lun;
11929
11930	softc = control_softc;
11931
11932	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11933	lun = softc->ctl_luns[targ_lun];
11934
11935	switch (io->io_hdr.msg_type) {
11936	case CTL_MSG_SERIALIZE:
11937		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
11938		break;
11939	case CTL_MSG_R2R: {
11940		const struct ctl_cmd_entry *entry;
11941
11942		/*
11943		 * This is only used in SER_ONLY mode.
11944		 */
11945		free_io = 0;
11946		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
11947		mtx_lock(&lun->lun_lock);
11948		if (ctl_scsiio_lun_check(lun,
11949		    entry, (struct ctl_scsiio *)io) != 0) {
11950			mtx_unlock(&lun->lun_lock);
11951			ctl_done(io);
11952			break;
11953		}
11954		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11955		mtx_unlock(&lun->lun_lock);
11956		ctl_enqueue_rtr(io);
11957		break;
11958	}
11959	case CTL_MSG_FINISH_IO:
11960		if (softc->ha_mode == CTL_HA_MODE_XFER) {
11961			free_io = 0;
11962			ctl_done(io);
11963		} else {
11964			free_io = 1;
11965			mtx_lock(&lun->lun_lock);
11966			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11967				     ooa_links);
11968			ctl_check_blocked(lun);
11969			mtx_unlock(&lun->lun_lock);
11970		}
11971		break;
11972	case CTL_MSG_PERS_ACTION:
11973		ctl_hndl_per_res_out_on_other_sc(
11974			(union ctl_ha_msg *)&io->presio.pr_msg);
11975		free_io = 1;
11976		break;
11977	case CTL_MSG_BAD_JUJU:
11978		free_io = 0;
11979		ctl_done(io);
11980		break;
11981	case CTL_MSG_DATAMOVE:
11982		/* Only used in XFER mode */
11983		free_io = 0;
11984		ctl_datamove_remote(io);
11985		break;
11986	case CTL_MSG_DATAMOVE_DONE:
11987		/* Only used in XFER mode */
11988		free_io = 0;
11989		io->scsiio.be_move_done(io);
11990		break;
11991	default:
11992		free_io = 1;
11993		printf("%s: Invalid message type %d\n",
11994		       __func__, io->io_hdr.msg_type);
11995		break;
11996	}
11997	if (free_io)
11998		ctl_free_io(io);
11999
12000}
12001
12002
12003/*
12004 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12005 * there is no match.
12006 */
12007static ctl_lun_error_pattern
12008ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12009{
12010	const struct ctl_cmd_entry *entry;
12011	ctl_lun_error_pattern filtered_pattern, pattern;
12012
12013	pattern = desc->error_pattern;
12014
12015	/*
12016	 * XXX KDM we need more data passed into this function to match a
12017	 * custom pattern, and we actually need to implement custom pattern
12018	 * matching.
12019	 */
12020	if (pattern & CTL_LUN_PAT_CMD)
12021		return (CTL_LUN_PAT_CMD);
12022
12023	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12024		return (CTL_LUN_PAT_ANY);
12025
12026	entry = ctl_get_cmd_entry(ctsio, NULL);
12027
12028	filtered_pattern = entry->pattern & pattern;
12029
12030	/*
12031	 * If the user requested specific flags in the pattern (e.g.
12032	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12033	 * flags.
12034	 *
12035	 * If the user did not specify any flags, it doesn't matter whether
12036	 * or not the command supports the flags.
12037	 */
12038	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12039	     (pattern & ~CTL_LUN_PAT_MASK))
12040		return (CTL_LUN_PAT_NONE);
12041
12042	/*
12043	 * If the user asked for a range check, see if the requested LBA
12044	 * range overlaps with this command's LBA range.
12045	 */
12046	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12047		uint64_t lba1;
12048		uint64_t len1;
12049		ctl_action action;
12050		int retval;
12051
12052		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12053		if (retval != 0)
12054			return (CTL_LUN_PAT_NONE);
12055
12056		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12057					      desc->lba_range.len, FALSE);
12058		/*
12059		 * A "pass" means that the LBA ranges don't overlap, so
12060		 * this doesn't match the user's range criteria.
12061		 */
12062		if (action == CTL_ACTION_PASS)
12063			return (CTL_LUN_PAT_NONE);
12064	}
12065
12066	return (filtered_pattern);
12067}
12068
12069static void
12070ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12071{
12072	struct ctl_error_desc *desc, *desc2;
12073
12074	mtx_assert(&lun->lun_lock, MA_OWNED);
12075
12076	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12077		ctl_lun_error_pattern pattern;
12078		/*
12079		 * Check to see whether this particular command matches
12080		 * the pattern in the descriptor.
12081		 */
12082		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12083		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12084			continue;
12085
12086		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12087		case CTL_LUN_INJ_ABORTED:
12088			ctl_set_aborted(&io->scsiio);
12089			break;
12090		case CTL_LUN_INJ_MEDIUM_ERR:
12091			ctl_set_medium_error(&io->scsiio);
12092			break;
12093		case CTL_LUN_INJ_UA:
12094			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12095			 * OCCURRED */
12096			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12097			break;
12098		case CTL_LUN_INJ_CUSTOM:
12099			/*
12100			 * We're assuming the user knows what he is doing.
12101			 * Just copy the sense information without doing
12102			 * checks.
12103			 */
12104			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12105			      MIN(sizeof(desc->custom_sense),
12106				  sizeof(io->scsiio.sense_data)));
12107			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12108			io->scsiio.sense_len = SSD_FULL_SIZE;
12109			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12110			break;
12111		case CTL_LUN_INJ_NONE:
12112		default:
12113			/*
12114			 * If this is an error injection type we don't know
12115			 * about, clear the continuous flag (if it is set)
12116			 * so it will get deleted below.
12117			 */
12118			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12119			break;
12120		}
12121		/*
12122		 * By default, each error injection action is a one-shot
12123		 */
12124		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12125			continue;
12126
12127		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12128
12129		free(desc, M_CTL);
12130	}
12131}
12132
12133#ifdef CTL_IO_DELAY
12134static void
12135ctl_datamove_timer_wakeup(void *arg)
12136{
12137	union ctl_io *io;
12138
12139	io = (union ctl_io *)arg;
12140
12141	ctl_datamove(io);
12142}
12143#endif /* CTL_IO_DELAY */
12144
12145void
12146ctl_datamove(union ctl_io *io)
12147{
12148	void (*fe_datamove)(union ctl_io *io);
12149
12150	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12151
12152	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12153
12154#ifdef CTL_TIME_IO
12155	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12156		char str[256];
12157		char path_str[64];
12158		struct sbuf sb;
12159
12160		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12161		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12162
12163		sbuf_cat(&sb, path_str);
12164		switch (io->io_hdr.io_type) {
12165		case CTL_IO_SCSI:
12166			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12167			sbuf_printf(&sb, "\n");
12168			sbuf_cat(&sb, path_str);
12169			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12170				    io->scsiio.tag_num, io->scsiio.tag_type);
12171			break;
12172		case CTL_IO_TASK:
12173			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12174				    "Tag Type: %d\n", io->taskio.task_action,
12175				    io->taskio.tag_num, io->taskio.tag_type);
12176			break;
12177		default:
12178			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12179			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12180			break;
12181		}
12182		sbuf_cat(&sb, path_str);
12183		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12184			    (intmax_t)time_uptime - io->io_hdr.start_time);
12185		sbuf_finish(&sb);
12186		printf("%s", sbuf_data(&sb));
12187	}
12188#endif /* CTL_TIME_IO */
12189
12190#ifdef CTL_IO_DELAY
12191	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12192		struct ctl_lun *lun;
12193
12194		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12195
12196		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12197	} else {
12198		struct ctl_lun *lun;
12199
12200		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12201		if ((lun != NULL)
12202		 && (lun->delay_info.datamove_delay > 0)) {
12203			struct callout *callout;
12204
12205			callout = (struct callout *)&io->io_hdr.timer_bytes;
12206			callout_init(callout, /*mpsafe*/ 1);
12207			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12208			callout_reset(callout,
12209				      lun->delay_info.datamove_delay * hz,
12210				      ctl_datamove_timer_wakeup, io);
12211			if (lun->delay_info.datamove_type ==
12212			    CTL_DELAY_TYPE_ONESHOT)
12213				lun->delay_info.datamove_delay = 0;
12214			return;
12215		}
12216	}
12217#endif
12218
12219	/*
12220	 * This command has been aborted.  Set the port status, so we fail
12221	 * the data move.
12222	 */
12223	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12224		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12225		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12226		       io->io_hdr.nexus.targ_port,
12227		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12228		       io->io_hdr.nexus.targ_lun);
12229		io->io_hdr.port_status = 31337;
12230		/*
12231		 * Note that the backend, in this case, will get the
12232		 * callback in its context.  In other cases it may get
12233		 * called in the frontend's interrupt thread context.
12234		 */
12235		io->scsiio.be_move_done(io);
12236		return;
12237	}
12238
12239	/* Don't confuse frontend with zero length data move. */
12240	if (io->scsiio.kern_data_len == 0) {
12241		io->scsiio.be_move_done(io);
12242		return;
12243	}
12244
12245	/*
12246	 * If we're in XFER mode and this I/O is from the other shelf
12247	 * controller, we need to send the DMA to the other side to
12248	 * actually transfer the data to/from the host.  In serialize only
12249	 * mode the transfer happens below CTL and ctl_datamove() is only
12250	 * called on the machine that originally received the I/O.
12251	 */
12252	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12253	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12254		union ctl_ha_msg msg;
12255		uint32_t sg_entries_sent;
12256		int do_sg_copy;
12257		int i;
12258
12259		memset(&msg, 0, sizeof(msg));
12260		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12261		msg.hdr.original_sc = io->io_hdr.original_sc;
12262		msg.hdr.serializing_sc = io;
12263		msg.hdr.nexus = io->io_hdr.nexus;
12264		msg.dt.flags = io->io_hdr.flags;
12265		/*
12266		 * We convert everything into a S/G list here.  We can't
12267		 * pass by reference, only by value between controllers.
12268		 * So we can't pass a pointer to the S/G list, only as many
12269		 * S/G entries as we can fit in here.  If it's possible for
12270		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12271		 * then we need to break this up into multiple transfers.
12272		 */
12273		if (io->scsiio.kern_sg_entries == 0) {
12274			msg.dt.kern_sg_entries = 1;
12275			/*
12276			 * If this is in cached memory, flush the cache
12277			 * before we send the DMA request to the other
12278			 * controller.  We want to do this in either the
12279			 * read or the write case.  The read case is
12280			 * straightforward.  In the write case, we want to
12281			 * make sure nothing is in the local cache that
12282			 * could overwrite the DMAed data.
12283			 */
12284			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12285				/*
12286				 * XXX KDM use bus_dmamap_sync() here.
12287				 */
12288			}
12289
12290			/*
12291			 * Convert to a physical address if this is a
12292			 * virtual address.
12293			 */
12294			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12295				msg.dt.sg_list[0].addr =
12296					io->scsiio.kern_data_ptr;
12297			} else {
12298				/*
12299				 * XXX KDM use busdma here!
12300				 */
12301#if 0
12302				msg.dt.sg_list[0].addr = (void *)
12303					vtophys(io->scsiio.kern_data_ptr);
12304#endif
12305			}
12306
12307			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12308			do_sg_copy = 0;
12309		} else {
12310			struct ctl_sg_entry *sgl;
12311
12312			do_sg_copy = 1;
12313			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12314			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12315			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12316				/*
12317				 * XXX KDM use bus_dmamap_sync() here.
12318				 */
12319			}
12320		}
12321
12322		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12323		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12324		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12325		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12326		msg.dt.sg_sequence = 0;
12327
12328		/*
12329		 * Loop until we've sent all of the S/G entries.  On the
12330		 * other end, we'll recompose these S/G entries into one
12331		 * contiguous list before passing it to the
12332		 */
12333		for (sg_entries_sent = 0; sg_entries_sent <
12334		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12335			msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list)/
12336				sizeof(msg.dt.sg_list[0])),
12337				msg.dt.kern_sg_entries - sg_entries_sent);
12338
12339			if (do_sg_copy != 0) {
12340				struct ctl_sg_entry *sgl;
12341				int j;
12342
12343				sgl = (struct ctl_sg_entry *)
12344					io->scsiio.kern_data_ptr;
12345				/*
12346				 * If this is in cached memory, flush the cache
12347				 * before we send the DMA request to the other
12348				 * controller.  We want to do this in either
12349				 * the * read or the write case.  The read
12350				 * case is straightforward.  In the write
12351				 * case, we want to make sure nothing is
12352				 * in the local cache that could overwrite
12353				 * the DMAed data.
12354				 */
12355
12356				for (i = sg_entries_sent, j = 0;
12357				     i < msg.dt.cur_sg_entries; i++, j++) {
12358					if ((io->io_hdr.flags &
12359					     CTL_FLAG_NO_DATASYNC) == 0) {
12360						/*
12361						 * XXX KDM use bus_dmamap_sync()
12362						 */
12363					}
12364					if ((io->io_hdr.flags &
12365					     CTL_FLAG_BUS_ADDR) == 0) {
12366						/*
12367						 * XXX KDM use busdma.
12368						 */
12369#if 0
12370						msg.dt.sg_list[j].addr =(void *)
12371						       vtophys(sgl[i].addr);
12372#endif
12373					} else {
12374						msg.dt.sg_list[j].addr =
12375							sgl[i].addr;
12376					}
12377					msg.dt.sg_list[j].len = sgl[i].len;
12378				}
12379			}
12380
12381			sg_entries_sent += msg.dt.cur_sg_entries;
12382			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12383				msg.dt.sg_last = 1;
12384			else
12385				msg.dt.sg_last = 0;
12386
12387			/*
12388			 * XXX KDM drop and reacquire the lock here?
12389			 */
12390			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12391			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12392				/*
12393				 * XXX do something here.
12394				 */
12395			}
12396
12397			msg.dt.sent_sg_entries = sg_entries_sent;
12398		}
12399		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12400		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12401			ctl_failover_io(io, /*have_lock*/ 0);
12402
12403	} else {
12404
12405		/*
12406		 * Lookup the fe_datamove() function for this particular
12407		 * front end.
12408		 */
12409		fe_datamove =
12410		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12411
12412		fe_datamove(io);
12413	}
12414}
12415
12416static void
12417ctl_send_datamove_done(union ctl_io *io, int have_lock)
12418{
12419	union ctl_ha_msg msg;
12420	int isc_status;
12421
12422	memset(&msg, 0, sizeof(msg));
12423
12424	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12425	msg.hdr.original_sc = io;
12426	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12427	msg.hdr.nexus = io->io_hdr.nexus;
12428	msg.hdr.status = io->io_hdr.status;
12429	msg.scsi.tag_num = io->scsiio.tag_num;
12430	msg.scsi.tag_type = io->scsiio.tag_type;
12431	msg.scsi.scsi_status = io->scsiio.scsi_status;
12432	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12433	       sizeof(io->scsiio.sense_data));
12434	msg.scsi.sense_len = io->scsiio.sense_len;
12435	msg.scsi.sense_residual = io->scsiio.sense_residual;
12436	msg.scsi.fetd_status = io->io_hdr.port_status;
12437	msg.scsi.residual = io->scsiio.residual;
12438	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12439
12440	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12441		ctl_failover_io(io, /*have_lock*/ have_lock);
12442		return;
12443	}
12444
12445	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12446	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12447		/* XXX do something if this fails */
12448	}
12449
12450}
12451
12452/*
12453 * The DMA to the remote side is done, now we need to tell the other side
12454 * we're done so it can continue with its data movement.
12455 */
12456static void
12457ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12458{
12459	union ctl_io *io;
12460
12461	io = rq->context;
12462
12463	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12464		printf("%s: ISC DMA write failed with error %d", __func__,
12465		       rq->ret);
12466		ctl_set_internal_failure(&io->scsiio,
12467					 /*sks_valid*/ 1,
12468					 /*retry_count*/ rq->ret);
12469	}
12470
12471	ctl_dt_req_free(rq);
12472
12473	/*
12474	 * In this case, we had to malloc the memory locally.  Free it.
12475	 */
12476	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12477		int i;
12478		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12479			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12480	}
12481	/*
12482	 * The data is in local and remote memory, so now we need to send
12483	 * status (good or back) back to the other side.
12484	 */
12485	ctl_send_datamove_done(io, /*have_lock*/ 0);
12486}
12487
12488/*
12489 * We've moved the data from the host/controller into local memory.  Now we
12490 * need to push it over to the remote controller's memory.
12491 */
12492static int
12493ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12494{
12495	int retval;
12496
12497	retval = 0;
12498
12499	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12500					  ctl_datamove_remote_write_cb);
12501
12502	return (retval);
12503}
12504
12505static void
12506ctl_datamove_remote_write(union ctl_io *io)
12507{
12508	int retval;
12509	void (*fe_datamove)(union ctl_io *io);
12510
12511	/*
12512	 * - Get the data from the host/HBA into local memory.
12513	 * - DMA memory from the local controller to the remote controller.
12514	 * - Send status back to the remote controller.
12515	 */
12516
12517	retval = ctl_datamove_remote_sgl_setup(io);
12518	if (retval != 0)
12519		return;
12520
12521	/* Switch the pointer over so the FETD knows what to do */
12522	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12523
12524	/*
12525	 * Use a custom move done callback, since we need to send completion
12526	 * back to the other controller, not to the backend on this side.
12527	 */
12528	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12529
12530	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12531
12532	fe_datamove(io);
12533
12534	return;
12535
12536}
12537
12538static int
12539ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12540{
12541#if 0
12542	char str[256];
12543	char path_str[64];
12544	struct sbuf sb;
12545#endif
12546
12547	/*
12548	 * In this case, we had to malloc the memory locally.  Free it.
12549	 */
12550	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12551		int i;
12552		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12553			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12554	}
12555
12556#if 0
12557	scsi_path_string(io, path_str, sizeof(path_str));
12558	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12559	sbuf_cat(&sb, path_str);
12560	scsi_command_string(&io->scsiio, NULL, &sb);
12561	sbuf_printf(&sb, "\n");
12562	sbuf_cat(&sb, path_str);
12563	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12564		    io->scsiio.tag_num, io->scsiio.tag_type);
12565	sbuf_cat(&sb, path_str);
12566	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12567		    io->io_hdr.flags, io->io_hdr.status);
12568	sbuf_finish(&sb);
12569	printk("%s", sbuf_data(&sb));
12570#endif
12571
12572
12573	/*
12574	 * The read is done, now we need to send status (good or bad) back
12575	 * to the other side.
12576	 */
12577	ctl_send_datamove_done(io, /*have_lock*/ 0);
12578
12579	return (0);
12580}
12581
12582static void
12583ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12584{
12585	union ctl_io *io;
12586	void (*fe_datamove)(union ctl_io *io);
12587
12588	io = rq->context;
12589
12590	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12591		printf("%s: ISC DMA read failed with error %d", __func__,
12592		       rq->ret);
12593		ctl_set_internal_failure(&io->scsiio,
12594					 /*sks_valid*/ 1,
12595					 /*retry_count*/ rq->ret);
12596	}
12597
12598	ctl_dt_req_free(rq);
12599
12600	/* Switch the pointer over so the FETD knows what to do */
12601	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12602
12603	/*
12604	 * Use a custom move done callback, since we need to send completion
12605	 * back to the other controller, not to the backend on this side.
12606	 */
12607	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12608
12609	/* XXX KDM add checks like the ones in ctl_datamove? */
12610
12611	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12612
12613	fe_datamove(io);
12614}
12615
12616static int
12617ctl_datamove_remote_sgl_setup(union ctl_io *io)
12618{
12619	struct ctl_sg_entry *local_sglist, *remote_sglist;
12620	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12621	struct ctl_softc *softc;
12622	int retval;
12623	int i;
12624
12625	retval = 0;
12626	softc = control_softc;
12627
12628	local_sglist = io->io_hdr.local_sglist;
12629	local_dma_sglist = io->io_hdr.local_dma_sglist;
12630	remote_sglist = io->io_hdr.remote_sglist;
12631	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12632
12633	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12634		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12635			local_sglist[i].len = remote_sglist[i].len;
12636
12637			/*
12638			 * XXX Detect the situation where the RS-level I/O
12639			 * redirector on the other side has already read the
12640			 * data off of the AOR RS on this side, and
12641			 * transferred it to remote (mirror) memory on the
12642			 * other side.  Since we already have the data in
12643			 * memory here, we just need to use it.
12644			 *
12645			 * XXX KDM this can probably be removed once we
12646			 * get the cache device code in and take the
12647			 * current AOR implementation out.
12648			 */
12649#ifdef NEEDTOPORT
12650			if ((remote_sglist[i].addr >=
12651			     (void *)vtophys(softc->mirr->addr))
12652			 && (remote_sglist[i].addr <
12653			     ((void *)vtophys(softc->mirr->addr) +
12654			     CacheMirrorOffset))) {
12655				local_sglist[i].addr = remote_sglist[i].addr -
12656					CacheMirrorOffset;
12657				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12658				     CTL_FLAG_DATA_IN)
12659					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12660			} else {
12661				local_sglist[i].addr = remote_sglist[i].addr +
12662					CacheMirrorOffset;
12663			}
12664#endif
12665#if 0
12666			printf("%s: local %p, remote %p, len %d\n",
12667			       __func__, local_sglist[i].addr,
12668			       remote_sglist[i].addr, local_sglist[i].len);
12669#endif
12670		}
12671	} else {
12672		uint32_t len_to_go;
12673
12674		/*
12675		 * In this case, we don't have automatically allocated
12676		 * memory for this I/O on this controller.  This typically
12677		 * happens with internal CTL I/O -- e.g. inquiry, mode
12678		 * sense, etc.  Anything coming from RAIDCore will have
12679		 * a mirror area available.
12680		 */
12681		len_to_go = io->scsiio.kern_data_len;
12682
12683		/*
12684		 * Clear the no datasync flag, we have to use malloced
12685		 * buffers.
12686		 */
12687		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12688
12689		/*
12690		 * The difficult thing here is that the size of the various
12691		 * S/G segments may be different than the size from the
12692		 * remote controller.  That'll make it harder when DMAing
12693		 * the data back to the other side.
12694		 */
12695		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12696		     sizeof(io->io_hdr.remote_sglist[0])) &&
12697		     (len_to_go > 0); i++) {
12698			local_sglist[i].len = MIN(len_to_go, 131072);
12699			CTL_SIZE_8B(local_dma_sglist[i].len,
12700				    local_sglist[i].len);
12701			local_sglist[i].addr =
12702				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
12703
12704			local_dma_sglist[i].addr = local_sglist[i].addr;
12705
12706			if (local_sglist[i].addr == NULL) {
12707				int j;
12708
12709				printf("malloc failed for %zd bytes!",
12710				       local_dma_sglist[i].len);
12711				for (j = 0; j < i; j++) {
12712					free(local_sglist[j].addr, M_CTL);
12713				}
12714				ctl_set_internal_failure(&io->scsiio,
12715							 /*sks_valid*/ 1,
12716							 /*retry_count*/ 4857);
12717				retval = 1;
12718				goto bailout_error;
12719
12720			}
12721			/* XXX KDM do we need a sync here? */
12722
12723			len_to_go -= local_sglist[i].len;
12724		}
12725		/*
12726		 * Reset the number of S/G entries accordingly.  The
12727		 * original number of S/G entries is available in
12728		 * rem_sg_entries.
12729		 */
12730		io->scsiio.kern_sg_entries = i;
12731
12732#if 0
12733		printf("%s: kern_sg_entries = %d\n", __func__,
12734		       io->scsiio.kern_sg_entries);
12735		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12736			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
12737			       local_sglist[i].addr, local_sglist[i].len,
12738			       local_dma_sglist[i].len);
12739#endif
12740	}
12741
12742
12743	return (retval);
12744
12745bailout_error:
12746
12747	ctl_send_datamove_done(io, /*have_lock*/ 0);
12748
12749	return (retval);
12750}
12751
12752static int
12753ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12754			 ctl_ha_dt_cb callback)
12755{
12756	struct ctl_ha_dt_req *rq;
12757	struct ctl_sg_entry *remote_sglist, *local_sglist;
12758	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
12759	uint32_t local_used, remote_used, total_used;
12760	int retval;
12761	int i, j;
12762
12763	retval = 0;
12764
12765	rq = ctl_dt_req_alloc();
12766
12767	/*
12768	 * If we failed to allocate the request, and if the DMA didn't fail
12769	 * anyway, set busy status.  This is just a resource allocation
12770	 * failure.
12771	 */
12772	if ((rq == NULL)
12773	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12774		ctl_set_busy(&io->scsiio);
12775
12776	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12777
12778		if (rq != NULL)
12779			ctl_dt_req_free(rq);
12780
12781		/*
12782		 * The data move failed.  We need to return status back
12783		 * to the other controller.  No point in trying to DMA
12784		 * data to the remote controller.
12785		 */
12786
12787		ctl_send_datamove_done(io, /*have_lock*/ 0);
12788
12789		retval = 1;
12790
12791		goto bailout;
12792	}
12793
12794	local_sglist = io->io_hdr.local_sglist;
12795	local_dma_sglist = io->io_hdr.local_dma_sglist;
12796	remote_sglist = io->io_hdr.remote_sglist;
12797	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12798	local_used = 0;
12799	remote_used = 0;
12800	total_used = 0;
12801
12802	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12803		rq->ret = CTL_HA_STATUS_SUCCESS;
12804		rq->context = io;
12805		callback(rq);
12806		goto bailout;
12807	}
12808
12809	/*
12810	 * Pull/push the data over the wire from/to the other controller.
12811	 * This takes into account the possibility that the local and
12812	 * remote sglists may not be identical in terms of the size of
12813	 * the elements and the number of elements.
12814	 *
12815	 * One fundamental assumption here is that the length allocated for
12816	 * both the local and remote sglists is identical.  Otherwise, we've
12817	 * essentially got a coding error of some sort.
12818	 */
12819	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12820		int isc_ret;
12821		uint32_t cur_len, dma_length;
12822		uint8_t *tmp_ptr;
12823
12824		rq->id = CTL_HA_DATA_CTL;
12825		rq->command = command;
12826		rq->context = io;
12827
12828		/*
12829		 * Both pointers should be aligned.  But it is possible
12830		 * that the allocation length is not.  They should both
12831		 * also have enough slack left over at the end, though,
12832		 * to round up to the next 8 byte boundary.
12833		 */
12834		cur_len = MIN(local_sglist[i].len - local_used,
12835			      remote_sglist[j].len - remote_used);
12836
12837		/*
12838		 * In this case, we have a size issue and need to decrease
12839		 * the size, except in the case where we actually have less
12840		 * than 8 bytes left.  In that case, we need to increase
12841		 * the DMA length to get the last bit.
12842		 */
12843		if ((cur_len & 0x7) != 0) {
12844			if (cur_len > 0x7) {
12845				cur_len = cur_len - (cur_len & 0x7);
12846				dma_length = cur_len;
12847			} else {
12848				CTL_SIZE_8B(dma_length, cur_len);
12849			}
12850
12851		} else
12852			dma_length = cur_len;
12853
12854		/*
12855		 * If we had to allocate memory for this I/O, instead of using
12856		 * the non-cached mirror memory, we'll need to flush the cache
12857		 * before trying to DMA to the other controller.
12858		 *
12859		 * We could end up doing this multiple times for the same
12860		 * segment if we have a larger local segment than remote
12861		 * segment.  That shouldn't be an issue.
12862		 */
12863		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12864			/*
12865			 * XXX KDM use bus_dmamap_sync() here.
12866			 */
12867		}
12868
12869		rq->size = dma_length;
12870
12871		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12872		tmp_ptr += local_used;
12873
12874		/* Use physical addresses when talking to ISC hardware */
12875		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12876			/* XXX KDM use busdma */
12877#if 0
12878			rq->local = vtophys(tmp_ptr);
12879#endif
12880		} else
12881			rq->local = tmp_ptr;
12882
12883		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12884		tmp_ptr += remote_used;
12885		rq->remote = tmp_ptr;
12886
12887		rq->callback = NULL;
12888
12889		local_used += cur_len;
12890		if (local_used >= local_sglist[i].len) {
12891			i++;
12892			local_used = 0;
12893		}
12894
12895		remote_used += cur_len;
12896		if (remote_used >= remote_sglist[j].len) {
12897			j++;
12898			remote_used = 0;
12899		}
12900		total_used += cur_len;
12901
12902		if (total_used >= io->scsiio.kern_data_len)
12903			rq->callback = callback;
12904
12905		if ((rq->size & 0x7) != 0) {
12906			printf("%s: warning: size %d is not on 8b boundary\n",
12907			       __func__, rq->size);
12908		}
12909		if (((uintptr_t)rq->local & 0x7) != 0) {
12910			printf("%s: warning: local %p not on 8b boundary\n",
12911			       __func__, rq->local);
12912		}
12913		if (((uintptr_t)rq->remote & 0x7) != 0) {
12914			printf("%s: warning: remote %p not on 8b boundary\n",
12915			       __func__, rq->local);
12916		}
12917#if 0
12918		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12919		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12920		       rq->local, rq->remote, rq->size);
12921#endif
12922
12923		isc_ret = ctl_dt_single(rq);
12924		if (isc_ret == CTL_HA_STATUS_WAIT)
12925			continue;
12926
12927		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12928			rq->ret = CTL_HA_STATUS_SUCCESS;
12929		} else {
12930			rq->ret = isc_ret;
12931		}
12932		callback(rq);
12933		goto bailout;
12934	}
12935
12936bailout:
12937	return (retval);
12938
12939}
12940
12941static void
12942ctl_datamove_remote_read(union ctl_io *io)
12943{
12944	int retval;
12945	int i;
12946
12947	/*
12948	 * This will send an error to the other controller in the case of a
12949	 * failure.
12950	 */
12951	retval = ctl_datamove_remote_sgl_setup(io);
12952	if (retval != 0)
12953		return;
12954
12955	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12956					  ctl_datamove_remote_read_cb);
12957	if ((retval != 0)
12958	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12959		/*
12960		 * Make sure we free memory if there was an error..  The
12961		 * ctl_datamove_remote_xfer() function will send the
12962		 * datamove done message, or call the callback with an
12963		 * error if there is a problem.
12964		 */
12965		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12966			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12967	}
12968
12969	return;
12970}
12971
12972/*
12973 * Process a datamove request from the other controller.  This is used for
12974 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12975 * first.  Once that is complete, the data gets DMAed into the remote
12976 * controller's memory.  For reads, we DMA from the remote controller's
12977 * memory into our memory first, and then move it out to the FETD.
12978 */
12979static void
12980ctl_datamove_remote(union ctl_io *io)
12981{
12982	struct ctl_softc *softc;
12983
12984	softc = control_softc;
12985
12986	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
12987
12988	/*
12989	 * Note that we look for an aborted I/O here, but don't do some of
12990	 * the other checks that ctl_datamove() normally does.
12991	 * We don't need to run the datamove delay code, since that should
12992	 * have been done if need be on the other controller.
12993	 */
12994	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12995		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12996		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12997		       io->io_hdr.nexus.targ_port,
12998		       io->io_hdr.nexus.targ_target.id,
12999		       io->io_hdr.nexus.targ_lun);
13000		io->io_hdr.port_status = 31338;
13001		ctl_send_datamove_done(io, /*have_lock*/ 0);
13002		return;
13003	}
13004
13005	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13006		ctl_datamove_remote_write(io);
13007	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13008		ctl_datamove_remote_read(io);
13009	} else {
13010		union ctl_ha_msg msg;
13011		struct scsi_sense_data *sense;
13012		uint8_t sks[3];
13013		int retry_count;
13014
13015		memset(&msg, 0, sizeof(msg));
13016
13017		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13018		msg.hdr.status = CTL_SCSI_ERROR;
13019		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13020
13021		retry_count = 4243;
13022
13023		sense = &msg.scsi.sense_data;
13024		sks[0] = SSD_SCS_VALID;
13025		sks[1] = (retry_count >> 8) & 0xff;
13026		sks[2] = retry_count & 0xff;
13027
13028		/* "Internal target failure" */
13029		scsi_set_sense_data(sense,
13030				    /*sense_format*/ SSD_TYPE_NONE,
13031				    /*current_error*/ 1,
13032				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13033				    /*asc*/ 0x44,
13034				    /*ascq*/ 0x00,
13035				    /*type*/ SSD_ELEM_SKS,
13036				    /*size*/ sizeof(sks),
13037				    /*data*/ sks,
13038				    SSD_ELEM_NONE);
13039
13040		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13041		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13042			ctl_failover_io(io, /*have_lock*/ 1);
13043			return;
13044		}
13045
13046		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13047		    CTL_HA_STATUS_SUCCESS) {
13048			/* XXX KDM what to do if this fails? */
13049		}
13050		return;
13051	}
13052
13053}
13054
13055static int
13056ctl_process_done(union ctl_io *io)
13057{
13058	struct ctl_lun *lun;
13059	struct ctl_softc *softc = control_softc;
13060	void (*fe_done)(union ctl_io *io);
13061	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13062
13063	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13064
13065	fe_done = softc->ctl_ports[targ_port]->fe_done;
13066
13067#ifdef CTL_TIME_IO
13068	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13069		char str[256];
13070		char path_str[64];
13071		struct sbuf sb;
13072
13073		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13074		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13075
13076		sbuf_cat(&sb, path_str);
13077		switch (io->io_hdr.io_type) {
13078		case CTL_IO_SCSI:
13079			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13080			sbuf_printf(&sb, "\n");
13081			sbuf_cat(&sb, path_str);
13082			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13083				    io->scsiio.tag_num, io->scsiio.tag_type);
13084			break;
13085		case CTL_IO_TASK:
13086			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13087				    "Tag Type: %d\n", io->taskio.task_action,
13088				    io->taskio.tag_num, io->taskio.tag_type);
13089			break;
13090		default:
13091			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13092			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13093			break;
13094		}
13095		sbuf_cat(&sb, path_str);
13096		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13097			    (intmax_t)time_uptime - io->io_hdr.start_time);
13098		sbuf_finish(&sb);
13099		printf("%s", sbuf_data(&sb));
13100	}
13101#endif /* CTL_TIME_IO */
13102
13103	switch (io->io_hdr.io_type) {
13104	case CTL_IO_SCSI:
13105		break;
13106	case CTL_IO_TASK:
13107		if (ctl_debug & CTL_DEBUG_INFO)
13108			ctl_io_error_print(io, NULL);
13109		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13110			ctl_free_io(io);
13111		else
13112			fe_done(io);
13113		return (CTL_RETVAL_COMPLETE);
13114	default:
13115		panic("ctl_process_done: invalid io type %d\n",
13116		      io->io_hdr.io_type);
13117		break; /* NOTREACHED */
13118	}
13119
13120	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13121	if (lun == NULL) {
13122		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13123				 io->io_hdr.nexus.targ_mapped_lun));
13124		goto bailout;
13125	}
13126
13127	mtx_lock(&lun->lun_lock);
13128
13129	/*
13130	 * Check to see if we have any errors to inject here.  We only
13131	 * inject errors for commands that don't already have errors set.
13132	 */
13133	if ((STAILQ_FIRST(&lun->error_list) != NULL) &&
13134	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13135	    ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13136		ctl_inject_error(lun, io);
13137
13138	/*
13139	 * XXX KDM how do we treat commands that aren't completed
13140	 * successfully?
13141	 *
13142	 * XXX KDM should we also track I/O latency?
13143	 */
13144	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13145	    io->io_hdr.io_type == CTL_IO_SCSI) {
13146#ifdef CTL_TIME_IO
13147		struct bintime cur_bt;
13148#endif
13149		int type;
13150
13151		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13152		    CTL_FLAG_DATA_IN)
13153			type = CTL_STATS_READ;
13154		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13155		    CTL_FLAG_DATA_OUT)
13156			type = CTL_STATS_WRITE;
13157		else
13158			type = CTL_STATS_NO_IO;
13159
13160		lun->stats.ports[targ_port].bytes[type] +=
13161		    io->scsiio.kern_total_len;
13162		lun->stats.ports[targ_port].operations[type]++;
13163#ifdef CTL_TIME_IO
13164		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13165		   &io->io_hdr.dma_bt);
13166		lun->stats.ports[targ_port].num_dmas[type] +=
13167		    io->io_hdr.num_dmas;
13168		getbintime(&cur_bt);
13169		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13170		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13171#endif
13172	}
13173
13174	/*
13175	 * Remove this from the OOA queue.
13176	 */
13177	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13178#ifdef CTL_TIME_IO
13179	if (TAILQ_EMPTY(&lun->ooa_queue))
13180		lun->last_busy = getsbinuptime();
13181#endif
13182
13183	/*
13184	 * Run through the blocked queue on this LUN and see if anything
13185	 * has become unblocked, now that this transaction is done.
13186	 */
13187	ctl_check_blocked(lun);
13188
13189	/*
13190	 * If the LUN has been invalidated, free it if there is nothing
13191	 * left on its OOA queue.
13192	 */
13193	if ((lun->flags & CTL_LUN_INVALID)
13194	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13195		mtx_unlock(&lun->lun_lock);
13196		mtx_lock(&softc->ctl_lock);
13197		ctl_free_lun(lun);
13198		mtx_unlock(&softc->ctl_lock);
13199	} else
13200		mtx_unlock(&lun->lun_lock);
13201
13202bailout:
13203
13204	/*
13205	 * If this command has been aborted, make sure we set the status
13206	 * properly.  The FETD is responsible for freeing the I/O and doing
13207	 * whatever it needs to do to clean up its state.
13208	 */
13209	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13210		ctl_set_task_aborted(&io->scsiio);
13211
13212	/*
13213	 * If enabled, print command error status.
13214	 */
13215	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13216	    (ctl_debug & CTL_DEBUG_INFO) != 0)
13217		ctl_io_error_print(io, NULL);
13218
13219	/*
13220	 * Tell the FETD or the other shelf controller we're done with this
13221	 * command.  Note that only SCSI commands get to this point.  Task
13222	 * management commands are completed above.
13223	 *
13224	 * We only send status to the other controller if we're in XFER
13225	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13226	 * received the I/O (from CTL's perspective), and so the status is
13227	 * generated there.
13228	 *
13229	 * XXX KDM if we hold the lock here, we could cause a deadlock
13230	 * if the frontend comes back in in this context to queue
13231	 * something.
13232	 */
13233	if ((softc->ha_mode == CTL_HA_MODE_XFER)
13234	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13235		union ctl_ha_msg msg;
13236
13237		memset(&msg, 0, sizeof(msg));
13238		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13239		msg.hdr.original_sc = io->io_hdr.original_sc;
13240		msg.hdr.nexus = io->io_hdr.nexus;
13241		msg.hdr.status = io->io_hdr.status;
13242		msg.scsi.scsi_status = io->scsiio.scsi_status;
13243		msg.scsi.tag_num = io->scsiio.tag_num;
13244		msg.scsi.tag_type = io->scsiio.tag_type;
13245		msg.scsi.sense_len = io->scsiio.sense_len;
13246		msg.scsi.sense_residual = io->scsiio.sense_residual;
13247		msg.scsi.residual = io->scsiio.residual;
13248		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13249		       sizeof(io->scsiio.sense_data));
13250		/*
13251		 * We copy this whether or not this is an I/O-related
13252		 * command.  Otherwise, we'd have to go and check to see
13253		 * whether it's a read/write command, and it really isn't
13254		 * worth it.
13255		 */
13256		memcpy(&msg.scsi.lbalen,
13257		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13258		       sizeof(msg.scsi.lbalen));
13259
13260		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13261				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13262			/* XXX do something here */
13263		}
13264
13265		ctl_free_io(io);
13266	} else
13267		fe_done(io);
13268
13269	return (CTL_RETVAL_COMPLETE);
13270}
13271
13272#ifdef CTL_WITH_CA
13273/*
13274 * Front end should call this if it doesn't do autosense.  When the request
13275 * sense comes back in from the initiator, we'll dequeue this and send it.
13276 */
13277int
13278ctl_queue_sense(union ctl_io *io)
13279{
13280	struct ctl_lun *lun;
13281	struct ctl_port *port;
13282	struct ctl_softc *softc;
13283	uint32_t initidx, targ_lun;
13284
13285	softc = control_softc;
13286
13287	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13288
13289	/*
13290	 * LUN lookup will likely move to the ctl_work_thread() once we
13291	 * have our new queueing infrastructure (that doesn't put things on
13292	 * a per-LUN queue initially).  That is so that we can handle
13293	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13294	 * can't deal with that right now.
13295	 */
13296	mtx_lock(&softc->ctl_lock);
13297
13298	/*
13299	 * If we don't have a LUN for this, just toss the sense
13300	 * information.
13301	 */
13302	port = ctl_io_port(&ctsio->io_hdr);
13303	targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13304	if ((targ_lun < CTL_MAX_LUNS)
13305	 && (softc->ctl_luns[targ_lun] != NULL))
13306		lun = softc->ctl_luns[targ_lun];
13307	else
13308		goto bailout;
13309
13310	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13311
13312	mtx_lock(&lun->lun_lock);
13313	/*
13314	 * Already have CA set for this LUN...toss the sense information.
13315	 */
13316	if (ctl_is_set(lun->have_ca, initidx)) {
13317		mtx_unlock(&lun->lun_lock);
13318		goto bailout;
13319	}
13320
13321	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13322	       MIN(sizeof(lun->pending_sense[initidx]),
13323	       sizeof(io->scsiio.sense_data)));
13324	ctl_set_mask(lun->have_ca, initidx);
13325	mtx_unlock(&lun->lun_lock);
13326
13327bailout:
13328	mtx_unlock(&softc->ctl_lock);
13329
13330	ctl_free_io(io);
13331
13332	return (CTL_RETVAL_COMPLETE);
13333}
13334#endif
13335
13336/*
13337 * Primary command inlet from frontend ports.  All SCSI and task I/O
13338 * requests must go through this function.
13339 */
13340int
13341ctl_queue(union ctl_io *io)
13342{
13343	struct ctl_port *port;
13344
13345	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13346
13347#ifdef CTL_TIME_IO
13348	io->io_hdr.start_time = time_uptime;
13349	getbintime(&io->io_hdr.start_bt);
13350#endif /* CTL_TIME_IO */
13351
13352	/* Map FE-specific LUN ID into global one. */
13353	port = ctl_io_port(&io->io_hdr);
13354	io->io_hdr.nexus.targ_mapped_lun =
13355	    ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13356
13357	switch (io->io_hdr.io_type) {
13358	case CTL_IO_SCSI:
13359	case CTL_IO_TASK:
13360		if (ctl_debug & CTL_DEBUG_CDB)
13361			ctl_io_print(io);
13362		ctl_enqueue_incoming(io);
13363		break;
13364	default:
13365		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13366		return (EINVAL);
13367	}
13368
13369	return (CTL_RETVAL_COMPLETE);
13370}
13371
13372#ifdef CTL_IO_DELAY
13373static void
13374ctl_done_timer_wakeup(void *arg)
13375{
13376	union ctl_io *io;
13377
13378	io = (union ctl_io *)arg;
13379	ctl_done(io);
13380}
13381#endif /* CTL_IO_DELAY */
13382
13383void
13384ctl_done(union ctl_io *io)
13385{
13386
13387	/*
13388	 * Enable this to catch duplicate completion issues.
13389	 */
13390#if 0
13391	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13392		printf("%s: type %d msg %d cdb %x iptl: "
13393		       "%d:%d:%d:%d tag 0x%04x "
13394		       "flag %#x status %x\n",
13395			__func__,
13396			io->io_hdr.io_type,
13397			io->io_hdr.msg_type,
13398			io->scsiio.cdb[0],
13399			io->io_hdr.nexus.initid.id,
13400			io->io_hdr.nexus.targ_port,
13401			io->io_hdr.nexus.targ_target.id,
13402			io->io_hdr.nexus.targ_lun,
13403			(io->io_hdr.io_type ==
13404			CTL_IO_TASK) ?
13405			io->taskio.tag_num :
13406			io->scsiio.tag_num,
13407		        io->io_hdr.flags,
13408			io->io_hdr.status);
13409	} else
13410		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13411#endif
13412
13413	/*
13414	 * This is an internal copy of an I/O, and should not go through
13415	 * the normal done processing logic.
13416	 */
13417	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13418		return;
13419
13420	/*
13421	 * We need to send a msg to the serializing shelf to finish the IO
13422	 * as well.  We don't send a finish message to the other shelf if
13423	 * this is a task management command.  Task management commands
13424	 * aren't serialized in the OOA queue, but rather just executed on
13425	 * both shelf controllers for commands that originated on that
13426	 * controller.
13427	 */
13428	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13429	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13430		union ctl_ha_msg msg_io;
13431
13432		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13433		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13434		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13435		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13436		}
13437		/* continue on to finish IO */
13438	}
13439#ifdef CTL_IO_DELAY
13440	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13441		struct ctl_lun *lun;
13442
13443		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13444
13445		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13446	} else {
13447		struct ctl_lun *lun;
13448
13449		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13450
13451		if ((lun != NULL)
13452		 && (lun->delay_info.done_delay > 0)) {
13453			struct callout *callout;
13454
13455			callout = (struct callout *)&io->io_hdr.timer_bytes;
13456			callout_init(callout, /*mpsafe*/ 1);
13457			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13458			callout_reset(callout,
13459				      lun->delay_info.done_delay * hz,
13460				      ctl_done_timer_wakeup, io);
13461			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13462				lun->delay_info.done_delay = 0;
13463			return;
13464		}
13465	}
13466#endif /* CTL_IO_DELAY */
13467
13468	ctl_enqueue_done(io);
13469}
13470
13471int
13472ctl_isc(struct ctl_scsiio *ctsio)
13473{
13474	struct ctl_lun *lun;
13475	int retval;
13476
13477	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13478
13479	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13480
13481	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13482
13483	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13484
13485	return (retval);
13486}
13487
13488
13489static void
13490ctl_work_thread(void *arg)
13491{
13492	struct ctl_thread *thr = (struct ctl_thread *)arg;
13493	struct ctl_softc *softc = thr->ctl_softc;
13494	union ctl_io *io;
13495	int retval;
13496
13497	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13498
13499	for (;;) {
13500		retval = 0;
13501
13502		/*
13503		 * We handle the queues in this order:
13504		 * - ISC
13505		 * - done queue (to free up resources, unblock other commands)
13506		 * - RtR queue
13507		 * - incoming queue
13508		 *
13509		 * If those queues are empty, we break out of the loop and
13510		 * go to sleep.
13511		 */
13512		mtx_lock(&thr->queue_lock);
13513		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13514		if (io != NULL) {
13515			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13516			mtx_unlock(&thr->queue_lock);
13517			ctl_handle_isc(io);
13518			continue;
13519		}
13520		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13521		if (io != NULL) {
13522			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13523			/* clear any blocked commands, call fe_done */
13524			mtx_unlock(&thr->queue_lock);
13525			retval = ctl_process_done(io);
13526			continue;
13527		}
13528		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13529		if (io != NULL) {
13530			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13531			mtx_unlock(&thr->queue_lock);
13532			if (io->io_hdr.io_type == CTL_IO_TASK)
13533				ctl_run_task(io);
13534			else
13535				ctl_scsiio_precheck(softc, &io->scsiio);
13536			continue;
13537		}
13538		if (!ctl_pause_rtr) {
13539			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13540			if (io != NULL) {
13541				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13542				mtx_unlock(&thr->queue_lock);
13543				retval = ctl_scsiio(&io->scsiio);
13544				if (retval != CTL_RETVAL_COMPLETE)
13545					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13546				continue;
13547			}
13548		}
13549
13550		/* Sleep until we have something to do. */
13551		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13552	}
13553}
13554
13555static void
13556ctl_lun_thread(void *arg)
13557{
13558	struct ctl_softc *softc = (struct ctl_softc *)arg;
13559	struct ctl_be_lun *be_lun;
13560	int retval;
13561
13562	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13563
13564	for (;;) {
13565		retval = 0;
13566		mtx_lock(&softc->ctl_lock);
13567		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13568		if (be_lun != NULL) {
13569			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13570			mtx_unlock(&softc->ctl_lock);
13571			ctl_create_lun(be_lun);
13572			continue;
13573		}
13574
13575		/* Sleep until we have something to do. */
13576		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13577		    PDROP | PRIBIO, "-", 0);
13578	}
13579}
13580
13581static void
13582ctl_thresh_thread(void *arg)
13583{
13584	struct ctl_softc *softc = (struct ctl_softc *)arg;
13585	struct ctl_lun *lun;
13586	struct ctl_be_lun *be_lun;
13587	struct scsi_da_rw_recovery_page *rwpage;
13588	struct ctl_logical_block_provisioning_page *page;
13589	const char *attr;
13590	uint64_t thres, val;
13591	int i, e;
13592
13593	CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13594
13595	for (;;) {
13596		mtx_lock(&softc->ctl_lock);
13597		STAILQ_FOREACH(lun, &softc->lun_list, links) {
13598			be_lun = lun->be_lun;
13599			if ((lun->flags & CTL_LUN_DISABLED) ||
13600			    (lun->flags & CTL_LUN_OFFLINE) ||
13601			    lun->backend->lun_attr == NULL)
13602				continue;
13603			rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
13604			if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
13605				continue;
13606			e = 0;
13607			page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
13608			for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13609				if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13610					continue;
13611				thres = scsi_4btoul(page->descr[i].count);
13612				thres <<= CTL_LBP_EXPONENT;
13613				switch (page->descr[i].resource) {
13614				case 0x01:
13615					attr = "blocksavail";
13616					break;
13617				case 0x02:
13618					attr = "blocksused";
13619					break;
13620				case 0xf1:
13621					attr = "poolblocksavail";
13622					break;
13623				case 0xf2:
13624					attr = "poolblocksused";
13625					break;
13626				default:
13627					continue;
13628				}
13629				mtx_unlock(&softc->ctl_lock); // XXX
13630				val = lun->backend->lun_attr(
13631				    lun->be_lun->be_lun, attr);
13632				mtx_lock(&softc->ctl_lock);
13633				if (val == UINT64_MAX)
13634					continue;
13635				if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13636				    == SLBPPD_ARMING_INC)
13637					e |= (val >= thres);
13638				else
13639					e |= (val <= thres);
13640			}
13641			mtx_lock(&lun->lun_lock);
13642			if (e) {
13643				if (lun->lasttpt == 0 ||
13644				    time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13645					lun->lasttpt = time_uptime;
13646					ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13647				}
13648			} else {
13649				lun->lasttpt = 0;
13650				ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13651			}
13652			mtx_unlock(&lun->lun_lock);
13653		}
13654		mtx_unlock(&softc->ctl_lock);
13655		pause("-", CTL_LBP_PERIOD * hz);
13656	}
13657}
13658
13659static void
13660ctl_enqueue_incoming(union ctl_io *io)
13661{
13662	struct ctl_softc *softc = control_softc;
13663	struct ctl_thread *thr;
13664	u_int idx;
13665
13666	idx = (io->io_hdr.nexus.targ_port * 127 +
13667	       io->io_hdr.nexus.initid.id) % worker_threads;
13668	thr = &softc->threads[idx];
13669	mtx_lock(&thr->queue_lock);
13670	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13671	mtx_unlock(&thr->queue_lock);
13672	wakeup(thr);
13673}
13674
13675static void
13676ctl_enqueue_rtr(union ctl_io *io)
13677{
13678	struct ctl_softc *softc = control_softc;
13679	struct ctl_thread *thr;
13680
13681	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13682	mtx_lock(&thr->queue_lock);
13683	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13684	mtx_unlock(&thr->queue_lock);
13685	wakeup(thr);
13686}
13687
13688static void
13689ctl_enqueue_done(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->done_queue, &io->io_hdr, links);
13697	mtx_unlock(&thr->queue_lock);
13698	wakeup(thr);
13699}
13700
13701#ifdef notyet
13702static void
13703ctl_enqueue_isc(union ctl_io *io)
13704{
13705	struct ctl_softc *softc = control_softc;
13706	struct ctl_thread *thr;
13707
13708	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13709	mtx_lock(&thr->queue_lock);
13710	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13711	mtx_unlock(&thr->queue_lock);
13712	wakeup(thr);
13713}
13714
13715/* Initialization and failover */
13716
13717void
13718ctl_init_isc_msg(void)
13719{
13720	printf("CTL: Still calling this thing\n");
13721}
13722
13723/*
13724 * Init component
13725 * 	Initializes component into configuration defined by bootMode
13726 *	(see hasc-sv.c)
13727 *  	returns hasc_Status:
13728 * 		OK
13729 *		ERROR - fatal error
13730 */
13731static ctl_ha_comp_status
13732ctl_isc_init(struct ctl_ha_component *c)
13733{
13734	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13735
13736	c->status = ret;
13737	return ret;
13738}
13739
13740/* Start component
13741 * 	Starts component in state requested. If component starts successfully,
13742 *	it must set its own state to the requestrd state
13743 *	When requested state is HASC_STATE_HA, the component may refine it
13744 * 	by adding _SLAVE or _MASTER flags.
13745 *	Currently allowed state transitions are:
13746 *	UNKNOWN->HA		- initial startup
13747 *	UNKNOWN->SINGLE - initial startup when no parter detected
13748 *	HA->SINGLE		- failover
13749 * returns ctl_ha_comp_status:
13750 * 		OK	- component successfully started in requested state
13751 *		FAILED  - could not start the requested state, failover may
13752 * 			  be possible
13753 *		ERROR	- fatal error detected, no future startup possible
13754 */
13755static ctl_ha_comp_status
13756ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13757{
13758	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13759
13760	printf("%s: go\n", __func__);
13761
13762	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13763	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13764		control_softc->is_single = 0;
13765		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13766		    != CTL_HA_STATUS_SUCCESS) {
13767			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13768			ret = CTL_HA_COMP_STATUS_ERROR;
13769		}
13770	} else if (CTL_HA_STATE_IS_HA(c->state)
13771		&& CTL_HA_STATE_IS_SINGLE(state)){
13772		// HA->SINGLE transition
13773	        ctl_failover();
13774		control_softc->is_single = 1;
13775	} else {
13776		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13777		       c->state, state);
13778		ret = CTL_HA_COMP_STATUS_ERROR;
13779	}
13780	if (CTL_HA_STATE_IS_SINGLE(state))
13781		control_softc->is_single = 1;
13782
13783	c->state = state;
13784	c->status = ret;
13785	return ret;
13786}
13787
13788/*
13789 * Quiesce component
13790 * The component must clear any error conditions (set status to OK) and
13791 * prepare itself to another Start call
13792 * returns ctl_ha_comp_status:
13793 * 	OK
13794 *	ERROR
13795 */
13796static ctl_ha_comp_status
13797ctl_isc_quiesce(struct ctl_ha_component *c)
13798{
13799	int ret = CTL_HA_COMP_STATUS_OK;
13800
13801	ctl_pause_rtr = 1;
13802	c->status = ret;
13803	return ret;
13804}
13805
13806struct ctl_ha_component ctl_ha_component_ctlisc =
13807{
13808	.name = "CTL ISC",
13809	.state = CTL_HA_STATE_UNKNOWN,
13810	.init = ctl_isc_init,
13811	.start = ctl_isc_start,
13812	.quiesce = ctl_isc_quiesce
13813};
13814#endif
13815
13816/*
13817 *  vim: ts=8
13818 */
13819