ctl.c revision 286806
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: head/sys/cam/ctl/ctl.c 286806 2015-08-15 13:34:38Z 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;
362SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
363    &worker_threads, 1, "Number of worker threads");
364static int ctl_debug = CTL_DEBUG_NONE;
365SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
366    &ctl_debug, 0, "Enabled debug flags");
367
368/*
369 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
370 * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
371 * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
372 * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
373 */
374#define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
375
376#ifdef notyet
377static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
378				  int param);
379static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
380#endif
381static int ctl_init(void);
382void ctl_shutdown(void);
383static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
384static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
385static void ctl_ioctl_online(void *arg);
386static void ctl_ioctl_offline(void *arg);
387static int ctl_ioctl_lun_enable(void *arg, int lun_id);
388static int ctl_ioctl_lun_disable(void *arg, int lun_id);
389static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
390static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
391static int ctl_ioctl_submit_wait(union ctl_io *io);
392static void ctl_ioctl_datamove(union ctl_io *io);
393static void ctl_ioctl_done(union ctl_io *io);
394static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
395			      struct ctl_ooa *ooa_hdr,
396			      struct ctl_ooa_entry *kern_entries);
397static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
398		     struct thread *td);
399static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
400			 struct ctl_be_lun *be_lun);
401static int ctl_free_lun(struct ctl_lun *lun);
402static void ctl_create_lun(struct ctl_be_lun *be_lun);
403static struct ctl_port * ctl_io_port(struct ctl_io_hdr *io_hdr);
404/**
405static void ctl_failover_change_pages(struct ctl_softc *softc,
406				      struct ctl_scsiio *ctsio, int master);
407**/
408
409static int ctl_do_mode_select(union ctl_io *io);
410static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
411			   uint64_t res_key, uint64_t sa_res_key,
412			   uint8_t type, uint32_t residx,
413			   struct ctl_scsiio *ctsio,
414			   struct scsi_per_res_out *cdb,
415			   struct scsi_per_res_out_parms* param);
416static void ctl_pro_preempt_other(struct ctl_lun *lun,
417				  union ctl_ha_msg *msg);
418static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
419static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
420static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
421static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
422static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
423static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
424static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
425					 int alloc_len);
426static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
427					 int alloc_len);
428static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
429static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
430static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
431static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
432static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
433static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
434    bool seq);
435static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
436static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
437    union ctl_io *pending_io, union ctl_io *ooa_io);
438static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
439				union ctl_io *starting_io);
440static int ctl_check_blocked(struct ctl_lun *lun);
441static int ctl_scsiio_lun_check(struct ctl_lun *lun,
442				const struct ctl_cmd_entry *entry,
443				struct ctl_scsiio *ctsio);
444//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
445#ifdef notyet
446static void ctl_failover(void);
447#endif
448static void ctl_clear_ua(struct ctl_softc *ctl_softc, uint32_t initidx,
449			 ctl_ua_type ua_type);
450static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
451			       struct ctl_scsiio *ctsio);
452static int ctl_scsiio(struct ctl_scsiio *ctsio);
453
454static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
455static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
456			    ctl_ua_type ua_type);
457static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
458			 ctl_ua_type ua_type);
459static int ctl_abort_task(union ctl_io *io);
460static int ctl_abort_task_set(union ctl_io *io);
461static int ctl_i_t_nexus_reset(union ctl_io *io);
462static void ctl_run_task(union ctl_io *io);
463#ifdef CTL_IO_DELAY
464static void ctl_datamove_timer_wakeup(void *arg);
465static void ctl_done_timer_wakeup(void *arg);
466#endif /* CTL_IO_DELAY */
467
468static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
469static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
470static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
471static void ctl_datamove_remote_write(union ctl_io *io);
472static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
473static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
474static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
475static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
476				    ctl_ha_dt_cb callback);
477static void ctl_datamove_remote_read(union ctl_io *io);
478static void ctl_datamove_remote(union ctl_io *io);
479static int ctl_process_done(union ctl_io *io);
480static void ctl_lun_thread(void *arg);
481static void ctl_thresh_thread(void *arg);
482static void ctl_work_thread(void *arg);
483static void ctl_enqueue_incoming(union ctl_io *io);
484static void ctl_enqueue_rtr(union ctl_io *io);
485static void ctl_enqueue_done(union ctl_io *io);
486#ifdef notyet
487static void ctl_enqueue_isc(union ctl_io *io);
488#endif
489static const struct ctl_cmd_entry *
490    ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
491static const struct ctl_cmd_entry *
492    ctl_validate_command(struct ctl_scsiio *ctsio);
493static int ctl_cmd_applicable(uint8_t lun_type,
494    const struct ctl_cmd_entry *entry);
495
496/*
497 * Load the serialization table.  This isn't very pretty, but is probably
498 * the easiest way to do it.
499 */
500#include "ctl_ser_table.c"
501
502/*
503 * We only need to define open, close and ioctl routines for this driver.
504 */
505static struct cdevsw ctl_cdevsw = {
506	.d_version =	D_VERSION,
507	.d_flags =	0,
508	.d_open =	ctl_open,
509	.d_close =	ctl_close,
510	.d_ioctl =	ctl_ioctl,
511	.d_name =	"ctl",
512};
513
514
515MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
516
517static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
518
519static moduledata_t ctl_moduledata = {
520	"ctl",
521	ctl_module_event_handler,
522	NULL
523};
524
525DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
526MODULE_VERSION(ctl, 1);
527
528static struct ctl_frontend ioctl_frontend =
529{
530	.name = "ioctl",
531};
532
533#ifdef notyet
534static void
535ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
536			    union ctl_ha_msg *msg_info)
537{
538	struct ctl_scsiio *ctsio;
539
540	if (msg_info->hdr.original_sc == NULL) {
541		printf("%s: original_sc == NULL!\n", __func__);
542		/* XXX KDM now what? */
543		return;
544	}
545
546	ctsio = &msg_info->hdr.original_sc->scsiio;
547	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
548	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
549	ctsio->io_hdr.status = msg_info->hdr.status;
550	ctsio->scsi_status = msg_info->scsi.scsi_status;
551	ctsio->sense_len = msg_info->scsi.sense_len;
552	ctsio->sense_residual = msg_info->scsi.sense_residual;
553	ctsio->residual = msg_info->scsi.residual;
554	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
555	       sizeof(ctsio->sense_data));
556	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
557	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
558	ctl_enqueue_isc((union ctl_io *)ctsio);
559}
560
561static void
562ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
563				union ctl_ha_msg *msg_info)
564{
565	struct ctl_scsiio *ctsio;
566
567	if (msg_info->hdr.serializing_sc == NULL) {
568		printf("%s: serializing_sc == NULL!\n", __func__);
569		/* XXX KDM now what? */
570		return;
571	}
572
573	ctsio = &msg_info->hdr.serializing_sc->scsiio;
574#if 0
575	/*
576	 * Attempt to catch the situation where an I/O has
577	 * been freed, and we're using it again.
578	 */
579	if (ctsio->io_hdr.io_type == 0xff) {
580		union ctl_io *tmp_io;
581		tmp_io = (union ctl_io *)ctsio;
582		printf("%s: %p use after free!\n", __func__,
583		       ctsio);
584		printf("%s: type %d msg %d cdb %x iptl: "
585		       "%d:%d:%d:%d tag 0x%04x "
586		       "flag %#x status %x\n",
587			__func__,
588			tmp_io->io_hdr.io_type,
589			tmp_io->io_hdr.msg_type,
590			tmp_io->scsiio.cdb[0],
591			tmp_io->io_hdr.nexus.initid.id,
592			tmp_io->io_hdr.nexus.targ_port,
593			tmp_io->io_hdr.nexus.targ_target.id,
594			tmp_io->io_hdr.nexus.targ_lun,
595			(tmp_io->io_hdr.io_type ==
596			CTL_IO_TASK) ?
597			tmp_io->taskio.tag_num :
598			tmp_io->scsiio.tag_num,
599		        tmp_io->io_hdr.flags,
600			tmp_io->io_hdr.status);
601	}
602#endif
603	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
604	ctl_enqueue_isc((union ctl_io *)ctsio);
605}
606
607/*
608 * ISC (Inter Shelf Communication) event handler.  Events from the HA
609 * subsystem come in here.
610 */
611static void
612ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
613{
614	struct ctl_softc *softc;
615	union ctl_io *io;
616	struct ctl_prio *presio;
617	ctl_ha_status isc_status;
618
619	softc = control_softc;
620	io = NULL;
621
622
623#if 0
624	printf("CTL: Isc Msg event %d\n", event);
625#endif
626	if (event == CTL_HA_EVT_MSG_RECV) {
627		union ctl_ha_msg msg_info;
628
629		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
630					     sizeof(msg_info), /*wait*/ 0);
631#if 0
632		printf("CTL: msg_type %d\n", msg_info.msg_type);
633#endif
634		if (isc_status != 0) {
635			printf("Error receiving message, status = %d\n",
636			       isc_status);
637			return;
638		}
639
640		switch (msg_info.hdr.msg_type) {
641		case CTL_MSG_SERIALIZE:
642#if 0
643			printf("Serialize\n");
644#endif
645			io = ctl_alloc_io_nowait(softc->othersc_pool);
646			if (io == NULL) {
647				printf("ctl_isc_event_handler: can't allocate "
648				       "ctl_io!\n");
649				/* Bad Juju */
650				/* Need to set busy and send msg back */
651				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
652				msg_info.hdr.status = CTL_SCSI_ERROR;
653				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
654				msg_info.scsi.sense_len = 0;
655			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
656				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
657				}
658				goto bailout;
659			}
660			ctl_zero_io(io);
661			// populate ctsio from msg_info
662			io->io_hdr.io_type = CTL_IO_SCSI;
663			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
664			io->io_hdr.original_sc = msg_info.hdr.original_sc;
665#if 0
666			printf("pOrig %x\n", (int)msg_info.original_sc);
667#endif
668			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
669					    CTL_FLAG_IO_ACTIVE;
670			/*
671			 * If we're in serialization-only mode, we don't
672			 * want to go through full done processing.  Thus
673			 * the COPY flag.
674			 *
675			 * XXX KDM add another flag that is more specific.
676			 */
677			if (softc->ha_mode == CTL_HA_MODE_SER_ONLY)
678				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
679			io->io_hdr.nexus = msg_info.hdr.nexus;
680#if 0
681			printf("targ %d, port %d, iid %d, lun %d\n",
682			       io->io_hdr.nexus.targ_target.id,
683			       io->io_hdr.nexus.targ_port,
684			       io->io_hdr.nexus.initid.id,
685			       io->io_hdr.nexus.targ_lun);
686#endif
687			io->scsiio.tag_num = msg_info.scsi.tag_num;
688			io->scsiio.tag_type = msg_info.scsi.tag_type;
689			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
690			       CTL_MAX_CDBLEN);
691			if (softc->ha_mode == CTL_HA_MODE_XFER) {
692				const struct ctl_cmd_entry *entry;
693
694				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
695				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
696				io->io_hdr.flags |=
697					entry->flags & CTL_FLAG_DATA_MASK;
698			}
699			ctl_enqueue_isc(io);
700			break;
701
702		/* Performed on the Originating SC, XFER mode only */
703		case CTL_MSG_DATAMOVE: {
704			struct ctl_sg_entry *sgl;
705			int i, j;
706
707			io = msg_info.hdr.original_sc;
708			if (io == NULL) {
709				printf("%s: original_sc == NULL!\n", __func__);
710				/* XXX KDM do something here */
711				break;
712			}
713			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
714			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
715			/*
716			 * Keep track of this, we need to send it back over
717			 * when the datamove is complete.
718			 */
719			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
720
721			if (msg_info.dt.sg_sequence == 0) {
722				/*
723				 * XXX KDM we use the preallocated S/G list
724				 * here, but we'll need to change this to
725				 * dynamic allocation if we need larger S/G
726				 * lists.
727				 */
728				if (msg_info.dt.kern_sg_entries >
729				    sizeof(io->io_hdr.remote_sglist) /
730				    sizeof(io->io_hdr.remote_sglist[0])) {
731					printf("%s: number of S/G entries "
732					    "needed %u > allocated num %zd\n",
733					    __func__,
734					    msg_info.dt.kern_sg_entries,
735					    sizeof(io->io_hdr.remote_sglist)/
736					    sizeof(io->io_hdr.remote_sglist[0]));
737
738					/*
739					 * XXX KDM send a message back to
740					 * the other side to shut down the
741					 * DMA.  The error will come back
742					 * through via the normal channel.
743					 */
744					break;
745				}
746				sgl = io->io_hdr.remote_sglist;
747				memset(sgl, 0,
748				       sizeof(io->io_hdr.remote_sglist));
749
750				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
751
752				io->scsiio.kern_sg_entries =
753					msg_info.dt.kern_sg_entries;
754				io->scsiio.rem_sg_entries =
755					msg_info.dt.kern_sg_entries;
756				io->scsiio.kern_data_len =
757					msg_info.dt.kern_data_len;
758				io->scsiio.kern_total_len =
759					msg_info.dt.kern_total_len;
760				io->scsiio.kern_data_resid =
761					msg_info.dt.kern_data_resid;
762				io->scsiio.kern_rel_offset =
763					msg_info.dt.kern_rel_offset;
764				/*
765				 * Clear out per-DMA flags.
766				 */
767				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
768				/*
769				 * Add per-DMA flags that are set for this
770				 * particular DMA request.
771				 */
772				io->io_hdr.flags |= msg_info.dt.flags &
773						    CTL_FLAG_RDMA_MASK;
774			} else
775				sgl = (struct ctl_sg_entry *)
776					io->scsiio.kern_data_ptr;
777
778			for (i = msg_info.dt.sent_sg_entries, j = 0;
779			     i < (msg_info.dt.sent_sg_entries +
780			     msg_info.dt.cur_sg_entries); i++, j++) {
781				sgl[i].addr = msg_info.dt.sg_list[j].addr;
782				sgl[i].len = msg_info.dt.sg_list[j].len;
783
784#if 0
785				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
786				       __func__,
787				       msg_info.dt.sg_list[j].addr,
788				       msg_info.dt.sg_list[j].len,
789				       sgl[i].addr, sgl[i].len, j, i);
790#endif
791			}
792#if 0
793			memcpy(&sgl[msg_info.dt.sent_sg_entries],
794			       msg_info.dt.sg_list,
795			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
796#endif
797
798			/*
799			 * If this is the last piece of the I/O, we've got
800			 * the full S/G list.  Queue processing in the thread.
801			 * Otherwise wait for the next piece.
802			 */
803			if (msg_info.dt.sg_last != 0)
804				ctl_enqueue_isc(io);
805			break;
806		}
807		/* Performed on the Serializing (primary) SC, XFER mode only */
808		case CTL_MSG_DATAMOVE_DONE: {
809			if (msg_info.hdr.serializing_sc == NULL) {
810				printf("%s: serializing_sc == NULL!\n",
811				       __func__);
812				/* XXX KDM now what? */
813				break;
814			}
815			/*
816			 * We grab the sense information here in case
817			 * there was a failure, so we can return status
818			 * back to the initiator.
819			 */
820			io = msg_info.hdr.serializing_sc;
821			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
822			io->io_hdr.status = msg_info.hdr.status;
823			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
824			io->scsiio.sense_len = msg_info.scsi.sense_len;
825			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
826			io->io_hdr.port_status = msg_info.scsi.fetd_status;
827			io->scsiio.residual = msg_info.scsi.residual;
828			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
829			       sizeof(io->scsiio.sense_data));
830			ctl_enqueue_isc(io);
831			break;
832		}
833
834		/* Preformed on Originating SC, SER_ONLY mode */
835		case CTL_MSG_R2R:
836			io = msg_info.hdr.original_sc;
837			if (io == NULL) {
838				printf("%s: Major Bummer\n", __func__);
839				return;
840			} else {
841#if 0
842				printf("pOrig %x\n",(int) ctsio);
843#endif
844			}
845			io->io_hdr.msg_type = CTL_MSG_R2R;
846			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
847			ctl_enqueue_isc(io);
848			break;
849
850		/*
851		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
852		 * mode.
853		 * Performed on the Originating (i.e. secondary) SC in XFER
854		 * mode
855		 */
856		case CTL_MSG_FINISH_IO:
857			if (softc->ha_mode == CTL_HA_MODE_XFER)
858				ctl_isc_handler_finish_xfer(softc,
859							    &msg_info);
860			else
861				ctl_isc_handler_finish_ser_only(softc,
862								&msg_info);
863			break;
864
865		/* Preformed on Originating SC */
866		case CTL_MSG_BAD_JUJU:
867			io = msg_info.hdr.original_sc;
868			if (io == NULL) {
869				printf("%s: Bad JUJU!, original_sc is NULL!\n",
870				       __func__);
871				break;
872			}
873			ctl_copy_sense_data(&msg_info, io);
874			/*
875			 * IO should have already been cleaned up on other
876			 * SC so clear this flag so we won't send a message
877			 * back to finish the IO there.
878			 */
879			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
880			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
881
882			/* io = msg_info.hdr.serializing_sc; */
883			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
884			ctl_enqueue_isc(io);
885			break;
886
887		/* Handle resets sent from the other side */
888		case CTL_MSG_MANAGE_TASKS: {
889			struct ctl_taskio *taskio;
890			taskio = (struct ctl_taskio *)ctl_alloc_io_nowait(
891			    softc->othersc_pool);
892			if (taskio == NULL) {
893				printf("ctl_isc_event_handler: can't allocate "
894				       "ctl_io!\n");
895				/* Bad Juju */
896				/* should I just call the proper reset func
897				   here??? */
898				goto bailout;
899			}
900			ctl_zero_io((union ctl_io *)taskio);
901			taskio->io_hdr.io_type = CTL_IO_TASK;
902			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
903			taskio->io_hdr.nexus = msg_info.hdr.nexus;
904			taskio->task_action = msg_info.task.task_action;
905			taskio->tag_num = msg_info.task.tag_num;
906			taskio->tag_type = msg_info.task.tag_type;
907#ifdef CTL_TIME_IO
908			taskio->io_hdr.start_time = time_uptime;
909			getbintime(&taskio->io_hdr.start_bt);
910#if 0
911			cs_prof_gettime(&taskio->io_hdr.start_ticks);
912#endif
913#endif /* CTL_TIME_IO */
914			ctl_run_task((union ctl_io *)taskio);
915			break;
916		}
917		/* Persistent Reserve action which needs attention */
918		case CTL_MSG_PERS_ACTION:
919			presio = (struct ctl_prio *)ctl_alloc_io_nowait(
920			    softc->othersc_pool);
921			if (presio == NULL) {
922				printf("ctl_isc_event_handler: can't allocate "
923				       "ctl_io!\n");
924				/* Bad Juju */
925				/* Need to set busy and send msg back */
926				goto bailout;
927			}
928			ctl_zero_io((union ctl_io *)presio);
929			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
930			presio->pr_msg = msg_info.pr;
931			ctl_enqueue_isc((union ctl_io *)presio);
932			break;
933		case CTL_MSG_SYNC_FE:
934			rcv_sync_msg = 1;
935			break;
936		default:
937		        printf("How did I get here?\n");
938		}
939	} else if (event == CTL_HA_EVT_MSG_SENT) {
940		if (param != CTL_HA_STATUS_SUCCESS) {
941			printf("Bad status from ctl_ha_msg_send status %d\n",
942			       param);
943		}
944		return;
945	} else if (event == CTL_HA_EVT_DISCONNECT) {
946		printf("CTL: Got a disconnect from Isc\n");
947		return;
948	} else {
949		printf("ctl_isc_event_handler: Unknown event %d\n", event);
950		return;
951	}
952
953bailout:
954	return;
955}
956
957static void
958ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
959{
960	struct scsi_sense_data *sense;
961
962	sense = &dest->scsiio.sense_data;
963	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
964	dest->scsiio.scsi_status = src->scsi.scsi_status;
965	dest->scsiio.sense_len = src->scsi.sense_len;
966	dest->io_hdr.status = src->hdr.status;
967}
968#endif
969
970static void
971ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
972{
973	ctl_ua_type *pu;
974
975	mtx_assert(&lun->lun_lock, MA_OWNED);
976	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
977	if (pu == NULL)
978		return;
979	pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
980}
981
982static void
983ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
984{
985	int i, j;
986
987	mtx_assert(&lun->lun_lock, MA_OWNED);
988	for (i = 0; i < CTL_MAX_PORTS; i++) {
989		if (lun->pending_ua[i] == NULL)
990			continue;
991		for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
992			if (i * CTL_MAX_INIT_PER_PORT + j == except)
993				continue;
994			lun->pending_ua[i][j] |= ua;
995		}
996	}
997}
998
999static void
1000ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1001{
1002	ctl_ua_type *pu;
1003
1004	mtx_assert(&lun->lun_lock, MA_OWNED);
1005	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1006	if (pu == NULL)
1007		return;
1008	pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1009}
1010
1011static void
1012ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1013{
1014	int i, j;
1015
1016	mtx_assert(&lun->lun_lock, MA_OWNED);
1017	for (i = 0; i < CTL_MAX_PORTS; i++) {
1018		if (lun->pending_ua[i] == NULL)
1019			continue;
1020		for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1021			if (i * CTL_MAX_INIT_PER_PORT + j == except)
1022				continue;
1023			lun->pending_ua[i][j] &= ~ua;
1024		}
1025	}
1026}
1027
1028static int
1029ctl_ha_state_sysctl(SYSCTL_HANDLER_ARGS)
1030{
1031	struct ctl_softc *softc = (struct ctl_softc *)arg1;
1032	struct ctl_lun *lun;
1033	int error, value;
1034
1035	if (softc->flags & CTL_FLAG_ACTIVE_SHELF)
1036		value = 0;
1037	else
1038		value = 1;
1039
1040	error = sysctl_handle_int(oidp, &value, 0, req);
1041	if ((error != 0) || (req->newptr == NULL))
1042		return (error);
1043
1044	mtx_lock(&softc->ctl_lock);
1045	if (value == 0)
1046		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1047	else
1048		softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1049	STAILQ_FOREACH(lun, &softc->lun_list, links) {
1050		mtx_lock(&lun->lun_lock);
1051		ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1052		mtx_unlock(&lun->lun_lock);
1053	}
1054	mtx_unlock(&softc->ctl_lock);
1055	return (0);
1056}
1057
1058static int
1059ctl_init(void)
1060{
1061	struct ctl_softc *softc;
1062	void *other_pool;
1063	struct ctl_port *port;
1064	int i, error, retval;
1065	//int isc_retval;
1066
1067	retval = 0;
1068	ctl_pause_rtr = 0;
1069        rcv_sync_msg = 0;
1070
1071	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1072			       M_WAITOK | M_ZERO);
1073	softc = control_softc;
1074
1075	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
1076			      "cam/ctl");
1077
1078	softc->dev->si_drv1 = softc;
1079
1080	/*
1081	 * By default, return a "bad LUN" peripheral qualifier for unknown
1082	 * LUNs.  The user can override this default using the tunable or
1083	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
1084	 */
1085	softc->inquiry_pq_no_lun = 1;
1086	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
1087			  &softc->inquiry_pq_no_lun);
1088	sysctl_ctx_init(&softc->sysctl_ctx);
1089	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1090		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1091		CTLFLAG_RD, 0, "CAM Target Layer");
1092
1093	if (softc->sysctl_tree == NULL) {
1094		printf("%s: unable to allocate sysctl tree\n", __func__);
1095		destroy_dev(softc->dev);
1096		free(control_softc, M_DEVBUF);
1097		control_softc = NULL;
1098		return (ENOMEM);
1099	}
1100
1101	SYSCTL_ADD_INT(&softc->sysctl_ctx,
1102		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
1103		       "inquiry_pq_no_lun", CTLFLAG_RW,
1104		       &softc->inquiry_pq_no_lun, 0,
1105		       "Report no lun possible for invalid LUNs");
1106
1107	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1108	softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1109	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1110	softc->open_count = 0;
1111
1112	/*
1113	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1114	 * the drive.
1115	 */
1116	softc->flags = CTL_FLAG_REAL_SYNC;
1117
1118	/*
1119	 * In Copan's HA scheme, the "master" and "slave" roles are
1120	 * figured out through the slot the controller is in.  Although it
1121	 * is an active/active system, someone has to be in charge.
1122	 */
1123	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1124	    OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1125	    "HA head ID (0 - no HA)");
1126	if (softc->ha_id == 0) {
1127		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1128		softc->is_single = 1;
1129		softc->port_offset = 0;
1130	} else
1131		softc->port_offset = (softc->ha_id - 1) * CTL_MAX_PORTS;
1132	softc->persis_offset = softc->port_offset * CTL_MAX_INIT_PER_PORT;
1133
1134	STAILQ_INIT(&softc->lun_list);
1135	STAILQ_INIT(&softc->pending_lun_queue);
1136	STAILQ_INIT(&softc->fe_list);
1137	STAILQ_INIT(&softc->port_list);
1138	STAILQ_INIT(&softc->be_list);
1139	ctl_tpc_init(softc);
1140
1141	if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1142	                    &other_pool) != 0)
1143	{
1144		printf("ctl: can't allocate %d entry other SC pool, "
1145		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1146		return (ENOMEM);
1147	}
1148	softc->othersc_pool = other_pool;
1149
1150	if (worker_threads <= 0)
1151		worker_threads = max(1, mp_ncpus / 4);
1152	if (worker_threads > CTL_MAX_THREADS)
1153		worker_threads = CTL_MAX_THREADS;
1154
1155	for (i = 0; i < worker_threads; i++) {
1156		struct ctl_thread *thr = &softc->threads[i];
1157
1158		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1159		thr->ctl_softc = softc;
1160		STAILQ_INIT(&thr->incoming_queue);
1161		STAILQ_INIT(&thr->rtr_queue);
1162		STAILQ_INIT(&thr->done_queue);
1163		STAILQ_INIT(&thr->isc_queue);
1164
1165		error = kproc_kthread_add(ctl_work_thread, thr,
1166		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1167		if (error != 0) {
1168			printf("error creating CTL work thread!\n");
1169			ctl_pool_free(other_pool);
1170			return (error);
1171		}
1172	}
1173	error = kproc_kthread_add(ctl_lun_thread, softc,
1174	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1175	if (error != 0) {
1176		printf("error creating CTL lun thread!\n");
1177		ctl_pool_free(other_pool);
1178		return (error);
1179	}
1180	error = kproc_kthread_add(ctl_thresh_thread, softc,
1181	    &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1182	if (error != 0) {
1183		printf("error creating CTL threshold thread!\n");
1184		ctl_pool_free(other_pool);
1185		return (error);
1186	}
1187
1188	/*
1189	 * Initialize the ioctl front end.
1190	 */
1191	ctl_frontend_register(&ioctl_frontend);
1192	port = &softc->ioctl_info.port;
1193	port->frontend = &ioctl_frontend;
1194	sprintf(softc->ioctl_info.port_name, "ioctl");
1195	port->port_type = CTL_PORT_IOCTL;
1196	port->num_requested_ctl_io = 100;
1197	port->port_name = softc->ioctl_info.port_name;
1198	port->port_online = ctl_ioctl_online;
1199	port->port_offline = ctl_ioctl_offline;
1200	port->onoff_arg = &softc->ioctl_info;
1201	port->lun_enable = ctl_ioctl_lun_enable;
1202	port->lun_disable = ctl_ioctl_lun_disable;
1203	port->targ_lun_arg = &softc->ioctl_info;
1204	port->fe_datamove = ctl_ioctl_datamove;
1205	port->fe_done = ctl_ioctl_done;
1206	port->max_targets = 15;
1207	port->max_target_id = 15;
1208
1209	if (ctl_port_register(&softc->ioctl_info.port) != 0) {
1210		printf("ctl: ioctl front end registration failed, will "
1211		       "continue anyway\n");
1212	}
1213
1214	SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1215	    OID_AUTO, "ha_state", CTLTYPE_INT | CTLFLAG_RWTUN,
1216	    softc, 0, ctl_ha_state_sysctl, "I", "HA state for this head");
1217
1218#ifdef CTL_IO_DELAY
1219	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1220		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1221		       sizeof(struct callout), CTL_TIMER_BYTES);
1222		return (EINVAL);
1223	}
1224#endif /* CTL_IO_DELAY */
1225
1226	return (0);
1227}
1228
1229void
1230ctl_shutdown(void)
1231{
1232	struct ctl_softc *softc;
1233	struct ctl_lun *lun, *next_lun;
1234
1235	softc = (struct ctl_softc *)control_softc;
1236
1237	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1238		printf("ctl: ioctl front end deregistration failed\n");
1239
1240	mtx_lock(&softc->ctl_lock);
1241
1242	/*
1243	 * Free up each LUN.
1244	 */
1245	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1246		next_lun = STAILQ_NEXT(lun, links);
1247		ctl_free_lun(lun);
1248	}
1249
1250	mtx_unlock(&softc->ctl_lock);
1251
1252	ctl_frontend_deregister(&ioctl_frontend);
1253
1254#if 0
1255	ctl_shutdown_thread(softc->work_thread);
1256	mtx_destroy(&softc->queue_lock);
1257#endif
1258
1259	ctl_tpc_shutdown(softc);
1260	uma_zdestroy(softc->io_zone);
1261	mtx_destroy(&softc->ctl_lock);
1262
1263	destroy_dev(softc->dev);
1264
1265	sysctl_ctx_free(&softc->sysctl_ctx);
1266
1267	free(control_softc, M_DEVBUF);
1268	control_softc = NULL;
1269}
1270
1271static int
1272ctl_module_event_handler(module_t mod, int what, void *arg)
1273{
1274
1275	switch (what) {
1276	case MOD_LOAD:
1277		return (ctl_init());
1278	case MOD_UNLOAD:
1279		return (EBUSY);
1280	default:
1281		return (EOPNOTSUPP);
1282	}
1283}
1284
1285/*
1286 * XXX KDM should we do some access checks here?  Bump a reference count to
1287 * prevent a CTL module from being unloaded while someone has it open?
1288 */
1289static int
1290ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1291{
1292	return (0);
1293}
1294
1295static int
1296ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1297{
1298	return (0);
1299}
1300
1301int
1302ctl_port_enable(ctl_port_type port_type)
1303{
1304	struct ctl_softc *softc = control_softc;
1305	struct ctl_port *port;
1306
1307	if (softc->is_single == 0) {
1308		union ctl_ha_msg msg_info;
1309		int isc_retval;
1310
1311#if 0
1312		printf("%s: HA mode, synchronizing frontend enable\n",
1313		        __func__);
1314#endif
1315		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1316	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1317		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1318			printf("Sync msg send error retval %d\n", isc_retval);
1319		}
1320		if (!rcv_sync_msg) {
1321			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1322			        sizeof(msg_info), 1);
1323		}
1324#if 0
1325        	printf("CTL:Frontend Enable\n");
1326	} else {
1327		printf("%s: single mode, skipping frontend synchronization\n",
1328		        __func__);
1329#endif
1330	}
1331
1332	STAILQ_FOREACH(port, &softc->port_list, links) {
1333		if (port_type & port->port_type)
1334		{
1335#if 0
1336			printf("port %d\n", port->targ_port);
1337#endif
1338			ctl_port_online(port);
1339		}
1340	}
1341
1342	return (0);
1343}
1344
1345int
1346ctl_port_disable(ctl_port_type port_type)
1347{
1348	struct ctl_softc *softc;
1349	struct ctl_port *port;
1350
1351	softc = control_softc;
1352
1353	STAILQ_FOREACH(port, &softc->port_list, links) {
1354		if (port_type & port->port_type)
1355			ctl_port_offline(port);
1356	}
1357
1358	return (0);
1359}
1360
1361/*
1362 * Returns 0 for success, 1 for failure.
1363 * Currently the only failure mode is if there aren't enough entries
1364 * allocated.  So, in case of a failure, look at num_entries_dropped,
1365 * reallocate and try again.
1366 */
1367int
1368ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1369	      int *num_entries_filled, int *num_entries_dropped,
1370	      ctl_port_type port_type, int no_virtual)
1371{
1372	struct ctl_softc *softc;
1373	struct ctl_port *port;
1374	int entries_dropped, entries_filled;
1375	int retval;
1376	int i;
1377
1378	softc = control_softc;
1379
1380	retval = 0;
1381	entries_filled = 0;
1382	entries_dropped = 0;
1383
1384	i = 0;
1385	mtx_lock(&softc->ctl_lock);
1386	STAILQ_FOREACH(port, &softc->port_list, links) {
1387		struct ctl_port_entry *entry;
1388
1389		if ((port->port_type & port_type) == 0)
1390			continue;
1391
1392		if ((no_virtual != 0)
1393		 && (port->virtual_port != 0))
1394			continue;
1395
1396		if (entries_filled >= num_entries_alloced) {
1397			entries_dropped++;
1398			continue;
1399		}
1400		entry = &entries[i];
1401
1402		entry->port_type = port->port_type;
1403		strlcpy(entry->port_name, port->port_name,
1404			sizeof(entry->port_name));
1405		entry->physical_port = port->physical_port;
1406		entry->virtual_port = port->virtual_port;
1407		entry->wwnn = port->wwnn;
1408		entry->wwpn = port->wwpn;
1409
1410		i++;
1411		entries_filled++;
1412	}
1413
1414	mtx_unlock(&softc->ctl_lock);
1415
1416	if (entries_dropped > 0)
1417		retval = 1;
1418
1419	*num_entries_dropped = entries_dropped;
1420	*num_entries_filled = entries_filled;
1421
1422	return (retval);
1423}
1424
1425static void
1426ctl_ioctl_online(void *arg)
1427{
1428	struct ctl_ioctl_info *ioctl_info;
1429
1430	ioctl_info = (struct ctl_ioctl_info *)arg;
1431
1432	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1433}
1434
1435static void
1436ctl_ioctl_offline(void *arg)
1437{
1438	struct ctl_ioctl_info *ioctl_info;
1439
1440	ioctl_info = (struct ctl_ioctl_info *)arg;
1441
1442	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1443}
1444
1445/*
1446 * Remove an initiator by port number and initiator ID.
1447 * Returns 0 for success, -1 for failure.
1448 */
1449int
1450ctl_remove_initiator(struct ctl_port *port, int iid)
1451{
1452	struct ctl_softc *softc = control_softc;
1453
1454	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1455
1456	if (iid > CTL_MAX_INIT_PER_PORT) {
1457		printf("%s: initiator ID %u > maximun %u!\n",
1458		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1459		return (-1);
1460	}
1461
1462	mtx_lock(&softc->ctl_lock);
1463	port->wwpn_iid[iid].in_use--;
1464	port->wwpn_iid[iid].last_use = time_uptime;
1465	mtx_unlock(&softc->ctl_lock);
1466
1467	return (0);
1468}
1469
1470/*
1471 * Add an initiator to the initiator map.
1472 * Returns iid for success, < 0 for failure.
1473 */
1474int
1475ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1476{
1477	struct ctl_softc *softc = control_softc;
1478	time_t best_time;
1479	int i, best;
1480
1481	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1482
1483	if (iid >= CTL_MAX_INIT_PER_PORT) {
1484		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1485		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1486		free(name, M_CTL);
1487		return (-1);
1488	}
1489
1490	mtx_lock(&softc->ctl_lock);
1491
1492	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1493		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1494			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1495				iid = i;
1496				break;
1497			}
1498			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1499			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1500				iid = i;
1501				break;
1502			}
1503		}
1504	}
1505
1506	if (iid < 0) {
1507		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1508			if (port->wwpn_iid[i].in_use == 0 &&
1509			    port->wwpn_iid[i].wwpn == 0 &&
1510			    port->wwpn_iid[i].name == NULL) {
1511				iid = i;
1512				break;
1513			}
1514		}
1515	}
1516
1517	if (iid < 0) {
1518		best = -1;
1519		best_time = INT32_MAX;
1520		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1521			if (port->wwpn_iid[i].in_use == 0) {
1522				if (port->wwpn_iid[i].last_use < best_time) {
1523					best = i;
1524					best_time = port->wwpn_iid[i].last_use;
1525				}
1526			}
1527		}
1528		iid = best;
1529	}
1530
1531	if (iid < 0) {
1532		mtx_unlock(&softc->ctl_lock);
1533		free(name, M_CTL);
1534		return (-2);
1535	}
1536
1537	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1538		/*
1539		 * This is not an error yet.
1540		 */
1541		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1542#if 0
1543			printf("%s: port %d iid %u WWPN %#jx arrived"
1544			    " again\n", __func__, port->targ_port,
1545			    iid, (uintmax_t)wwpn);
1546#endif
1547			goto take;
1548		}
1549		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1550		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1551#if 0
1552			printf("%s: port %d iid %u name '%s' arrived"
1553			    " again\n", __func__, port->targ_port,
1554			    iid, name);
1555#endif
1556			goto take;
1557		}
1558
1559		/*
1560		 * This is an error, but what do we do about it?  The
1561		 * driver is telling us we have a new WWPN for this
1562		 * initiator ID, so we pretty much need to use it.
1563		 */
1564		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1565		    " but WWPN %#jx '%s' is still at that address\n",
1566		    __func__, port->targ_port, iid, wwpn, name,
1567		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1568		    port->wwpn_iid[iid].name);
1569
1570		/*
1571		 * XXX KDM clear have_ca and ua_pending on each LUN for
1572		 * this initiator.
1573		 */
1574	}
1575take:
1576	free(port->wwpn_iid[iid].name, M_CTL);
1577	port->wwpn_iid[iid].name = name;
1578	port->wwpn_iid[iid].wwpn = wwpn;
1579	port->wwpn_iid[iid].in_use++;
1580	mtx_unlock(&softc->ctl_lock);
1581
1582	return (iid);
1583}
1584
1585static int
1586ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1587{
1588	int len;
1589
1590	switch (port->port_type) {
1591	case CTL_PORT_FC:
1592	{
1593		struct scsi_transportid_fcp *id =
1594		    (struct scsi_transportid_fcp *)buf;
1595		if (port->wwpn_iid[iid].wwpn == 0)
1596			return (0);
1597		memset(id, 0, sizeof(*id));
1598		id->format_protocol = SCSI_PROTO_FC;
1599		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1600		return (sizeof(*id));
1601	}
1602	case CTL_PORT_ISCSI:
1603	{
1604		struct scsi_transportid_iscsi_port *id =
1605		    (struct scsi_transportid_iscsi_port *)buf;
1606		if (port->wwpn_iid[iid].name == NULL)
1607			return (0);
1608		memset(id, 0, 256);
1609		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1610		    SCSI_PROTO_ISCSI;
1611		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1612		len = roundup2(min(len, 252), 4);
1613		scsi_ulto2b(len, id->additional_length);
1614		return (sizeof(*id) + len);
1615	}
1616	case CTL_PORT_SAS:
1617	{
1618		struct scsi_transportid_sas *id =
1619		    (struct scsi_transportid_sas *)buf;
1620		if (port->wwpn_iid[iid].wwpn == 0)
1621			return (0);
1622		memset(id, 0, sizeof(*id));
1623		id->format_protocol = SCSI_PROTO_SAS;
1624		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1625		return (sizeof(*id));
1626	}
1627	default:
1628	{
1629		struct scsi_transportid_spi *id =
1630		    (struct scsi_transportid_spi *)buf;
1631		memset(id, 0, sizeof(*id));
1632		id->format_protocol = SCSI_PROTO_SPI;
1633		scsi_ulto2b(iid, id->scsi_addr);
1634		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1635		return (sizeof(*id));
1636	}
1637	}
1638}
1639
1640static int
1641ctl_ioctl_lun_enable(void *arg, int lun_id)
1642{
1643	return (0);
1644}
1645
1646static int
1647ctl_ioctl_lun_disable(void *arg, int lun_id)
1648{
1649	return (0);
1650}
1651
1652/*
1653 * Data movement routine for the CTL ioctl frontend port.
1654 */
1655static int
1656ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1657{
1658	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1659	struct ctl_sg_entry ext_entry, kern_entry;
1660	int ext_sglen, ext_sg_entries, kern_sg_entries;
1661	int ext_sg_start, ext_offset;
1662	int len_to_copy, len_copied;
1663	int kern_watermark, ext_watermark;
1664	int ext_sglist_malloced;
1665	int i, j;
1666
1667	ext_sglist_malloced = 0;
1668	ext_sg_start = 0;
1669	ext_offset = 0;
1670
1671	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1672
1673	/*
1674	 * If this flag is set, fake the data transfer.
1675	 */
1676	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1677		ctsio->ext_data_filled = ctsio->ext_data_len;
1678		goto bailout;
1679	}
1680
1681	/*
1682	 * To simplify things here, if we have a single buffer, stick it in
1683	 * a S/G entry and just make it a single entry S/G list.
1684	 */
1685	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1686		int len_seen;
1687
1688		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1689
1690		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1691							   M_WAITOK);
1692		ext_sglist_malloced = 1;
1693		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1694				   ext_sglen) != 0) {
1695			ctl_set_internal_failure(ctsio,
1696						 /*sks_valid*/ 0,
1697						 /*retry_count*/ 0);
1698			goto bailout;
1699		}
1700		ext_sg_entries = ctsio->ext_sg_entries;
1701		len_seen = 0;
1702		for (i = 0; i < ext_sg_entries; i++) {
1703			if ((len_seen + ext_sglist[i].len) >=
1704			     ctsio->ext_data_filled) {
1705				ext_sg_start = i;
1706				ext_offset = ctsio->ext_data_filled - len_seen;
1707				break;
1708			}
1709			len_seen += ext_sglist[i].len;
1710		}
1711	} else {
1712		ext_sglist = &ext_entry;
1713		ext_sglist->addr = ctsio->ext_data_ptr;
1714		ext_sglist->len = ctsio->ext_data_len;
1715		ext_sg_entries = 1;
1716		ext_sg_start = 0;
1717		ext_offset = ctsio->ext_data_filled;
1718	}
1719
1720	if (ctsio->kern_sg_entries > 0) {
1721		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1722		kern_sg_entries = ctsio->kern_sg_entries;
1723	} else {
1724		kern_sglist = &kern_entry;
1725		kern_sglist->addr = ctsio->kern_data_ptr;
1726		kern_sglist->len = ctsio->kern_data_len;
1727		kern_sg_entries = 1;
1728	}
1729
1730
1731	kern_watermark = 0;
1732	ext_watermark = ext_offset;
1733	len_copied = 0;
1734	for (i = ext_sg_start, j = 0;
1735	     i < ext_sg_entries && j < kern_sg_entries;) {
1736		uint8_t *ext_ptr, *kern_ptr;
1737
1738		len_to_copy = MIN(ext_sglist[i].len - ext_watermark,
1739				  kern_sglist[j].len - kern_watermark);
1740
1741		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1742		ext_ptr = ext_ptr + ext_watermark;
1743		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1744			/*
1745			 * XXX KDM fix this!
1746			 */
1747			panic("need to implement bus address support");
1748#if 0
1749			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1750#endif
1751		} else
1752			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1753		kern_ptr = kern_ptr + kern_watermark;
1754
1755		kern_watermark += len_to_copy;
1756		ext_watermark += len_to_copy;
1757
1758		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1759		     CTL_FLAG_DATA_IN) {
1760			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1761					 "bytes to user\n", len_to_copy));
1762			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1763					 "to %p\n", kern_ptr, ext_ptr));
1764			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1765				ctl_set_internal_failure(ctsio,
1766							 /*sks_valid*/ 0,
1767							 /*retry_count*/ 0);
1768				goto bailout;
1769			}
1770		} else {
1771			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1772					 "bytes from user\n", len_to_copy));
1773			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1774					 "to %p\n", ext_ptr, kern_ptr));
1775			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1776				ctl_set_internal_failure(ctsio,
1777							 /*sks_valid*/ 0,
1778							 /*retry_count*/0);
1779				goto bailout;
1780			}
1781		}
1782
1783		len_copied += len_to_copy;
1784
1785		if (ext_sglist[i].len == ext_watermark) {
1786			i++;
1787			ext_watermark = 0;
1788		}
1789
1790		if (kern_sglist[j].len == kern_watermark) {
1791			j++;
1792			kern_watermark = 0;
1793		}
1794	}
1795
1796	ctsio->ext_data_filled += len_copied;
1797
1798	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1799			 "kern_sg_entries: %d\n", ext_sg_entries,
1800			 kern_sg_entries));
1801	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1802			 "kern_data_len = %d\n", ctsio->ext_data_len,
1803			 ctsio->kern_data_len));
1804
1805
1806	/* XXX KDM set residual?? */
1807bailout:
1808
1809	if (ext_sglist_malloced != 0)
1810		free(ext_sglist, M_CTL);
1811
1812	return (CTL_RETVAL_COMPLETE);
1813}
1814
1815/*
1816 * Serialize a command that went down the "wrong" side, and so was sent to
1817 * this controller for execution.  The logic is a little different than the
1818 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1819 * sent back to the other side, but in the success case, we execute the
1820 * command on this side (XFER mode) or tell the other side to execute it
1821 * (SER_ONLY mode).
1822 */
1823static int
1824ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1825{
1826	struct ctl_softc *softc;
1827	union ctl_ha_msg msg_info;
1828	struct ctl_lun *lun;
1829	int retval = 0;
1830	uint32_t targ_lun;
1831
1832	softc = control_softc;
1833
1834	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1835	lun = softc->ctl_luns[targ_lun];
1836	if (lun==NULL)
1837	{
1838		/*
1839		 * Why isn't LUN defined? The other side wouldn't
1840		 * send a cmd if the LUN is undefined.
1841		 */
1842		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1843
1844		/* "Logical unit not supported" */
1845		ctl_set_sense_data(&msg_info.scsi.sense_data,
1846				   lun,
1847				   /*sense_format*/SSD_TYPE_NONE,
1848				   /*current_error*/ 1,
1849				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1850				   /*asc*/ 0x25,
1851				   /*ascq*/ 0x00,
1852				   SSD_ELEM_NONE);
1853
1854		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1855		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1856		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1857		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1858		msg_info.hdr.serializing_sc = NULL;
1859		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1860	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1861				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1862		}
1863		return(1);
1864
1865	}
1866
1867	mtx_lock(&lun->lun_lock);
1868    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1869
1870	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1871		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1872		 ooa_links))) {
1873	case CTL_ACTION_BLOCK:
1874		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1875		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1876				  blocked_links);
1877		break;
1878	case CTL_ACTION_PASS:
1879	case CTL_ACTION_SKIP:
1880		if (softc->ha_mode == CTL_HA_MODE_XFER) {
1881			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1882			ctl_enqueue_rtr((union ctl_io *)ctsio);
1883		} else {
1884
1885			/* send msg back to other side */
1886			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1887			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1888			msg_info.hdr.msg_type = CTL_MSG_R2R;
1889#if 0
1890			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1891#endif
1892		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1893			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1894			}
1895		}
1896		break;
1897	case CTL_ACTION_OVERLAP:
1898		/* OVERLAPPED COMMANDS ATTEMPTED */
1899		ctl_set_sense_data(&msg_info.scsi.sense_data,
1900				   lun,
1901				   /*sense_format*/SSD_TYPE_NONE,
1902				   /*current_error*/ 1,
1903				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1904				   /*asc*/ 0x4E,
1905				   /*ascq*/ 0x00,
1906				   SSD_ELEM_NONE);
1907
1908		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1909		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1910		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1911		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1912		msg_info.hdr.serializing_sc = NULL;
1913		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1914#if 0
1915		printf("BAD JUJU:Major Bummer Overlap\n");
1916#endif
1917		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1918		retval = 1;
1919		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1920		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1921		}
1922		break;
1923	case CTL_ACTION_OVERLAP_TAG:
1924		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1925		ctl_set_sense_data(&msg_info.scsi.sense_data,
1926				   lun,
1927				   /*sense_format*/SSD_TYPE_NONE,
1928				   /*current_error*/ 1,
1929				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1930				   /*asc*/ 0x4D,
1931				   /*ascq*/ ctsio->tag_num & 0xff,
1932				   SSD_ELEM_NONE);
1933
1934		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1935		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1936		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1937		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1938		msg_info.hdr.serializing_sc = NULL;
1939		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1940#if 0
1941		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1942#endif
1943		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1944		retval = 1;
1945		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1946		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1947		}
1948		break;
1949	case CTL_ACTION_ERROR:
1950	default:
1951		/* "Internal target failure" */
1952		ctl_set_sense_data(&msg_info.scsi.sense_data,
1953				   lun,
1954				   /*sense_format*/SSD_TYPE_NONE,
1955				   /*current_error*/ 1,
1956				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1957				   /*asc*/ 0x44,
1958				   /*ascq*/ 0x00,
1959				   SSD_ELEM_NONE);
1960
1961		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1962		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1963		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1964		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1965		msg_info.hdr.serializing_sc = NULL;
1966		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1967#if 0
1968		printf("BAD JUJU:Major Bummer HW Error\n");
1969#endif
1970		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1971		retval = 1;
1972		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1973		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1974		}
1975		break;
1976	}
1977	mtx_unlock(&lun->lun_lock);
1978	return (retval);
1979}
1980
1981static int
1982ctl_ioctl_submit_wait(union ctl_io *io)
1983{
1984	struct ctl_fe_ioctl_params params;
1985	ctl_fe_ioctl_state last_state;
1986	int done, retval;
1987
1988	retval = 0;
1989
1990	bzero(&params, sizeof(params));
1991
1992	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1993	cv_init(&params.sem, "ctlioccv");
1994	params.state = CTL_IOCTL_INPROG;
1995	last_state = params.state;
1996
1997	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1998
1999	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
2000
2001	/* This shouldn't happen */
2002	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
2003		return (retval);
2004
2005	done = 0;
2006
2007	do {
2008		mtx_lock(&params.ioctl_mtx);
2009		/*
2010		 * Check the state here, and don't sleep if the state has
2011		 * already changed (i.e. wakeup has already occured, but we
2012		 * weren't waiting yet).
2013		 */
2014		if (params.state == last_state) {
2015			/* XXX KDM cv_wait_sig instead? */
2016			cv_wait(&params.sem, &params.ioctl_mtx);
2017		}
2018		last_state = params.state;
2019
2020		switch (params.state) {
2021		case CTL_IOCTL_INPROG:
2022			/* Why did we wake up? */
2023			/* XXX KDM error here? */
2024			mtx_unlock(&params.ioctl_mtx);
2025			break;
2026		case CTL_IOCTL_DATAMOVE:
2027			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
2028
2029			/*
2030			 * change last_state back to INPROG to avoid
2031			 * deadlock on subsequent data moves.
2032			 */
2033			params.state = last_state = CTL_IOCTL_INPROG;
2034
2035			mtx_unlock(&params.ioctl_mtx);
2036			ctl_ioctl_do_datamove(&io->scsiio);
2037			/*
2038			 * Note that in some cases, most notably writes,
2039			 * this will queue the I/O and call us back later.
2040			 * In other cases, generally reads, this routine
2041			 * will immediately call back and wake us up,
2042			 * probably using our own context.
2043			 */
2044			io->scsiio.be_move_done(io);
2045			break;
2046		case CTL_IOCTL_DONE:
2047			mtx_unlock(&params.ioctl_mtx);
2048			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
2049			done = 1;
2050			break;
2051		default:
2052			mtx_unlock(&params.ioctl_mtx);
2053			/* XXX KDM error here? */
2054			break;
2055		}
2056	} while (done == 0);
2057
2058	mtx_destroy(&params.ioctl_mtx);
2059	cv_destroy(&params.sem);
2060
2061	return (CTL_RETVAL_COMPLETE);
2062}
2063
2064static void
2065ctl_ioctl_datamove(union ctl_io *io)
2066{
2067	struct ctl_fe_ioctl_params *params;
2068
2069	params = (struct ctl_fe_ioctl_params *)
2070		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2071
2072	mtx_lock(&params->ioctl_mtx);
2073	params->state = CTL_IOCTL_DATAMOVE;
2074	cv_broadcast(&params->sem);
2075	mtx_unlock(&params->ioctl_mtx);
2076}
2077
2078static void
2079ctl_ioctl_done(union ctl_io *io)
2080{
2081	struct ctl_fe_ioctl_params *params;
2082
2083	params = (struct ctl_fe_ioctl_params *)
2084		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2085
2086	mtx_lock(&params->ioctl_mtx);
2087	params->state = CTL_IOCTL_DONE;
2088	cv_broadcast(&params->sem);
2089	mtx_unlock(&params->ioctl_mtx);
2090}
2091
2092/*
2093 * Returns 0 for success, errno for failure.
2094 */
2095static int
2096ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2097		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2098{
2099	union ctl_io *io;
2100	int retval;
2101
2102	retval = 0;
2103
2104	mtx_lock(&lun->lun_lock);
2105	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2106	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2107	     ooa_links)) {
2108		struct ctl_ooa_entry *entry;
2109
2110		/*
2111		 * If we've got more than we can fit, just count the
2112		 * remaining entries.
2113		 */
2114		if (*cur_fill_num >= ooa_hdr->alloc_num)
2115			continue;
2116
2117		entry = &kern_entries[*cur_fill_num];
2118
2119		entry->tag_num = io->scsiio.tag_num;
2120		entry->lun_num = lun->lun;
2121#ifdef CTL_TIME_IO
2122		entry->start_bt = io->io_hdr.start_bt;
2123#endif
2124		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2125		entry->cdb_len = io->scsiio.cdb_len;
2126		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2127			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2128
2129		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2130			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2131
2132		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2133			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2134
2135		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2136			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2137
2138		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2139			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2140	}
2141	mtx_unlock(&lun->lun_lock);
2142
2143	return (retval);
2144}
2145
2146static void *
2147ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2148		 size_t error_str_len)
2149{
2150	void *kptr;
2151
2152	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2153
2154	if (copyin(user_addr, kptr, len) != 0) {
2155		snprintf(error_str, error_str_len, "Error copying %d bytes "
2156			 "from user address %p to kernel address %p", len,
2157			 user_addr, kptr);
2158		free(kptr, M_CTL);
2159		return (NULL);
2160	}
2161
2162	return (kptr);
2163}
2164
2165static void
2166ctl_free_args(int num_args, struct ctl_be_arg *args)
2167{
2168	int i;
2169
2170	if (args == NULL)
2171		return;
2172
2173	for (i = 0; i < num_args; i++) {
2174		free(args[i].kname, M_CTL);
2175		free(args[i].kvalue, M_CTL);
2176	}
2177
2178	free(args, M_CTL);
2179}
2180
2181static struct ctl_be_arg *
2182ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2183		char *error_str, size_t error_str_len)
2184{
2185	struct ctl_be_arg *args;
2186	int i;
2187
2188	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2189				error_str, error_str_len);
2190
2191	if (args == NULL)
2192		goto bailout;
2193
2194	for (i = 0; i < num_args; i++) {
2195		args[i].kname = NULL;
2196		args[i].kvalue = NULL;
2197	}
2198
2199	for (i = 0; i < num_args; i++) {
2200		uint8_t *tmpptr;
2201
2202		args[i].kname = ctl_copyin_alloc(args[i].name,
2203			args[i].namelen, error_str, error_str_len);
2204		if (args[i].kname == NULL)
2205			goto bailout;
2206
2207		if (args[i].kname[args[i].namelen - 1] != '\0') {
2208			snprintf(error_str, error_str_len, "Argument %d "
2209				 "name is not NUL-terminated", i);
2210			goto bailout;
2211		}
2212
2213		if (args[i].flags & CTL_BEARG_RD) {
2214			tmpptr = ctl_copyin_alloc(args[i].value,
2215				args[i].vallen, error_str, error_str_len);
2216			if (tmpptr == NULL)
2217				goto bailout;
2218			if ((args[i].flags & CTL_BEARG_ASCII)
2219			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2220				snprintf(error_str, error_str_len, "Argument "
2221				    "%d value is not NUL-terminated", i);
2222				goto bailout;
2223			}
2224			args[i].kvalue = tmpptr;
2225		} else {
2226			args[i].kvalue = malloc(args[i].vallen,
2227			    M_CTL, M_WAITOK | M_ZERO);
2228		}
2229	}
2230
2231	return (args);
2232bailout:
2233
2234	ctl_free_args(num_args, args);
2235
2236	return (NULL);
2237}
2238
2239static void
2240ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2241{
2242	int i;
2243
2244	for (i = 0; i < num_args; i++) {
2245		if (args[i].flags & CTL_BEARG_WR)
2246			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2247	}
2248}
2249
2250/*
2251 * Escape characters that are illegal or not recommended in XML.
2252 */
2253int
2254ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2255{
2256	char *end = str + size;
2257	int retval;
2258
2259	retval = 0;
2260
2261	for (; *str && str < end; str++) {
2262		switch (*str) {
2263		case '&':
2264			retval = sbuf_printf(sb, "&amp;");
2265			break;
2266		case '>':
2267			retval = sbuf_printf(sb, "&gt;");
2268			break;
2269		case '<':
2270			retval = sbuf_printf(sb, "&lt;");
2271			break;
2272		default:
2273			retval = sbuf_putc(sb, *str);
2274			break;
2275		}
2276
2277		if (retval != 0)
2278			break;
2279
2280	}
2281
2282	return (retval);
2283}
2284
2285static void
2286ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2287{
2288	struct scsi_vpd_id_descriptor *desc;
2289	int i;
2290
2291	if (id == NULL || id->len < 4)
2292		return;
2293	desc = (struct scsi_vpd_id_descriptor *)id->data;
2294	switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2295	case SVPD_ID_TYPE_T10:
2296		sbuf_printf(sb, "t10.");
2297		break;
2298	case SVPD_ID_TYPE_EUI64:
2299		sbuf_printf(sb, "eui.");
2300		break;
2301	case SVPD_ID_TYPE_NAA:
2302		sbuf_printf(sb, "naa.");
2303		break;
2304	case SVPD_ID_TYPE_SCSI_NAME:
2305		break;
2306	}
2307	switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2308	case SVPD_ID_CODESET_BINARY:
2309		for (i = 0; i < desc->length; i++)
2310			sbuf_printf(sb, "%02x", desc->identifier[i]);
2311		break;
2312	case SVPD_ID_CODESET_ASCII:
2313		sbuf_printf(sb, "%.*s", (int)desc->length,
2314		    (char *)desc->identifier);
2315		break;
2316	case SVPD_ID_CODESET_UTF8:
2317		sbuf_printf(sb, "%s", (char *)desc->identifier);
2318		break;
2319	}
2320}
2321
2322static int
2323ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2324	  struct thread *td)
2325{
2326	struct ctl_softc *softc;
2327	int retval;
2328
2329	softc = control_softc;
2330
2331	retval = 0;
2332
2333	switch (cmd) {
2334	case CTL_IO: {
2335		union ctl_io *io;
2336		void *pool_tmp;
2337
2338		/*
2339		 * If we haven't been "enabled", don't allow any SCSI I/O
2340		 * to this FETD.
2341		 */
2342		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2343			retval = EPERM;
2344			break;
2345		}
2346
2347		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2348
2349		/*
2350		 * Need to save the pool reference so it doesn't get
2351		 * spammed by the user's ctl_io.
2352		 */
2353		pool_tmp = io->io_hdr.pool;
2354		memcpy(io, (void *)addr, sizeof(*io));
2355		io->io_hdr.pool = pool_tmp;
2356
2357		/*
2358		 * No status yet, so make sure the status is set properly.
2359		 */
2360		io->io_hdr.status = CTL_STATUS_NONE;
2361
2362		/*
2363		 * The user sets the initiator ID, target and LUN IDs.
2364		 */
2365		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2366		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2367		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2368		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2369			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2370
2371		retval = ctl_ioctl_submit_wait(io);
2372
2373		if (retval != 0) {
2374			ctl_free_io(io);
2375			break;
2376		}
2377
2378		memcpy((void *)addr, io, sizeof(*io));
2379
2380		/* return this to our pool */
2381		ctl_free_io(io);
2382
2383		break;
2384	}
2385	case CTL_ENABLE_PORT:
2386	case CTL_DISABLE_PORT:
2387	case CTL_SET_PORT_WWNS: {
2388		struct ctl_port *port;
2389		struct ctl_port_entry *entry;
2390
2391		entry = (struct ctl_port_entry *)addr;
2392
2393		mtx_lock(&softc->ctl_lock);
2394		STAILQ_FOREACH(port, &softc->port_list, links) {
2395			int action, done;
2396
2397			action = 0;
2398			done = 0;
2399
2400			if ((entry->port_type == CTL_PORT_NONE)
2401			 && (entry->targ_port == port->targ_port)) {
2402				/*
2403				 * If the user only wants to enable or
2404				 * disable or set WWNs on a specific port,
2405				 * do the operation and we're done.
2406				 */
2407				action = 1;
2408				done = 1;
2409			} else if (entry->port_type & port->port_type) {
2410				/*
2411				 * Compare the user's type mask with the
2412				 * particular frontend type to see if we
2413				 * have a match.
2414				 */
2415				action = 1;
2416				done = 0;
2417
2418				/*
2419				 * Make sure the user isn't trying to set
2420				 * WWNs on multiple ports at the same time.
2421				 */
2422				if (cmd == CTL_SET_PORT_WWNS) {
2423					printf("%s: Can't set WWNs on "
2424					       "multiple ports\n", __func__);
2425					retval = EINVAL;
2426					break;
2427				}
2428			}
2429			if (action != 0) {
2430				/*
2431				 * XXX KDM we have to drop the lock here,
2432				 * because the online/offline operations
2433				 * can potentially block.  We need to
2434				 * reference count the frontends so they
2435				 * can't go away,
2436				 */
2437				mtx_unlock(&softc->ctl_lock);
2438
2439				if (cmd == CTL_ENABLE_PORT) {
2440					ctl_port_online(port);
2441				} else if (cmd == CTL_DISABLE_PORT) {
2442					ctl_port_offline(port);
2443				}
2444
2445				mtx_lock(&softc->ctl_lock);
2446
2447				if (cmd == CTL_SET_PORT_WWNS)
2448					ctl_port_set_wwns(port,
2449					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2450					    1 : 0, entry->wwnn,
2451					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2452					    1 : 0, entry->wwpn);
2453			}
2454			if (done != 0)
2455				break;
2456		}
2457		mtx_unlock(&softc->ctl_lock);
2458		break;
2459	}
2460	case CTL_GET_PORT_LIST: {
2461		struct ctl_port *port;
2462		struct ctl_port_list *list;
2463		int i;
2464
2465		list = (struct ctl_port_list *)addr;
2466
2467		if (list->alloc_len != (list->alloc_num *
2468		    sizeof(struct ctl_port_entry))) {
2469			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2470			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2471			       "%zu\n", __func__, list->alloc_len,
2472			       list->alloc_num, sizeof(struct ctl_port_entry));
2473			retval = EINVAL;
2474			break;
2475		}
2476		list->fill_len = 0;
2477		list->fill_num = 0;
2478		list->dropped_num = 0;
2479		i = 0;
2480		mtx_lock(&softc->ctl_lock);
2481		STAILQ_FOREACH(port, &softc->port_list, links) {
2482			struct ctl_port_entry entry, *list_entry;
2483
2484			if (list->fill_num >= list->alloc_num) {
2485				list->dropped_num++;
2486				continue;
2487			}
2488
2489			entry.port_type = port->port_type;
2490			strlcpy(entry.port_name, port->port_name,
2491				sizeof(entry.port_name));
2492			entry.targ_port = port->targ_port;
2493			entry.physical_port = port->physical_port;
2494			entry.virtual_port = port->virtual_port;
2495			entry.wwnn = port->wwnn;
2496			entry.wwpn = port->wwpn;
2497			if (port->status & CTL_PORT_STATUS_ONLINE)
2498				entry.online = 1;
2499			else
2500				entry.online = 0;
2501
2502			list_entry = &list->entries[i];
2503
2504			retval = copyout(&entry, list_entry, sizeof(entry));
2505			if (retval != 0) {
2506				printf("%s: CTL_GET_PORT_LIST: copyout "
2507				       "returned %d\n", __func__, retval);
2508				break;
2509			}
2510			i++;
2511			list->fill_num++;
2512			list->fill_len += sizeof(entry);
2513		}
2514		mtx_unlock(&softc->ctl_lock);
2515
2516		/*
2517		 * If this is non-zero, we had a copyout fault, so there's
2518		 * probably no point in attempting to set the status inside
2519		 * the structure.
2520		 */
2521		if (retval != 0)
2522			break;
2523
2524		if (list->dropped_num > 0)
2525			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2526		else
2527			list->status = CTL_PORT_LIST_OK;
2528		break;
2529	}
2530	case CTL_DUMP_OOA: {
2531		struct ctl_lun *lun;
2532		union ctl_io *io;
2533		char printbuf[128];
2534		struct sbuf sb;
2535
2536		mtx_lock(&softc->ctl_lock);
2537		printf("Dumping OOA queues:\n");
2538		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2539			mtx_lock(&lun->lun_lock);
2540			for (io = (union ctl_io *)TAILQ_FIRST(
2541			     &lun->ooa_queue); io != NULL;
2542			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2543			     ooa_links)) {
2544				sbuf_new(&sb, printbuf, sizeof(printbuf),
2545					 SBUF_FIXEDLEN);
2546				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2547					    (intmax_t)lun->lun,
2548					    io->scsiio.tag_num,
2549					    (io->io_hdr.flags &
2550					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2551					    (io->io_hdr.flags &
2552					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2553					    (io->io_hdr.flags &
2554					    CTL_FLAG_ABORT) ? " ABORT" : "",
2555			                    (io->io_hdr.flags &
2556		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2557				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2558				sbuf_finish(&sb);
2559				printf("%s\n", sbuf_data(&sb));
2560			}
2561			mtx_unlock(&lun->lun_lock);
2562		}
2563		printf("OOA queues dump done\n");
2564		mtx_unlock(&softc->ctl_lock);
2565		break;
2566	}
2567	case CTL_GET_OOA: {
2568		struct ctl_lun *lun;
2569		struct ctl_ooa *ooa_hdr;
2570		struct ctl_ooa_entry *entries;
2571		uint32_t cur_fill_num;
2572
2573		ooa_hdr = (struct ctl_ooa *)addr;
2574
2575		if ((ooa_hdr->alloc_len == 0)
2576		 || (ooa_hdr->alloc_num == 0)) {
2577			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2578			       "must be non-zero\n", __func__,
2579			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2580			retval = EINVAL;
2581			break;
2582		}
2583
2584		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2585		    sizeof(struct ctl_ooa_entry))) {
2586			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2587			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2588			       __func__, ooa_hdr->alloc_len,
2589			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2590			retval = EINVAL;
2591			break;
2592		}
2593
2594		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2595		if (entries == NULL) {
2596			printf("%s: could not allocate %d bytes for OOA "
2597			       "dump\n", __func__, ooa_hdr->alloc_len);
2598			retval = ENOMEM;
2599			break;
2600		}
2601
2602		mtx_lock(&softc->ctl_lock);
2603		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2604		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2605		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2606			mtx_unlock(&softc->ctl_lock);
2607			free(entries, M_CTL);
2608			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2609			       __func__, (uintmax_t)ooa_hdr->lun_num);
2610			retval = EINVAL;
2611			break;
2612		}
2613
2614		cur_fill_num = 0;
2615
2616		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2617			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2618				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2619					ooa_hdr, entries);
2620				if (retval != 0)
2621					break;
2622			}
2623			if (retval != 0) {
2624				mtx_unlock(&softc->ctl_lock);
2625				free(entries, M_CTL);
2626				break;
2627			}
2628		} else {
2629			lun = softc->ctl_luns[ooa_hdr->lun_num];
2630
2631			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2632						    entries);
2633		}
2634		mtx_unlock(&softc->ctl_lock);
2635
2636		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2637		ooa_hdr->fill_len = ooa_hdr->fill_num *
2638			sizeof(struct ctl_ooa_entry);
2639		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2640		if (retval != 0) {
2641			printf("%s: error copying out %d bytes for OOA dump\n",
2642			       __func__, ooa_hdr->fill_len);
2643		}
2644
2645		getbintime(&ooa_hdr->cur_bt);
2646
2647		if (cur_fill_num > ooa_hdr->alloc_num) {
2648			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2649			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2650		} else {
2651			ooa_hdr->dropped_num = 0;
2652			ooa_hdr->status = CTL_OOA_OK;
2653		}
2654
2655		free(entries, M_CTL);
2656		break;
2657	}
2658	case CTL_CHECK_OOA: {
2659		union ctl_io *io;
2660		struct ctl_lun *lun;
2661		struct ctl_ooa_info *ooa_info;
2662
2663
2664		ooa_info = (struct ctl_ooa_info *)addr;
2665
2666		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2667			ooa_info->status = CTL_OOA_INVALID_LUN;
2668			break;
2669		}
2670		mtx_lock(&softc->ctl_lock);
2671		lun = softc->ctl_luns[ooa_info->lun_id];
2672		if (lun == NULL) {
2673			mtx_unlock(&softc->ctl_lock);
2674			ooa_info->status = CTL_OOA_INVALID_LUN;
2675			break;
2676		}
2677		mtx_lock(&lun->lun_lock);
2678		mtx_unlock(&softc->ctl_lock);
2679		ooa_info->num_entries = 0;
2680		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2681		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2682		     &io->io_hdr, ooa_links)) {
2683			ooa_info->num_entries++;
2684		}
2685		mtx_unlock(&lun->lun_lock);
2686
2687		ooa_info->status = CTL_OOA_SUCCESS;
2688
2689		break;
2690	}
2691	case CTL_DELAY_IO: {
2692		struct ctl_io_delay_info *delay_info;
2693#ifdef CTL_IO_DELAY
2694		struct ctl_lun *lun;
2695#endif /* CTL_IO_DELAY */
2696
2697		delay_info = (struct ctl_io_delay_info *)addr;
2698
2699#ifdef CTL_IO_DELAY
2700		mtx_lock(&softc->ctl_lock);
2701
2702		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2703		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2704			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2705		} else {
2706			lun = softc->ctl_luns[delay_info->lun_id];
2707			mtx_lock(&lun->lun_lock);
2708
2709			delay_info->status = CTL_DELAY_STATUS_OK;
2710
2711			switch (delay_info->delay_type) {
2712			case CTL_DELAY_TYPE_CONT:
2713				break;
2714			case CTL_DELAY_TYPE_ONESHOT:
2715				break;
2716			default:
2717				delay_info->status =
2718					CTL_DELAY_STATUS_INVALID_TYPE;
2719				break;
2720			}
2721
2722			switch (delay_info->delay_loc) {
2723			case CTL_DELAY_LOC_DATAMOVE:
2724				lun->delay_info.datamove_type =
2725					delay_info->delay_type;
2726				lun->delay_info.datamove_delay =
2727					delay_info->delay_secs;
2728				break;
2729			case CTL_DELAY_LOC_DONE:
2730				lun->delay_info.done_type =
2731					delay_info->delay_type;
2732				lun->delay_info.done_delay =
2733					delay_info->delay_secs;
2734				break;
2735			default:
2736				delay_info->status =
2737					CTL_DELAY_STATUS_INVALID_LOC;
2738				break;
2739			}
2740			mtx_unlock(&lun->lun_lock);
2741		}
2742
2743		mtx_unlock(&softc->ctl_lock);
2744#else
2745		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2746#endif /* CTL_IO_DELAY */
2747		break;
2748	}
2749	case CTL_REALSYNC_SET: {
2750		int *syncstate;
2751
2752		syncstate = (int *)addr;
2753
2754		mtx_lock(&softc->ctl_lock);
2755		switch (*syncstate) {
2756		case 0:
2757			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2758			break;
2759		case 1:
2760			softc->flags |= CTL_FLAG_REAL_SYNC;
2761			break;
2762		default:
2763			retval = EINVAL;
2764			break;
2765		}
2766		mtx_unlock(&softc->ctl_lock);
2767		break;
2768	}
2769	case CTL_REALSYNC_GET: {
2770		int *syncstate;
2771
2772		syncstate = (int*)addr;
2773
2774		mtx_lock(&softc->ctl_lock);
2775		if (softc->flags & CTL_FLAG_REAL_SYNC)
2776			*syncstate = 1;
2777		else
2778			*syncstate = 0;
2779		mtx_unlock(&softc->ctl_lock);
2780
2781		break;
2782	}
2783	case CTL_SETSYNC:
2784	case CTL_GETSYNC: {
2785		struct ctl_sync_info *sync_info;
2786		struct ctl_lun *lun;
2787
2788		sync_info = (struct ctl_sync_info *)addr;
2789
2790		mtx_lock(&softc->ctl_lock);
2791		lun = softc->ctl_luns[sync_info->lun_id];
2792		if (lun == NULL) {
2793			mtx_unlock(&softc->ctl_lock);
2794			sync_info->status = CTL_GS_SYNC_NO_LUN;
2795		}
2796		/*
2797		 * Get or set the sync interval.  We're not bounds checking
2798		 * in the set case, hopefully the user won't do something
2799		 * silly.
2800		 */
2801		mtx_lock(&lun->lun_lock);
2802		mtx_unlock(&softc->ctl_lock);
2803		if (cmd == CTL_GETSYNC)
2804			sync_info->sync_interval = lun->sync_interval;
2805		else
2806			lun->sync_interval = sync_info->sync_interval;
2807		mtx_unlock(&lun->lun_lock);
2808
2809		sync_info->status = CTL_GS_SYNC_OK;
2810
2811		break;
2812	}
2813	case CTL_GETSTATS: {
2814		struct ctl_stats *stats;
2815		struct ctl_lun *lun;
2816		int i;
2817
2818		stats = (struct ctl_stats *)addr;
2819
2820		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2821		     stats->alloc_len) {
2822			stats->status = CTL_SS_NEED_MORE_SPACE;
2823			stats->num_luns = softc->num_luns;
2824			break;
2825		}
2826		/*
2827		 * XXX KDM no locking here.  If the LUN list changes,
2828		 * things can blow up.
2829		 */
2830		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2831		     i++, lun = STAILQ_NEXT(lun, links)) {
2832			retval = copyout(&lun->stats, &stats->lun_stats[i],
2833					 sizeof(lun->stats));
2834			if (retval != 0)
2835				break;
2836		}
2837		stats->num_luns = softc->num_luns;
2838		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2839				 softc->num_luns;
2840		stats->status = CTL_SS_OK;
2841#ifdef CTL_TIME_IO
2842		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2843#else
2844		stats->flags = CTL_STATS_FLAG_NONE;
2845#endif
2846		getnanouptime(&stats->timestamp);
2847		break;
2848	}
2849	case CTL_ERROR_INJECT: {
2850		struct ctl_error_desc *err_desc, *new_err_desc;
2851		struct ctl_lun *lun;
2852
2853		err_desc = (struct ctl_error_desc *)addr;
2854
2855		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2856				      M_WAITOK | M_ZERO);
2857		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2858
2859		mtx_lock(&softc->ctl_lock);
2860		lun = softc->ctl_luns[err_desc->lun_id];
2861		if (lun == NULL) {
2862			mtx_unlock(&softc->ctl_lock);
2863			free(new_err_desc, M_CTL);
2864			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2865			       __func__, (uintmax_t)err_desc->lun_id);
2866			retval = EINVAL;
2867			break;
2868		}
2869		mtx_lock(&lun->lun_lock);
2870		mtx_unlock(&softc->ctl_lock);
2871
2872		/*
2873		 * We could do some checking here to verify the validity
2874		 * of the request, but given the complexity of error
2875		 * injection requests, the checking logic would be fairly
2876		 * complex.
2877		 *
2878		 * For now, if the request is invalid, it just won't get
2879		 * executed and might get deleted.
2880		 */
2881		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2882
2883		/*
2884		 * XXX KDM check to make sure the serial number is unique,
2885		 * in case we somehow manage to wrap.  That shouldn't
2886		 * happen for a very long time, but it's the right thing to
2887		 * do.
2888		 */
2889		new_err_desc->serial = lun->error_serial;
2890		err_desc->serial = lun->error_serial;
2891		lun->error_serial++;
2892
2893		mtx_unlock(&lun->lun_lock);
2894		break;
2895	}
2896	case CTL_ERROR_INJECT_DELETE: {
2897		struct ctl_error_desc *delete_desc, *desc, *desc2;
2898		struct ctl_lun *lun;
2899		int delete_done;
2900
2901		delete_desc = (struct ctl_error_desc *)addr;
2902		delete_done = 0;
2903
2904		mtx_lock(&softc->ctl_lock);
2905		lun = softc->ctl_luns[delete_desc->lun_id];
2906		if (lun == NULL) {
2907			mtx_unlock(&softc->ctl_lock);
2908			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2909			       __func__, (uintmax_t)delete_desc->lun_id);
2910			retval = EINVAL;
2911			break;
2912		}
2913		mtx_lock(&lun->lun_lock);
2914		mtx_unlock(&softc->ctl_lock);
2915		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2916			if (desc->serial != delete_desc->serial)
2917				continue;
2918
2919			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2920				      links);
2921			free(desc, M_CTL);
2922			delete_done = 1;
2923		}
2924		mtx_unlock(&lun->lun_lock);
2925		if (delete_done == 0) {
2926			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2927			       "error serial %ju on LUN %u\n", __func__,
2928			       delete_desc->serial, delete_desc->lun_id);
2929			retval = EINVAL;
2930			break;
2931		}
2932		break;
2933	}
2934	case CTL_DUMP_STRUCTS: {
2935		int i, j, k;
2936		struct ctl_port *port;
2937		struct ctl_frontend *fe;
2938
2939		mtx_lock(&softc->ctl_lock);
2940		printf("CTL Persistent Reservation information start:\n");
2941		for (i = 0; i < CTL_MAX_LUNS; i++) {
2942			struct ctl_lun *lun;
2943
2944			lun = softc->ctl_luns[i];
2945
2946			if ((lun == NULL)
2947			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2948				continue;
2949
2950			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2951				if (lun->pr_keys[j] == NULL)
2952					continue;
2953				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2954					if (lun->pr_keys[j][k] == 0)
2955						continue;
2956					printf("  LUN %d port %d iid %d key "
2957					       "%#jx\n", i, j, k,
2958					       (uintmax_t)lun->pr_keys[j][k]);
2959				}
2960			}
2961		}
2962		printf("CTL Persistent Reservation information end\n");
2963		printf("CTL Ports:\n");
2964		STAILQ_FOREACH(port, &softc->port_list, links) {
2965			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2966			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2967			       port->frontend->name, port->port_type,
2968			       port->physical_port, port->virtual_port,
2969			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2970			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2971				if (port->wwpn_iid[j].in_use == 0 &&
2972				    port->wwpn_iid[j].wwpn == 0 &&
2973				    port->wwpn_iid[j].name == NULL)
2974					continue;
2975
2976				printf("    iid %u use %d WWPN %#jx '%s'\n",
2977				    j, port->wwpn_iid[j].in_use,
2978				    (uintmax_t)port->wwpn_iid[j].wwpn,
2979				    port->wwpn_iid[j].name);
2980			}
2981		}
2982		printf("CTL Port information end\n");
2983		mtx_unlock(&softc->ctl_lock);
2984		/*
2985		 * XXX KDM calling this without a lock.  We'd likely want
2986		 * to drop the lock before calling the frontend's dump
2987		 * routine anyway.
2988		 */
2989		printf("CTL Frontends:\n");
2990		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2991			printf("  Frontend '%s'\n", fe->name);
2992			if (fe->fe_dump != NULL)
2993				fe->fe_dump();
2994		}
2995		printf("CTL Frontend information end\n");
2996		break;
2997	}
2998	case CTL_LUN_REQ: {
2999		struct ctl_lun_req *lun_req;
3000		struct ctl_backend_driver *backend;
3001
3002		lun_req = (struct ctl_lun_req *)addr;
3003
3004		backend = ctl_backend_find(lun_req->backend);
3005		if (backend == NULL) {
3006			lun_req->status = CTL_LUN_ERROR;
3007			snprintf(lun_req->error_str,
3008				 sizeof(lun_req->error_str),
3009				 "Backend \"%s\" not found.",
3010				 lun_req->backend);
3011			break;
3012		}
3013		if (lun_req->num_be_args > 0) {
3014			lun_req->kern_be_args = ctl_copyin_args(
3015				lun_req->num_be_args,
3016				lun_req->be_args,
3017				lun_req->error_str,
3018				sizeof(lun_req->error_str));
3019			if (lun_req->kern_be_args == NULL) {
3020				lun_req->status = CTL_LUN_ERROR;
3021				break;
3022			}
3023		}
3024
3025		retval = backend->ioctl(dev, cmd, addr, flag, td);
3026
3027		if (lun_req->num_be_args > 0) {
3028			ctl_copyout_args(lun_req->num_be_args,
3029				      lun_req->kern_be_args);
3030			ctl_free_args(lun_req->num_be_args,
3031				      lun_req->kern_be_args);
3032		}
3033		break;
3034	}
3035	case CTL_LUN_LIST: {
3036		struct sbuf *sb;
3037		struct ctl_lun *lun;
3038		struct ctl_lun_list *list;
3039		struct ctl_option *opt;
3040
3041		list = (struct ctl_lun_list *)addr;
3042
3043		/*
3044		 * Allocate a fixed length sbuf here, based on the length
3045		 * of the user's buffer.  We could allocate an auto-extending
3046		 * buffer, and then tell the user how much larger our
3047		 * amount of data is than his buffer, but that presents
3048		 * some problems:
3049		 *
3050		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3051		 *     we can't hold a lock while calling them with an
3052		 *     auto-extending buffer.
3053 		 *
3054		 * 2.  There is not currently a LUN reference counting
3055		 *     mechanism, outside of outstanding transactions on
3056		 *     the LUN's OOA queue.  So a LUN could go away on us
3057		 *     while we're getting the LUN number, backend-specific
3058		 *     information, etc.  Thus, given the way things
3059		 *     currently work, we need to hold the CTL lock while
3060		 *     grabbing LUN information.
3061		 *
3062		 * So, from the user's standpoint, the best thing to do is
3063		 * allocate what he thinks is a reasonable buffer length,
3064		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3065		 * double the buffer length and try again.  (And repeat
3066		 * that until he succeeds.)
3067		 */
3068		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3069		if (sb == NULL) {
3070			list->status = CTL_LUN_LIST_ERROR;
3071			snprintf(list->error_str, sizeof(list->error_str),
3072				 "Unable to allocate %d bytes for LUN list",
3073				 list->alloc_len);
3074			break;
3075		}
3076
3077		sbuf_printf(sb, "<ctllunlist>\n");
3078
3079		mtx_lock(&softc->ctl_lock);
3080		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3081			mtx_lock(&lun->lun_lock);
3082			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3083					     (uintmax_t)lun->lun);
3084
3085			/*
3086			 * Bail out as soon as we see that we've overfilled
3087			 * the buffer.
3088			 */
3089			if (retval != 0)
3090				break;
3091
3092			retval = sbuf_printf(sb, "\t<backend_type>%s"
3093					     "</backend_type>\n",
3094					     (lun->backend == NULL) ?  "none" :
3095					     lun->backend->name);
3096
3097			if (retval != 0)
3098				break;
3099
3100			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3101					     lun->be_lun->lun_type);
3102
3103			if (retval != 0)
3104				break;
3105
3106			if (lun->backend == NULL) {
3107				retval = sbuf_printf(sb, "</lun>\n");
3108				if (retval != 0)
3109					break;
3110				continue;
3111			}
3112
3113			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3114					     (lun->be_lun->maxlba > 0) ?
3115					     lun->be_lun->maxlba + 1 : 0);
3116
3117			if (retval != 0)
3118				break;
3119
3120			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3121					     lun->be_lun->blocksize);
3122
3123			if (retval != 0)
3124				break;
3125
3126			retval = sbuf_printf(sb, "\t<serial_number>");
3127
3128			if (retval != 0)
3129				break;
3130
3131			retval = ctl_sbuf_printf_esc(sb,
3132			    lun->be_lun->serial_num,
3133			    sizeof(lun->be_lun->serial_num));
3134
3135			if (retval != 0)
3136				break;
3137
3138			retval = sbuf_printf(sb, "</serial_number>\n");
3139
3140			if (retval != 0)
3141				break;
3142
3143			retval = sbuf_printf(sb, "\t<device_id>");
3144
3145			if (retval != 0)
3146				break;
3147
3148			retval = ctl_sbuf_printf_esc(sb,
3149			    lun->be_lun->device_id,
3150			    sizeof(lun->be_lun->device_id));
3151
3152			if (retval != 0)
3153				break;
3154
3155			retval = sbuf_printf(sb, "</device_id>\n");
3156
3157			if (retval != 0)
3158				break;
3159
3160			if (lun->backend->lun_info != NULL) {
3161				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3162				if (retval != 0)
3163					break;
3164			}
3165			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3166				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3167				    opt->name, opt->value, opt->name);
3168				if (retval != 0)
3169					break;
3170			}
3171
3172			retval = sbuf_printf(sb, "</lun>\n");
3173
3174			if (retval != 0)
3175				break;
3176			mtx_unlock(&lun->lun_lock);
3177		}
3178		if (lun != NULL)
3179			mtx_unlock(&lun->lun_lock);
3180		mtx_unlock(&softc->ctl_lock);
3181
3182		if ((retval != 0)
3183		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3184			retval = 0;
3185			sbuf_delete(sb);
3186			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3187			snprintf(list->error_str, sizeof(list->error_str),
3188				 "Out of space, %d bytes is too small",
3189				 list->alloc_len);
3190			break;
3191		}
3192
3193		sbuf_finish(sb);
3194
3195		retval = copyout(sbuf_data(sb), list->lun_xml,
3196				 sbuf_len(sb) + 1);
3197
3198		list->fill_len = sbuf_len(sb) + 1;
3199		list->status = CTL_LUN_LIST_OK;
3200		sbuf_delete(sb);
3201		break;
3202	}
3203	case CTL_ISCSI: {
3204		struct ctl_iscsi *ci;
3205		struct ctl_frontend *fe;
3206
3207		ci = (struct ctl_iscsi *)addr;
3208
3209		fe = ctl_frontend_find("iscsi");
3210		if (fe == NULL) {
3211			ci->status = CTL_ISCSI_ERROR;
3212			snprintf(ci->error_str, sizeof(ci->error_str),
3213			    "Frontend \"iscsi\" not found.");
3214			break;
3215		}
3216
3217		retval = fe->ioctl(dev, cmd, addr, flag, td);
3218		break;
3219	}
3220	case CTL_PORT_REQ: {
3221		struct ctl_req *req;
3222		struct ctl_frontend *fe;
3223
3224		req = (struct ctl_req *)addr;
3225
3226		fe = ctl_frontend_find(req->driver);
3227		if (fe == NULL) {
3228			req->status = CTL_LUN_ERROR;
3229			snprintf(req->error_str, sizeof(req->error_str),
3230			    "Frontend \"%s\" not found.", req->driver);
3231			break;
3232		}
3233		if (req->num_args > 0) {
3234			req->kern_args = ctl_copyin_args(req->num_args,
3235			    req->args, req->error_str, sizeof(req->error_str));
3236			if (req->kern_args == NULL) {
3237				req->status = CTL_LUN_ERROR;
3238				break;
3239			}
3240		}
3241
3242		retval = fe->ioctl(dev, cmd, addr, flag, td);
3243
3244		if (req->num_args > 0) {
3245			ctl_copyout_args(req->num_args, req->kern_args);
3246			ctl_free_args(req->num_args, req->kern_args);
3247		}
3248		break;
3249	}
3250	case CTL_PORT_LIST: {
3251		struct sbuf *sb;
3252		struct ctl_port *port;
3253		struct ctl_lun_list *list;
3254		struct ctl_option *opt;
3255		int j;
3256		uint32_t plun;
3257
3258		list = (struct ctl_lun_list *)addr;
3259
3260		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3261		if (sb == NULL) {
3262			list->status = CTL_LUN_LIST_ERROR;
3263			snprintf(list->error_str, sizeof(list->error_str),
3264				 "Unable to allocate %d bytes for LUN list",
3265				 list->alloc_len);
3266			break;
3267		}
3268
3269		sbuf_printf(sb, "<ctlportlist>\n");
3270
3271		mtx_lock(&softc->ctl_lock);
3272		STAILQ_FOREACH(port, &softc->port_list, links) {
3273			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3274					     (uintmax_t)port->targ_port);
3275
3276			/*
3277			 * Bail out as soon as we see that we've overfilled
3278			 * the buffer.
3279			 */
3280			if (retval != 0)
3281				break;
3282
3283			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3284			    "</frontend_type>\n", port->frontend->name);
3285			if (retval != 0)
3286				break;
3287
3288			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3289					     port->port_type);
3290			if (retval != 0)
3291				break;
3292
3293			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3294			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3295			if (retval != 0)
3296				break;
3297
3298			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3299			    port->port_name);
3300			if (retval != 0)
3301				break;
3302
3303			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3304			    port->physical_port);
3305			if (retval != 0)
3306				break;
3307
3308			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3309			    port->virtual_port);
3310			if (retval != 0)
3311				break;
3312
3313			if (port->target_devid != NULL) {
3314				sbuf_printf(sb, "\t<target>");
3315				ctl_id_sbuf(port->target_devid, sb);
3316				sbuf_printf(sb, "</target>\n");
3317			}
3318
3319			if (port->port_devid != NULL) {
3320				sbuf_printf(sb, "\t<port>");
3321				ctl_id_sbuf(port->port_devid, sb);
3322				sbuf_printf(sb, "</port>\n");
3323			}
3324
3325			if (port->port_info != NULL) {
3326				retval = port->port_info(port->onoff_arg, sb);
3327				if (retval != 0)
3328					break;
3329			}
3330			STAILQ_FOREACH(opt, &port->options, links) {
3331				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3332				    opt->name, opt->value, opt->name);
3333				if (retval != 0)
3334					break;
3335			}
3336
3337			if (port->lun_map != NULL) {
3338				sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3339				for (j = 0; j < CTL_MAX_LUNS; j++) {
3340					plun = ctl_lun_map_from_port(port, j);
3341					if (plun >= CTL_MAX_LUNS)
3342						continue;
3343					sbuf_printf(sb,
3344					    "\t<lun id=\"%u\">%u</lun>\n",
3345					    j, plun);
3346				}
3347			}
3348
3349			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3350				if (port->wwpn_iid[j].in_use == 0 ||
3351				    (port->wwpn_iid[j].wwpn == 0 &&
3352				     port->wwpn_iid[j].name == NULL))
3353					continue;
3354
3355				if (port->wwpn_iid[j].name != NULL)
3356					retval = sbuf_printf(sb,
3357					    "\t<initiator id=\"%u\">%s</initiator>\n",
3358					    j, port->wwpn_iid[j].name);
3359				else
3360					retval = sbuf_printf(sb,
3361					    "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3362					    j, port->wwpn_iid[j].wwpn);
3363				if (retval != 0)
3364					break;
3365			}
3366			if (retval != 0)
3367				break;
3368
3369			retval = sbuf_printf(sb, "</targ_port>\n");
3370			if (retval != 0)
3371				break;
3372		}
3373		mtx_unlock(&softc->ctl_lock);
3374
3375		if ((retval != 0)
3376		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3377			retval = 0;
3378			sbuf_delete(sb);
3379			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3380			snprintf(list->error_str, sizeof(list->error_str),
3381				 "Out of space, %d bytes is too small",
3382				 list->alloc_len);
3383			break;
3384		}
3385
3386		sbuf_finish(sb);
3387
3388		retval = copyout(sbuf_data(sb), list->lun_xml,
3389				 sbuf_len(sb) + 1);
3390
3391		list->fill_len = sbuf_len(sb) + 1;
3392		list->status = CTL_LUN_LIST_OK;
3393		sbuf_delete(sb);
3394		break;
3395	}
3396	case CTL_LUN_MAP: {
3397		struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3398		struct ctl_port *port;
3399
3400		mtx_lock(&softc->ctl_lock);
3401		if (lm->port >= CTL_MAX_PORTS ||
3402		    (port = softc->ctl_ports[lm->port]) == NULL) {
3403			mtx_unlock(&softc->ctl_lock);
3404			return (ENXIO);
3405		}
3406		mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3407		if (lm->plun < CTL_MAX_LUNS) {
3408			if (lm->lun == UINT32_MAX)
3409				retval = ctl_lun_map_unset(port, lm->plun);
3410			else if (lm->lun < CTL_MAX_LUNS &&
3411			    softc->ctl_luns[lm->lun] != NULL)
3412				retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3413			else
3414				return (ENXIO);
3415		} else if (lm->plun == UINT32_MAX) {
3416			if (lm->lun == UINT32_MAX)
3417				retval = ctl_lun_map_deinit(port);
3418			else
3419				retval = ctl_lun_map_init(port);
3420		} else
3421			return (ENXIO);
3422		break;
3423	}
3424	default: {
3425		/* XXX KDM should we fix this? */
3426#if 0
3427		struct ctl_backend_driver *backend;
3428		unsigned int type;
3429		int found;
3430
3431		found = 0;
3432
3433		/*
3434		 * We encode the backend type as the ioctl type for backend
3435		 * ioctls.  So parse it out here, and then search for a
3436		 * backend of this type.
3437		 */
3438		type = _IOC_TYPE(cmd);
3439
3440		STAILQ_FOREACH(backend, &softc->be_list, links) {
3441			if (backend->type == type) {
3442				found = 1;
3443				break;
3444			}
3445		}
3446		if (found == 0) {
3447			printf("ctl: unknown ioctl command %#lx or backend "
3448			       "%d\n", cmd, type);
3449			retval = EINVAL;
3450			break;
3451		}
3452		retval = backend->ioctl(dev, cmd, addr, flag, td);
3453#endif
3454		retval = ENOTTY;
3455		break;
3456	}
3457	}
3458	return (retval);
3459}
3460
3461uint32_t
3462ctl_get_initindex(struct ctl_nexus *nexus)
3463{
3464	if (nexus->targ_port < CTL_MAX_PORTS)
3465		return (nexus->initid.id +
3466			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3467	else
3468		return (nexus->initid.id +
3469		       ((nexus->targ_port - CTL_MAX_PORTS) *
3470			CTL_MAX_INIT_PER_PORT));
3471}
3472
3473uint32_t
3474ctl_get_resindex(struct ctl_nexus *nexus)
3475{
3476	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3477}
3478
3479uint32_t
3480ctl_port_idx(int port_num)
3481{
3482	if (port_num < CTL_MAX_PORTS)
3483		return(port_num);
3484	else
3485		return(port_num - CTL_MAX_PORTS);
3486}
3487
3488int
3489ctl_lun_map_init(struct ctl_port *port)
3490{
3491	struct ctl_softc *softc = control_softc;
3492	struct ctl_lun *lun;
3493	uint32_t i;
3494
3495	if (port->lun_map == NULL)
3496		port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
3497		    M_CTL, M_NOWAIT);
3498	if (port->lun_map == NULL)
3499		return (ENOMEM);
3500	for (i = 0; i < CTL_MAX_LUNS; i++)
3501		port->lun_map[i] = UINT32_MAX;
3502	if (port->status & CTL_PORT_STATUS_ONLINE) {
3503		STAILQ_FOREACH(lun, &softc->lun_list, links)
3504			port->lun_disable(port->targ_lun_arg, lun->lun);
3505	}
3506	return (0);
3507}
3508
3509int
3510ctl_lun_map_deinit(struct ctl_port *port)
3511{
3512	struct ctl_softc *softc = control_softc;
3513	struct ctl_lun *lun;
3514
3515	if (port->lun_map == NULL)
3516		return (0);
3517	free(port->lun_map, M_CTL);
3518	port->lun_map = NULL;
3519	if (port->status & CTL_PORT_STATUS_ONLINE) {
3520		STAILQ_FOREACH(lun, &softc->lun_list, links)
3521			port->lun_enable(port->targ_lun_arg, lun->lun);
3522	}
3523	return (0);
3524}
3525
3526int
3527ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3528{
3529	int status;
3530	uint32_t old;
3531
3532	if (port->lun_map == NULL) {
3533		status = ctl_lun_map_init(port);
3534		if (status != 0)
3535			return (status);
3536	}
3537	old = port->lun_map[plun];
3538	port->lun_map[plun] = glun;
3539	if ((port->status & CTL_PORT_STATUS_ONLINE) && old >= CTL_MAX_LUNS)
3540		port->lun_enable(port->targ_lun_arg, plun);
3541	return (0);
3542}
3543
3544int
3545ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3546{
3547	uint32_t old;
3548
3549	if (port->lun_map == NULL)
3550		return (0);
3551	old = port->lun_map[plun];
3552	port->lun_map[plun] = UINT32_MAX;
3553	if ((port->status & CTL_PORT_STATUS_ONLINE) && old < CTL_MAX_LUNS)
3554		port->lun_disable(port->targ_lun_arg, plun);
3555	return (0);
3556}
3557
3558uint32_t
3559ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3560{
3561
3562	if (port == NULL)
3563		return (UINT32_MAX);
3564	if (port->lun_map == NULL || lun_id >= CTL_MAX_LUNS)
3565		return (lun_id);
3566	return (port->lun_map[lun_id]);
3567}
3568
3569uint32_t
3570ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3571{
3572	uint32_t i;
3573
3574	if (port == NULL)
3575		return (UINT32_MAX);
3576	if (port->lun_map == NULL)
3577		return (lun_id);
3578	for (i = 0; i < CTL_MAX_LUNS; i++) {
3579		if (port->lun_map[i] == lun_id)
3580			return (i);
3581	}
3582	return (UINT32_MAX);
3583}
3584
3585static struct ctl_port *
3586ctl_io_port(struct ctl_io_hdr *io_hdr)
3587{
3588	int port_num;
3589
3590	port_num = io_hdr->nexus.targ_port;
3591	return (control_softc->ctl_ports[ctl_port_idx(port_num)]);
3592}
3593
3594/*
3595 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3596 * that are a power of 2.
3597 */
3598int
3599ctl_ffz(uint32_t *mask, uint32_t size)
3600{
3601	uint32_t num_chunks, num_pieces;
3602	int i, j;
3603
3604	num_chunks = (size >> 5);
3605	if (num_chunks == 0)
3606		num_chunks++;
3607	num_pieces = MIN((sizeof(uint32_t) * 8), size);
3608
3609	for (i = 0; i < num_chunks; i++) {
3610		for (j = 0; j < num_pieces; j++) {
3611			if ((mask[i] & (1 << j)) == 0)
3612				return ((i << 5) + j);
3613		}
3614	}
3615
3616	return (-1);
3617}
3618
3619int
3620ctl_set_mask(uint32_t *mask, uint32_t bit)
3621{
3622	uint32_t chunk, piece;
3623
3624	chunk = bit >> 5;
3625	piece = bit % (sizeof(uint32_t) * 8);
3626
3627	if ((mask[chunk] & (1 << piece)) != 0)
3628		return (-1);
3629	else
3630		mask[chunk] |= (1 << piece);
3631
3632	return (0);
3633}
3634
3635int
3636ctl_clear_mask(uint32_t *mask, uint32_t bit)
3637{
3638	uint32_t chunk, piece;
3639
3640	chunk = bit >> 5;
3641	piece = bit % (sizeof(uint32_t) * 8);
3642
3643	if ((mask[chunk] & (1 << piece)) == 0)
3644		return (-1);
3645	else
3646		mask[chunk] &= ~(1 << piece);
3647
3648	return (0);
3649}
3650
3651int
3652ctl_is_set(uint32_t *mask, uint32_t bit)
3653{
3654	uint32_t chunk, piece;
3655
3656	chunk = bit >> 5;
3657	piece = bit % (sizeof(uint32_t) * 8);
3658
3659	if ((mask[chunk] & (1 << piece)) == 0)
3660		return (0);
3661	else
3662		return (1);
3663}
3664
3665static uint64_t
3666ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3667{
3668	uint64_t *t;
3669
3670	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3671	if (t == NULL)
3672		return (0);
3673	return (t[residx % CTL_MAX_INIT_PER_PORT]);
3674}
3675
3676static void
3677ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3678{
3679	uint64_t *t;
3680
3681	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3682	if (t == NULL)
3683		return;
3684	t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3685}
3686
3687static void
3688ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3689{
3690	uint64_t *p;
3691	u_int i;
3692
3693	i = residx/CTL_MAX_INIT_PER_PORT;
3694	if (lun->pr_keys[i] != NULL)
3695		return;
3696	mtx_unlock(&lun->lun_lock);
3697	p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3698	    M_WAITOK | M_ZERO);
3699	mtx_lock(&lun->lun_lock);
3700	if (lun->pr_keys[i] == NULL)
3701		lun->pr_keys[i] = p;
3702	else
3703		free(p, M_CTL);
3704}
3705
3706static void
3707ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3708{
3709	uint64_t *t;
3710
3711	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3712	KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3713	t[residx % CTL_MAX_INIT_PER_PORT] = key;
3714}
3715
3716/*
3717 * ctl_softc, pool_name, total_ctl_io are passed in.
3718 * npool is passed out.
3719 */
3720int
3721ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3722		uint32_t total_ctl_io, void **npool)
3723{
3724#ifdef IO_POOLS
3725	struct ctl_io_pool *pool;
3726
3727	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3728					    M_NOWAIT | M_ZERO);
3729	if (pool == NULL)
3730		return (ENOMEM);
3731
3732	snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3733	pool->ctl_softc = ctl_softc;
3734	pool->zone = uma_zsecond_create(pool->name, NULL,
3735	    NULL, NULL, NULL, ctl_softc->io_zone);
3736	/* uma_prealloc(pool->zone, total_ctl_io); */
3737
3738	*npool = pool;
3739#else
3740	*npool = ctl_softc->io_zone;
3741#endif
3742	return (0);
3743}
3744
3745void
3746ctl_pool_free(struct ctl_io_pool *pool)
3747{
3748
3749	if (pool == NULL)
3750		return;
3751
3752#ifdef IO_POOLS
3753	uma_zdestroy(pool->zone);
3754	free(pool, M_CTL);
3755#endif
3756}
3757
3758union ctl_io *
3759ctl_alloc_io(void *pool_ref)
3760{
3761	union ctl_io *io;
3762#ifdef IO_POOLS
3763	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3764
3765	io = uma_zalloc(pool->zone, M_WAITOK);
3766#else
3767	io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3768#endif
3769	if (io != NULL)
3770		io->io_hdr.pool = pool_ref;
3771	return (io);
3772}
3773
3774union ctl_io *
3775ctl_alloc_io_nowait(void *pool_ref)
3776{
3777	union ctl_io *io;
3778#ifdef IO_POOLS
3779	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3780
3781	io = uma_zalloc(pool->zone, M_NOWAIT);
3782#else
3783	io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3784#endif
3785	if (io != NULL)
3786		io->io_hdr.pool = pool_ref;
3787	return (io);
3788}
3789
3790void
3791ctl_free_io(union ctl_io *io)
3792{
3793#ifdef IO_POOLS
3794	struct ctl_io_pool *pool;
3795#endif
3796
3797	if (io == NULL)
3798		return;
3799
3800#ifdef IO_POOLS
3801	pool = (struct ctl_io_pool *)io->io_hdr.pool;
3802	uma_zfree(pool->zone, io);
3803#else
3804	uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3805#endif
3806}
3807
3808void
3809ctl_zero_io(union ctl_io *io)
3810{
3811	void *pool_ref;
3812
3813	if (io == NULL)
3814		return;
3815
3816	/*
3817	 * May need to preserve linked list pointers at some point too.
3818	 */
3819	pool_ref = io->io_hdr.pool;
3820	memset(io, 0, sizeof(*io));
3821	io->io_hdr.pool = pool_ref;
3822}
3823
3824/*
3825 * This routine is currently used for internal copies of ctl_ios that need
3826 * to persist for some reason after we've already returned status to the
3827 * FETD.  (Thus the flag set.)
3828 *
3829 * XXX XXX
3830 * Note that this makes a blind copy of all fields in the ctl_io, except
3831 * for the pool reference.  This includes any memory that has been
3832 * allocated!  That memory will no longer be valid after done has been
3833 * called, so this would be VERY DANGEROUS for command that actually does
3834 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3835 * start and stop commands, which don't transfer any data, so this is not a
3836 * problem.  If it is used for anything else, the caller would also need to
3837 * allocate data buffer space and this routine would need to be modified to
3838 * copy the data buffer(s) as well.
3839 */
3840void
3841ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3842{
3843	void *pool_ref;
3844
3845	if ((src == NULL)
3846	 || (dest == NULL))
3847		return;
3848
3849	/*
3850	 * May need to preserve linked list pointers at some point too.
3851	 */
3852	pool_ref = dest->io_hdr.pool;
3853
3854	memcpy(dest, src, MIN(sizeof(*src), sizeof(*dest)));
3855
3856	dest->io_hdr.pool = pool_ref;
3857	/*
3858	 * We need to know that this is an internal copy, and doesn't need
3859	 * to get passed back to the FETD that allocated it.
3860	 */
3861	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3862}
3863
3864int
3865ctl_expand_number(const char *buf, uint64_t *num)
3866{
3867	char *endptr;
3868	uint64_t number;
3869	unsigned shift;
3870
3871	number = strtoq(buf, &endptr, 0);
3872
3873	switch (tolower((unsigned char)*endptr)) {
3874	case 'e':
3875		shift = 60;
3876		break;
3877	case 'p':
3878		shift = 50;
3879		break;
3880	case 't':
3881		shift = 40;
3882		break;
3883	case 'g':
3884		shift = 30;
3885		break;
3886	case 'm':
3887		shift = 20;
3888		break;
3889	case 'k':
3890		shift = 10;
3891		break;
3892	case 'b':
3893	case '\0': /* No unit. */
3894		*num = number;
3895		return (0);
3896	default:
3897		/* Unrecognized unit. */
3898		return (-1);
3899	}
3900
3901	if ((number << shift) >> shift != number) {
3902		/* Overflow */
3903		return (-1);
3904	}
3905	*num = number << shift;
3906	return (0);
3907}
3908
3909
3910/*
3911 * This routine could be used in the future to load default and/or saved
3912 * mode page parameters for a particuar lun.
3913 */
3914static int
3915ctl_init_page_index(struct ctl_lun *lun)
3916{
3917	int i;
3918	struct ctl_page_index *page_index;
3919	const char *value;
3920	uint64_t ival;
3921
3922	memcpy(&lun->mode_pages.index, page_index_template,
3923	       sizeof(page_index_template));
3924
3925	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3926
3927		page_index = &lun->mode_pages.index[i];
3928		/*
3929		 * If this is a disk-only mode page, there's no point in
3930		 * setting it up.  For some pages, we have to have some
3931		 * basic information about the disk in order to calculate the
3932		 * mode page data.
3933		 */
3934		if ((lun->be_lun->lun_type != T_DIRECT)
3935		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3936			continue;
3937
3938		switch (page_index->page_code & SMPH_PC_MASK) {
3939		case SMS_RW_ERROR_RECOVERY_PAGE: {
3940			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3941				panic("subpage is incorrect!");
3942			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3943			       &rw_er_page_default,
3944			       sizeof(rw_er_page_default));
3945			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3946			       &rw_er_page_changeable,
3947			       sizeof(rw_er_page_changeable));
3948			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3949			       &rw_er_page_default,
3950			       sizeof(rw_er_page_default));
3951			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3952			       &rw_er_page_default,
3953			       sizeof(rw_er_page_default));
3954			page_index->page_data =
3955				(uint8_t *)lun->mode_pages.rw_er_page;
3956			break;
3957		}
3958		case SMS_FORMAT_DEVICE_PAGE: {
3959			struct scsi_format_page *format_page;
3960
3961			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3962				panic("subpage is incorrect!");
3963
3964			/*
3965			 * Sectors per track are set above.  Bytes per
3966			 * sector need to be set here on a per-LUN basis.
3967			 */
3968			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3969			       &format_page_default,
3970			       sizeof(format_page_default));
3971			memcpy(&lun->mode_pages.format_page[
3972			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3973			       sizeof(format_page_changeable));
3974			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3975			       &format_page_default,
3976			       sizeof(format_page_default));
3977			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3978			       &format_page_default,
3979			       sizeof(format_page_default));
3980
3981			format_page = &lun->mode_pages.format_page[
3982				CTL_PAGE_CURRENT];
3983			scsi_ulto2b(lun->be_lun->blocksize,
3984				    format_page->bytes_per_sector);
3985
3986			format_page = &lun->mode_pages.format_page[
3987				CTL_PAGE_DEFAULT];
3988			scsi_ulto2b(lun->be_lun->blocksize,
3989				    format_page->bytes_per_sector);
3990
3991			format_page = &lun->mode_pages.format_page[
3992				CTL_PAGE_SAVED];
3993			scsi_ulto2b(lun->be_lun->blocksize,
3994				    format_page->bytes_per_sector);
3995
3996			page_index->page_data =
3997				(uint8_t *)lun->mode_pages.format_page;
3998			break;
3999		}
4000		case SMS_RIGID_DISK_PAGE: {
4001			struct scsi_rigid_disk_page *rigid_disk_page;
4002			uint32_t sectors_per_cylinder;
4003			uint64_t cylinders;
4004#ifndef	__XSCALE__
4005			int shift;
4006#endif /* !__XSCALE__ */
4007
4008			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4009				panic("invalid subpage value %d",
4010				      page_index->subpage);
4011
4012			/*
4013			 * Rotation rate and sectors per track are set
4014			 * above.  We calculate the cylinders here based on
4015			 * capacity.  Due to the number of heads and
4016			 * sectors per track we're using, smaller arrays
4017			 * may turn out to have 0 cylinders.  Linux and
4018			 * FreeBSD don't pay attention to these mode pages
4019			 * to figure out capacity, but Solaris does.  It
4020			 * seems to deal with 0 cylinders just fine, and
4021			 * works out a fake geometry based on the capacity.
4022			 */
4023			memcpy(&lun->mode_pages.rigid_disk_page[
4024			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4025			       sizeof(rigid_disk_page_default));
4026			memcpy(&lun->mode_pages.rigid_disk_page[
4027			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4028			       sizeof(rigid_disk_page_changeable));
4029
4030			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4031				CTL_DEFAULT_HEADS;
4032
4033			/*
4034			 * The divide method here will be more accurate,
4035			 * probably, but results in floating point being
4036			 * used in the kernel on i386 (__udivdi3()).  On the
4037			 * XScale, though, __udivdi3() is implemented in
4038			 * software.
4039			 *
4040			 * The shift method for cylinder calculation is
4041			 * accurate if sectors_per_cylinder is a power of
4042			 * 2.  Otherwise it might be slightly off -- you
4043			 * might have a bit of a truncation problem.
4044			 */
4045#ifdef	__XSCALE__
4046			cylinders = (lun->be_lun->maxlba + 1) /
4047				sectors_per_cylinder;
4048#else
4049			for (shift = 31; shift > 0; shift--) {
4050				if (sectors_per_cylinder & (1 << shift))
4051					break;
4052			}
4053			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4054#endif
4055
4056			/*
4057			 * We've basically got 3 bytes, or 24 bits for the
4058			 * cylinder size in the mode page.  If we're over,
4059			 * just round down to 2^24.
4060			 */
4061			if (cylinders > 0xffffff)
4062				cylinders = 0xffffff;
4063
4064			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4065				CTL_PAGE_DEFAULT];
4066			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4067
4068			if ((value = ctl_get_opt(&lun->be_lun->options,
4069			    "rpm")) != NULL) {
4070				scsi_ulto2b(strtol(value, NULL, 0),
4071				     rigid_disk_page->rotation_rate);
4072			}
4073
4074			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4075			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4076			       sizeof(rigid_disk_page_default));
4077			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4078			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4079			       sizeof(rigid_disk_page_default));
4080
4081			page_index->page_data =
4082				(uint8_t *)lun->mode_pages.rigid_disk_page;
4083			break;
4084		}
4085		case SMS_CACHING_PAGE: {
4086			struct scsi_caching_page *caching_page;
4087
4088			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4089				panic("invalid subpage value %d",
4090				      page_index->subpage);
4091			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4092			       &caching_page_default,
4093			       sizeof(caching_page_default));
4094			memcpy(&lun->mode_pages.caching_page[
4095			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4096			       sizeof(caching_page_changeable));
4097			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4098			       &caching_page_default,
4099			       sizeof(caching_page_default));
4100			caching_page = &lun->mode_pages.caching_page[
4101			    CTL_PAGE_SAVED];
4102			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4103			if (value != NULL && strcmp(value, "off") == 0)
4104				caching_page->flags1 &= ~SCP_WCE;
4105			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4106			if (value != NULL && strcmp(value, "off") == 0)
4107				caching_page->flags1 |= SCP_RCD;
4108			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4109			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4110			       sizeof(caching_page_default));
4111			page_index->page_data =
4112				(uint8_t *)lun->mode_pages.caching_page;
4113			break;
4114		}
4115		case SMS_CONTROL_MODE_PAGE: {
4116			struct scsi_control_page *control_page;
4117
4118			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4119				panic("invalid subpage value %d",
4120				      page_index->subpage);
4121
4122			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4123			       &control_page_default,
4124			       sizeof(control_page_default));
4125			memcpy(&lun->mode_pages.control_page[
4126			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4127			       sizeof(control_page_changeable));
4128			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4129			       &control_page_default,
4130			       sizeof(control_page_default));
4131			control_page = &lun->mode_pages.control_page[
4132			    CTL_PAGE_SAVED];
4133			value = ctl_get_opt(&lun->be_lun->options, "reordering");
4134			if (value != NULL && strcmp(value, "unrestricted") == 0) {
4135				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4136				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4137			}
4138			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4139			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4140			       sizeof(control_page_default));
4141			page_index->page_data =
4142				(uint8_t *)lun->mode_pages.control_page;
4143			break;
4144
4145		}
4146		case SMS_INFO_EXCEPTIONS_PAGE: {
4147			switch (page_index->subpage) {
4148			case SMS_SUBPAGE_PAGE_0:
4149				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4150				       &ie_page_default,
4151				       sizeof(ie_page_default));
4152				memcpy(&lun->mode_pages.ie_page[
4153				       CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4154				       sizeof(ie_page_changeable));
4155				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4156				       &ie_page_default,
4157				       sizeof(ie_page_default));
4158				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4159				       &ie_page_default,
4160				       sizeof(ie_page_default));
4161				page_index->page_data =
4162					(uint8_t *)lun->mode_pages.ie_page;
4163				break;
4164			case 0x02: {
4165				struct ctl_logical_block_provisioning_page *page;
4166
4167				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4168				       &lbp_page_default,
4169				       sizeof(lbp_page_default));
4170				memcpy(&lun->mode_pages.lbp_page[
4171				       CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4172				       sizeof(lbp_page_changeable));
4173				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4174				       &lbp_page_default,
4175				       sizeof(lbp_page_default));
4176				page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4177				value = ctl_get_opt(&lun->be_lun->options,
4178				    "avail-threshold");
4179				if (value != NULL &&
4180				    ctl_expand_number(value, &ival) == 0) {
4181					page->descr[0].flags |= SLBPPD_ENABLED |
4182					    SLBPPD_ARMING_DEC;
4183					if (lun->be_lun->blocksize)
4184						ival /= lun->be_lun->blocksize;
4185					else
4186						ival /= 512;
4187					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4188					    page->descr[0].count);
4189				}
4190				value = ctl_get_opt(&lun->be_lun->options,
4191				    "used-threshold");
4192				if (value != NULL &&
4193				    ctl_expand_number(value, &ival) == 0) {
4194					page->descr[1].flags |= SLBPPD_ENABLED |
4195					    SLBPPD_ARMING_INC;
4196					if (lun->be_lun->blocksize)
4197						ival /= lun->be_lun->blocksize;
4198					else
4199						ival /= 512;
4200					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4201					    page->descr[1].count);
4202				}
4203				value = ctl_get_opt(&lun->be_lun->options,
4204				    "pool-avail-threshold");
4205				if (value != NULL &&
4206				    ctl_expand_number(value, &ival) == 0) {
4207					page->descr[2].flags |= SLBPPD_ENABLED |
4208					    SLBPPD_ARMING_DEC;
4209					if (lun->be_lun->blocksize)
4210						ival /= lun->be_lun->blocksize;
4211					else
4212						ival /= 512;
4213					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4214					    page->descr[2].count);
4215				}
4216				value = ctl_get_opt(&lun->be_lun->options,
4217				    "pool-used-threshold");
4218				if (value != NULL &&
4219				    ctl_expand_number(value, &ival) == 0) {
4220					page->descr[3].flags |= SLBPPD_ENABLED |
4221					    SLBPPD_ARMING_INC;
4222					if (lun->be_lun->blocksize)
4223						ival /= lun->be_lun->blocksize;
4224					else
4225						ival /= 512;
4226					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4227					    page->descr[3].count);
4228				}
4229				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4230				       &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4231				       sizeof(lbp_page_default));
4232				page_index->page_data =
4233					(uint8_t *)lun->mode_pages.lbp_page;
4234			}}
4235			break;
4236		}
4237		case SMS_VENDOR_SPECIFIC_PAGE:{
4238			switch (page_index->subpage) {
4239			case DBGCNF_SUBPAGE_CODE: {
4240				struct copan_debugconf_subpage *current_page,
4241							       *saved_page;
4242
4243				memcpy(&lun->mode_pages.debugconf_subpage[
4244				       CTL_PAGE_CURRENT],
4245				       &debugconf_page_default,
4246				       sizeof(debugconf_page_default));
4247				memcpy(&lun->mode_pages.debugconf_subpage[
4248				       CTL_PAGE_CHANGEABLE],
4249				       &debugconf_page_changeable,
4250				       sizeof(debugconf_page_changeable));
4251				memcpy(&lun->mode_pages.debugconf_subpage[
4252				       CTL_PAGE_DEFAULT],
4253				       &debugconf_page_default,
4254				       sizeof(debugconf_page_default));
4255				memcpy(&lun->mode_pages.debugconf_subpage[
4256				       CTL_PAGE_SAVED],
4257				       &debugconf_page_default,
4258				       sizeof(debugconf_page_default));
4259				page_index->page_data =
4260					(uint8_t *)lun->mode_pages.debugconf_subpage;
4261
4262				current_page = (struct copan_debugconf_subpage *)
4263					(page_index->page_data +
4264					 (page_index->page_len *
4265					  CTL_PAGE_CURRENT));
4266				saved_page = (struct copan_debugconf_subpage *)
4267					(page_index->page_data +
4268					 (page_index->page_len *
4269					  CTL_PAGE_SAVED));
4270				break;
4271			}
4272			default:
4273				panic("invalid subpage value %d",
4274				      page_index->subpage);
4275				break;
4276			}
4277   			break;
4278		}
4279		default:
4280			panic("invalid page value %d",
4281			      page_index->page_code & SMPH_PC_MASK);
4282			break;
4283    	}
4284	}
4285
4286	return (CTL_RETVAL_COMPLETE);
4287}
4288
4289static int
4290ctl_init_log_page_index(struct ctl_lun *lun)
4291{
4292	struct ctl_page_index *page_index;
4293	int i, j, k, prev;
4294
4295	memcpy(&lun->log_pages.index, log_page_index_template,
4296	       sizeof(log_page_index_template));
4297
4298	prev = -1;
4299	for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4300
4301		page_index = &lun->log_pages.index[i];
4302		/*
4303		 * If this is a disk-only mode page, there's no point in
4304		 * setting it up.  For some pages, we have to have some
4305		 * basic information about the disk in order to calculate the
4306		 * mode page data.
4307		 */
4308		if ((lun->be_lun->lun_type != T_DIRECT)
4309		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4310			continue;
4311
4312		if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4313		     lun->backend->lun_attr == NULL)
4314			continue;
4315
4316		if (page_index->page_code != prev) {
4317			lun->log_pages.pages_page[j] = page_index->page_code;
4318			prev = page_index->page_code;
4319			j++;
4320		}
4321		lun->log_pages.subpages_page[k*2] = page_index->page_code;
4322		lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4323		k++;
4324	}
4325	lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4326	lun->log_pages.index[0].page_len = j;
4327	lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4328	lun->log_pages.index[1].page_len = k * 2;
4329	lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4330	lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4331	lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4332	lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4333
4334	return (CTL_RETVAL_COMPLETE);
4335}
4336
4337static int
4338hex2bin(const char *str, uint8_t *buf, int buf_size)
4339{
4340	int i;
4341	u_char c;
4342
4343	memset(buf, 0, buf_size);
4344	while (isspace(str[0]))
4345		str++;
4346	if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4347		str += 2;
4348	buf_size *= 2;
4349	for (i = 0; str[i] != 0 && i < buf_size; i++) {
4350		c = str[i];
4351		if (isdigit(c))
4352			c -= '0';
4353		else if (isalpha(c))
4354			c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4355		else
4356			break;
4357		if (c >= 16)
4358			break;
4359		if ((i & 1) == 0)
4360			buf[i / 2] |= (c << 4);
4361		else
4362			buf[i / 2] |= c;
4363	}
4364	return ((i + 1) / 2);
4365}
4366
4367/*
4368 * LUN allocation.
4369 *
4370 * Requirements:
4371 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4372 *   wants us to allocate the LUN and he can block.
4373 * - ctl_softc is always set
4374 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4375 *
4376 * Returns 0 for success, non-zero (errno) for failure.
4377 */
4378static int
4379ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4380	      struct ctl_be_lun *const be_lun)
4381{
4382	struct ctl_lun *nlun, *lun;
4383	struct scsi_vpd_id_descriptor *desc;
4384	struct scsi_vpd_id_t10 *t10id;
4385	const char *eui, *naa, *scsiname, *vendor, *value;
4386	int lun_number, i, lun_malloced;
4387	int devidlen, idlen1, idlen2 = 0, len;
4388
4389	if (be_lun == NULL)
4390		return (EINVAL);
4391
4392	/*
4393	 * We currently only support Direct Access or Processor LUN types.
4394	 */
4395	switch (be_lun->lun_type) {
4396	case T_DIRECT:
4397		break;
4398	case T_PROCESSOR:
4399		break;
4400	case T_SEQUENTIAL:
4401	case T_CHANGER:
4402	default:
4403		be_lun->lun_config_status(be_lun->be_lun,
4404					  CTL_LUN_CONFIG_FAILURE);
4405		break;
4406	}
4407	if (ctl_lun == NULL) {
4408		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4409		lun_malloced = 1;
4410	} else {
4411		lun_malloced = 0;
4412		lun = ctl_lun;
4413	}
4414
4415	memset(lun, 0, sizeof(*lun));
4416	if (lun_malloced)
4417		lun->flags = CTL_LUN_MALLOCED;
4418
4419	/* Generate LUN ID. */
4420	devidlen = max(CTL_DEVID_MIN_LEN,
4421	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4422	idlen1 = sizeof(*t10id) + devidlen;
4423	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4424	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4425	if (scsiname != NULL) {
4426		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4427		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4428	}
4429	eui = ctl_get_opt(&be_lun->options, "eui");
4430	if (eui != NULL) {
4431		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4432	}
4433	naa = ctl_get_opt(&be_lun->options, "naa");
4434	if (naa != NULL) {
4435		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4436	}
4437	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4438	    M_CTL, M_WAITOK | M_ZERO);
4439	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4440	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4441	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4442	desc->length = idlen1;
4443	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4444	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4445	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4446		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4447	} else {
4448		strncpy(t10id->vendor, vendor,
4449		    min(sizeof(t10id->vendor), strlen(vendor)));
4450	}
4451	strncpy((char *)t10id->vendor_spec_id,
4452	    (char *)be_lun->device_id, devidlen);
4453	if (scsiname != NULL) {
4454		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4455		    desc->length);
4456		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4457		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4458		    SVPD_ID_TYPE_SCSI_NAME;
4459		desc->length = idlen2;
4460		strlcpy(desc->identifier, scsiname, idlen2);
4461	}
4462	if (eui != NULL) {
4463		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4464		    desc->length);
4465		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4466		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4467		    SVPD_ID_TYPE_EUI64;
4468		desc->length = hex2bin(eui, desc->identifier, 16);
4469		desc->length = desc->length > 12 ? 16 :
4470		    (desc->length > 8 ? 12 : 8);
4471		len -= 16 - desc->length;
4472	}
4473	if (naa != NULL) {
4474		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4475		    desc->length);
4476		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4477		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4478		    SVPD_ID_TYPE_NAA;
4479		desc->length = hex2bin(naa, desc->identifier, 16);
4480		desc->length = desc->length > 8 ? 16 : 8;
4481		len -= 16 - desc->length;
4482	}
4483	lun->lun_devid->len = len;
4484
4485	mtx_lock(&ctl_softc->ctl_lock);
4486	/*
4487	 * See if the caller requested a particular LUN number.  If so, see
4488	 * if it is available.  Otherwise, allocate the first available LUN.
4489	 */
4490	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4491		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4492		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4493			mtx_unlock(&ctl_softc->ctl_lock);
4494			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4495				printf("ctl: requested LUN ID %d is higher "
4496				       "than CTL_MAX_LUNS - 1 (%d)\n",
4497				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4498			} else {
4499				/*
4500				 * XXX KDM return an error, or just assign
4501				 * another LUN ID in this case??
4502				 */
4503				printf("ctl: requested LUN ID %d is already "
4504				       "in use\n", be_lun->req_lun_id);
4505			}
4506			if (lun->flags & CTL_LUN_MALLOCED)
4507				free(lun, M_CTL);
4508			be_lun->lun_config_status(be_lun->be_lun,
4509						  CTL_LUN_CONFIG_FAILURE);
4510			return (ENOSPC);
4511		}
4512		lun_number = be_lun->req_lun_id;
4513	} else {
4514		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4515		if (lun_number == -1) {
4516			mtx_unlock(&ctl_softc->ctl_lock);
4517			printf("ctl: can't allocate LUN, out of LUNs\n");
4518			if (lun->flags & CTL_LUN_MALLOCED)
4519				free(lun, M_CTL);
4520			be_lun->lun_config_status(be_lun->be_lun,
4521						  CTL_LUN_CONFIG_FAILURE);
4522			return (ENOSPC);
4523		}
4524	}
4525	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4526
4527	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4528	lun->lun = lun_number;
4529	lun->be_lun = be_lun;
4530	/*
4531	 * The processor LUN is always enabled.  Disk LUNs come on line
4532	 * disabled, and must be enabled by the backend.
4533	 */
4534	lun->flags |= CTL_LUN_DISABLED;
4535	lun->backend = be_lun->be;
4536	be_lun->ctl_lun = lun;
4537	be_lun->lun_id = lun_number;
4538	atomic_add_int(&be_lun->be->num_luns, 1);
4539	if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4540		lun->flags |= CTL_LUN_OFFLINE;
4541
4542	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4543		lun->flags |= CTL_LUN_STOPPED;
4544
4545	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4546		lun->flags |= CTL_LUN_INOPERABLE;
4547
4548	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4549		lun->flags |= CTL_LUN_PRIMARY_SC;
4550
4551	value = ctl_get_opt(&be_lun->options, "readonly");
4552	if (value != NULL && strcmp(value, "on") == 0)
4553		lun->flags |= CTL_LUN_READONLY;
4554
4555	lun->serseq = CTL_LUN_SERSEQ_OFF;
4556	if (be_lun->flags & CTL_LUN_FLAG_SERSEQ_READ)
4557		lun->serseq = CTL_LUN_SERSEQ_READ;
4558	value = ctl_get_opt(&be_lun->options, "serseq");
4559	if (value != NULL && strcmp(value, "on") == 0)
4560		lun->serseq = CTL_LUN_SERSEQ_ON;
4561	else if (value != NULL && strcmp(value, "read") == 0)
4562		lun->serseq = CTL_LUN_SERSEQ_READ;
4563	else if (value != NULL && strcmp(value, "off") == 0)
4564		lun->serseq = CTL_LUN_SERSEQ_OFF;
4565
4566	lun->ctl_softc = ctl_softc;
4567#ifdef CTL_TIME_IO
4568	lun->last_busy = getsbinuptime();
4569#endif
4570	TAILQ_INIT(&lun->ooa_queue);
4571	TAILQ_INIT(&lun->blocked_queue);
4572	STAILQ_INIT(&lun->error_list);
4573	ctl_tpc_lun_init(lun);
4574
4575	/*
4576	 * Initialize the mode and log page index.
4577	 */
4578	ctl_init_page_index(lun);
4579	ctl_init_log_page_index(lun);
4580
4581	/*
4582	 * Now, before we insert this lun on the lun list, set the lun
4583	 * inventory changed UA for all other luns.
4584	 */
4585	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4586		mtx_lock(&nlun->lun_lock);
4587		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4588		mtx_unlock(&nlun->lun_lock);
4589	}
4590
4591	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4592
4593	ctl_softc->ctl_luns[lun_number] = lun;
4594
4595	ctl_softc->num_luns++;
4596
4597	/* Setup statistics gathering */
4598	lun->stats.device_type = be_lun->lun_type;
4599	lun->stats.lun_number = lun_number;
4600	if (lun->stats.device_type == T_DIRECT)
4601		lun->stats.blocksize = be_lun->blocksize;
4602	else
4603		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4604	for (i = 0;i < CTL_MAX_PORTS;i++)
4605		lun->stats.ports[i].targ_port = i;
4606
4607	mtx_unlock(&ctl_softc->ctl_lock);
4608
4609	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4610	return (0);
4611}
4612
4613/*
4614 * Delete a LUN.
4615 * Assumptions:
4616 * - LUN has already been marked invalid and any pending I/O has been taken
4617 *   care of.
4618 */
4619static int
4620ctl_free_lun(struct ctl_lun *lun)
4621{
4622	struct ctl_softc *softc;
4623	struct ctl_lun *nlun;
4624	int i;
4625
4626	softc = lun->ctl_softc;
4627
4628	mtx_assert(&softc->ctl_lock, MA_OWNED);
4629
4630	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4631
4632	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4633
4634	softc->ctl_luns[lun->lun] = NULL;
4635
4636	if (!TAILQ_EMPTY(&lun->ooa_queue))
4637		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4638
4639	softc->num_luns--;
4640
4641	/*
4642	 * Tell the backend to free resources, if this LUN has a backend.
4643	 */
4644	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4645	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4646
4647	ctl_tpc_lun_shutdown(lun);
4648	mtx_destroy(&lun->lun_lock);
4649	free(lun->lun_devid, M_CTL);
4650	for (i = 0; i < CTL_MAX_PORTS; i++)
4651		free(lun->pending_ua[i], M_CTL);
4652	for (i = 0; i < 2 * CTL_MAX_PORTS; i++)
4653		free(lun->pr_keys[i], M_CTL);
4654	free(lun->write_buffer, M_CTL);
4655	if (lun->flags & CTL_LUN_MALLOCED)
4656		free(lun, M_CTL);
4657
4658	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4659		mtx_lock(&nlun->lun_lock);
4660		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4661		mtx_unlock(&nlun->lun_lock);
4662	}
4663
4664	return (0);
4665}
4666
4667static void
4668ctl_create_lun(struct ctl_be_lun *be_lun)
4669{
4670	struct ctl_softc *softc;
4671
4672	softc = control_softc;
4673
4674	/*
4675	 * ctl_alloc_lun() should handle all potential failure cases.
4676	 */
4677	ctl_alloc_lun(softc, NULL, be_lun);
4678}
4679
4680int
4681ctl_add_lun(struct ctl_be_lun *be_lun)
4682{
4683	struct ctl_softc *softc = control_softc;
4684
4685	mtx_lock(&softc->ctl_lock);
4686	STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4687	mtx_unlock(&softc->ctl_lock);
4688	wakeup(&softc->pending_lun_queue);
4689
4690	return (0);
4691}
4692
4693int
4694ctl_enable_lun(struct ctl_be_lun *be_lun)
4695{
4696	struct ctl_softc *softc;
4697	struct ctl_port *port, *nport;
4698	struct ctl_lun *lun;
4699	int retval;
4700
4701	lun = (struct ctl_lun *)be_lun->ctl_lun;
4702	softc = lun->ctl_softc;
4703
4704	mtx_lock(&softc->ctl_lock);
4705	mtx_lock(&lun->lun_lock);
4706	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4707		/*
4708		 * eh?  Why did we get called if the LUN is already
4709		 * enabled?
4710		 */
4711		mtx_unlock(&lun->lun_lock);
4712		mtx_unlock(&softc->ctl_lock);
4713		return (0);
4714	}
4715	lun->flags &= ~CTL_LUN_DISABLED;
4716	mtx_unlock(&lun->lun_lock);
4717
4718	for (port = STAILQ_FIRST(&softc->port_list); port != NULL; port = nport) {
4719		nport = STAILQ_NEXT(port, links);
4720		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4721		    port->lun_map != NULL)
4722			continue;
4723
4724		/*
4725		 * Drop the lock while we call the FETD's enable routine.
4726		 * This can lead to a callback into CTL (at least in the
4727		 * case of the internal initiator frontend.
4728		 */
4729		mtx_unlock(&softc->ctl_lock);
4730		retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4731		mtx_lock(&softc->ctl_lock);
4732		if (retval != 0) {
4733			printf("%s: FETD %s port %d returned error "
4734			       "%d for lun_enable on lun %jd\n",
4735			       __func__, port->port_name, port->targ_port,
4736			       retval, (intmax_t)lun->lun);
4737		}
4738	}
4739
4740	mtx_unlock(&softc->ctl_lock);
4741
4742	return (0);
4743}
4744
4745int
4746ctl_disable_lun(struct ctl_be_lun *be_lun)
4747{
4748	struct ctl_softc *softc;
4749	struct ctl_port *port;
4750	struct ctl_lun *lun;
4751	int retval;
4752
4753	lun = (struct ctl_lun *)be_lun->ctl_lun;
4754	softc = lun->ctl_softc;
4755
4756	mtx_lock(&softc->ctl_lock);
4757	mtx_lock(&lun->lun_lock);
4758	if (lun->flags & CTL_LUN_DISABLED) {
4759		mtx_unlock(&lun->lun_lock);
4760		mtx_unlock(&softc->ctl_lock);
4761		return (0);
4762	}
4763	lun->flags |= CTL_LUN_DISABLED;
4764	mtx_unlock(&lun->lun_lock);
4765
4766	STAILQ_FOREACH(port, &softc->port_list, links) {
4767		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4768		    port->lun_map != NULL)
4769			continue;
4770		mtx_unlock(&softc->ctl_lock);
4771		/*
4772		 * Drop the lock before we call the frontend's disable
4773		 * routine, to avoid lock order reversals.
4774		 *
4775		 * XXX KDM what happens if the frontend list changes while
4776		 * we're traversing it?  It's unlikely, but should be handled.
4777		 */
4778		retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4779		mtx_lock(&softc->ctl_lock);
4780		if (retval != 0) {
4781			printf("%s: FETD %s port %d returned error "
4782			       "%d for lun_disable on lun %jd\n",
4783			       __func__, port->port_name, port->targ_port,
4784			       retval, (intmax_t)lun->lun);
4785		}
4786	}
4787
4788	mtx_unlock(&softc->ctl_lock);
4789
4790	return (0);
4791}
4792
4793int
4794ctl_start_lun(struct ctl_be_lun *be_lun)
4795{
4796	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4797
4798	mtx_lock(&lun->lun_lock);
4799	lun->flags &= ~CTL_LUN_STOPPED;
4800	mtx_unlock(&lun->lun_lock);
4801	return (0);
4802}
4803
4804int
4805ctl_stop_lun(struct ctl_be_lun *be_lun)
4806{
4807	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4808
4809	mtx_lock(&lun->lun_lock);
4810	lun->flags |= CTL_LUN_STOPPED;
4811	mtx_unlock(&lun->lun_lock);
4812	return (0);
4813}
4814
4815int
4816ctl_lun_offline(struct ctl_be_lun *be_lun)
4817{
4818	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4819
4820	mtx_lock(&lun->lun_lock);
4821	lun->flags |= CTL_LUN_OFFLINE;
4822	mtx_unlock(&lun->lun_lock);
4823	return (0);
4824}
4825
4826int
4827ctl_lun_online(struct ctl_be_lun *be_lun)
4828{
4829	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4830
4831	mtx_lock(&lun->lun_lock);
4832	lun->flags &= ~CTL_LUN_OFFLINE;
4833	mtx_unlock(&lun->lun_lock);
4834	return (0);
4835}
4836
4837int
4838ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4839{
4840	struct ctl_softc *softc;
4841	struct ctl_lun *lun;
4842
4843	lun = (struct ctl_lun *)be_lun->ctl_lun;
4844	softc = lun->ctl_softc;
4845
4846	mtx_lock(&lun->lun_lock);
4847
4848	/*
4849	 * The LUN needs to be disabled before it can be marked invalid.
4850	 */
4851	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4852		mtx_unlock(&lun->lun_lock);
4853		return (-1);
4854	}
4855	/*
4856	 * Mark the LUN invalid.
4857	 */
4858	lun->flags |= CTL_LUN_INVALID;
4859
4860	/*
4861	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4862	 * If we have something in the OOA queue, we'll free it when the
4863	 * last I/O completes.
4864	 */
4865	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4866		mtx_unlock(&lun->lun_lock);
4867		mtx_lock(&softc->ctl_lock);
4868		ctl_free_lun(lun);
4869		mtx_unlock(&softc->ctl_lock);
4870	} else
4871		mtx_unlock(&lun->lun_lock);
4872
4873	return (0);
4874}
4875
4876int
4877ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4878{
4879	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4880
4881	mtx_lock(&lun->lun_lock);
4882	lun->flags |= CTL_LUN_INOPERABLE;
4883	mtx_unlock(&lun->lun_lock);
4884	return (0);
4885}
4886
4887int
4888ctl_lun_operable(struct ctl_be_lun *be_lun)
4889{
4890	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4891
4892	mtx_lock(&lun->lun_lock);
4893	lun->flags &= ~CTL_LUN_INOPERABLE;
4894	mtx_unlock(&lun->lun_lock);
4895	return (0);
4896}
4897
4898void
4899ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4900{
4901	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4902
4903	mtx_lock(&lun->lun_lock);
4904	ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGED);
4905	mtx_unlock(&lun->lun_lock);
4906}
4907
4908/*
4909 * Backend "memory move is complete" callback for requests that never
4910 * make it down to say RAIDCore's configuration code.
4911 */
4912int
4913ctl_config_move_done(union ctl_io *io)
4914{
4915	int retval;
4916
4917	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4918	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4919	    ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
4920
4921	if ((io->io_hdr.port_status != 0) &&
4922	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4923	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4924		/*
4925		 * For hardware error sense keys, the sense key
4926		 * specific value is defined to be a retry count,
4927		 * but we use it to pass back an internal FETD
4928		 * error code.  XXX KDM  Hopefully the FETD is only
4929		 * using 16 bits for an error code, since that's
4930		 * all the space we have in the sks field.
4931		 */
4932		ctl_set_internal_failure(&io->scsiio,
4933					 /*sks_valid*/ 1,
4934					 /*retry_count*/
4935					 io->io_hdr.port_status);
4936	}
4937
4938	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
4939	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4940	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
4941	    ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4942		/*
4943		 * XXX KDM just assuming a single pointer here, and not a
4944		 * S/G list.  If we start using S/G lists for config data,
4945		 * we'll need to know how to clean them up here as well.
4946		 */
4947		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4948			free(io->scsiio.kern_data_ptr, M_CTL);
4949		ctl_done(io);
4950		retval = CTL_RETVAL_COMPLETE;
4951	} else {
4952		/*
4953		 * XXX KDM now we need to continue data movement.  Some
4954		 * options:
4955		 * - call ctl_scsiio() again?  We don't do this for data
4956		 *   writes, because for those at least we know ahead of
4957		 *   time where the write will go and how long it is.  For
4958		 *   config writes, though, that information is largely
4959		 *   contained within the write itself, thus we need to
4960		 *   parse out the data again.
4961		 *
4962		 * - Call some other function once the data is in?
4963		 */
4964		if (ctl_debug & CTL_DEBUG_CDB_DATA)
4965			ctl_data_print(io);
4966
4967		/*
4968		 * XXX KDM call ctl_scsiio() again for now, and check flag
4969		 * bits to see whether we're allocated or not.
4970		 */
4971		retval = ctl_scsiio(&io->scsiio);
4972	}
4973	return (retval);
4974}
4975
4976/*
4977 * This gets called by a backend driver when it is done with a
4978 * data_submit method.
4979 */
4980void
4981ctl_data_submit_done(union ctl_io *io)
4982{
4983	/*
4984	 * If the IO_CONT flag is set, we need to call the supplied
4985	 * function to continue processing the I/O, instead of completing
4986	 * the I/O just yet.
4987	 *
4988	 * If there is an error, though, we don't want to keep processing.
4989	 * Instead, just send status back to the initiator.
4990	 */
4991	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4992	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4993	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4994	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4995		io->scsiio.io_cont(io);
4996		return;
4997	}
4998	ctl_done(io);
4999}
5000
5001/*
5002 * This gets called by a backend driver when it is done with a
5003 * configuration write.
5004 */
5005void
5006ctl_config_write_done(union ctl_io *io)
5007{
5008	uint8_t *buf;
5009
5010	/*
5011	 * If the IO_CONT flag is set, we need to call the supplied
5012	 * function to continue processing the I/O, instead of completing
5013	 * the I/O just yet.
5014	 *
5015	 * If there is an error, though, we don't want to keep processing.
5016	 * Instead, just send status back to the initiator.
5017	 */
5018	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5019	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5020	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5021	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5022		io->scsiio.io_cont(io);
5023		return;
5024	}
5025	/*
5026	 * Since a configuration write can be done for commands that actually
5027	 * have data allocated, like write buffer, and commands that have
5028	 * no data, like start/stop unit, we need to check here.
5029	 */
5030	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5031		buf = io->scsiio.kern_data_ptr;
5032	else
5033		buf = NULL;
5034	ctl_done(io);
5035	if (buf)
5036		free(buf, M_CTL);
5037}
5038
5039void
5040ctl_config_read_done(union ctl_io *io)
5041{
5042	uint8_t *buf;
5043
5044	/*
5045	 * If there is some error -- we are done, skip data transfer.
5046	 */
5047	if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5048	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5049	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5050		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5051			buf = io->scsiio.kern_data_ptr;
5052		else
5053			buf = NULL;
5054		ctl_done(io);
5055		if (buf)
5056			free(buf, M_CTL);
5057		return;
5058	}
5059
5060	/*
5061	 * If the IO_CONT flag is set, we need to call the supplied
5062	 * function to continue processing the I/O, instead of completing
5063	 * the I/O just yet.
5064	 */
5065	if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5066		io->scsiio.io_cont(io);
5067		return;
5068	}
5069
5070	ctl_datamove(io);
5071}
5072
5073/*
5074 * SCSI release command.
5075 */
5076int
5077ctl_scsi_release(struct ctl_scsiio *ctsio)
5078{
5079	int length, longid, thirdparty_id, resv_id;
5080	struct ctl_lun *lun;
5081	uint32_t residx;
5082
5083	length = 0;
5084	resv_id = 0;
5085
5086	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5087
5088	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5089	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5090
5091	switch (ctsio->cdb[0]) {
5092	case RELEASE_10: {
5093		struct scsi_release_10 *cdb;
5094
5095		cdb = (struct scsi_release_10 *)ctsio->cdb;
5096
5097		if (cdb->byte2 & SR10_LONGID)
5098			longid = 1;
5099		else
5100			thirdparty_id = cdb->thirdparty_id;
5101
5102		resv_id = cdb->resv_id;
5103		length = scsi_2btoul(cdb->length);
5104		break;
5105	}
5106	}
5107
5108
5109	/*
5110	 * XXX KDM right now, we only support LUN reservation.  We don't
5111	 * support 3rd party reservations, or extent reservations, which
5112	 * might actually need the parameter list.  If we've gotten this
5113	 * far, we've got a LUN reservation.  Anything else got kicked out
5114	 * above.  So, according to SPC, ignore the length.
5115	 */
5116	length = 0;
5117
5118	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5119	 && (length > 0)) {
5120		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5121		ctsio->kern_data_len = length;
5122		ctsio->kern_total_len = length;
5123		ctsio->kern_data_resid = 0;
5124		ctsio->kern_rel_offset = 0;
5125		ctsio->kern_sg_entries = 0;
5126		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5127		ctsio->be_move_done = ctl_config_move_done;
5128		ctl_datamove((union ctl_io *)ctsio);
5129
5130		return (CTL_RETVAL_COMPLETE);
5131	}
5132
5133	if (length > 0)
5134		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5135
5136	mtx_lock(&lun->lun_lock);
5137
5138	/*
5139	 * According to SPC, it is not an error for an intiator to attempt
5140	 * to release a reservation on a LUN that isn't reserved, or that
5141	 * is reserved by another initiator.  The reservation can only be
5142	 * released, though, by the initiator who made it or by one of
5143	 * several reset type events.
5144	 */
5145	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5146			lun->flags &= ~CTL_LUN_RESERVED;
5147
5148	mtx_unlock(&lun->lun_lock);
5149
5150	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5151		free(ctsio->kern_data_ptr, M_CTL);
5152		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5153	}
5154
5155	ctl_set_success(ctsio);
5156	ctl_done((union ctl_io *)ctsio);
5157	return (CTL_RETVAL_COMPLETE);
5158}
5159
5160int
5161ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5162{
5163	int extent, thirdparty, longid;
5164	int resv_id, length;
5165	uint64_t thirdparty_id;
5166	struct ctl_lun *lun;
5167	uint32_t residx;
5168
5169	extent = 0;
5170	thirdparty = 0;
5171	longid = 0;
5172	resv_id = 0;
5173	length = 0;
5174	thirdparty_id = 0;
5175
5176	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5177
5178	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5179	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5180
5181	switch (ctsio->cdb[0]) {
5182	case RESERVE_10: {
5183		struct scsi_reserve_10 *cdb;
5184
5185		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5186
5187		if (cdb->byte2 & SR10_LONGID)
5188			longid = 1;
5189		else
5190			thirdparty_id = cdb->thirdparty_id;
5191
5192		resv_id = cdb->resv_id;
5193		length = scsi_2btoul(cdb->length);
5194		break;
5195	}
5196	}
5197
5198	/*
5199	 * XXX KDM right now, we only support LUN reservation.  We don't
5200	 * support 3rd party reservations, or extent reservations, which
5201	 * might actually need the parameter list.  If we've gotten this
5202	 * far, we've got a LUN reservation.  Anything else got kicked out
5203	 * above.  So, according to SPC, ignore the length.
5204	 */
5205	length = 0;
5206
5207	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5208	 && (length > 0)) {
5209		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5210		ctsio->kern_data_len = length;
5211		ctsio->kern_total_len = length;
5212		ctsio->kern_data_resid = 0;
5213		ctsio->kern_rel_offset = 0;
5214		ctsio->kern_sg_entries = 0;
5215		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5216		ctsio->be_move_done = ctl_config_move_done;
5217		ctl_datamove((union ctl_io *)ctsio);
5218
5219		return (CTL_RETVAL_COMPLETE);
5220	}
5221
5222	if (length > 0)
5223		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5224
5225	mtx_lock(&lun->lun_lock);
5226	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5227		ctl_set_reservation_conflict(ctsio);
5228		goto bailout;
5229	}
5230
5231	lun->flags |= CTL_LUN_RESERVED;
5232	lun->res_idx = residx;
5233
5234	ctl_set_success(ctsio);
5235
5236bailout:
5237	mtx_unlock(&lun->lun_lock);
5238
5239	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5240		free(ctsio->kern_data_ptr, M_CTL);
5241		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5242	}
5243
5244	ctl_done((union ctl_io *)ctsio);
5245	return (CTL_RETVAL_COMPLETE);
5246}
5247
5248int
5249ctl_start_stop(struct ctl_scsiio *ctsio)
5250{
5251	struct scsi_start_stop_unit *cdb;
5252	struct ctl_lun *lun;
5253	int retval;
5254
5255	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5256
5257	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5258	retval = 0;
5259
5260	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5261
5262	/*
5263	 * XXX KDM
5264	 * We don't support the immediate bit on a stop unit.  In order to
5265	 * do that, we would need to code up a way to know that a stop is
5266	 * pending, and hold off any new commands until it completes, one
5267	 * way or another.  Then we could accept or reject those commands
5268	 * depending on its status.  We would almost need to do the reverse
5269	 * of what we do below for an immediate start -- return the copy of
5270	 * the ctl_io to the FETD with status to send to the host (and to
5271	 * free the copy!) and then free the original I/O once the stop
5272	 * actually completes.  That way, the OOA queue mechanism can work
5273	 * to block commands that shouldn't proceed.  Another alternative
5274	 * would be to put the copy in the queue in place of the original,
5275	 * and return the original back to the caller.  That could be
5276	 * slightly safer..
5277	 */
5278	if ((cdb->byte2 & SSS_IMMED)
5279	 && ((cdb->how & SSS_START) == 0)) {
5280		ctl_set_invalid_field(ctsio,
5281				      /*sks_valid*/ 1,
5282				      /*command*/ 1,
5283				      /*field*/ 1,
5284				      /*bit_valid*/ 1,
5285				      /*bit*/ 0);
5286		ctl_done((union ctl_io *)ctsio);
5287		return (CTL_RETVAL_COMPLETE);
5288	}
5289
5290	if ((lun->flags & CTL_LUN_PR_RESERVED)
5291	 && ((cdb->how & SSS_START)==0)) {
5292		uint32_t residx;
5293
5294		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5295		if (ctl_get_prkey(lun, residx) == 0
5296		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5297
5298			ctl_set_reservation_conflict(ctsio);
5299			ctl_done((union ctl_io *)ctsio);
5300			return (CTL_RETVAL_COMPLETE);
5301		}
5302	}
5303
5304	/*
5305	 * If there is no backend on this device, we can't start or stop
5306	 * it.  In theory we shouldn't get any start/stop commands in the
5307	 * first place at this level if the LUN doesn't have a backend.
5308	 * That should get stopped by the command decode code.
5309	 */
5310	if (lun->backend == NULL) {
5311		ctl_set_invalid_opcode(ctsio);
5312		ctl_done((union ctl_io *)ctsio);
5313		return (CTL_RETVAL_COMPLETE);
5314	}
5315
5316	/*
5317	 * XXX KDM Copan-specific offline behavior.
5318	 * Figure out a reasonable way to port this?
5319	 */
5320#ifdef NEEDTOPORT
5321	mtx_lock(&lun->lun_lock);
5322
5323	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5324	 && (lun->flags & CTL_LUN_OFFLINE)) {
5325		/*
5326		 * If the LUN is offline, and the on/offline bit isn't set,
5327		 * reject the start or stop.  Otherwise, let it through.
5328		 */
5329		mtx_unlock(&lun->lun_lock);
5330		ctl_set_lun_not_ready(ctsio);
5331		ctl_done((union ctl_io *)ctsio);
5332	} else {
5333		mtx_unlock(&lun->lun_lock);
5334#endif /* NEEDTOPORT */
5335		/*
5336		 * This could be a start or a stop when we're online,
5337		 * or a stop/offline or start/online.  A start or stop when
5338		 * we're offline is covered in the case above.
5339		 */
5340		/*
5341		 * In the non-immediate case, we send the request to
5342		 * the backend and return status to the user when
5343		 * it is done.
5344		 *
5345		 * In the immediate case, we allocate a new ctl_io
5346		 * to hold a copy of the request, and send that to
5347		 * the backend.  We then set good status on the
5348		 * user's request and return it immediately.
5349		 */
5350		if (cdb->byte2 & SSS_IMMED) {
5351			union ctl_io *new_io;
5352
5353			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5354			ctl_copy_io((union ctl_io *)ctsio, new_io);
5355			retval = lun->backend->config_write(new_io);
5356			ctl_set_success(ctsio);
5357			ctl_done((union ctl_io *)ctsio);
5358		} else {
5359			retval = lun->backend->config_write(
5360				(union ctl_io *)ctsio);
5361		}
5362#ifdef NEEDTOPORT
5363	}
5364#endif
5365	return (retval);
5366}
5367
5368/*
5369 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5370 * we don't really do anything with the LBA and length fields if the user
5371 * passes them in.  Instead we'll just flush out the cache for the entire
5372 * LUN.
5373 */
5374int
5375ctl_sync_cache(struct ctl_scsiio *ctsio)
5376{
5377	struct ctl_lun *lun;
5378	struct ctl_softc *softc;
5379	struct ctl_lba_len_flags *lbalen;
5380	uint64_t starting_lba;
5381	uint32_t block_count;
5382	int retval;
5383	uint8_t byte2;
5384
5385	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5386
5387	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5388	softc = lun->ctl_softc;
5389	retval = 0;
5390
5391	switch (ctsio->cdb[0]) {
5392	case SYNCHRONIZE_CACHE: {
5393		struct scsi_sync_cache *cdb;
5394		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5395
5396		starting_lba = scsi_4btoul(cdb->begin_lba);
5397		block_count = scsi_2btoul(cdb->lb_count);
5398		byte2 = cdb->byte2;
5399		break;
5400	}
5401	case SYNCHRONIZE_CACHE_16: {
5402		struct scsi_sync_cache_16 *cdb;
5403		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5404
5405		starting_lba = scsi_8btou64(cdb->begin_lba);
5406		block_count = scsi_4btoul(cdb->lb_count);
5407		byte2 = cdb->byte2;
5408		break;
5409	}
5410	default:
5411		ctl_set_invalid_opcode(ctsio);
5412		ctl_done((union ctl_io *)ctsio);
5413		goto bailout;
5414		break; /* NOTREACHED */
5415	}
5416
5417	/*
5418	 * We check the LBA and length, but don't do anything with them.
5419	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5420	 * get flushed.  This check will just help satisfy anyone who wants
5421	 * to see an error for an out of range LBA.
5422	 */
5423	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5424		ctl_set_lba_out_of_range(ctsio);
5425		ctl_done((union ctl_io *)ctsio);
5426		goto bailout;
5427	}
5428
5429	/*
5430	 * If this LUN has no backend, we can't flush the cache anyway.
5431	 */
5432	if (lun->backend == NULL) {
5433		ctl_set_invalid_opcode(ctsio);
5434		ctl_done((union ctl_io *)ctsio);
5435		goto bailout;
5436	}
5437
5438	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5439	lbalen->lba = starting_lba;
5440	lbalen->len = block_count;
5441	lbalen->flags = byte2;
5442
5443	/*
5444	 * Check to see whether we're configured to send the SYNCHRONIZE
5445	 * CACHE command directly to the back end.
5446	 */
5447	mtx_lock(&lun->lun_lock);
5448	if ((softc->flags & CTL_FLAG_REAL_SYNC)
5449	 && (++(lun->sync_count) >= lun->sync_interval)) {
5450		lun->sync_count = 0;
5451		mtx_unlock(&lun->lun_lock);
5452		retval = lun->backend->config_write((union ctl_io *)ctsio);
5453	} else {
5454		mtx_unlock(&lun->lun_lock);
5455		ctl_set_success(ctsio);
5456		ctl_done((union ctl_io *)ctsio);
5457	}
5458
5459bailout:
5460
5461	return (retval);
5462}
5463
5464int
5465ctl_format(struct ctl_scsiio *ctsio)
5466{
5467	struct scsi_format *cdb;
5468	struct ctl_lun *lun;
5469	int length, defect_list_len;
5470
5471	CTL_DEBUG_PRINT(("ctl_format\n"));
5472
5473	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5474
5475	cdb = (struct scsi_format *)ctsio->cdb;
5476
5477	length = 0;
5478	if (cdb->byte2 & SF_FMTDATA) {
5479		if (cdb->byte2 & SF_LONGLIST)
5480			length = sizeof(struct scsi_format_header_long);
5481		else
5482			length = sizeof(struct scsi_format_header_short);
5483	}
5484
5485	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5486	 && (length > 0)) {
5487		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5488		ctsio->kern_data_len = length;
5489		ctsio->kern_total_len = length;
5490		ctsio->kern_data_resid = 0;
5491		ctsio->kern_rel_offset = 0;
5492		ctsio->kern_sg_entries = 0;
5493		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5494		ctsio->be_move_done = ctl_config_move_done;
5495		ctl_datamove((union ctl_io *)ctsio);
5496
5497		return (CTL_RETVAL_COMPLETE);
5498	}
5499
5500	defect_list_len = 0;
5501
5502	if (cdb->byte2 & SF_FMTDATA) {
5503		if (cdb->byte2 & SF_LONGLIST) {
5504			struct scsi_format_header_long *header;
5505
5506			header = (struct scsi_format_header_long *)
5507				ctsio->kern_data_ptr;
5508
5509			defect_list_len = scsi_4btoul(header->defect_list_len);
5510			if (defect_list_len != 0) {
5511				ctl_set_invalid_field(ctsio,
5512						      /*sks_valid*/ 1,
5513						      /*command*/ 0,
5514						      /*field*/ 2,
5515						      /*bit_valid*/ 0,
5516						      /*bit*/ 0);
5517				goto bailout;
5518			}
5519		} else {
5520			struct scsi_format_header_short *header;
5521
5522			header = (struct scsi_format_header_short *)
5523				ctsio->kern_data_ptr;
5524
5525			defect_list_len = scsi_2btoul(header->defect_list_len);
5526			if (defect_list_len != 0) {
5527				ctl_set_invalid_field(ctsio,
5528						      /*sks_valid*/ 1,
5529						      /*command*/ 0,
5530						      /*field*/ 2,
5531						      /*bit_valid*/ 0,
5532						      /*bit*/ 0);
5533				goto bailout;
5534			}
5535		}
5536	}
5537
5538	/*
5539	 * The format command will clear out the "Medium format corrupted"
5540	 * status if set by the configuration code.  That status is really
5541	 * just a way to notify the host that we have lost the media, and
5542	 * get them to issue a command that will basically make them think
5543	 * they're blowing away the media.
5544	 */
5545	mtx_lock(&lun->lun_lock);
5546	lun->flags &= ~CTL_LUN_INOPERABLE;
5547	mtx_unlock(&lun->lun_lock);
5548
5549	ctl_set_success(ctsio);
5550bailout:
5551
5552	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5553		free(ctsio->kern_data_ptr, M_CTL);
5554		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5555	}
5556
5557	ctl_done((union ctl_io *)ctsio);
5558	return (CTL_RETVAL_COMPLETE);
5559}
5560
5561int
5562ctl_read_buffer(struct ctl_scsiio *ctsio)
5563{
5564	struct scsi_read_buffer *cdb;
5565	struct ctl_lun *lun;
5566	int buffer_offset, len;
5567	static uint8_t descr[4];
5568	static uint8_t echo_descr[4] = { 0 };
5569
5570	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5571
5572	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5573	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5574
5575	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5576	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5577	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5578		ctl_set_invalid_field(ctsio,
5579				      /*sks_valid*/ 1,
5580				      /*command*/ 1,
5581				      /*field*/ 1,
5582				      /*bit_valid*/ 1,
5583				      /*bit*/ 4);
5584		ctl_done((union ctl_io *)ctsio);
5585		return (CTL_RETVAL_COMPLETE);
5586	}
5587
5588	len = scsi_3btoul(cdb->length);
5589	buffer_offset = scsi_3btoul(cdb->offset);
5590
5591	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5592		ctl_set_invalid_field(ctsio,
5593				      /*sks_valid*/ 1,
5594				      /*command*/ 1,
5595				      /*field*/ 6,
5596				      /*bit_valid*/ 0,
5597				      /*bit*/ 0);
5598		ctl_done((union ctl_io *)ctsio);
5599		return (CTL_RETVAL_COMPLETE);
5600	}
5601
5602	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5603		descr[0] = 0;
5604		scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5605		ctsio->kern_data_ptr = descr;
5606		len = min(len, sizeof(descr));
5607	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5608		ctsio->kern_data_ptr = echo_descr;
5609		len = min(len, sizeof(echo_descr));
5610	} else {
5611		if (lun->write_buffer == NULL) {
5612			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5613			    M_CTL, M_WAITOK);
5614		}
5615		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5616	}
5617	ctsio->kern_data_len = len;
5618	ctsio->kern_total_len = len;
5619	ctsio->kern_data_resid = 0;
5620	ctsio->kern_rel_offset = 0;
5621	ctsio->kern_sg_entries = 0;
5622	ctl_set_success(ctsio);
5623	ctsio->be_move_done = ctl_config_move_done;
5624	ctl_datamove((union ctl_io *)ctsio);
5625	return (CTL_RETVAL_COMPLETE);
5626}
5627
5628int
5629ctl_write_buffer(struct ctl_scsiio *ctsio)
5630{
5631	struct scsi_write_buffer *cdb;
5632	struct ctl_lun *lun;
5633	int buffer_offset, len;
5634
5635	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5636
5637	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5638	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5639
5640	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5641		ctl_set_invalid_field(ctsio,
5642				      /*sks_valid*/ 1,
5643				      /*command*/ 1,
5644				      /*field*/ 1,
5645				      /*bit_valid*/ 1,
5646				      /*bit*/ 4);
5647		ctl_done((union ctl_io *)ctsio);
5648		return (CTL_RETVAL_COMPLETE);
5649	}
5650
5651	len = scsi_3btoul(cdb->length);
5652	buffer_offset = scsi_3btoul(cdb->offset);
5653
5654	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5655		ctl_set_invalid_field(ctsio,
5656				      /*sks_valid*/ 1,
5657				      /*command*/ 1,
5658				      /*field*/ 6,
5659				      /*bit_valid*/ 0,
5660				      /*bit*/ 0);
5661		ctl_done((union ctl_io *)ctsio);
5662		return (CTL_RETVAL_COMPLETE);
5663	}
5664
5665	/*
5666	 * If we've got a kernel request that hasn't been malloced yet,
5667	 * malloc it and tell the caller the data buffer is here.
5668	 */
5669	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5670		if (lun->write_buffer == NULL) {
5671			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5672			    M_CTL, M_WAITOK);
5673		}
5674		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5675		ctsio->kern_data_len = len;
5676		ctsio->kern_total_len = len;
5677		ctsio->kern_data_resid = 0;
5678		ctsio->kern_rel_offset = 0;
5679		ctsio->kern_sg_entries = 0;
5680		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5681		ctsio->be_move_done = ctl_config_move_done;
5682		ctl_datamove((union ctl_io *)ctsio);
5683
5684		return (CTL_RETVAL_COMPLETE);
5685	}
5686
5687	ctl_set_success(ctsio);
5688	ctl_done((union ctl_io *)ctsio);
5689	return (CTL_RETVAL_COMPLETE);
5690}
5691
5692int
5693ctl_write_same(struct ctl_scsiio *ctsio)
5694{
5695	struct ctl_lun *lun;
5696	struct ctl_lba_len_flags *lbalen;
5697	uint64_t lba;
5698	uint32_t num_blocks;
5699	int len, retval;
5700	uint8_t byte2;
5701
5702	retval = CTL_RETVAL_COMPLETE;
5703
5704	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5705
5706	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5707
5708	switch (ctsio->cdb[0]) {
5709	case WRITE_SAME_10: {
5710		struct scsi_write_same_10 *cdb;
5711
5712		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5713
5714		lba = scsi_4btoul(cdb->addr);
5715		num_blocks = scsi_2btoul(cdb->length);
5716		byte2 = cdb->byte2;
5717		break;
5718	}
5719	case WRITE_SAME_16: {
5720		struct scsi_write_same_16 *cdb;
5721
5722		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5723
5724		lba = scsi_8btou64(cdb->addr);
5725		num_blocks = scsi_4btoul(cdb->length);
5726		byte2 = cdb->byte2;
5727		break;
5728	}
5729	default:
5730		/*
5731		 * We got a command we don't support.  This shouldn't
5732		 * happen, commands should be filtered out above us.
5733		 */
5734		ctl_set_invalid_opcode(ctsio);
5735		ctl_done((union ctl_io *)ctsio);
5736
5737		return (CTL_RETVAL_COMPLETE);
5738		break; /* NOTREACHED */
5739	}
5740
5741	/* NDOB and ANCHOR flags can be used only together with UNMAP */
5742	if ((byte2 & SWS_UNMAP) == 0 &&
5743	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5744		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5745		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5746		ctl_done((union ctl_io *)ctsio);
5747		return (CTL_RETVAL_COMPLETE);
5748	}
5749
5750	/*
5751	 * The first check is to make sure we're in bounds, the second
5752	 * check is to catch wrap-around problems.  If the lba + num blocks
5753	 * is less than the lba, then we've wrapped around and the block
5754	 * range is invalid anyway.
5755	 */
5756	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5757	 || ((lba + num_blocks) < lba)) {
5758		ctl_set_lba_out_of_range(ctsio);
5759		ctl_done((union ctl_io *)ctsio);
5760		return (CTL_RETVAL_COMPLETE);
5761	}
5762
5763	/* Zero number of blocks means "to the last logical block" */
5764	if (num_blocks == 0) {
5765		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5766			ctl_set_invalid_field(ctsio,
5767					      /*sks_valid*/ 0,
5768					      /*command*/ 1,
5769					      /*field*/ 0,
5770					      /*bit_valid*/ 0,
5771					      /*bit*/ 0);
5772			ctl_done((union ctl_io *)ctsio);
5773			return (CTL_RETVAL_COMPLETE);
5774		}
5775		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5776	}
5777
5778	len = lun->be_lun->blocksize;
5779
5780	/*
5781	 * If we've got a kernel request that hasn't been malloced yet,
5782	 * malloc it and tell the caller the data buffer is here.
5783	 */
5784	if ((byte2 & SWS_NDOB) == 0 &&
5785	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5786		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5787		ctsio->kern_data_len = len;
5788		ctsio->kern_total_len = len;
5789		ctsio->kern_data_resid = 0;
5790		ctsio->kern_rel_offset = 0;
5791		ctsio->kern_sg_entries = 0;
5792		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5793		ctsio->be_move_done = ctl_config_move_done;
5794		ctl_datamove((union ctl_io *)ctsio);
5795
5796		return (CTL_RETVAL_COMPLETE);
5797	}
5798
5799	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5800	lbalen->lba = lba;
5801	lbalen->len = num_blocks;
5802	lbalen->flags = byte2;
5803	retval = lun->backend->config_write((union ctl_io *)ctsio);
5804
5805	return (retval);
5806}
5807
5808int
5809ctl_unmap(struct ctl_scsiio *ctsio)
5810{
5811	struct ctl_lun *lun;
5812	struct scsi_unmap *cdb;
5813	struct ctl_ptr_len_flags *ptrlen;
5814	struct scsi_unmap_header *hdr;
5815	struct scsi_unmap_desc *buf, *end, *endnz, *range;
5816	uint64_t lba;
5817	uint32_t num_blocks;
5818	int len, retval;
5819	uint8_t byte2;
5820
5821	retval = CTL_RETVAL_COMPLETE;
5822
5823	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5824
5825	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5826	cdb = (struct scsi_unmap *)ctsio->cdb;
5827
5828	len = scsi_2btoul(cdb->length);
5829	byte2 = cdb->byte2;
5830
5831	/*
5832	 * If we've got a kernel request that hasn't been malloced yet,
5833	 * malloc it and tell the caller the data buffer is here.
5834	 */
5835	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5836		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5837		ctsio->kern_data_len = len;
5838		ctsio->kern_total_len = len;
5839		ctsio->kern_data_resid = 0;
5840		ctsio->kern_rel_offset = 0;
5841		ctsio->kern_sg_entries = 0;
5842		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5843		ctsio->be_move_done = ctl_config_move_done;
5844		ctl_datamove((union ctl_io *)ctsio);
5845
5846		return (CTL_RETVAL_COMPLETE);
5847	}
5848
5849	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5850	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5851	if (len < sizeof (*hdr) ||
5852	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5853	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5854	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5855		ctl_set_invalid_field(ctsio,
5856				      /*sks_valid*/ 0,
5857				      /*command*/ 0,
5858				      /*field*/ 0,
5859				      /*bit_valid*/ 0,
5860				      /*bit*/ 0);
5861		goto done;
5862	}
5863	len = scsi_2btoul(hdr->desc_length);
5864	buf = (struct scsi_unmap_desc *)(hdr + 1);
5865	end = buf + len / sizeof(*buf);
5866
5867	endnz = buf;
5868	for (range = buf; range < end; range++) {
5869		lba = scsi_8btou64(range->lba);
5870		num_blocks = scsi_4btoul(range->length);
5871		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5872		 || ((lba + num_blocks) < lba)) {
5873			ctl_set_lba_out_of_range(ctsio);
5874			ctl_done((union ctl_io *)ctsio);
5875			return (CTL_RETVAL_COMPLETE);
5876		}
5877		if (num_blocks != 0)
5878			endnz = range + 1;
5879	}
5880
5881	/*
5882	 * Block backend can not handle zero last range.
5883	 * Filter it out and return if there is nothing left.
5884	 */
5885	len = (uint8_t *)endnz - (uint8_t *)buf;
5886	if (len == 0) {
5887		ctl_set_success(ctsio);
5888		goto done;
5889	}
5890
5891	mtx_lock(&lun->lun_lock);
5892	ptrlen = (struct ctl_ptr_len_flags *)
5893	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5894	ptrlen->ptr = (void *)buf;
5895	ptrlen->len = len;
5896	ptrlen->flags = byte2;
5897	ctl_check_blocked(lun);
5898	mtx_unlock(&lun->lun_lock);
5899
5900	retval = lun->backend->config_write((union ctl_io *)ctsio);
5901	return (retval);
5902
5903done:
5904	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5905		free(ctsio->kern_data_ptr, M_CTL);
5906		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5907	}
5908	ctl_done((union ctl_io *)ctsio);
5909	return (CTL_RETVAL_COMPLETE);
5910}
5911
5912/*
5913 * Note that this function currently doesn't actually do anything inside
5914 * CTL to enforce things if the DQue bit is turned on.
5915 *
5916 * Also note that this function can't be used in the default case, because
5917 * the DQue bit isn't set in the changeable mask for the control mode page
5918 * anyway.  This is just here as an example for how to implement a page
5919 * handler, and a placeholder in case we want to allow the user to turn
5920 * tagged queueing on and off.
5921 *
5922 * The D_SENSE bit handling is functional, however, and will turn
5923 * descriptor sense on and off for a given LUN.
5924 */
5925int
5926ctl_control_page_handler(struct ctl_scsiio *ctsio,
5927			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5928{
5929	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5930	struct ctl_lun *lun;
5931	int set_ua;
5932	uint32_t initidx;
5933
5934	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5935	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5936	set_ua = 0;
5937
5938	user_cp = (struct scsi_control_page *)page_ptr;
5939	current_cp = (struct scsi_control_page *)
5940		(page_index->page_data + (page_index->page_len *
5941		CTL_PAGE_CURRENT));
5942	saved_cp = (struct scsi_control_page *)
5943		(page_index->page_data + (page_index->page_len *
5944		CTL_PAGE_SAVED));
5945
5946	mtx_lock(&lun->lun_lock);
5947	if (((current_cp->rlec & SCP_DSENSE) == 0)
5948	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5949		/*
5950		 * Descriptor sense is currently turned off and the user
5951		 * wants to turn it on.
5952		 */
5953		current_cp->rlec |= SCP_DSENSE;
5954		saved_cp->rlec |= SCP_DSENSE;
5955		lun->flags |= CTL_LUN_SENSE_DESC;
5956		set_ua = 1;
5957	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
5958		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
5959		/*
5960		 * Descriptor sense is currently turned on, and the user
5961		 * wants to turn it off.
5962		 */
5963		current_cp->rlec &= ~SCP_DSENSE;
5964		saved_cp->rlec &= ~SCP_DSENSE;
5965		lun->flags &= ~CTL_LUN_SENSE_DESC;
5966		set_ua = 1;
5967	}
5968	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
5969	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
5970		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5971		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5972		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5973		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5974		set_ua = 1;
5975	}
5976	if ((current_cp->eca_and_aen & SCP_SWP) !=
5977	    (user_cp->eca_and_aen & SCP_SWP)) {
5978		current_cp->eca_and_aen &= ~SCP_SWP;
5979		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5980		saved_cp->eca_and_aen &= ~SCP_SWP;
5981		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5982		set_ua = 1;
5983	}
5984	if (set_ua != 0)
5985		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5986	mtx_unlock(&lun->lun_lock);
5987
5988	return (0);
5989}
5990
5991int
5992ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
5993		     struct ctl_page_index *page_index, uint8_t *page_ptr)
5994{
5995	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
5996	struct ctl_lun *lun;
5997	int set_ua;
5998	uint32_t initidx;
5999
6000	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6001	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6002	set_ua = 0;
6003
6004	user_cp = (struct scsi_caching_page *)page_ptr;
6005	current_cp = (struct scsi_caching_page *)
6006		(page_index->page_data + (page_index->page_len *
6007		CTL_PAGE_CURRENT));
6008	saved_cp = (struct scsi_caching_page *)
6009		(page_index->page_data + (page_index->page_len *
6010		CTL_PAGE_SAVED));
6011
6012	mtx_lock(&lun->lun_lock);
6013	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6014	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6015		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6016		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6017		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6018		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6019		set_ua = 1;
6020	}
6021	if (set_ua != 0)
6022		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6023	mtx_unlock(&lun->lun_lock);
6024
6025	return (0);
6026}
6027
6028int
6029ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6030				struct ctl_page_index *page_index,
6031				uint8_t *page_ptr)
6032{
6033	uint8_t *c;
6034	int i;
6035
6036	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6037	ctl_time_io_secs =
6038		(c[0] << 8) |
6039		(c[1] << 0) |
6040		0;
6041	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6042	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6043	printf("page data:");
6044	for (i=0; i<8; i++)
6045		printf(" %.2x",page_ptr[i]);
6046	printf("\n");
6047	return (0);
6048}
6049
6050int
6051ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6052			       struct ctl_page_index *page_index,
6053			       int pc)
6054{
6055	struct copan_debugconf_subpage *page;
6056
6057	page = (struct copan_debugconf_subpage *)page_index->page_data +
6058		(page_index->page_len * pc);
6059
6060	switch (pc) {
6061	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6062	case SMS_PAGE_CTRL_DEFAULT >> 6:
6063	case SMS_PAGE_CTRL_SAVED >> 6:
6064		/*
6065		 * We don't update the changable or default bits for this page.
6066		 */
6067		break;
6068	case SMS_PAGE_CTRL_CURRENT >> 6:
6069		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6070		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6071		break;
6072	default:
6073#ifdef NEEDTOPORT
6074		EPRINT(0, "Invalid PC %d!!", pc);
6075#endif /* NEEDTOPORT */
6076		break;
6077	}
6078	return (0);
6079}
6080
6081
6082static int
6083ctl_do_mode_select(union ctl_io *io)
6084{
6085	struct scsi_mode_page_header *page_header;
6086	struct ctl_page_index *page_index;
6087	struct ctl_scsiio *ctsio;
6088	int control_dev, page_len;
6089	int page_len_offset, page_len_size;
6090	union ctl_modepage_info *modepage_info;
6091	struct ctl_lun *lun;
6092	int *len_left, *len_used;
6093	int retval, i;
6094
6095	ctsio = &io->scsiio;
6096	page_index = NULL;
6097	page_len = 0;
6098	retval = CTL_RETVAL_COMPLETE;
6099
6100	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6101
6102	if (lun->be_lun->lun_type != T_DIRECT)
6103		control_dev = 1;
6104	else
6105		control_dev = 0;
6106
6107	modepage_info = (union ctl_modepage_info *)
6108		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6109	len_left = &modepage_info->header.len_left;
6110	len_used = &modepage_info->header.len_used;
6111
6112do_next_page:
6113
6114	page_header = (struct scsi_mode_page_header *)
6115		(ctsio->kern_data_ptr + *len_used);
6116
6117	if (*len_left == 0) {
6118		free(ctsio->kern_data_ptr, M_CTL);
6119		ctl_set_success(ctsio);
6120		ctl_done((union ctl_io *)ctsio);
6121		return (CTL_RETVAL_COMPLETE);
6122	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6123
6124		free(ctsio->kern_data_ptr, M_CTL);
6125		ctl_set_param_len_error(ctsio);
6126		ctl_done((union ctl_io *)ctsio);
6127		return (CTL_RETVAL_COMPLETE);
6128
6129	} else if ((page_header->page_code & SMPH_SPF)
6130		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6131
6132		free(ctsio->kern_data_ptr, M_CTL);
6133		ctl_set_param_len_error(ctsio);
6134		ctl_done((union ctl_io *)ctsio);
6135		return (CTL_RETVAL_COMPLETE);
6136	}
6137
6138
6139	/*
6140	 * XXX KDM should we do something with the block descriptor?
6141	 */
6142	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6143
6144		if ((control_dev != 0)
6145		 && (lun->mode_pages.index[i].page_flags &
6146		     CTL_PAGE_FLAG_DISK_ONLY))
6147			continue;
6148
6149		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6150		    (page_header->page_code & SMPH_PC_MASK))
6151			continue;
6152
6153		/*
6154		 * If neither page has a subpage code, then we've got a
6155		 * match.
6156		 */
6157		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6158		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6159			page_index = &lun->mode_pages.index[i];
6160			page_len = page_header->page_length;
6161			break;
6162		}
6163
6164		/*
6165		 * If both pages have subpages, then the subpage numbers
6166		 * have to match.
6167		 */
6168		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6169		  && (page_header->page_code & SMPH_SPF)) {
6170			struct scsi_mode_page_header_sp *sph;
6171
6172			sph = (struct scsi_mode_page_header_sp *)page_header;
6173
6174			if (lun->mode_pages.index[i].subpage ==
6175			    sph->subpage) {
6176				page_index = &lun->mode_pages.index[i];
6177				page_len = scsi_2btoul(sph->page_length);
6178				break;
6179			}
6180		}
6181	}
6182
6183	/*
6184	 * If we couldn't find the page, or if we don't have a mode select
6185	 * handler for it, send back an error to the user.
6186	 */
6187	if ((page_index == NULL)
6188	 || (page_index->select_handler == NULL)) {
6189		ctl_set_invalid_field(ctsio,
6190				      /*sks_valid*/ 1,
6191				      /*command*/ 0,
6192				      /*field*/ *len_used,
6193				      /*bit_valid*/ 0,
6194				      /*bit*/ 0);
6195		free(ctsio->kern_data_ptr, M_CTL);
6196		ctl_done((union ctl_io *)ctsio);
6197		return (CTL_RETVAL_COMPLETE);
6198	}
6199
6200	if (page_index->page_code & SMPH_SPF) {
6201		page_len_offset = 2;
6202		page_len_size = 2;
6203	} else {
6204		page_len_size = 1;
6205		page_len_offset = 1;
6206	}
6207
6208	/*
6209	 * If the length the initiator gives us isn't the one we specify in
6210	 * the mode page header, or if they didn't specify enough data in
6211	 * the CDB to avoid truncating this page, kick out the request.
6212	 */
6213	if ((page_len != (page_index->page_len - page_len_offset -
6214			  page_len_size))
6215	 || (*len_left < page_index->page_len)) {
6216
6217
6218		ctl_set_invalid_field(ctsio,
6219				      /*sks_valid*/ 1,
6220				      /*command*/ 0,
6221				      /*field*/ *len_used + page_len_offset,
6222				      /*bit_valid*/ 0,
6223				      /*bit*/ 0);
6224		free(ctsio->kern_data_ptr, M_CTL);
6225		ctl_done((union ctl_io *)ctsio);
6226		return (CTL_RETVAL_COMPLETE);
6227	}
6228
6229	/*
6230	 * Run through the mode page, checking to make sure that the bits
6231	 * the user changed are actually legal for him to change.
6232	 */
6233	for (i = 0; i < page_index->page_len; i++) {
6234		uint8_t *user_byte, *change_mask, *current_byte;
6235		int bad_bit;
6236		int j;
6237
6238		user_byte = (uint8_t *)page_header + i;
6239		change_mask = page_index->page_data +
6240			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6241		current_byte = page_index->page_data +
6242			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6243
6244		/*
6245		 * Check to see whether the user set any bits in this byte
6246		 * that he is not allowed to set.
6247		 */
6248		if ((*user_byte & ~(*change_mask)) ==
6249		    (*current_byte & ~(*change_mask)))
6250			continue;
6251
6252		/*
6253		 * Go through bit by bit to determine which one is illegal.
6254		 */
6255		bad_bit = 0;
6256		for (j = 7; j >= 0; j--) {
6257			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6258			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6259				bad_bit = i;
6260				break;
6261			}
6262		}
6263		ctl_set_invalid_field(ctsio,
6264				      /*sks_valid*/ 1,
6265				      /*command*/ 0,
6266				      /*field*/ *len_used + i,
6267				      /*bit_valid*/ 1,
6268				      /*bit*/ bad_bit);
6269		free(ctsio->kern_data_ptr, M_CTL);
6270		ctl_done((union ctl_io *)ctsio);
6271		return (CTL_RETVAL_COMPLETE);
6272	}
6273
6274	/*
6275	 * Decrement these before we call the page handler, since we may
6276	 * end up getting called back one way or another before the handler
6277	 * returns to this context.
6278	 */
6279	*len_left -= page_index->page_len;
6280	*len_used += page_index->page_len;
6281
6282	retval = page_index->select_handler(ctsio, page_index,
6283					    (uint8_t *)page_header);
6284
6285	/*
6286	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6287	 * wait until this queued command completes to finish processing
6288	 * the mode page.  If it returns anything other than
6289	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6290	 * already set the sense information, freed the data pointer, and
6291	 * completed the io for us.
6292	 */
6293	if (retval != CTL_RETVAL_COMPLETE)
6294		goto bailout_no_done;
6295
6296	/*
6297	 * If the initiator sent us more than one page, parse the next one.
6298	 */
6299	if (*len_left > 0)
6300		goto do_next_page;
6301
6302	ctl_set_success(ctsio);
6303	free(ctsio->kern_data_ptr, M_CTL);
6304	ctl_done((union ctl_io *)ctsio);
6305
6306bailout_no_done:
6307
6308	return (CTL_RETVAL_COMPLETE);
6309
6310}
6311
6312int
6313ctl_mode_select(struct ctl_scsiio *ctsio)
6314{
6315	int param_len, pf, sp;
6316	int header_size, bd_len;
6317	int len_left, len_used;
6318	struct ctl_page_index *page_index;
6319	struct ctl_lun *lun;
6320	int control_dev, page_len;
6321	union ctl_modepage_info *modepage_info;
6322	int retval;
6323
6324	pf = 0;
6325	sp = 0;
6326	page_len = 0;
6327	len_used = 0;
6328	len_left = 0;
6329	retval = 0;
6330	bd_len = 0;
6331	page_index = NULL;
6332
6333	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6334
6335	if (lun->be_lun->lun_type != T_DIRECT)
6336		control_dev = 1;
6337	else
6338		control_dev = 0;
6339
6340	switch (ctsio->cdb[0]) {
6341	case MODE_SELECT_6: {
6342		struct scsi_mode_select_6 *cdb;
6343
6344		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6345
6346		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6347		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6348
6349		param_len = cdb->length;
6350		header_size = sizeof(struct scsi_mode_header_6);
6351		break;
6352	}
6353	case MODE_SELECT_10: {
6354		struct scsi_mode_select_10 *cdb;
6355
6356		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6357
6358		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6359		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6360
6361		param_len = scsi_2btoul(cdb->length);
6362		header_size = sizeof(struct scsi_mode_header_10);
6363		break;
6364	}
6365	default:
6366		ctl_set_invalid_opcode(ctsio);
6367		ctl_done((union ctl_io *)ctsio);
6368		return (CTL_RETVAL_COMPLETE);
6369		break; /* NOTREACHED */
6370	}
6371
6372	/*
6373	 * From SPC-3:
6374	 * "A parameter list length of zero indicates that the Data-Out Buffer
6375	 * shall be empty. This condition shall not be considered as an error."
6376	 */
6377	if (param_len == 0) {
6378		ctl_set_success(ctsio);
6379		ctl_done((union ctl_io *)ctsio);
6380		return (CTL_RETVAL_COMPLETE);
6381	}
6382
6383	/*
6384	 * Since we'll hit this the first time through, prior to
6385	 * allocation, we don't need to free a data buffer here.
6386	 */
6387	if (param_len < header_size) {
6388		ctl_set_param_len_error(ctsio);
6389		ctl_done((union ctl_io *)ctsio);
6390		return (CTL_RETVAL_COMPLETE);
6391	}
6392
6393	/*
6394	 * Allocate the data buffer and grab the user's data.  In theory,
6395	 * we shouldn't have to sanity check the parameter list length here
6396	 * because the maximum size is 64K.  We should be able to malloc
6397	 * that much without too many problems.
6398	 */
6399	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6400		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6401		ctsio->kern_data_len = param_len;
6402		ctsio->kern_total_len = param_len;
6403		ctsio->kern_data_resid = 0;
6404		ctsio->kern_rel_offset = 0;
6405		ctsio->kern_sg_entries = 0;
6406		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6407		ctsio->be_move_done = ctl_config_move_done;
6408		ctl_datamove((union ctl_io *)ctsio);
6409
6410		return (CTL_RETVAL_COMPLETE);
6411	}
6412
6413	switch (ctsio->cdb[0]) {
6414	case MODE_SELECT_6: {
6415		struct scsi_mode_header_6 *mh6;
6416
6417		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6418		bd_len = mh6->blk_desc_len;
6419		break;
6420	}
6421	case MODE_SELECT_10: {
6422		struct scsi_mode_header_10 *mh10;
6423
6424		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6425		bd_len = scsi_2btoul(mh10->blk_desc_len);
6426		break;
6427	}
6428	default:
6429		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6430		break;
6431	}
6432
6433	if (param_len < (header_size + bd_len)) {
6434		free(ctsio->kern_data_ptr, M_CTL);
6435		ctl_set_param_len_error(ctsio);
6436		ctl_done((union ctl_io *)ctsio);
6437		return (CTL_RETVAL_COMPLETE);
6438	}
6439
6440	/*
6441	 * Set the IO_CONT flag, so that if this I/O gets passed to
6442	 * ctl_config_write_done(), it'll get passed back to
6443	 * ctl_do_mode_select() for further processing, or completion if
6444	 * we're all done.
6445	 */
6446	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6447	ctsio->io_cont = ctl_do_mode_select;
6448
6449	modepage_info = (union ctl_modepage_info *)
6450		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6451
6452	memset(modepage_info, 0, sizeof(*modepage_info));
6453
6454	len_left = param_len - header_size - bd_len;
6455	len_used = header_size + bd_len;
6456
6457	modepage_info->header.len_left = len_left;
6458	modepage_info->header.len_used = len_used;
6459
6460	return (ctl_do_mode_select((union ctl_io *)ctsio));
6461}
6462
6463int
6464ctl_mode_sense(struct ctl_scsiio *ctsio)
6465{
6466	struct ctl_lun *lun;
6467	int pc, page_code, dbd, llba, subpage;
6468	int alloc_len, page_len, header_len, total_len;
6469	struct scsi_mode_block_descr *block_desc;
6470	struct ctl_page_index *page_index;
6471	int control_dev;
6472
6473	dbd = 0;
6474	llba = 0;
6475	block_desc = NULL;
6476	page_index = NULL;
6477
6478	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6479
6480	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6481
6482	if (lun->be_lun->lun_type != T_DIRECT)
6483		control_dev = 1;
6484	else
6485		control_dev = 0;
6486
6487	switch (ctsio->cdb[0]) {
6488	case MODE_SENSE_6: {
6489		struct scsi_mode_sense_6 *cdb;
6490
6491		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6492
6493		header_len = sizeof(struct scsi_mode_hdr_6);
6494		if (cdb->byte2 & SMS_DBD)
6495			dbd = 1;
6496		else
6497			header_len += sizeof(struct scsi_mode_block_descr);
6498
6499		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6500		page_code = cdb->page & SMS_PAGE_CODE;
6501		subpage = cdb->subpage;
6502		alloc_len = cdb->length;
6503		break;
6504	}
6505	case MODE_SENSE_10: {
6506		struct scsi_mode_sense_10 *cdb;
6507
6508		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6509
6510		header_len = sizeof(struct scsi_mode_hdr_10);
6511
6512		if (cdb->byte2 & SMS_DBD)
6513			dbd = 1;
6514		else
6515			header_len += sizeof(struct scsi_mode_block_descr);
6516		if (cdb->byte2 & SMS10_LLBAA)
6517			llba = 1;
6518		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6519		page_code = cdb->page & SMS_PAGE_CODE;
6520		subpage = cdb->subpage;
6521		alloc_len = scsi_2btoul(cdb->length);
6522		break;
6523	}
6524	default:
6525		ctl_set_invalid_opcode(ctsio);
6526		ctl_done((union ctl_io *)ctsio);
6527		return (CTL_RETVAL_COMPLETE);
6528		break; /* NOTREACHED */
6529	}
6530
6531	/*
6532	 * We have to make a first pass through to calculate the size of
6533	 * the pages that match the user's query.  Then we allocate enough
6534	 * memory to hold it, and actually copy the data into the buffer.
6535	 */
6536	switch (page_code) {
6537	case SMS_ALL_PAGES_PAGE: {
6538		int i;
6539
6540		page_len = 0;
6541
6542		/*
6543		 * At the moment, values other than 0 and 0xff here are
6544		 * reserved according to SPC-3.
6545		 */
6546		if ((subpage != SMS_SUBPAGE_PAGE_0)
6547		 && (subpage != SMS_SUBPAGE_ALL)) {
6548			ctl_set_invalid_field(ctsio,
6549					      /*sks_valid*/ 1,
6550					      /*command*/ 1,
6551					      /*field*/ 3,
6552					      /*bit_valid*/ 0,
6553					      /*bit*/ 0);
6554			ctl_done((union ctl_io *)ctsio);
6555			return (CTL_RETVAL_COMPLETE);
6556		}
6557
6558		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6559			if ((control_dev != 0)
6560			 && (lun->mode_pages.index[i].page_flags &
6561			     CTL_PAGE_FLAG_DISK_ONLY))
6562				continue;
6563
6564			/*
6565			 * We don't use this subpage if the user didn't
6566			 * request all subpages.
6567			 */
6568			if ((lun->mode_pages.index[i].subpage != 0)
6569			 && (subpage == SMS_SUBPAGE_PAGE_0))
6570				continue;
6571
6572#if 0
6573			printf("found page %#x len %d\n",
6574			       lun->mode_pages.index[i].page_code &
6575			       SMPH_PC_MASK,
6576			       lun->mode_pages.index[i].page_len);
6577#endif
6578			page_len += lun->mode_pages.index[i].page_len;
6579		}
6580		break;
6581	}
6582	default: {
6583		int i;
6584
6585		page_len = 0;
6586
6587		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6588			/* Look for the right page code */
6589			if ((lun->mode_pages.index[i].page_code &
6590			     SMPH_PC_MASK) != page_code)
6591				continue;
6592
6593			/* Look for the right subpage or the subpage wildcard*/
6594			if ((lun->mode_pages.index[i].subpage != subpage)
6595			 && (subpage != SMS_SUBPAGE_ALL))
6596				continue;
6597
6598			/* Make sure the page is supported for this dev type */
6599			if ((control_dev != 0)
6600			 && (lun->mode_pages.index[i].page_flags &
6601			     CTL_PAGE_FLAG_DISK_ONLY))
6602				continue;
6603
6604#if 0
6605			printf("found page %#x len %d\n",
6606			       lun->mode_pages.index[i].page_code &
6607			       SMPH_PC_MASK,
6608			       lun->mode_pages.index[i].page_len);
6609#endif
6610
6611			page_len += lun->mode_pages.index[i].page_len;
6612		}
6613
6614		if (page_len == 0) {
6615			ctl_set_invalid_field(ctsio,
6616					      /*sks_valid*/ 1,
6617					      /*command*/ 1,
6618					      /*field*/ 2,
6619					      /*bit_valid*/ 1,
6620					      /*bit*/ 5);
6621			ctl_done((union ctl_io *)ctsio);
6622			return (CTL_RETVAL_COMPLETE);
6623		}
6624		break;
6625	}
6626	}
6627
6628	total_len = header_len + page_len;
6629#if 0
6630	printf("header_len = %d, page_len = %d, total_len = %d\n",
6631	       header_len, page_len, total_len);
6632#endif
6633
6634	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6635	ctsio->kern_sg_entries = 0;
6636	ctsio->kern_data_resid = 0;
6637	ctsio->kern_rel_offset = 0;
6638	if (total_len < alloc_len) {
6639		ctsio->residual = alloc_len - total_len;
6640		ctsio->kern_data_len = total_len;
6641		ctsio->kern_total_len = total_len;
6642	} else {
6643		ctsio->residual = 0;
6644		ctsio->kern_data_len = alloc_len;
6645		ctsio->kern_total_len = alloc_len;
6646	}
6647
6648	switch (ctsio->cdb[0]) {
6649	case MODE_SENSE_6: {
6650		struct scsi_mode_hdr_6 *header;
6651
6652		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6653
6654		header->datalen = MIN(total_len - 1, 254);
6655		if (control_dev == 0) {
6656			header->dev_specific = 0x10; /* DPOFUA */
6657			if ((lun->flags & CTL_LUN_READONLY) ||
6658			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6659			    .eca_and_aen & SCP_SWP) != 0)
6660				    header->dev_specific |= 0x80; /* WP */
6661		}
6662		if (dbd)
6663			header->block_descr_len = 0;
6664		else
6665			header->block_descr_len =
6666				sizeof(struct scsi_mode_block_descr);
6667		block_desc = (struct scsi_mode_block_descr *)&header[1];
6668		break;
6669	}
6670	case MODE_SENSE_10: {
6671		struct scsi_mode_hdr_10 *header;
6672		int datalen;
6673
6674		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6675
6676		datalen = MIN(total_len - 2, 65533);
6677		scsi_ulto2b(datalen, header->datalen);
6678		if (control_dev == 0) {
6679			header->dev_specific = 0x10; /* DPOFUA */
6680			if ((lun->flags & CTL_LUN_READONLY) ||
6681			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6682			    .eca_and_aen & SCP_SWP) != 0)
6683				    header->dev_specific |= 0x80; /* WP */
6684		}
6685		if (dbd)
6686			scsi_ulto2b(0, header->block_descr_len);
6687		else
6688			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6689				    header->block_descr_len);
6690		block_desc = (struct scsi_mode_block_descr *)&header[1];
6691		break;
6692	}
6693	default:
6694		panic("invalid CDB type %#x", ctsio->cdb[0]);
6695		break; /* NOTREACHED */
6696	}
6697
6698	/*
6699	 * If we've got a disk, use its blocksize in the block
6700	 * descriptor.  Otherwise, just set it to 0.
6701	 */
6702	if (dbd == 0) {
6703		if (control_dev == 0)
6704			scsi_ulto3b(lun->be_lun->blocksize,
6705				    block_desc->block_len);
6706		else
6707			scsi_ulto3b(0, block_desc->block_len);
6708	}
6709
6710	switch (page_code) {
6711	case SMS_ALL_PAGES_PAGE: {
6712		int i, data_used;
6713
6714		data_used = header_len;
6715		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6716			struct ctl_page_index *page_index;
6717
6718			page_index = &lun->mode_pages.index[i];
6719
6720			if ((control_dev != 0)
6721			 && (page_index->page_flags &
6722			    CTL_PAGE_FLAG_DISK_ONLY))
6723				continue;
6724
6725			/*
6726			 * We don't use this subpage if the user didn't
6727			 * request all subpages.  We already checked (above)
6728			 * to make sure the user only specified a subpage
6729			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6730			 */
6731			if ((page_index->subpage != 0)
6732			 && (subpage == SMS_SUBPAGE_PAGE_0))
6733				continue;
6734
6735			/*
6736			 * Call the handler, if it exists, to update the
6737			 * page to the latest values.
6738			 */
6739			if (page_index->sense_handler != NULL)
6740				page_index->sense_handler(ctsio, page_index,pc);
6741
6742			memcpy(ctsio->kern_data_ptr + data_used,
6743			       page_index->page_data +
6744			       (page_index->page_len * pc),
6745			       page_index->page_len);
6746			data_used += page_index->page_len;
6747		}
6748		break;
6749	}
6750	default: {
6751		int i, data_used;
6752
6753		data_used = header_len;
6754
6755		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6756			struct ctl_page_index *page_index;
6757
6758			page_index = &lun->mode_pages.index[i];
6759
6760			/* Look for the right page code */
6761			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6762				continue;
6763
6764			/* Look for the right subpage or the subpage wildcard*/
6765			if ((page_index->subpage != subpage)
6766			 && (subpage != SMS_SUBPAGE_ALL))
6767				continue;
6768
6769			/* Make sure the page is supported for this dev type */
6770			if ((control_dev != 0)
6771			 && (page_index->page_flags &
6772			     CTL_PAGE_FLAG_DISK_ONLY))
6773				continue;
6774
6775			/*
6776			 * Call the handler, if it exists, to update the
6777			 * page to the latest values.
6778			 */
6779			if (page_index->sense_handler != NULL)
6780				page_index->sense_handler(ctsio, page_index,pc);
6781
6782			memcpy(ctsio->kern_data_ptr + data_used,
6783			       page_index->page_data +
6784			       (page_index->page_len * pc),
6785			       page_index->page_len);
6786			data_used += page_index->page_len;
6787		}
6788		break;
6789	}
6790	}
6791
6792	ctl_set_success(ctsio);
6793	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6794	ctsio->be_move_done = ctl_config_move_done;
6795	ctl_datamove((union ctl_io *)ctsio);
6796	return (CTL_RETVAL_COMPLETE);
6797}
6798
6799int
6800ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6801			       struct ctl_page_index *page_index,
6802			       int pc)
6803{
6804	struct ctl_lun *lun;
6805	struct scsi_log_param_header *phdr;
6806	uint8_t *data;
6807	uint64_t val;
6808
6809	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6810	data = page_index->page_data;
6811
6812	if (lun->backend->lun_attr != NULL &&
6813	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6814	     != UINT64_MAX) {
6815		phdr = (struct scsi_log_param_header *)data;
6816		scsi_ulto2b(0x0001, phdr->param_code);
6817		phdr->param_control = SLP_LBIN | SLP_LP;
6818		phdr->param_len = 8;
6819		data = (uint8_t *)(phdr + 1);
6820		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6821		data[4] = 0x02; /* per-pool */
6822		data += phdr->param_len;
6823	}
6824
6825	if (lun->backend->lun_attr != NULL &&
6826	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6827	     != UINT64_MAX) {
6828		phdr = (struct scsi_log_param_header *)data;
6829		scsi_ulto2b(0x0002, phdr->param_code);
6830		phdr->param_control = SLP_LBIN | SLP_LP;
6831		phdr->param_len = 8;
6832		data = (uint8_t *)(phdr + 1);
6833		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6834		data[4] = 0x01; /* per-LUN */
6835		data += phdr->param_len;
6836	}
6837
6838	if (lun->backend->lun_attr != NULL &&
6839	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6840	     != UINT64_MAX) {
6841		phdr = (struct scsi_log_param_header *)data;
6842		scsi_ulto2b(0x00f1, phdr->param_code);
6843		phdr->param_control = SLP_LBIN | SLP_LP;
6844		phdr->param_len = 8;
6845		data = (uint8_t *)(phdr + 1);
6846		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6847		data[4] = 0x02; /* per-pool */
6848		data += phdr->param_len;
6849	}
6850
6851	if (lun->backend->lun_attr != NULL &&
6852	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6853	     != UINT64_MAX) {
6854		phdr = (struct scsi_log_param_header *)data;
6855		scsi_ulto2b(0x00f2, phdr->param_code);
6856		phdr->param_control = SLP_LBIN | SLP_LP;
6857		phdr->param_len = 8;
6858		data = (uint8_t *)(phdr + 1);
6859		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6860		data[4] = 0x02; /* per-pool */
6861		data += phdr->param_len;
6862	}
6863
6864	page_index->page_len = data - page_index->page_data;
6865	return (0);
6866}
6867
6868int
6869ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6870			       struct ctl_page_index *page_index,
6871			       int pc)
6872{
6873	struct ctl_lun *lun;
6874	struct stat_page *data;
6875	uint64_t rn, wn, rb, wb;
6876	struct bintime rt, wt;
6877	int i;
6878
6879	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6880	data = (struct stat_page *)page_index->page_data;
6881
6882	scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6883	data->sap.hdr.param_control = SLP_LBIN;
6884	data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6885	    sizeof(struct scsi_log_param_header);
6886	rn = wn = rb = wb = 0;
6887	bintime_clear(&rt);
6888	bintime_clear(&wt);
6889	for (i = 0; i < CTL_MAX_PORTS; i++) {
6890		rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6891		wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6892		rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6893		wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6894		bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6895		bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6896	}
6897	scsi_u64to8b(rn, data->sap.read_num);
6898	scsi_u64to8b(wn, data->sap.write_num);
6899	if (lun->stats.blocksize > 0) {
6900		scsi_u64to8b(wb / lun->stats.blocksize,
6901		    data->sap.recvieved_lba);
6902		scsi_u64to8b(rb / lun->stats.blocksize,
6903		    data->sap.transmitted_lba);
6904	}
6905	scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6906	    data->sap.read_int);
6907	scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6908	    data->sap.write_int);
6909	scsi_u64to8b(0, data->sap.weighted_num);
6910	scsi_u64to8b(0, data->sap.weighted_int);
6911	scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6912	data->it.hdr.param_control = SLP_LBIN;
6913	data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6914	    sizeof(struct scsi_log_param_header);
6915#ifdef CTL_TIME_IO
6916	scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6917#endif
6918	scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6919	data->it.hdr.param_control = SLP_LBIN;
6920	data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6921	    sizeof(struct scsi_log_param_header);
6922	scsi_ulto4b(3, data->ti.exponent);
6923	scsi_ulto4b(1, data->ti.integer);
6924
6925	page_index->page_len = sizeof(*data);
6926	return (0);
6927}
6928
6929int
6930ctl_log_sense(struct ctl_scsiio *ctsio)
6931{
6932	struct ctl_lun *lun;
6933	int i, pc, page_code, subpage;
6934	int alloc_len, total_len;
6935	struct ctl_page_index *page_index;
6936	struct scsi_log_sense *cdb;
6937	struct scsi_log_header *header;
6938
6939	CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6940
6941	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6942	cdb = (struct scsi_log_sense *)ctsio->cdb;
6943	pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6944	page_code = cdb->page & SLS_PAGE_CODE;
6945	subpage = cdb->subpage;
6946	alloc_len = scsi_2btoul(cdb->length);
6947
6948	page_index = NULL;
6949	for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6950		page_index = &lun->log_pages.index[i];
6951
6952		/* Look for the right page code */
6953		if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6954			continue;
6955
6956		/* Look for the right subpage or the subpage wildcard*/
6957		if (page_index->subpage != subpage)
6958			continue;
6959
6960		break;
6961	}
6962	if (i >= CTL_NUM_LOG_PAGES) {
6963		ctl_set_invalid_field(ctsio,
6964				      /*sks_valid*/ 1,
6965				      /*command*/ 1,
6966				      /*field*/ 2,
6967				      /*bit_valid*/ 0,
6968				      /*bit*/ 0);
6969		ctl_done((union ctl_io *)ctsio);
6970		return (CTL_RETVAL_COMPLETE);
6971	}
6972
6973	total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6974
6975	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6976	ctsio->kern_sg_entries = 0;
6977	ctsio->kern_data_resid = 0;
6978	ctsio->kern_rel_offset = 0;
6979	if (total_len < alloc_len) {
6980		ctsio->residual = alloc_len - total_len;
6981		ctsio->kern_data_len = total_len;
6982		ctsio->kern_total_len = total_len;
6983	} else {
6984		ctsio->residual = 0;
6985		ctsio->kern_data_len = alloc_len;
6986		ctsio->kern_total_len = alloc_len;
6987	}
6988
6989	header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6990	header->page = page_index->page_code;
6991	if (page_index->subpage) {
6992		header->page |= SL_SPF;
6993		header->subpage = page_index->subpage;
6994	}
6995	scsi_ulto2b(page_index->page_len, header->datalen);
6996
6997	/*
6998	 * Call the handler, if it exists, to update the
6999	 * page to the latest values.
7000	 */
7001	if (page_index->sense_handler != NULL)
7002		page_index->sense_handler(ctsio, page_index, pc);
7003
7004	memcpy(header + 1, page_index->page_data, page_index->page_len);
7005
7006	ctl_set_success(ctsio);
7007	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7008	ctsio->be_move_done = ctl_config_move_done;
7009	ctl_datamove((union ctl_io *)ctsio);
7010	return (CTL_RETVAL_COMPLETE);
7011}
7012
7013int
7014ctl_read_capacity(struct ctl_scsiio *ctsio)
7015{
7016	struct scsi_read_capacity *cdb;
7017	struct scsi_read_capacity_data *data;
7018	struct ctl_lun *lun;
7019	uint32_t lba;
7020
7021	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7022
7023	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7024
7025	lba = scsi_4btoul(cdb->addr);
7026	if (((cdb->pmi & SRC_PMI) == 0)
7027	 && (lba != 0)) {
7028		ctl_set_invalid_field(/*ctsio*/ ctsio,
7029				      /*sks_valid*/ 1,
7030				      /*command*/ 1,
7031				      /*field*/ 2,
7032				      /*bit_valid*/ 0,
7033				      /*bit*/ 0);
7034		ctl_done((union ctl_io *)ctsio);
7035		return (CTL_RETVAL_COMPLETE);
7036	}
7037
7038	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7039
7040	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7041	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7042	ctsio->residual = 0;
7043	ctsio->kern_data_len = sizeof(*data);
7044	ctsio->kern_total_len = sizeof(*data);
7045	ctsio->kern_data_resid = 0;
7046	ctsio->kern_rel_offset = 0;
7047	ctsio->kern_sg_entries = 0;
7048
7049	/*
7050	 * If the maximum LBA is greater than 0xfffffffe, the user must
7051	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7052	 * serivce action set.
7053	 */
7054	if (lun->be_lun->maxlba > 0xfffffffe)
7055		scsi_ulto4b(0xffffffff, data->addr);
7056	else
7057		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7058
7059	/*
7060	 * XXX KDM this may not be 512 bytes...
7061	 */
7062	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7063
7064	ctl_set_success(ctsio);
7065	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7066	ctsio->be_move_done = ctl_config_move_done;
7067	ctl_datamove((union ctl_io *)ctsio);
7068	return (CTL_RETVAL_COMPLETE);
7069}
7070
7071int
7072ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7073{
7074	struct scsi_read_capacity_16 *cdb;
7075	struct scsi_read_capacity_data_long *data;
7076	struct ctl_lun *lun;
7077	uint64_t lba;
7078	uint32_t alloc_len;
7079
7080	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7081
7082	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7083
7084	alloc_len = scsi_4btoul(cdb->alloc_len);
7085	lba = scsi_8btou64(cdb->addr);
7086
7087	if ((cdb->reladr & SRC16_PMI)
7088	 && (lba != 0)) {
7089		ctl_set_invalid_field(/*ctsio*/ ctsio,
7090				      /*sks_valid*/ 1,
7091				      /*command*/ 1,
7092				      /*field*/ 2,
7093				      /*bit_valid*/ 0,
7094				      /*bit*/ 0);
7095		ctl_done((union ctl_io *)ctsio);
7096		return (CTL_RETVAL_COMPLETE);
7097	}
7098
7099	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7100
7101	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7102	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7103
7104	if (sizeof(*data) < alloc_len) {
7105		ctsio->residual = alloc_len - sizeof(*data);
7106		ctsio->kern_data_len = sizeof(*data);
7107		ctsio->kern_total_len = sizeof(*data);
7108	} else {
7109		ctsio->residual = 0;
7110		ctsio->kern_data_len = alloc_len;
7111		ctsio->kern_total_len = alloc_len;
7112	}
7113	ctsio->kern_data_resid = 0;
7114	ctsio->kern_rel_offset = 0;
7115	ctsio->kern_sg_entries = 0;
7116
7117	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7118	/* XXX KDM this may not be 512 bytes... */
7119	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7120	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7121	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7122	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7123		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7124
7125	ctl_set_success(ctsio);
7126	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7127	ctsio->be_move_done = ctl_config_move_done;
7128	ctl_datamove((union ctl_io *)ctsio);
7129	return (CTL_RETVAL_COMPLETE);
7130}
7131
7132int
7133ctl_get_lba_status(struct ctl_scsiio *ctsio)
7134{
7135	struct scsi_get_lba_status *cdb;
7136	struct scsi_get_lba_status_data *data;
7137	struct ctl_lun *lun;
7138	struct ctl_lba_len_flags *lbalen;
7139	uint64_t lba;
7140	uint32_t alloc_len, total_len;
7141	int retval;
7142
7143	CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7144
7145	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7146	cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7147	lba = scsi_8btou64(cdb->addr);
7148	alloc_len = scsi_4btoul(cdb->alloc_len);
7149
7150	if (lba > lun->be_lun->maxlba) {
7151		ctl_set_lba_out_of_range(ctsio);
7152		ctl_done((union ctl_io *)ctsio);
7153		return (CTL_RETVAL_COMPLETE);
7154	}
7155
7156	total_len = sizeof(*data) + sizeof(data->descr[0]);
7157	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7158	data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7159
7160	if (total_len < alloc_len) {
7161		ctsio->residual = alloc_len - total_len;
7162		ctsio->kern_data_len = total_len;
7163		ctsio->kern_total_len = total_len;
7164	} else {
7165		ctsio->residual = 0;
7166		ctsio->kern_data_len = alloc_len;
7167		ctsio->kern_total_len = alloc_len;
7168	}
7169	ctsio->kern_data_resid = 0;
7170	ctsio->kern_rel_offset = 0;
7171	ctsio->kern_sg_entries = 0;
7172
7173	/* Fill dummy data in case backend can't tell anything. */
7174	scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7175	scsi_u64to8b(lba, data->descr[0].addr);
7176	scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7177	    data->descr[0].length);
7178	data->descr[0].status = 0; /* Mapped or unknown. */
7179
7180	ctl_set_success(ctsio);
7181	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7182	ctsio->be_move_done = ctl_config_move_done;
7183
7184	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7185	lbalen->lba = lba;
7186	lbalen->len = total_len;
7187	lbalen->flags = 0;
7188	retval = lun->backend->config_read((union ctl_io *)ctsio);
7189	return (CTL_RETVAL_COMPLETE);
7190}
7191
7192int
7193ctl_read_defect(struct ctl_scsiio *ctsio)
7194{
7195	struct scsi_read_defect_data_10 *ccb10;
7196	struct scsi_read_defect_data_12 *ccb12;
7197	struct scsi_read_defect_data_hdr_10 *data10;
7198	struct scsi_read_defect_data_hdr_12 *data12;
7199	uint32_t alloc_len, data_len;
7200	uint8_t format;
7201
7202	CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7203
7204	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7205		ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7206		format = ccb10->format;
7207		alloc_len = scsi_2btoul(ccb10->alloc_length);
7208		data_len = sizeof(*data10);
7209	} else {
7210		ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7211		format = ccb12->format;
7212		alloc_len = scsi_4btoul(ccb12->alloc_length);
7213		data_len = sizeof(*data12);
7214	}
7215	if (alloc_len == 0) {
7216		ctl_set_success(ctsio);
7217		ctl_done((union ctl_io *)ctsio);
7218		return (CTL_RETVAL_COMPLETE);
7219	}
7220
7221	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7222	if (data_len < alloc_len) {
7223		ctsio->residual = alloc_len - data_len;
7224		ctsio->kern_data_len = data_len;
7225		ctsio->kern_total_len = data_len;
7226	} else {
7227		ctsio->residual = 0;
7228		ctsio->kern_data_len = alloc_len;
7229		ctsio->kern_total_len = alloc_len;
7230	}
7231	ctsio->kern_data_resid = 0;
7232	ctsio->kern_rel_offset = 0;
7233	ctsio->kern_sg_entries = 0;
7234
7235	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7236		data10 = (struct scsi_read_defect_data_hdr_10 *)
7237		    ctsio->kern_data_ptr;
7238		data10->format = format;
7239		scsi_ulto2b(0, data10->length);
7240	} else {
7241		data12 = (struct scsi_read_defect_data_hdr_12 *)
7242		    ctsio->kern_data_ptr;
7243		data12->format = format;
7244		scsi_ulto2b(0, data12->generation);
7245		scsi_ulto4b(0, data12->length);
7246	}
7247
7248	ctl_set_success(ctsio);
7249	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7250	ctsio->be_move_done = ctl_config_move_done;
7251	ctl_datamove((union ctl_io *)ctsio);
7252	return (CTL_RETVAL_COMPLETE);
7253}
7254
7255int
7256ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7257{
7258	struct scsi_maintenance_in *cdb;
7259	int retval;
7260	int alloc_len, ext, total_len = 0, g, p, pc, pg, gs, os;
7261	int num_target_port_groups, num_target_ports;
7262	struct ctl_lun *lun;
7263	struct ctl_softc *softc;
7264	struct ctl_port *port;
7265	struct scsi_target_group_data *rtg_ptr;
7266	struct scsi_target_group_data_extended *rtg_ext_ptr;
7267	struct scsi_target_port_group_descriptor *tpg_desc;
7268
7269	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7270
7271	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7272	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7273	softc = lun->ctl_softc;
7274
7275	retval = CTL_RETVAL_COMPLETE;
7276
7277	switch (cdb->byte2 & STG_PDF_MASK) {
7278	case STG_PDF_LENGTH:
7279		ext = 0;
7280		break;
7281	case STG_PDF_EXTENDED:
7282		ext = 1;
7283		break;
7284	default:
7285		ctl_set_invalid_field(/*ctsio*/ ctsio,
7286				      /*sks_valid*/ 1,
7287				      /*command*/ 1,
7288				      /*field*/ 2,
7289				      /*bit_valid*/ 1,
7290				      /*bit*/ 5);
7291		ctl_done((union ctl_io *)ctsio);
7292		return(retval);
7293	}
7294
7295	if (softc->is_single)
7296		num_target_port_groups = 1;
7297	else
7298		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7299	num_target_ports = 0;
7300	mtx_lock(&softc->ctl_lock);
7301	STAILQ_FOREACH(port, &softc->port_list, links) {
7302		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7303			continue;
7304		if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7305			continue;
7306		num_target_ports++;
7307	}
7308	mtx_unlock(&softc->ctl_lock);
7309
7310	if (ext)
7311		total_len = sizeof(struct scsi_target_group_data_extended);
7312	else
7313		total_len = sizeof(struct scsi_target_group_data);
7314	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7315		num_target_port_groups +
7316	    sizeof(struct scsi_target_port_descriptor) *
7317		num_target_ports * num_target_port_groups;
7318
7319	alloc_len = scsi_4btoul(cdb->length);
7320
7321	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7322
7323	ctsio->kern_sg_entries = 0;
7324
7325	if (total_len < alloc_len) {
7326		ctsio->residual = alloc_len - total_len;
7327		ctsio->kern_data_len = total_len;
7328		ctsio->kern_total_len = total_len;
7329	} else {
7330		ctsio->residual = 0;
7331		ctsio->kern_data_len = alloc_len;
7332		ctsio->kern_total_len = alloc_len;
7333	}
7334	ctsio->kern_data_resid = 0;
7335	ctsio->kern_rel_offset = 0;
7336
7337	if (ext) {
7338		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7339		    ctsio->kern_data_ptr;
7340		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7341		rtg_ext_ptr->format_type = 0x10;
7342		rtg_ext_ptr->implicit_transition_time = 0;
7343		tpg_desc = &rtg_ext_ptr->groups[0];
7344	} else {
7345		rtg_ptr = (struct scsi_target_group_data *)
7346		    ctsio->kern_data_ptr;
7347		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7348		tpg_desc = &rtg_ptr->groups[0];
7349	}
7350
7351	mtx_lock(&softc->ctl_lock);
7352	pg = softc->port_offset / CTL_MAX_PORTS;
7353	if (softc->flags & CTL_FLAG_ACTIVE_SHELF) {
7354		if (softc->ha_mode == CTL_HA_MODE_ACT_STBY) {
7355			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7356			os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7357		} else if (lun->flags & CTL_LUN_PRIMARY_SC) {
7358			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7359			os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7360		} else {
7361			gs = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7362			os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7363		}
7364	} else {
7365		gs = TPG_ASYMMETRIC_ACCESS_STANDBY;
7366		os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7367	}
7368	for (g = 0; g < num_target_port_groups; g++) {
7369		tpg_desc->pref_state = (g == pg) ? gs : os;
7370		tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP;
7371		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7372		tpg_desc->status = TPG_IMPLICIT;
7373		pc = 0;
7374		STAILQ_FOREACH(port, &softc->port_list, links) {
7375			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7376				continue;
7377			if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7378				continue;
7379			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7380			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7381			    relative_target_port_identifier);
7382			pc++;
7383		}
7384		tpg_desc->target_port_count = pc;
7385		tpg_desc = (struct scsi_target_port_group_descriptor *)
7386		    &tpg_desc->descriptors[pc];
7387	}
7388	mtx_unlock(&softc->ctl_lock);
7389
7390	ctl_set_success(ctsio);
7391	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7392	ctsio->be_move_done = ctl_config_move_done;
7393	ctl_datamove((union ctl_io *)ctsio);
7394	return(retval);
7395}
7396
7397int
7398ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7399{
7400	struct ctl_lun *lun;
7401	struct scsi_report_supported_opcodes *cdb;
7402	const struct ctl_cmd_entry *entry, *sentry;
7403	struct scsi_report_supported_opcodes_all *all;
7404	struct scsi_report_supported_opcodes_descr *descr;
7405	struct scsi_report_supported_opcodes_one *one;
7406	int retval;
7407	int alloc_len, total_len;
7408	int opcode, service_action, i, j, num;
7409
7410	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7411
7412	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7413	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7414
7415	retval = CTL_RETVAL_COMPLETE;
7416
7417	opcode = cdb->requested_opcode;
7418	service_action = scsi_2btoul(cdb->requested_service_action);
7419	switch (cdb->options & RSO_OPTIONS_MASK) {
7420	case RSO_OPTIONS_ALL:
7421		num = 0;
7422		for (i = 0; i < 256; i++) {
7423			entry = &ctl_cmd_table[i];
7424			if (entry->flags & CTL_CMD_FLAG_SA5) {
7425				for (j = 0; j < 32; j++) {
7426					sentry = &((const struct ctl_cmd_entry *)
7427					    entry->execute)[j];
7428					if (ctl_cmd_applicable(
7429					    lun->be_lun->lun_type, sentry))
7430						num++;
7431				}
7432			} else {
7433				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7434				    entry))
7435					num++;
7436			}
7437		}
7438		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7439		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7440		break;
7441	case RSO_OPTIONS_OC:
7442		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7443			ctl_set_invalid_field(/*ctsio*/ ctsio,
7444					      /*sks_valid*/ 1,
7445					      /*command*/ 1,
7446					      /*field*/ 2,
7447					      /*bit_valid*/ 1,
7448					      /*bit*/ 2);
7449			ctl_done((union ctl_io *)ctsio);
7450			return (CTL_RETVAL_COMPLETE);
7451		}
7452		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7453		break;
7454	case RSO_OPTIONS_OC_SA:
7455		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7456		    service_action >= 32) {
7457			ctl_set_invalid_field(/*ctsio*/ ctsio,
7458					      /*sks_valid*/ 1,
7459					      /*command*/ 1,
7460					      /*field*/ 2,
7461					      /*bit_valid*/ 1,
7462					      /*bit*/ 2);
7463			ctl_done((union ctl_io *)ctsio);
7464			return (CTL_RETVAL_COMPLETE);
7465		}
7466		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7467		break;
7468	default:
7469		ctl_set_invalid_field(/*ctsio*/ ctsio,
7470				      /*sks_valid*/ 1,
7471				      /*command*/ 1,
7472				      /*field*/ 2,
7473				      /*bit_valid*/ 1,
7474				      /*bit*/ 2);
7475		ctl_done((union ctl_io *)ctsio);
7476		return (CTL_RETVAL_COMPLETE);
7477	}
7478
7479	alloc_len = scsi_4btoul(cdb->length);
7480
7481	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7482
7483	ctsio->kern_sg_entries = 0;
7484
7485	if (total_len < alloc_len) {
7486		ctsio->residual = alloc_len - total_len;
7487		ctsio->kern_data_len = total_len;
7488		ctsio->kern_total_len = total_len;
7489	} else {
7490		ctsio->residual = 0;
7491		ctsio->kern_data_len = alloc_len;
7492		ctsio->kern_total_len = alloc_len;
7493	}
7494	ctsio->kern_data_resid = 0;
7495	ctsio->kern_rel_offset = 0;
7496
7497	switch (cdb->options & RSO_OPTIONS_MASK) {
7498	case RSO_OPTIONS_ALL:
7499		all = (struct scsi_report_supported_opcodes_all *)
7500		    ctsio->kern_data_ptr;
7501		num = 0;
7502		for (i = 0; i < 256; i++) {
7503			entry = &ctl_cmd_table[i];
7504			if (entry->flags & CTL_CMD_FLAG_SA5) {
7505				for (j = 0; j < 32; j++) {
7506					sentry = &((const struct ctl_cmd_entry *)
7507					    entry->execute)[j];
7508					if (!ctl_cmd_applicable(
7509					    lun->be_lun->lun_type, sentry))
7510						continue;
7511					descr = &all->descr[num++];
7512					descr->opcode = i;
7513					scsi_ulto2b(j, descr->service_action);
7514					descr->flags = RSO_SERVACTV;
7515					scsi_ulto2b(sentry->length,
7516					    descr->cdb_length);
7517				}
7518			} else {
7519				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7520				    entry))
7521					continue;
7522				descr = &all->descr[num++];
7523				descr->opcode = i;
7524				scsi_ulto2b(0, descr->service_action);
7525				descr->flags = 0;
7526				scsi_ulto2b(entry->length, descr->cdb_length);
7527			}
7528		}
7529		scsi_ulto4b(
7530		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7531		    all->length);
7532		break;
7533	case RSO_OPTIONS_OC:
7534		one = (struct scsi_report_supported_opcodes_one *)
7535		    ctsio->kern_data_ptr;
7536		entry = &ctl_cmd_table[opcode];
7537		goto fill_one;
7538	case RSO_OPTIONS_OC_SA:
7539		one = (struct scsi_report_supported_opcodes_one *)
7540		    ctsio->kern_data_ptr;
7541		entry = &ctl_cmd_table[opcode];
7542		entry = &((const struct ctl_cmd_entry *)
7543		    entry->execute)[service_action];
7544fill_one:
7545		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7546			one->support = 3;
7547			scsi_ulto2b(entry->length, one->cdb_length);
7548			one->cdb_usage[0] = opcode;
7549			memcpy(&one->cdb_usage[1], entry->usage,
7550			    entry->length - 1);
7551		} else
7552			one->support = 1;
7553		break;
7554	}
7555
7556	ctl_set_success(ctsio);
7557	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7558	ctsio->be_move_done = ctl_config_move_done;
7559	ctl_datamove((union ctl_io *)ctsio);
7560	return(retval);
7561}
7562
7563int
7564ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7565{
7566	struct scsi_report_supported_tmf *cdb;
7567	struct scsi_report_supported_tmf_data *data;
7568	int retval;
7569	int alloc_len, total_len;
7570
7571	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7572
7573	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7574
7575	retval = CTL_RETVAL_COMPLETE;
7576
7577	total_len = sizeof(struct scsi_report_supported_tmf_data);
7578	alloc_len = scsi_4btoul(cdb->length);
7579
7580	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7581
7582	ctsio->kern_sg_entries = 0;
7583
7584	if (total_len < alloc_len) {
7585		ctsio->residual = alloc_len - total_len;
7586		ctsio->kern_data_len = total_len;
7587		ctsio->kern_total_len = total_len;
7588	} else {
7589		ctsio->residual = 0;
7590		ctsio->kern_data_len = alloc_len;
7591		ctsio->kern_total_len = alloc_len;
7592	}
7593	ctsio->kern_data_resid = 0;
7594	ctsio->kern_rel_offset = 0;
7595
7596	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7597	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7598	data->byte2 |= RST_ITNRS;
7599
7600	ctl_set_success(ctsio);
7601	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7602	ctsio->be_move_done = ctl_config_move_done;
7603	ctl_datamove((union ctl_io *)ctsio);
7604	return (retval);
7605}
7606
7607int
7608ctl_report_timestamp(struct ctl_scsiio *ctsio)
7609{
7610	struct scsi_report_timestamp *cdb;
7611	struct scsi_report_timestamp_data *data;
7612	struct timeval tv;
7613	int64_t timestamp;
7614	int retval;
7615	int alloc_len, total_len;
7616
7617	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7618
7619	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7620
7621	retval = CTL_RETVAL_COMPLETE;
7622
7623	total_len = sizeof(struct scsi_report_timestamp_data);
7624	alloc_len = scsi_4btoul(cdb->length);
7625
7626	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7627
7628	ctsio->kern_sg_entries = 0;
7629
7630	if (total_len < alloc_len) {
7631		ctsio->residual = alloc_len - total_len;
7632		ctsio->kern_data_len = total_len;
7633		ctsio->kern_total_len = total_len;
7634	} else {
7635		ctsio->residual = 0;
7636		ctsio->kern_data_len = alloc_len;
7637		ctsio->kern_total_len = alloc_len;
7638	}
7639	ctsio->kern_data_resid = 0;
7640	ctsio->kern_rel_offset = 0;
7641
7642	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7643	scsi_ulto2b(sizeof(*data) - 2, data->length);
7644	data->origin = RTS_ORIG_OUTSIDE;
7645	getmicrotime(&tv);
7646	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7647	scsi_ulto4b(timestamp >> 16, data->timestamp);
7648	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7649
7650	ctl_set_success(ctsio);
7651	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7652	ctsio->be_move_done = ctl_config_move_done;
7653	ctl_datamove((union ctl_io *)ctsio);
7654	return (retval);
7655}
7656
7657int
7658ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7659{
7660	struct scsi_per_res_in *cdb;
7661	int alloc_len, total_len = 0;
7662	/* struct scsi_per_res_in_rsrv in_data; */
7663	struct ctl_lun *lun;
7664	struct ctl_softc *softc;
7665	uint64_t key;
7666
7667	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7668
7669	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7670
7671	alloc_len = scsi_2btoul(cdb->length);
7672
7673	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7674	softc = lun->ctl_softc;
7675
7676retry:
7677	mtx_lock(&lun->lun_lock);
7678	switch (cdb->action) {
7679	case SPRI_RK: /* read keys */
7680		total_len = sizeof(struct scsi_per_res_in_keys) +
7681			lun->pr_key_count *
7682			sizeof(struct scsi_per_res_key);
7683		break;
7684	case SPRI_RR: /* read reservation */
7685		if (lun->flags & CTL_LUN_PR_RESERVED)
7686			total_len = sizeof(struct scsi_per_res_in_rsrv);
7687		else
7688			total_len = sizeof(struct scsi_per_res_in_header);
7689		break;
7690	case SPRI_RC: /* report capabilities */
7691		total_len = sizeof(struct scsi_per_res_cap);
7692		break;
7693	case SPRI_RS: /* read full status */
7694		total_len = sizeof(struct scsi_per_res_in_header) +
7695		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7696		    lun->pr_key_count;
7697		break;
7698	default:
7699		panic("Invalid PR type %x", cdb->action);
7700	}
7701	mtx_unlock(&lun->lun_lock);
7702
7703	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7704
7705	if (total_len < alloc_len) {
7706		ctsio->residual = alloc_len - total_len;
7707		ctsio->kern_data_len = total_len;
7708		ctsio->kern_total_len = total_len;
7709	} else {
7710		ctsio->residual = 0;
7711		ctsio->kern_data_len = alloc_len;
7712		ctsio->kern_total_len = alloc_len;
7713	}
7714
7715	ctsio->kern_data_resid = 0;
7716	ctsio->kern_rel_offset = 0;
7717	ctsio->kern_sg_entries = 0;
7718
7719	mtx_lock(&lun->lun_lock);
7720	switch (cdb->action) {
7721	case SPRI_RK: { // read keys
7722        struct scsi_per_res_in_keys *res_keys;
7723		int i, key_count;
7724
7725		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7726
7727		/*
7728		 * We had to drop the lock to allocate our buffer, which
7729		 * leaves time for someone to come in with another
7730		 * persistent reservation.  (That is unlikely, though,
7731		 * since this should be the only persistent reservation
7732		 * command active right now.)
7733		 */
7734		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7735		    (lun->pr_key_count *
7736		     sizeof(struct scsi_per_res_key)))){
7737			mtx_unlock(&lun->lun_lock);
7738			free(ctsio->kern_data_ptr, M_CTL);
7739			printf("%s: reservation length changed, retrying\n",
7740			       __func__);
7741			goto retry;
7742		}
7743
7744		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7745
7746		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7747			     lun->pr_key_count, res_keys->header.length);
7748
7749		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7750			if ((key = ctl_get_prkey(lun, i)) == 0)
7751				continue;
7752
7753			/*
7754			 * We used lun->pr_key_count to calculate the
7755			 * size to allocate.  If it turns out the number of
7756			 * initiators with the registered flag set is
7757			 * larger than that (i.e. they haven't been kept in
7758			 * sync), we've got a problem.
7759			 */
7760			if (key_count >= lun->pr_key_count) {
7761#ifdef NEEDTOPORT
7762				csevent_log(CSC_CTL | CSC_SHELF_SW |
7763					    CTL_PR_ERROR,
7764					    csevent_LogType_Fault,
7765					    csevent_AlertLevel_Yellow,
7766					    csevent_FRU_ShelfController,
7767					    csevent_FRU_Firmware,
7768				        csevent_FRU_Unknown,
7769					    "registered keys %d >= key "
7770					    "count %d", key_count,
7771					    lun->pr_key_count);
7772#endif
7773				key_count++;
7774				continue;
7775			}
7776			scsi_u64to8b(key, res_keys->keys[key_count].key);
7777			key_count++;
7778		}
7779		break;
7780	}
7781	case SPRI_RR: { // read reservation
7782		struct scsi_per_res_in_rsrv *res;
7783		int tmp_len, header_only;
7784
7785		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7786
7787		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7788
7789		if (lun->flags & CTL_LUN_PR_RESERVED)
7790		{
7791			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7792			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7793				    res->header.length);
7794			header_only = 0;
7795		} else {
7796			tmp_len = sizeof(struct scsi_per_res_in_header);
7797			scsi_ulto4b(0, res->header.length);
7798			header_only = 1;
7799		}
7800
7801		/*
7802		 * We had to drop the lock to allocate our buffer, which
7803		 * leaves time for someone to come in with another
7804		 * persistent reservation.  (That is unlikely, though,
7805		 * since this should be the only persistent reservation
7806		 * command active right now.)
7807		 */
7808		if (tmp_len != total_len) {
7809			mtx_unlock(&lun->lun_lock);
7810			free(ctsio->kern_data_ptr, M_CTL);
7811			printf("%s: reservation status changed, retrying\n",
7812			       __func__);
7813			goto retry;
7814		}
7815
7816		/*
7817		 * No reservation held, so we're done.
7818		 */
7819		if (header_only != 0)
7820			break;
7821
7822		/*
7823		 * If the registration is an All Registrants type, the key
7824		 * is 0, since it doesn't really matter.
7825		 */
7826		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7827			scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7828			    res->data.reservation);
7829		}
7830		res->data.scopetype = lun->res_type;
7831		break;
7832	}
7833	case SPRI_RC:     //report capabilities
7834	{
7835		struct scsi_per_res_cap *res_cap;
7836		uint16_t type_mask;
7837
7838		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7839		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7840		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7841		type_mask = SPRI_TM_WR_EX_AR |
7842			    SPRI_TM_EX_AC_RO |
7843			    SPRI_TM_WR_EX_RO |
7844			    SPRI_TM_EX_AC |
7845			    SPRI_TM_WR_EX |
7846			    SPRI_TM_EX_AC_AR;
7847		scsi_ulto2b(type_mask, res_cap->type_mask);
7848		break;
7849	}
7850	case SPRI_RS: { // read full status
7851		struct scsi_per_res_in_full *res_status;
7852		struct scsi_per_res_in_full_desc *res_desc;
7853		struct ctl_port *port;
7854		int i, len;
7855
7856		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7857
7858		/*
7859		 * We had to drop the lock to allocate our buffer, which
7860		 * leaves time for someone to come in with another
7861		 * persistent reservation.  (That is unlikely, though,
7862		 * since this should be the only persistent reservation
7863		 * command active right now.)
7864		 */
7865		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7866		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7867		     lun->pr_key_count)){
7868			mtx_unlock(&lun->lun_lock);
7869			free(ctsio->kern_data_ptr, M_CTL);
7870			printf("%s: reservation length changed, retrying\n",
7871			       __func__);
7872			goto retry;
7873		}
7874
7875		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7876
7877		res_desc = &res_status->desc[0];
7878		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7879			if ((key = ctl_get_prkey(lun, i)) == 0)
7880				continue;
7881
7882			scsi_u64to8b(key, res_desc->res_key.key);
7883			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7884			    (lun->pr_res_idx == i ||
7885			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7886				res_desc->flags = SPRI_FULL_R_HOLDER;
7887				res_desc->scopetype = lun->res_type;
7888			}
7889			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7890			    res_desc->rel_trgt_port_id);
7891			len = 0;
7892			port = softc->ctl_ports[
7893			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7894			if (port != NULL)
7895				len = ctl_create_iid(port,
7896				    i % CTL_MAX_INIT_PER_PORT,
7897				    res_desc->transport_id);
7898			scsi_ulto4b(len, res_desc->additional_length);
7899			res_desc = (struct scsi_per_res_in_full_desc *)
7900			    &res_desc->transport_id[len];
7901		}
7902		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7903		    res_status->header.length);
7904		break;
7905	}
7906	default:
7907		/*
7908		 * This is a bug, because we just checked for this above,
7909		 * and should have returned an error.
7910		 */
7911		panic("Invalid PR type %x", cdb->action);
7912		break; /* NOTREACHED */
7913	}
7914	mtx_unlock(&lun->lun_lock);
7915
7916	ctl_set_success(ctsio);
7917	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7918	ctsio->be_move_done = ctl_config_move_done;
7919	ctl_datamove((union ctl_io *)ctsio);
7920	return (CTL_RETVAL_COMPLETE);
7921}
7922
7923static void
7924ctl_est_res_ua(struct ctl_lun *lun, uint32_t residx, ctl_ua_type ua)
7925{
7926	int off = lun->ctl_softc->persis_offset;
7927
7928	if (residx >= off && residx < off + CTL_MAX_INITIATORS)
7929		ctl_est_ua(lun, residx - off, ua);
7930}
7931
7932/*
7933 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7934 * it should return.
7935 */
7936static int
7937ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7938		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7939		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7940		struct scsi_per_res_out_parms* param)
7941{
7942	union ctl_ha_msg persis_io;
7943	int retval, i;
7944	int isc_retval;
7945
7946	retval = 0;
7947
7948	mtx_lock(&lun->lun_lock);
7949	if (sa_res_key == 0) {
7950		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7951			/* validate scope and type */
7952			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7953			     SPR_LU_SCOPE) {
7954				mtx_unlock(&lun->lun_lock);
7955				ctl_set_invalid_field(/*ctsio*/ ctsio,
7956						      /*sks_valid*/ 1,
7957						      /*command*/ 1,
7958						      /*field*/ 2,
7959						      /*bit_valid*/ 1,
7960						      /*bit*/ 4);
7961				ctl_done((union ctl_io *)ctsio);
7962				return (1);
7963			}
7964
7965		        if (type>8 || type==2 || type==4 || type==0) {
7966				mtx_unlock(&lun->lun_lock);
7967				ctl_set_invalid_field(/*ctsio*/ ctsio,
7968       	           				      /*sks_valid*/ 1,
7969						      /*command*/ 1,
7970						      /*field*/ 2,
7971						      /*bit_valid*/ 1,
7972						      /*bit*/ 0);
7973				ctl_done((union ctl_io *)ctsio);
7974				return (1);
7975		        }
7976
7977			/*
7978			 * Unregister everybody else and build UA for
7979			 * them
7980			 */
7981			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7982				if (i == residx || ctl_get_prkey(lun, i) == 0)
7983					continue;
7984
7985				ctl_clr_prkey(lun, i);
7986				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7987			}
7988			lun->pr_key_count = 1;
7989			lun->res_type = type;
7990			if (lun->res_type != SPR_TYPE_WR_EX_AR
7991			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7992				lun->pr_res_idx = residx;
7993
7994			/* send msg to other side */
7995			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7996			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7997			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7998			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7999			persis_io.pr.pr_info.res_type = type;
8000			memcpy(persis_io.pr.pr_info.sa_res_key,
8001			       param->serv_act_res_key,
8002			       sizeof(param->serv_act_res_key));
8003			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8004			     &persis_io, sizeof(persis_io), 0)) >
8005			     CTL_HA_STATUS_SUCCESS) {
8006				printf("CTL:Persis Out error returned "
8007				       "from ctl_ha_msg_send %d\n",
8008				       isc_retval);
8009			}
8010		} else {
8011			/* not all registrants */
8012			mtx_unlock(&lun->lun_lock);
8013			free(ctsio->kern_data_ptr, M_CTL);
8014			ctl_set_invalid_field(ctsio,
8015					      /*sks_valid*/ 1,
8016					      /*command*/ 0,
8017					      /*field*/ 8,
8018					      /*bit_valid*/ 0,
8019					      /*bit*/ 0);
8020			ctl_done((union ctl_io *)ctsio);
8021			return (1);
8022		}
8023	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8024		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8025		int found = 0;
8026
8027		if (res_key == sa_res_key) {
8028			/* special case */
8029			/*
8030			 * The spec implies this is not good but doesn't
8031			 * say what to do. There are two choices either
8032			 * generate a res conflict or check condition
8033			 * with illegal field in parameter data. Since
8034			 * that is what is done when the sa_res_key is
8035			 * zero I'll take that approach since this has
8036			 * to do with the sa_res_key.
8037			 */
8038			mtx_unlock(&lun->lun_lock);
8039			free(ctsio->kern_data_ptr, M_CTL);
8040			ctl_set_invalid_field(ctsio,
8041					      /*sks_valid*/ 1,
8042					      /*command*/ 0,
8043					      /*field*/ 8,
8044					      /*bit_valid*/ 0,
8045					      /*bit*/ 0);
8046			ctl_done((union ctl_io *)ctsio);
8047			return (1);
8048		}
8049
8050		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8051			if (ctl_get_prkey(lun, i) != sa_res_key)
8052				continue;
8053
8054			found = 1;
8055			ctl_clr_prkey(lun, i);
8056			lun->pr_key_count--;
8057			ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8058		}
8059		if (!found) {
8060			mtx_unlock(&lun->lun_lock);
8061			free(ctsio->kern_data_ptr, M_CTL);
8062			ctl_set_reservation_conflict(ctsio);
8063			ctl_done((union ctl_io *)ctsio);
8064			return (CTL_RETVAL_COMPLETE);
8065		}
8066		/* send msg to other side */
8067		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8068		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8069		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8070		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8071		persis_io.pr.pr_info.res_type = type;
8072		memcpy(persis_io.pr.pr_info.sa_res_key,
8073		       param->serv_act_res_key,
8074		       sizeof(param->serv_act_res_key));
8075		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8076		     &persis_io, sizeof(persis_io), 0)) >
8077		     CTL_HA_STATUS_SUCCESS) {
8078			printf("CTL:Persis Out error returned from "
8079			       "ctl_ha_msg_send %d\n", isc_retval);
8080		}
8081	} else {
8082		/* Reserved but not all registrants */
8083		/* sa_res_key is res holder */
8084		if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
8085			/* validate scope and type */
8086			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8087			     SPR_LU_SCOPE) {
8088				mtx_unlock(&lun->lun_lock);
8089				ctl_set_invalid_field(/*ctsio*/ ctsio,
8090						      /*sks_valid*/ 1,
8091						      /*command*/ 1,
8092						      /*field*/ 2,
8093						      /*bit_valid*/ 1,
8094						      /*bit*/ 4);
8095				ctl_done((union ctl_io *)ctsio);
8096				return (1);
8097			}
8098
8099			if (type>8 || type==2 || type==4 || type==0) {
8100				mtx_unlock(&lun->lun_lock);
8101				ctl_set_invalid_field(/*ctsio*/ ctsio,
8102						      /*sks_valid*/ 1,
8103						      /*command*/ 1,
8104						      /*field*/ 2,
8105						      /*bit_valid*/ 1,
8106						      /*bit*/ 0);
8107				ctl_done((union ctl_io *)ctsio);
8108				return (1);
8109			}
8110
8111			/*
8112			 * Do the following:
8113			 * if sa_res_key != res_key remove all
8114			 * registrants w/sa_res_key and generate UA
8115			 * for these registrants(Registrations
8116			 * Preempted) if it wasn't an exclusive
8117			 * reservation generate UA(Reservations
8118			 * Preempted) for all other registered nexuses
8119			 * if the type has changed. Establish the new
8120			 * reservation and holder. If res_key and
8121			 * sa_res_key are the same do the above
8122			 * except don't unregister the res holder.
8123			 */
8124
8125			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8126				if (i == residx || ctl_get_prkey(lun, i) == 0)
8127					continue;
8128
8129				if (sa_res_key == ctl_get_prkey(lun, i)) {
8130					ctl_clr_prkey(lun, i);
8131					lun->pr_key_count--;
8132					ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8133				} else if (type != lun->res_type
8134					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8135					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8136					ctl_est_res_ua(lun, i, CTL_UA_RES_RELEASE);
8137				}
8138			}
8139			lun->res_type = type;
8140			if (lun->res_type != SPR_TYPE_WR_EX_AR
8141			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8142				lun->pr_res_idx = residx;
8143			else
8144				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8145
8146			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8147			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8148			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8149			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8150			persis_io.pr.pr_info.res_type = type;
8151			memcpy(persis_io.pr.pr_info.sa_res_key,
8152			       param->serv_act_res_key,
8153			       sizeof(param->serv_act_res_key));
8154			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8155			     &persis_io, sizeof(persis_io), 0)) >
8156			     CTL_HA_STATUS_SUCCESS) {
8157				printf("CTL:Persis Out error returned "
8158				       "from ctl_ha_msg_send %d\n",
8159				       isc_retval);
8160			}
8161		} else {
8162			/*
8163			 * sa_res_key is not the res holder just
8164			 * remove registrants
8165			 */
8166			int found=0;
8167
8168			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8169				if (sa_res_key != ctl_get_prkey(lun, i))
8170					continue;
8171
8172				found = 1;
8173				ctl_clr_prkey(lun, i);
8174				lun->pr_key_count--;
8175				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8176			}
8177
8178			if (!found) {
8179				mtx_unlock(&lun->lun_lock);
8180				free(ctsio->kern_data_ptr, M_CTL);
8181				ctl_set_reservation_conflict(ctsio);
8182				ctl_done((union ctl_io *)ctsio);
8183		        	return (1);
8184			}
8185			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8186			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8187			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8188			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8189			persis_io.pr.pr_info.res_type = type;
8190			memcpy(persis_io.pr.pr_info.sa_res_key,
8191			       param->serv_act_res_key,
8192			       sizeof(param->serv_act_res_key));
8193			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8194			     &persis_io, sizeof(persis_io), 0)) >
8195			     CTL_HA_STATUS_SUCCESS) {
8196				printf("CTL:Persis Out error returned "
8197				       "from ctl_ha_msg_send %d\n",
8198				isc_retval);
8199			}
8200		}
8201	}
8202
8203	lun->PRGeneration++;
8204	mtx_unlock(&lun->lun_lock);
8205
8206	return (retval);
8207}
8208
8209static void
8210ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8211{
8212	uint64_t sa_res_key;
8213	int i;
8214
8215	sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8216
8217	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8218	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8219	 || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8220		if (sa_res_key == 0) {
8221			/*
8222			 * Unregister everybody else and build UA for
8223			 * them
8224			 */
8225			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8226				if (i == msg->pr.pr_info.residx ||
8227				    ctl_get_prkey(lun, i) == 0)
8228					continue;
8229
8230				ctl_clr_prkey(lun, i);
8231				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8232			}
8233
8234			lun->pr_key_count = 1;
8235			lun->res_type = msg->pr.pr_info.res_type;
8236			if (lun->res_type != SPR_TYPE_WR_EX_AR
8237			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8238				lun->pr_res_idx = msg->pr.pr_info.residx;
8239		} else {
8240		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8241				if (sa_res_key == ctl_get_prkey(lun, i))
8242					continue;
8243
8244				ctl_clr_prkey(lun, i);
8245				lun->pr_key_count--;
8246				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8247			}
8248		}
8249	} else {
8250		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8251			if (i == msg->pr.pr_info.residx ||
8252			    ctl_get_prkey(lun, i) == 0)
8253				continue;
8254
8255			if (sa_res_key == ctl_get_prkey(lun, i)) {
8256				ctl_clr_prkey(lun, i);
8257				lun->pr_key_count--;
8258				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8259			} else if (msg->pr.pr_info.res_type != lun->res_type
8260				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8261				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8262				ctl_est_res_ua(lun, i, CTL_UA_RES_RELEASE);
8263			}
8264		}
8265		lun->res_type = msg->pr.pr_info.res_type;
8266		if (lun->res_type != SPR_TYPE_WR_EX_AR
8267		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8268			lun->pr_res_idx = msg->pr.pr_info.residx;
8269		else
8270			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8271	}
8272	lun->PRGeneration++;
8273
8274}
8275
8276
8277int
8278ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8279{
8280	int retval;
8281	int isc_retval;
8282	u_int32_t param_len;
8283	struct scsi_per_res_out *cdb;
8284	struct ctl_lun *lun;
8285	struct scsi_per_res_out_parms* param;
8286	struct ctl_softc *softc;
8287	uint32_t residx;
8288	uint64_t res_key, sa_res_key, key;
8289	uint8_t type;
8290	union ctl_ha_msg persis_io;
8291	int    i;
8292
8293	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8294
8295	retval = CTL_RETVAL_COMPLETE;
8296
8297	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8298	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8299	softc = lun->ctl_softc;
8300
8301	/*
8302	 * We only support whole-LUN scope.  The scope & type are ignored for
8303	 * register, register and ignore existing key and clear.
8304	 * We sometimes ignore scope and type on preempts too!!
8305	 * Verify reservation type here as well.
8306	 */
8307	type = cdb->scope_type & SPR_TYPE_MASK;
8308	if ((cdb->action == SPRO_RESERVE)
8309	 || (cdb->action == SPRO_RELEASE)) {
8310		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8311			ctl_set_invalid_field(/*ctsio*/ ctsio,
8312					      /*sks_valid*/ 1,
8313					      /*command*/ 1,
8314					      /*field*/ 2,
8315					      /*bit_valid*/ 1,
8316					      /*bit*/ 4);
8317			ctl_done((union ctl_io *)ctsio);
8318			return (CTL_RETVAL_COMPLETE);
8319		}
8320
8321		if (type>8 || type==2 || type==4 || type==0) {
8322			ctl_set_invalid_field(/*ctsio*/ ctsio,
8323					      /*sks_valid*/ 1,
8324					      /*command*/ 1,
8325					      /*field*/ 2,
8326					      /*bit_valid*/ 1,
8327					      /*bit*/ 0);
8328			ctl_done((union ctl_io *)ctsio);
8329			return (CTL_RETVAL_COMPLETE);
8330		}
8331	}
8332
8333	param_len = scsi_4btoul(cdb->length);
8334
8335	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8336		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8337		ctsio->kern_data_len = param_len;
8338		ctsio->kern_total_len = param_len;
8339		ctsio->kern_data_resid = 0;
8340		ctsio->kern_rel_offset = 0;
8341		ctsio->kern_sg_entries = 0;
8342		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8343		ctsio->be_move_done = ctl_config_move_done;
8344		ctl_datamove((union ctl_io *)ctsio);
8345
8346		return (CTL_RETVAL_COMPLETE);
8347	}
8348
8349	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8350
8351	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8352	res_key = scsi_8btou64(param->res_key.key);
8353	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8354
8355	/*
8356	 * Validate the reservation key here except for SPRO_REG_IGNO
8357	 * This must be done for all other service actions
8358	 */
8359	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8360		mtx_lock(&lun->lun_lock);
8361		if ((key = ctl_get_prkey(lun, residx)) != 0) {
8362			if (res_key != key) {
8363				/*
8364				 * The current key passed in doesn't match
8365				 * the one the initiator previously
8366				 * registered.
8367				 */
8368				mtx_unlock(&lun->lun_lock);
8369				free(ctsio->kern_data_ptr, M_CTL);
8370				ctl_set_reservation_conflict(ctsio);
8371				ctl_done((union ctl_io *)ctsio);
8372				return (CTL_RETVAL_COMPLETE);
8373			}
8374		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8375			/*
8376			 * We are not registered
8377			 */
8378			mtx_unlock(&lun->lun_lock);
8379			free(ctsio->kern_data_ptr, M_CTL);
8380			ctl_set_reservation_conflict(ctsio);
8381			ctl_done((union ctl_io *)ctsio);
8382			return (CTL_RETVAL_COMPLETE);
8383		} else if (res_key != 0) {
8384			/*
8385			 * We are not registered and trying to register but
8386			 * the register key isn't zero.
8387			 */
8388			mtx_unlock(&lun->lun_lock);
8389			free(ctsio->kern_data_ptr, M_CTL);
8390			ctl_set_reservation_conflict(ctsio);
8391			ctl_done((union ctl_io *)ctsio);
8392			return (CTL_RETVAL_COMPLETE);
8393		}
8394		mtx_unlock(&lun->lun_lock);
8395	}
8396
8397	switch (cdb->action & SPRO_ACTION_MASK) {
8398	case SPRO_REGISTER:
8399	case SPRO_REG_IGNO: {
8400
8401#if 0
8402		printf("Registration received\n");
8403#endif
8404
8405		/*
8406		 * We don't support any of these options, as we report in
8407		 * the read capabilities request (see
8408		 * ctl_persistent_reserve_in(), above).
8409		 */
8410		if ((param->flags & SPR_SPEC_I_PT)
8411		 || (param->flags & SPR_ALL_TG_PT)
8412		 || (param->flags & SPR_APTPL)) {
8413			int bit_ptr;
8414
8415			if (param->flags & SPR_APTPL)
8416				bit_ptr = 0;
8417			else if (param->flags & SPR_ALL_TG_PT)
8418				bit_ptr = 2;
8419			else /* SPR_SPEC_I_PT */
8420				bit_ptr = 3;
8421
8422			free(ctsio->kern_data_ptr, M_CTL);
8423			ctl_set_invalid_field(ctsio,
8424					      /*sks_valid*/ 1,
8425					      /*command*/ 0,
8426					      /*field*/ 20,
8427					      /*bit_valid*/ 1,
8428					      /*bit*/ bit_ptr);
8429			ctl_done((union ctl_io *)ctsio);
8430			return (CTL_RETVAL_COMPLETE);
8431		}
8432
8433		mtx_lock(&lun->lun_lock);
8434
8435		/*
8436		 * The initiator wants to clear the
8437		 * key/unregister.
8438		 */
8439		if (sa_res_key == 0) {
8440			if ((res_key == 0
8441			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8442			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8443			  && ctl_get_prkey(lun, residx) == 0)) {
8444				mtx_unlock(&lun->lun_lock);
8445				goto done;
8446			}
8447
8448			ctl_clr_prkey(lun, residx);
8449			lun->pr_key_count--;
8450
8451			if (residx == lun->pr_res_idx) {
8452				lun->flags &= ~CTL_LUN_PR_RESERVED;
8453				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8454
8455				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8456				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8457				 && lun->pr_key_count) {
8458					/*
8459					 * If the reservation is a registrants
8460					 * only type we need to generate a UA
8461					 * for other registered inits.  The
8462					 * sense code should be RESERVATIONS
8463					 * RELEASED
8464					 */
8465
8466					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8467						if (ctl_get_prkey(lun, i +
8468						    softc->persis_offset) == 0)
8469							continue;
8470						ctl_est_ua(lun, i,
8471						    CTL_UA_RES_RELEASE);
8472					}
8473				}
8474				lun->res_type = 0;
8475			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8476				if (lun->pr_key_count==0) {
8477					lun->flags &= ~CTL_LUN_PR_RESERVED;
8478					lun->res_type = 0;
8479					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8480				}
8481			}
8482			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8483			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8484			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8485			persis_io.pr.pr_info.residx = residx;
8486			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8487			     &persis_io, sizeof(persis_io), 0 )) >
8488			     CTL_HA_STATUS_SUCCESS) {
8489				printf("CTL:Persis Out error returned from "
8490				       "ctl_ha_msg_send %d\n", isc_retval);
8491			}
8492		} else /* sa_res_key != 0 */ {
8493
8494			/*
8495			 * If we aren't registered currently then increment
8496			 * the key count and set the registered flag.
8497			 */
8498			ctl_alloc_prkey(lun, residx);
8499			if (ctl_get_prkey(lun, residx) == 0)
8500				lun->pr_key_count++;
8501			ctl_set_prkey(lun, residx, sa_res_key);
8502
8503			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8504			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8505			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8506			persis_io.pr.pr_info.residx = residx;
8507			memcpy(persis_io.pr.pr_info.sa_res_key,
8508			       param->serv_act_res_key,
8509			       sizeof(param->serv_act_res_key));
8510			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8511			     &persis_io, sizeof(persis_io), 0)) >
8512			     CTL_HA_STATUS_SUCCESS) {
8513				printf("CTL:Persis Out error returned from "
8514				       "ctl_ha_msg_send %d\n", isc_retval);
8515			}
8516		}
8517		lun->PRGeneration++;
8518		mtx_unlock(&lun->lun_lock);
8519
8520		break;
8521	}
8522	case SPRO_RESERVE:
8523#if 0
8524                printf("Reserve executed type %d\n", type);
8525#endif
8526		mtx_lock(&lun->lun_lock);
8527		if (lun->flags & CTL_LUN_PR_RESERVED) {
8528			/*
8529			 * if this isn't the reservation holder and it's
8530			 * not a "all registrants" type or if the type is
8531			 * different then we have a conflict
8532			 */
8533			if ((lun->pr_res_idx != residx
8534			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8535			 || lun->res_type != type) {
8536				mtx_unlock(&lun->lun_lock);
8537				free(ctsio->kern_data_ptr, M_CTL);
8538				ctl_set_reservation_conflict(ctsio);
8539				ctl_done((union ctl_io *)ctsio);
8540				return (CTL_RETVAL_COMPLETE);
8541			}
8542			mtx_unlock(&lun->lun_lock);
8543		} else /* create a reservation */ {
8544			/*
8545			 * If it's not an "all registrants" type record
8546			 * reservation holder
8547			 */
8548			if (type != SPR_TYPE_WR_EX_AR
8549			 && type != SPR_TYPE_EX_AC_AR)
8550				lun->pr_res_idx = residx; /* Res holder */
8551			else
8552				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8553
8554			lun->flags |= CTL_LUN_PR_RESERVED;
8555			lun->res_type = type;
8556
8557			mtx_unlock(&lun->lun_lock);
8558
8559			/* send msg to other side */
8560			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8561			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8562			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8563			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8564			persis_io.pr.pr_info.res_type = type;
8565			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8566			     &persis_io, sizeof(persis_io), 0)) >
8567			     CTL_HA_STATUS_SUCCESS) {
8568				printf("CTL:Persis Out error returned from "
8569				       "ctl_ha_msg_send %d\n", isc_retval);
8570			}
8571		}
8572		break;
8573
8574	case SPRO_RELEASE:
8575		mtx_lock(&lun->lun_lock);
8576		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8577			/* No reservation exists return good status */
8578			mtx_unlock(&lun->lun_lock);
8579			goto done;
8580		}
8581		/*
8582		 * Is this nexus a reservation holder?
8583		 */
8584		if (lun->pr_res_idx != residx
8585		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8586			/*
8587			 * not a res holder return good status but
8588			 * do nothing
8589			 */
8590			mtx_unlock(&lun->lun_lock);
8591			goto done;
8592		}
8593
8594		if (lun->res_type != type) {
8595			mtx_unlock(&lun->lun_lock);
8596			free(ctsio->kern_data_ptr, M_CTL);
8597			ctl_set_illegal_pr_release(ctsio);
8598			ctl_done((union ctl_io *)ctsio);
8599			return (CTL_RETVAL_COMPLETE);
8600		}
8601
8602		/* okay to release */
8603		lun->flags &= ~CTL_LUN_PR_RESERVED;
8604		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8605		lun->res_type = 0;
8606
8607		/*
8608		 * if this isn't an exclusive access
8609		 * res generate UA for all other
8610		 * registrants.
8611		 */
8612		if (type != SPR_TYPE_EX_AC
8613		 && type != SPR_TYPE_WR_EX) {
8614			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8615				if (i == residx ||
8616				    ctl_get_prkey(lun,
8617				     i + softc->persis_offset) == 0)
8618					continue;
8619				ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8620			}
8621		}
8622		mtx_unlock(&lun->lun_lock);
8623		/* Send msg to other side */
8624		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8625		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8626		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8627		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8628		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8629			printf("CTL:Persis Out error returned from "
8630			       "ctl_ha_msg_send %d\n", isc_retval);
8631		}
8632		break;
8633
8634	case SPRO_CLEAR:
8635		/* send msg to other side */
8636
8637		mtx_lock(&lun->lun_lock);
8638		lun->flags &= ~CTL_LUN_PR_RESERVED;
8639		lun->res_type = 0;
8640		lun->pr_key_count = 0;
8641		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8642
8643		ctl_clr_prkey(lun, residx);
8644		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8645			if (ctl_get_prkey(lun, i) != 0) {
8646				ctl_clr_prkey(lun, i);
8647				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8648			}
8649		lun->PRGeneration++;
8650		mtx_unlock(&lun->lun_lock);
8651		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8652		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8653		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8654		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8655		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8656			printf("CTL:Persis Out error returned from "
8657			       "ctl_ha_msg_send %d\n", isc_retval);
8658		}
8659		break;
8660
8661	case SPRO_PREEMPT:
8662	case SPRO_PRE_ABO: {
8663		int nretval;
8664
8665		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8666					  residx, ctsio, cdb, param);
8667		if (nretval != 0)
8668			return (CTL_RETVAL_COMPLETE);
8669		break;
8670	}
8671	default:
8672		panic("Invalid PR type %x", cdb->action);
8673	}
8674
8675done:
8676	free(ctsio->kern_data_ptr, M_CTL);
8677	ctl_set_success(ctsio);
8678	ctl_done((union ctl_io *)ctsio);
8679
8680	return (retval);
8681}
8682
8683/*
8684 * This routine is for handling a message from the other SC pertaining to
8685 * persistent reserve out. All the error checking will have been done
8686 * so only perorming the action need be done here to keep the two
8687 * in sync.
8688 */
8689static void
8690ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8691{
8692	struct ctl_lun *lun;
8693	struct ctl_softc *softc;
8694	int i;
8695	uint32_t targ_lun;
8696
8697	softc = control_softc;
8698
8699	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8700	lun = softc->ctl_luns[targ_lun];
8701	mtx_lock(&lun->lun_lock);
8702	switch(msg->pr.pr_info.action) {
8703	case CTL_PR_REG_KEY:
8704		ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8705		if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8706			lun->pr_key_count++;
8707		ctl_set_prkey(lun, msg->pr.pr_info.residx,
8708		    scsi_8btou64(msg->pr.pr_info.sa_res_key));
8709		lun->PRGeneration++;
8710		break;
8711
8712	case CTL_PR_UNREG_KEY:
8713		ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8714		lun->pr_key_count--;
8715
8716		/* XXX Need to see if the reservation has been released */
8717		/* if so do we need to generate UA? */
8718		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8719			lun->flags &= ~CTL_LUN_PR_RESERVED;
8720			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8721
8722			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8723			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8724			 && lun->pr_key_count) {
8725				/*
8726				 * If the reservation is a registrants
8727				 * only type we need to generate a UA
8728				 * for other registered inits.  The
8729				 * sense code should be RESERVATIONS
8730				 * RELEASED
8731				 */
8732
8733				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8734					if (ctl_get_prkey(lun, i +
8735					    softc->persis_offset) == 0)
8736						continue;
8737
8738					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8739				}
8740			}
8741			lun->res_type = 0;
8742		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8743			if (lun->pr_key_count==0) {
8744				lun->flags &= ~CTL_LUN_PR_RESERVED;
8745				lun->res_type = 0;
8746				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8747			}
8748		}
8749		lun->PRGeneration++;
8750		break;
8751
8752	case CTL_PR_RESERVE:
8753		lun->flags |= CTL_LUN_PR_RESERVED;
8754		lun->res_type = msg->pr.pr_info.res_type;
8755		lun->pr_res_idx = msg->pr.pr_info.residx;
8756
8757		break;
8758
8759	case CTL_PR_RELEASE:
8760		/*
8761		 * if this isn't an exclusive access res generate UA for all
8762		 * other registrants.
8763		 */
8764		if (lun->res_type != SPR_TYPE_EX_AC
8765		 && lun->res_type != SPR_TYPE_WR_EX) {
8766			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8767				if (ctl_get_prkey(lun, i + softc->persis_offset) != 0)
8768					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8769		}
8770
8771		lun->flags &= ~CTL_LUN_PR_RESERVED;
8772		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8773		lun->res_type = 0;
8774		break;
8775
8776	case CTL_PR_PREEMPT:
8777		ctl_pro_preempt_other(lun, msg);
8778		break;
8779	case CTL_PR_CLEAR:
8780		lun->flags &= ~CTL_LUN_PR_RESERVED;
8781		lun->res_type = 0;
8782		lun->pr_key_count = 0;
8783		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8784
8785		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8786			if (ctl_get_prkey(lun, i) == 0)
8787				continue;
8788			ctl_clr_prkey(lun, i);
8789			ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8790		}
8791		lun->PRGeneration++;
8792		break;
8793	}
8794
8795	mtx_unlock(&lun->lun_lock);
8796}
8797
8798int
8799ctl_read_write(struct ctl_scsiio *ctsio)
8800{
8801	struct ctl_lun *lun;
8802	struct ctl_lba_len_flags *lbalen;
8803	uint64_t lba;
8804	uint32_t num_blocks;
8805	int flags, retval;
8806	int isread;
8807
8808	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8809
8810	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8811
8812	flags = 0;
8813	retval = CTL_RETVAL_COMPLETE;
8814
8815	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8816	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8817	switch (ctsio->cdb[0]) {
8818	case READ_6:
8819	case WRITE_6: {
8820		struct scsi_rw_6 *cdb;
8821
8822		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8823
8824		lba = scsi_3btoul(cdb->addr);
8825		/* only 5 bits are valid in the most significant address byte */
8826		lba &= 0x1fffff;
8827		num_blocks = cdb->length;
8828		/*
8829		 * This is correct according to SBC-2.
8830		 */
8831		if (num_blocks == 0)
8832			num_blocks = 256;
8833		break;
8834	}
8835	case READ_10:
8836	case WRITE_10: {
8837		struct scsi_rw_10 *cdb;
8838
8839		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8840		if (cdb->byte2 & SRW10_FUA)
8841			flags |= CTL_LLF_FUA;
8842		if (cdb->byte2 & SRW10_DPO)
8843			flags |= CTL_LLF_DPO;
8844		lba = scsi_4btoul(cdb->addr);
8845		num_blocks = scsi_2btoul(cdb->length);
8846		break;
8847	}
8848	case WRITE_VERIFY_10: {
8849		struct scsi_write_verify_10 *cdb;
8850
8851		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8852		flags |= CTL_LLF_FUA;
8853		if (cdb->byte2 & SWV_DPO)
8854			flags |= CTL_LLF_DPO;
8855		lba = scsi_4btoul(cdb->addr);
8856		num_blocks = scsi_2btoul(cdb->length);
8857		break;
8858	}
8859	case READ_12:
8860	case WRITE_12: {
8861		struct scsi_rw_12 *cdb;
8862
8863		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8864		if (cdb->byte2 & SRW12_FUA)
8865			flags |= CTL_LLF_FUA;
8866		if (cdb->byte2 & SRW12_DPO)
8867			flags |= CTL_LLF_DPO;
8868		lba = scsi_4btoul(cdb->addr);
8869		num_blocks = scsi_4btoul(cdb->length);
8870		break;
8871	}
8872	case WRITE_VERIFY_12: {
8873		struct scsi_write_verify_12 *cdb;
8874
8875		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8876		flags |= CTL_LLF_FUA;
8877		if (cdb->byte2 & SWV_DPO)
8878			flags |= CTL_LLF_DPO;
8879		lba = scsi_4btoul(cdb->addr);
8880		num_blocks = scsi_4btoul(cdb->length);
8881		break;
8882	}
8883	case READ_16:
8884	case WRITE_16: {
8885		struct scsi_rw_16 *cdb;
8886
8887		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8888		if (cdb->byte2 & SRW12_FUA)
8889			flags |= CTL_LLF_FUA;
8890		if (cdb->byte2 & SRW12_DPO)
8891			flags |= CTL_LLF_DPO;
8892		lba = scsi_8btou64(cdb->addr);
8893		num_blocks = scsi_4btoul(cdb->length);
8894		break;
8895	}
8896	case WRITE_ATOMIC_16: {
8897		struct scsi_rw_16 *cdb;
8898
8899		if (lun->be_lun->atomicblock == 0) {
8900			ctl_set_invalid_opcode(ctsio);
8901			ctl_done((union ctl_io *)ctsio);
8902			return (CTL_RETVAL_COMPLETE);
8903		}
8904
8905		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8906		if (cdb->byte2 & SRW12_FUA)
8907			flags |= CTL_LLF_FUA;
8908		if (cdb->byte2 & SRW12_DPO)
8909			flags |= CTL_LLF_DPO;
8910		lba = scsi_8btou64(cdb->addr);
8911		num_blocks = scsi_4btoul(cdb->length);
8912		if (num_blocks > lun->be_lun->atomicblock) {
8913			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8914			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8915			    /*bit*/ 0);
8916			ctl_done((union ctl_io *)ctsio);
8917			return (CTL_RETVAL_COMPLETE);
8918		}
8919		break;
8920	}
8921	case WRITE_VERIFY_16: {
8922		struct scsi_write_verify_16 *cdb;
8923
8924		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8925		flags |= CTL_LLF_FUA;
8926		if (cdb->byte2 & SWV_DPO)
8927			flags |= CTL_LLF_DPO;
8928		lba = scsi_8btou64(cdb->addr);
8929		num_blocks = scsi_4btoul(cdb->length);
8930		break;
8931	}
8932	default:
8933		/*
8934		 * We got a command we don't support.  This shouldn't
8935		 * happen, commands should be filtered out above us.
8936		 */
8937		ctl_set_invalid_opcode(ctsio);
8938		ctl_done((union ctl_io *)ctsio);
8939
8940		return (CTL_RETVAL_COMPLETE);
8941		break; /* NOTREACHED */
8942	}
8943
8944	/*
8945	 * The first check is to make sure we're in bounds, the second
8946	 * check is to catch wrap-around problems.  If the lba + num blocks
8947	 * is less than the lba, then we've wrapped around and the block
8948	 * range is invalid anyway.
8949	 */
8950	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8951	 || ((lba + num_blocks) < lba)) {
8952		ctl_set_lba_out_of_range(ctsio);
8953		ctl_done((union ctl_io *)ctsio);
8954		return (CTL_RETVAL_COMPLETE);
8955	}
8956
8957	/*
8958	 * According to SBC-3, a transfer length of 0 is not an error.
8959	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8960	 * translates to 256 blocks for those commands.
8961	 */
8962	if (num_blocks == 0) {
8963		ctl_set_success(ctsio);
8964		ctl_done((union ctl_io *)ctsio);
8965		return (CTL_RETVAL_COMPLETE);
8966	}
8967
8968	/* Set FUA and/or DPO if caches are disabled. */
8969	if (isread) {
8970		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8971		    SCP_RCD) != 0)
8972			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8973	} else {
8974		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8975		    SCP_WCE) == 0)
8976			flags |= CTL_LLF_FUA;
8977	}
8978
8979	lbalen = (struct ctl_lba_len_flags *)
8980	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8981	lbalen->lba = lba;
8982	lbalen->len = num_blocks;
8983	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8984
8985	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8986	ctsio->kern_rel_offset = 0;
8987
8988	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8989
8990	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8991
8992	return (retval);
8993}
8994
8995static int
8996ctl_cnw_cont(union ctl_io *io)
8997{
8998	struct ctl_scsiio *ctsio;
8999	struct ctl_lun *lun;
9000	struct ctl_lba_len_flags *lbalen;
9001	int retval;
9002
9003	ctsio = &io->scsiio;
9004	ctsio->io_hdr.status = CTL_STATUS_NONE;
9005	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9006	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9007	lbalen = (struct ctl_lba_len_flags *)
9008	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9009	lbalen->flags &= ~CTL_LLF_COMPARE;
9010	lbalen->flags |= CTL_LLF_WRITE;
9011
9012	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9013	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9014	return (retval);
9015}
9016
9017int
9018ctl_cnw(struct ctl_scsiio *ctsio)
9019{
9020	struct ctl_lun *lun;
9021	struct ctl_lba_len_flags *lbalen;
9022	uint64_t lba;
9023	uint32_t num_blocks;
9024	int flags, retval;
9025
9026	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9027
9028	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9029
9030	flags = 0;
9031	retval = CTL_RETVAL_COMPLETE;
9032
9033	switch (ctsio->cdb[0]) {
9034	case COMPARE_AND_WRITE: {
9035		struct scsi_compare_and_write *cdb;
9036
9037		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9038		if (cdb->byte2 & SRW10_FUA)
9039			flags |= CTL_LLF_FUA;
9040		if (cdb->byte2 & SRW10_DPO)
9041			flags |= CTL_LLF_DPO;
9042		lba = scsi_8btou64(cdb->addr);
9043		num_blocks = cdb->length;
9044		break;
9045	}
9046	default:
9047		/*
9048		 * We got a command we don't support.  This shouldn't
9049		 * happen, commands should be filtered out above us.
9050		 */
9051		ctl_set_invalid_opcode(ctsio);
9052		ctl_done((union ctl_io *)ctsio);
9053
9054		return (CTL_RETVAL_COMPLETE);
9055		break; /* NOTREACHED */
9056	}
9057
9058	/*
9059	 * The first check is to make sure we're in bounds, the second
9060	 * check is to catch wrap-around problems.  If the lba + num blocks
9061	 * is less than the lba, then we've wrapped around and the block
9062	 * range is invalid anyway.
9063	 */
9064	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9065	 || ((lba + num_blocks) < lba)) {
9066		ctl_set_lba_out_of_range(ctsio);
9067		ctl_done((union ctl_io *)ctsio);
9068		return (CTL_RETVAL_COMPLETE);
9069	}
9070
9071	/*
9072	 * According to SBC-3, a transfer length of 0 is not an error.
9073	 */
9074	if (num_blocks == 0) {
9075		ctl_set_success(ctsio);
9076		ctl_done((union ctl_io *)ctsio);
9077		return (CTL_RETVAL_COMPLETE);
9078	}
9079
9080	/* Set FUA if write cache is disabled. */
9081	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9082	    SCP_WCE) == 0)
9083		flags |= CTL_LLF_FUA;
9084
9085	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9086	ctsio->kern_rel_offset = 0;
9087
9088	/*
9089	 * Set the IO_CONT flag, so that if this I/O gets passed to
9090	 * ctl_data_submit_done(), it'll get passed back to
9091	 * ctl_ctl_cnw_cont() for further processing.
9092	 */
9093	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9094	ctsio->io_cont = ctl_cnw_cont;
9095
9096	lbalen = (struct ctl_lba_len_flags *)
9097	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9098	lbalen->lba = lba;
9099	lbalen->len = num_blocks;
9100	lbalen->flags = CTL_LLF_COMPARE | flags;
9101
9102	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9103	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9104	return (retval);
9105}
9106
9107int
9108ctl_verify(struct ctl_scsiio *ctsio)
9109{
9110	struct ctl_lun *lun;
9111	struct ctl_lba_len_flags *lbalen;
9112	uint64_t lba;
9113	uint32_t num_blocks;
9114	int bytchk, flags;
9115	int retval;
9116
9117	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9118
9119	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9120
9121	bytchk = 0;
9122	flags = CTL_LLF_FUA;
9123	retval = CTL_RETVAL_COMPLETE;
9124
9125	switch (ctsio->cdb[0]) {
9126	case VERIFY_10: {
9127		struct scsi_verify_10 *cdb;
9128
9129		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9130		if (cdb->byte2 & SVFY_BYTCHK)
9131			bytchk = 1;
9132		if (cdb->byte2 & SVFY_DPO)
9133			flags |= CTL_LLF_DPO;
9134		lba = scsi_4btoul(cdb->addr);
9135		num_blocks = scsi_2btoul(cdb->length);
9136		break;
9137	}
9138	case VERIFY_12: {
9139		struct scsi_verify_12 *cdb;
9140
9141		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9142		if (cdb->byte2 & SVFY_BYTCHK)
9143			bytchk = 1;
9144		if (cdb->byte2 & SVFY_DPO)
9145			flags |= CTL_LLF_DPO;
9146		lba = scsi_4btoul(cdb->addr);
9147		num_blocks = scsi_4btoul(cdb->length);
9148		break;
9149	}
9150	case VERIFY_16: {
9151		struct scsi_rw_16 *cdb;
9152
9153		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9154		if (cdb->byte2 & SVFY_BYTCHK)
9155			bytchk = 1;
9156		if (cdb->byte2 & SVFY_DPO)
9157			flags |= CTL_LLF_DPO;
9158		lba = scsi_8btou64(cdb->addr);
9159		num_blocks = scsi_4btoul(cdb->length);
9160		break;
9161	}
9162	default:
9163		/*
9164		 * We got a command we don't support.  This shouldn't
9165		 * happen, commands should be filtered out above us.
9166		 */
9167		ctl_set_invalid_opcode(ctsio);
9168		ctl_done((union ctl_io *)ctsio);
9169		return (CTL_RETVAL_COMPLETE);
9170	}
9171
9172	/*
9173	 * The first check is to make sure we're in bounds, the second
9174	 * check is to catch wrap-around problems.  If the lba + num blocks
9175	 * is less than the lba, then we've wrapped around and the block
9176	 * range is invalid anyway.
9177	 */
9178	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9179	 || ((lba + num_blocks) < lba)) {
9180		ctl_set_lba_out_of_range(ctsio);
9181		ctl_done((union ctl_io *)ctsio);
9182		return (CTL_RETVAL_COMPLETE);
9183	}
9184
9185	/*
9186	 * According to SBC-3, a transfer length of 0 is not an error.
9187	 */
9188	if (num_blocks == 0) {
9189		ctl_set_success(ctsio);
9190		ctl_done((union ctl_io *)ctsio);
9191		return (CTL_RETVAL_COMPLETE);
9192	}
9193
9194	lbalen = (struct ctl_lba_len_flags *)
9195	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9196	lbalen->lba = lba;
9197	lbalen->len = num_blocks;
9198	if (bytchk) {
9199		lbalen->flags = CTL_LLF_COMPARE | flags;
9200		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9201	} else {
9202		lbalen->flags = CTL_LLF_VERIFY | flags;
9203		ctsio->kern_total_len = 0;
9204	}
9205	ctsio->kern_rel_offset = 0;
9206
9207	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9208	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9209	return (retval);
9210}
9211
9212int
9213ctl_report_luns(struct ctl_scsiio *ctsio)
9214{
9215	struct ctl_softc *softc = control_softc;
9216	struct scsi_report_luns *cdb;
9217	struct scsi_report_luns_data *lun_data;
9218	struct ctl_lun *lun, *request_lun;
9219	struct ctl_port *port;
9220	int num_luns, retval;
9221	uint32_t alloc_len, lun_datalen;
9222	int num_filled, well_known;
9223	uint32_t initidx, targ_lun_id, lun_id;
9224
9225	retval = CTL_RETVAL_COMPLETE;
9226	well_known = 0;
9227
9228	cdb = (struct scsi_report_luns *)ctsio->cdb;
9229	port = ctl_io_port(&ctsio->io_hdr);
9230
9231	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9232
9233	mtx_lock(&softc->ctl_lock);
9234	num_luns = 0;
9235	for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
9236		if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
9237			num_luns++;
9238	}
9239	mtx_unlock(&softc->ctl_lock);
9240
9241	switch (cdb->select_report) {
9242	case RPL_REPORT_DEFAULT:
9243	case RPL_REPORT_ALL:
9244		break;
9245	case RPL_REPORT_WELLKNOWN:
9246		well_known = 1;
9247		num_luns = 0;
9248		break;
9249	default:
9250		ctl_set_invalid_field(ctsio,
9251				      /*sks_valid*/ 1,
9252				      /*command*/ 1,
9253				      /*field*/ 2,
9254				      /*bit_valid*/ 0,
9255				      /*bit*/ 0);
9256		ctl_done((union ctl_io *)ctsio);
9257		return (retval);
9258		break; /* NOTREACHED */
9259	}
9260
9261	alloc_len = scsi_4btoul(cdb->length);
9262	/*
9263	 * The initiator has to allocate at least 16 bytes for this request,
9264	 * so he can at least get the header and the first LUN.  Otherwise
9265	 * we reject the request (per SPC-3 rev 14, section 6.21).
9266	 */
9267	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9268	    sizeof(struct scsi_report_luns_lundata))) {
9269		ctl_set_invalid_field(ctsio,
9270				      /*sks_valid*/ 1,
9271				      /*command*/ 1,
9272				      /*field*/ 6,
9273				      /*bit_valid*/ 0,
9274				      /*bit*/ 0);
9275		ctl_done((union ctl_io *)ctsio);
9276		return (retval);
9277	}
9278
9279	request_lun = (struct ctl_lun *)
9280		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9281
9282	lun_datalen = sizeof(*lun_data) +
9283		(num_luns * sizeof(struct scsi_report_luns_lundata));
9284
9285	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9286	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9287	ctsio->kern_sg_entries = 0;
9288
9289	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9290
9291	mtx_lock(&softc->ctl_lock);
9292	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9293		lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9294		if (lun_id >= CTL_MAX_LUNS)
9295			continue;
9296		lun = softc->ctl_luns[lun_id];
9297		if (lun == NULL)
9298			continue;
9299
9300		if (targ_lun_id <= 0xff) {
9301			/*
9302			 * Peripheral addressing method, bus number 0.
9303			 */
9304			lun_data->luns[num_filled].lundata[0] =
9305				RPL_LUNDATA_ATYP_PERIPH;
9306			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9307			num_filled++;
9308		} else if (targ_lun_id <= 0x3fff) {
9309			/*
9310			 * Flat addressing method.
9311			 */
9312			lun_data->luns[num_filled].lundata[0] =
9313				RPL_LUNDATA_ATYP_FLAT | (targ_lun_id >> 8);
9314			lun_data->luns[num_filled].lundata[1] =
9315				(targ_lun_id & 0xff);
9316			num_filled++;
9317		} else if (targ_lun_id <= 0xffffff) {
9318			/*
9319			 * Extended flat addressing method.
9320			 */
9321			lun_data->luns[num_filled].lundata[0] =
9322			    RPL_LUNDATA_ATYP_EXTLUN | 0x12;
9323			scsi_ulto3b(targ_lun_id,
9324			    &lun_data->luns[num_filled].lundata[1]);
9325			num_filled++;
9326		} else {
9327			printf("ctl_report_luns: bogus LUN number %jd, "
9328			       "skipping\n", (intmax_t)targ_lun_id);
9329		}
9330		/*
9331		 * According to SPC-3, rev 14 section 6.21:
9332		 *
9333		 * "The execution of a REPORT LUNS command to any valid and
9334		 * installed logical unit shall clear the REPORTED LUNS DATA
9335		 * HAS CHANGED unit attention condition for all logical
9336		 * units of that target with respect to the requesting
9337		 * initiator. A valid and installed logical unit is one
9338		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9339		 * INQUIRY data (see 6.4.2)."
9340		 *
9341		 * If request_lun is NULL, the LUN this report luns command
9342		 * was issued to is either disabled or doesn't exist. In that
9343		 * case, we shouldn't clear any pending lun change unit
9344		 * attention.
9345		 */
9346		if (request_lun != NULL) {
9347			mtx_lock(&lun->lun_lock);
9348			ctl_clr_ua(lun, initidx, CTL_UA_RES_RELEASE);
9349			mtx_unlock(&lun->lun_lock);
9350		}
9351	}
9352	mtx_unlock(&softc->ctl_lock);
9353
9354	/*
9355	 * It's quite possible that we've returned fewer LUNs than we allocated
9356	 * space for.  Trim it.
9357	 */
9358	lun_datalen = sizeof(*lun_data) +
9359		(num_filled * sizeof(struct scsi_report_luns_lundata));
9360
9361	if (lun_datalen < alloc_len) {
9362		ctsio->residual = alloc_len - lun_datalen;
9363		ctsio->kern_data_len = lun_datalen;
9364		ctsio->kern_total_len = lun_datalen;
9365	} else {
9366		ctsio->residual = 0;
9367		ctsio->kern_data_len = alloc_len;
9368		ctsio->kern_total_len = alloc_len;
9369	}
9370	ctsio->kern_data_resid = 0;
9371	ctsio->kern_rel_offset = 0;
9372	ctsio->kern_sg_entries = 0;
9373
9374	/*
9375	 * We set this to the actual data length, regardless of how much
9376	 * space we actually have to return results.  If the user looks at
9377	 * this value, he'll know whether or not he allocated enough space
9378	 * and reissue the command if necessary.  We don't support well
9379	 * known logical units, so if the user asks for that, return none.
9380	 */
9381	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9382
9383	/*
9384	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9385	 * this request.
9386	 */
9387	ctl_set_success(ctsio);
9388	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9389	ctsio->be_move_done = ctl_config_move_done;
9390	ctl_datamove((union ctl_io *)ctsio);
9391	return (retval);
9392}
9393
9394int
9395ctl_request_sense(struct ctl_scsiio *ctsio)
9396{
9397	struct scsi_request_sense *cdb;
9398	struct scsi_sense_data *sense_ptr;
9399	struct ctl_softc *ctl_softc;
9400	struct ctl_lun *lun;
9401	uint32_t initidx;
9402	int have_error;
9403	scsi_sense_data_type sense_format;
9404	ctl_ua_type ua_type;
9405
9406	cdb = (struct scsi_request_sense *)ctsio->cdb;
9407
9408	ctl_softc = control_softc;
9409	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9410
9411	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9412
9413	/*
9414	 * Determine which sense format the user wants.
9415	 */
9416	if (cdb->byte2 & SRS_DESC)
9417		sense_format = SSD_TYPE_DESC;
9418	else
9419		sense_format = SSD_TYPE_FIXED;
9420
9421	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9422	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9423	ctsio->kern_sg_entries = 0;
9424
9425	/*
9426	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9427	 * larger than the largest allowed value for the length field in the
9428	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9429	 */
9430	ctsio->residual = 0;
9431	ctsio->kern_data_len = cdb->length;
9432	ctsio->kern_total_len = cdb->length;
9433
9434	ctsio->kern_data_resid = 0;
9435	ctsio->kern_rel_offset = 0;
9436	ctsio->kern_sg_entries = 0;
9437
9438	/*
9439	 * If we don't have a LUN, we don't have any pending sense.
9440	 */
9441	if (lun == NULL)
9442		goto no_sense;
9443
9444	have_error = 0;
9445	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9446	/*
9447	 * Check for pending sense, and then for pending unit attentions.
9448	 * Pending sense gets returned first, then pending unit attentions.
9449	 */
9450	mtx_lock(&lun->lun_lock);
9451#ifdef CTL_WITH_CA
9452	if (ctl_is_set(lun->have_ca, initidx)) {
9453		scsi_sense_data_type stored_format;
9454
9455		/*
9456		 * Check to see which sense format was used for the stored
9457		 * sense data.
9458		 */
9459		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9460
9461		/*
9462		 * If the user requested a different sense format than the
9463		 * one we stored, then we need to convert it to the other
9464		 * format.  If we're going from descriptor to fixed format
9465		 * sense data, we may lose things in translation, depending
9466		 * on what options were used.
9467		 *
9468		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9469		 * for some reason we'll just copy it out as-is.
9470		 */
9471		if ((stored_format == SSD_TYPE_FIXED)
9472		 && (sense_format == SSD_TYPE_DESC))
9473			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9474			    &lun->pending_sense[initidx],
9475			    (struct scsi_sense_data_desc *)sense_ptr);
9476		else if ((stored_format == SSD_TYPE_DESC)
9477		      && (sense_format == SSD_TYPE_FIXED))
9478			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9479			    &lun->pending_sense[initidx],
9480			    (struct scsi_sense_data_fixed *)sense_ptr);
9481		else
9482			memcpy(sense_ptr, &lun->pending_sense[initidx],
9483			       MIN(sizeof(*sense_ptr),
9484			       sizeof(lun->pending_sense[initidx])));
9485
9486		ctl_clear_mask(lun->have_ca, initidx);
9487		have_error = 1;
9488	} else
9489#endif
9490	{
9491		ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9492		if (ua_type != CTL_UA_NONE)
9493			have_error = 1;
9494		if (ua_type == CTL_UA_LUN_CHANGE) {
9495			mtx_unlock(&lun->lun_lock);
9496			mtx_lock(&ctl_softc->ctl_lock);
9497			ctl_clear_ua(ctl_softc, initidx, ua_type);
9498			mtx_unlock(&ctl_softc->ctl_lock);
9499			mtx_lock(&lun->lun_lock);
9500		}
9501
9502	}
9503	mtx_unlock(&lun->lun_lock);
9504
9505	/*
9506	 * We already have a pending error, return it.
9507	 */
9508	if (have_error != 0) {
9509		/*
9510		 * We report the SCSI status as OK, since the status of the
9511		 * request sense command itself is OK.
9512		 * We report 0 for the sense length, because we aren't doing
9513		 * autosense in this case.  We're reporting sense as
9514		 * parameter data.
9515		 */
9516		ctl_set_success(ctsio);
9517		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9518		ctsio->be_move_done = ctl_config_move_done;
9519		ctl_datamove((union ctl_io *)ctsio);
9520		return (CTL_RETVAL_COMPLETE);
9521	}
9522
9523no_sense:
9524
9525	/*
9526	 * No sense information to report, so we report that everything is
9527	 * okay.
9528	 */
9529	ctl_set_sense_data(sense_ptr,
9530			   lun,
9531			   sense_format,
9532			   /*current_error*/ 1,
9533			   /*sense_key*/ SSD_KEY_NO_SENSE,
9534			   /*asc*/ 0x00,
9535			   /*ascq*/ 0x00,
9536			   SSD_ELEM_NONE);
9537
9538	/*
9539	 * We report 0 for the sense length, because we aren't doing
9540	 * autosense in this case.  We're reporting sense as parameter data.
9541	 */
9542	ctl_set_success(ctsio);
9543	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9544	ctsio->be_move_done = ctl_config_move_done;
9545	ctl_datamove((union ctl_io *)ctsio);
9546	return (CTL_RETVAL_COMPLETE);
9547}
9548
9549int
9550ctl_tur(struct ctl_scsiio *ctsio)
9551{
9552
9553	CTL_DEBUG_PRINT(("ctl_tur\n"));
9554
9555	ctl_set_success(ctsio);
9556	ctl_done((union ctl_io *)ctsio);
9557
9558	return (CTL_RETVAL_COMPLETE);
9559}
9560
9561#ifdef notyet
9562static int
9563ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9564{
9565
9566}
9567#endif
9568
9569/*
9570 * SCSI VPD page 0x00, the Supported VPD Pages page.
9571 */
9572static int
9573ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9574{
9575	struct scsi_vpd_supported_pages *pages;
9576	int sup_page_size;
9577	struct ctl_lun *lun;
9578	int p;
9579
9580	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9581
9582	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9583	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9584	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9585	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9586	ctsio->kern_sg_entries = 0;
9587
9588	if (sup_page_size < alloc_len) {
9589		ctsio->residual = alloc_len - sup_page_size;
9590		ctsio->kern_data_len = sup_page_size;
9591		ctsio->kern_total_len = sup_page_size;
9592	} else {
9593		ctsio->residual = 0;
9594		ctsio->kern_data_len = alloc_len;
9595		ctsio->kern_total_len = alloc_len;
9596	}
9597	ctsio->kern_data_resid = 0;
9598	ctsio->kern_rel_offset = 0;
9599	ctsio->kern_sg_entries = 0;
9600
9601	/*
9602	 * The control device is always connected.  The disk device, on the
9603	 * other hand, may not be online all the time.  Need to change this
9604	 * to figure out whether the disk device is actually online or not.
9605	 */
9606	if (lun != NULL)
9607		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9608				lun->be_lun->lun_type;
9609	else
9610		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9611
9612	p = 0;
9613	/* Supported VPD pages */
9614	pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9615	/* Serial Number */
9616	pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9617	/* Device Identification */
9618	pages->page_list[p++] = SVPD_DEVICE_ID;
9619	/* Extended INQUIRY Data */
9620	pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9621	/* Mode Page Policy */
9622	pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9623	/* SCSI Ports */
9624	pages->page_list[p++] = SVPD_SCSI_PORTS;
9625	/* Third-party Copy */
9626	pages->page_list[p++] = SVPD_SCSI_TPC;
9627	if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9628		/* Block limits */
9629		pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9630		/* Block Device Characteristics */
9631		pages->page_list[p++] = SVPD_BDC;
9632		/* Logical Block Provisioning */
9633		pages->page_list[p++] = SVPD_LBP;
9634	}
9635	pages->length = p;
9636
9637	ctl_set_success(ctsio);
9638	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9639	ctsio->be_move_done = ctl_config_move_done;
9640	ctl_datamove((union ctl_io *)ctsio);
9641	return (CTL_RETVAL_COMPLETE);
9642}
9643
9644/*
9645 * SCSI VPD page 0x80, the Unit Serial Number page.
9646 */
9647static int
9648ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9649{
9650	struct scsi_vpd_unit_serial_number *sn_ptr;
9651	struct ctl_lun *lun;
9652	int data_len;
9653
9654	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9655
9656	data_len = 4 + CTL_SN_LEN;
9657	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9658	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9659	if (data_len < alloc_len) {
9660		ctsio->residual = alloc_len - data_len;
9661		ctsio->kern_data_len = data_len;
9662		ctsio->kern_total_len = data_len;
9663	} else {
9664		ctsio->residual = 0;
9665		ctsio->kern_data_len = alloc_len;
9666		ctsio->kern_total_len = alloc_len;
9667	}
9668	ctsio->kern_data_resid = 0;
9669	ctsio->kern_rel_offset = 0;
9670	ctsio->kern_sg_entries = 0;
9671
9672	/*
9673	 * The control device is always connected.  The disk device, on the
9674	 * other hand, may not be online all the time.  Need to change this
9675	 * to figure out whether the disk device is actually online or not.
9676	 */
9677	if (lun != NULL)
9678		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9679				  lun->be_lun->lun_type;
9680	else
9681		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9682
9683	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9684	sn_ptr->length = CTL_SN_LEN;
9685	/*
9686	 * If we don't have a LUN, we just leave the serial number as
9687	 * all spaces.
9688	 */
9689	if (lun != NULL) {
9690		strncpy((char *)sn_ptr->serial_num,
9691			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9692	} else
9693		memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9694
9695	ctl_set_success(ctsio);
9696	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9697	ctsio->be_move_done = ctl_config_move_done;
9698	ctl_datamove((union ctl_io *)ctsio);
9699	return (CTL_RETVAL_COMPLETE);
9700}
9701
9702
9703/*
9704 * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9705 */
9706static int
9707ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9708{
9709	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9710	struct ctl_lun *lun;
9711	int data_len;
9712
9713	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9714
9715	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9716	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9717	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9718	ctsio->kern_sg_entries = 0;
9719
9720	if (data_len < alloc_len) {
9721		ctsio->residual = alloc_len - data_len;
9722		ctsio->kern_data_len = data_len;
9723		ctsio->kern_total_len = data_len;
9724	} else {
9725		ctsio->residual = 0;
9726		ctsio->kern_data_len = alloc_len;
9727		ctsio->kern_total_len = alloc_len;
9728	}
9729	ctsio->kern_data_resid = 0;
9730	ctsio->kern_rel_offset = 0;
9731	ctsio->kern_sg_entries = 0;
9732
9733	/*
9734	 * The control device is always connected.  The disk device, on the
9735	 * other hand, may not be online all the time.
9736	 */
9737	if (lun != NULL)
9738		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9739				     lun->be_lun->lun_type;
9740	else
9741		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9742	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9743	scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9744	/*
9745	 * We support head of queue, ordered and simple tags.
9746	 */
9747	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9748	/*
9749	 * Volatile cache supported.
9750	 */
9751	eid_ptr->flags3 = SVPD_EID_V_SUP;
9752
9753	/*
9754	 * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9755	 * attention for a particular IT nexus on all LUNs once we report
9756	 * it to that nexus once.  This bit is required as of SPC-4.
9757	 */
9758	eid_ptr->flags4 = SVPD_EID_LUICLT;
9759
9760	/*
9761	 * XXX KDM in order to correctly answer this, we would need
9762	 * information from the SIM to determine how much sense data it
9763	 * can send.  So this would really be a path inquiry field, most
9764	 * likely.  This can be set to a maximum of 252 according to SPC-4,
9765	 * but the hardware may or may not be able to support that much.
9766	 * 0 just means that the maximum sense data length is not reported.
9767	 */
9768	eid_ptr->max_sense_length = 0;
9769
9770	ctl_set_success(ctsio);
9771	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9772	ctsio->be_move_done = ctl_config_move_done;
9773	ctl_datamove((union ctl_io *)ctsio);
9774	return (CTL_RETVAL_COMPLETE);
9775}
9776
9777static int
9778ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9779{
9780	struct scsi_vpd_mode_page_policy *mpp_ptr;
9781	struct ctl_lun *lun;
9782	int data_len;
9783
9784	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9785
9786	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9787	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9788
9789	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9790	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9791	ctsio->kern_sg_entries = 0;
9792
9793	if (data_len < alloc_len) {
9794		ctsio->residual = alloc_len - data_len;
9795		ctsio->kern_data_len = data_len;
9796		ctsio->kern_total_len = data_len;
9797	} else {
9798		ctsio->residual = 0;
9799		ctsio->kern_data_len = alloc_len;
9800		ctsio->kern_total_len = alloc_len;
9801	}
9802	ctsio->kern_data_resid = 0;
9803	ctsio->kern_rel_offset = 0;
9804	ctsio->kern_sg_entries = 0;
9805
9806	/*
9807	 * The control device is always connected.  The disk device, on the
9808	 * other hand, may not be online all the time.
9809	 */
9810	if (lun != NULL)
9811		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9812				     lun->be_lun->lun_type;
9813	else
9814		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9815	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9816	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9817	mpp_ptr->descr[0].page_code = 0x3f;
9818	mpp_ptr->descr[0].subpage_code = 0xff;
9819	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9820
9821	ctl_set_success(ctsio);
9822	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9823	ctsio->be_move_done = ctl_config_move_done;
9824	ctl_datamove((union ctl_io *)ctsio);
9825	return (CTL_RETVAL_COMPLETE);
9826}
9827
9828/*
9829 * SCSI VPD page 0x83, the Device Identification page.
9830 */
9831static int
9832ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9833{
9834	struct scsi_vpd_device_id *devid_ptr;
9835	struct scsi_vpd_id_descriptor *desc;
9836	struct ctl_softc *softc;
9837	struct ctl_lun *lun;
9838	struct ctl_port *port;
9839	int data_len;
9840	uint8_t proto;
9841
9842	softc = control_softc;
9843
9844	port = softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9845	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9846
9847	data_len = sizeof(struct scsi_vpd_device_id) +
9848	    sizeof(struct scsi_vpd_id_descriptor) +
9849		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9850	    sizeof(struct scsi_vpd_id_descriptor) +
9851		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9852	if (lun && lun->lun_devid)
9853		data_len += lun->lun_devid->len;
9854	if (port->port_devid)
9855		data_len += port->port_devid->len;
9856	if (port->target_devid)
9857		data_len += port->target_devid->len;
9858
9859	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9860	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9861	ctsio->kern_sg_entries = 0;
9862
9863	if (data_len < alloc_len) {
9864		ctsio->residual = alloc_len - data_len;
9865		ctsio->kern_data_len = data_len;
9866		ctsio->kern_total_len = data_len;
9867	} else {
9868		ctsio->residual = 0;
9869		ctsio->kern_data_len = alloc_len;
9870		ctsio->kern_total_len = alloc_len;
9871	}
9872	ctsio->kern_data_resid = 0;
9873	ctsio->kern_rel_offset = 0;
9874	ctsio->kern_sg_entries = 0;
9875
9876	/*
9877	 * The control device is always connected.  The disk device, on the
9878	 * other hand, may not be online all the time.
9879	 */
9880	if (lun != NULL)
9881		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9882				     lun->be_lun->lun_type;
9883	else
9884		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9885	devid_ptr->page_code = SVPD_DEVICE_ID;
9886	scsi_ulto2b(data_len - 4, devid_ptr->length);
9887
9888	if (port->port_type == CTL_PORT_FC)
9889		proto = SCSI_PROTO_FC << 4;
9890	else if (port->port_type == CTL_PORT_ISCSI)
9891		proto = SCSI_PROTO_ISCSI << 4;
9892	else
9893		proto = SCSI_PROTO_SPI << 4;
9894	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9895
9896	/*
9897	 * We're using a LUN association here.  i.e., this device ID is a
9898	 * per-LUN identifier.
9899	 */
9900	if (lun && lun->lun_devid) {
9901		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9902		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9903		    lun->lun_devid->len);
9904	}
9905
9906	/*
9907	 * This is for the WWPN which is a port association.
9908	 */
9909	if (port->port_devid) {
9910		memcpy(desc, port->port_devid->data, port->port_devid->len);
9911		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9912		    port->port_devid->len);
9913	}
9914
9915	/*
9916	 * This is for the Relative Target Port(type 4h) identifier
9917	 */
9918	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9919	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9920	    SVPD_ID_TYPE_RELTARG;
9921	desc->length = 4;
9922	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9923	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9924	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9925
9926	/*
9927	 * This is for the Target Port Group(type 5h) identifier
9928	 */
9929	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9930	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9931	    SVPD_ID_TYPE_TPORTGRP;
9932	desc->length = 4;
9933	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9934	    &desc->identifier[2]);
9935	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9936	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9937
9938	/*
9939	 * This is for the Target identifier
9940	 */
9941	if (port->target_devid) {
9942		memcpy(desc, port->target_devid->data, port->target_devid->len);
9943	}
9944
9945	ctl_set_success(ctsio);
9946	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9947	ctsio->be_move_done = ctl_config_move_done;
9948	ctl_datamove((union ctl_io *)ctsio);
9949	return (CTL_RETVAL_COMPLETE);
9950}
9951
9952static int
9953ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9954{
9955	struct ctl_softc *softc = control_softc;
9956	struct scsi_vpd_scsi_ports *sp;
9957	struct scsi_vpd_port_designation *pd;
9958	struct scsi_vpd_port_designation_cont *pdc;
9959	struct ctl_lun *lun;
9960	struct ctl_port *port;
9961	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
9962	int num_target_port_groups;
9963
9964	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9965
9966	if (softc->is_single)
9967		num_target_port_groups = 1;
9968	else
9969		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
9970	num_target_ports = 0;
9971	iid_len = 0;
9972	id_len = 0;
9973	mtx_lock(&softc->ctl_lock);
9974	STAILQ_FOREACH(port, &softc->port_list, links) {
9975		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9976			continue;
9977		if (lun != NULL &&
9978		    ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9979			continue;
9980		num_target_ports++;
9981		if (port->init_devid)
9982			iid_len += port->init_devid->len;
9983		if (port->port_devid)
9984			id_len += port->port_devid->len;
9985	}
9986	mtx_unlock(&softc->ctl_lock);
9987
9988	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
9989	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9990	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9991	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9992	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9993	ctsio->kern_sg_entries = 0;
9994
9995	if (data_len < alloc_len) {
9996		ctsio->residual = alloc_len - data_len;
9997		ctsio->kern_data_len = data_len;
9998		ctsio->kern_total_len = data_len;
9999	} else {
10000		ctsio->residual = 0;
10001		ctsio->kern_data_len = alloc_len;
10002		ctsio->kern_total_len = alloc_len;
10003	}
10004	ctsio->kern_data_resid = 0;
10005	ctsio->kern_rel_offset = 0;
10006	ctsio->kern_sg_entries = 0;
10007
10008	/*
10009	 * The control device is always connected.  The disk device, on the
10010	 * other hand, may not be online all the time.  Need to change this
10011	 * to figure out whether the disk device is actually online or not.
10012	 */
10013	if (lun != NULL)
10014		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10015				  lun->be_lun->lun_type;
10016	else
10017		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10018
10019	sp->page_code = SVPD_SCSI_PORTS;
10020	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10021	    sp->page_length);
10022	pd = &sp->design[0];
10023
10024	mtx_lock(&softc->ctl_lock);
10025	pg = softc->port_offset / CTL_MAX_PORTS;
10026	for (g = 0; g < num_target_port_groups; g++) {
10027		STAILQ_FOREACH(port, &softc->port_list, links) {
10028			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10029				continue;
10030			if (lun != NULL &&
10031			    ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
10032				continue;
10033			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10034			scsi_ulto2b(p, pd->relative_port_id);
10035			if (port->init_devid && g == pg) {
10036				iid_len = port->init_devid->len;
10037				memcpy(pd->initiator_transportid,
10038				    port->init_devid->data, port->init_devid->len);
10039			} else
10040				iid_len = 0;
10041			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10042			pdc = (struct scsi_vpd_port_designation_cont *)
10043			    (&pd->initiator_transportid[iid_len]);
10044			if (port->port_devid && g == pg) {
10045				id_len = port->port_devid->len;
10046				memcpy(pdc->target_port_descriptors,
10047				    port->port_devid->data, port->port_devid->len);
10048			} else
10049				id_len = 0;
10050			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10051			pd = (struct scsi_vpd_port_designation *)
10052			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10053		}
10054	}
10055	mtx_unlock(&softc->ctl_lock);
10056
10057	ctl_set_success(ctsio);
10058	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10059	ctsio->be_move_done = ctl_config_move_done;
10060	ctl_datamove((union ctl_io *)ctsio);
10061	return (CTL_RETVAL_COMPLETE);
10062}
10063
10064static int
10065ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10066{
10067	struct scsi_vpd_block_limits *bl_ptr;
10068	struct ctl_lun *lun;
10069	int bs;
10070
10071	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10072
10073	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10074	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10075	ctsio->kern_sg_entries = 0;
10076
10077	if (sizeof(*bl_ptr) < alloc_len) {
10078		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10079		ctsio->kern_data_len = sizeof(*bl_ptr);
10080		ctsio->kern_total_len = sizeof(*bl_ptr);
10081	} else {
10082		ctsio->residual = 0;
10083		ctsio->kern_data_len = alloc_len;
10084		ctsio->kern_total_len = alloc_len;
10085	}
10086	ctsio->kern_data_resid = 0;
10087	ctsio->kern_rel_offset = 0;
10088	ctsio->kern_sg_entries = 0;
10089
10090	/*
10091	 * The control device is always connected.  The disk device, on the
10092	 * other hand, may not be online all the time.  Need to change this
10093	 * to figure out whether the disk device is actually online or not.
10094	 */
10095	if (lun != NULL)
10096		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10097				  lun->be_lun->lun_type;
10098	else
10099		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10100
10101	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10102	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
10103	bl_ptr->max_cmp_write_len = 0xff;
10104	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10105	if (lun != NULL) {
10106		bs = lun->be_lun->blocksize;
10107		scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
10108		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10109			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10110			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10111			if (lun->be_lun->ublockexp != 0) {
10112				scsi_ulto4b((1 << lun->be_lun->ublockexp),
10113				    bl_ptr->opt_unmap_grain);
10114				scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
10115				    bl_ptr->unmap_grain_align);
10116			}
10117		}
10118		scsi_ulto4b(lun->be_lun->atomicblock,
10119		    bl_ptr->max_atomic_transfer_length);
10120		scsi_ulto4b(0, bl_ptr->atomic_alignment);
10121		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10122	}
10123	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10124
10125	ctl_set_success(ctsio);
10126	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10127	ctsio->be_move_done = ctl_config_move_done;
10128	ctl_datamove((union ctl_io *)ctsio);
10129	return (CTL_RETVAL_COMPLETE);
10130}
10131
10132static int
10133ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10134{
10135	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10136	struct ctl_lun *lun;
10137	const char *value;
10138	u_int i;
10139
10140	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10141
10142	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10143	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10144	ctsio->kern_sg_entries = 0;
10145
10146	if (sizeof(*bdc_ptr) < alloc_len) {
10147		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10148		ctsio->kern_data_len = sizeof(*bdc_ptr);
10149		ctsio->kern_total_len = sizeof(*bdc_ptr);
10150	} else {
10151		ctsio->residual = 0;
10152		ctsio->kern_data_len = alloc_len;
10153		ctsio->kern_total_len = alloc_len;
10154	}
10155	ctsio->kern_data_resid = 0;
10156	ctsio->kern_rel_offset = 0;
10157	ctsio->kern_sg_entries = 0;
10158
10159	/*
10160	 * The control device is always connected.  The disk device, on the
10161	 * other hand, may not be online all the time.  Need to change this
10162	 * to figure out whether the disk device is actually online or not.
10163	 */
10164	if (lun != NULL)
10165		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10166				  lun->be_lun->lun_type;
10167	else
10168		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10169	bdc_ptr->page_code = SVPD_BDC;
10170	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10171	if (lun != NULL &&
10172	    (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
10173		i = strtol(value, NULL, 0);
10174	else
10175		i = CTL_DEFAULT_ROTATION_RATE;
10176	scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
10177	if (lun != NULL &&
10178	    (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
10179		i = strtol(value, NULL, 0);
10180	else
10181		i = 0;
10182	bdc_ptr->wab_wac_ff = (i & 0x0f);
10183	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10184
10185	ctl_set_success(ctsio);
10186	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10187	ctsio->be_move_done = ctl_config_move_done;
10188	ctl_datamove((union ctl_io *)ctsio);
10189	return (CTL_RETVAL_COMPLETE);
10190}
10191
10192static int
10193ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10194{
10195	struct scsi_vpd_logical_block_prov *lbp_ptr;
10196	struct ctl_lun *lun;
10197
10198	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10199
10200	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10201	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10202	ctsio->kern_sg_entries = 0;
10203
10204	if (sizeof(*lbp_ptr) < alloc_len) {
10205		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10206		ctsio->kern_data_len = sizeof(*lbp_ptr);
10207		ctsio->kern_total_len = sizeof(*lbp_ptr);
10208	} else {
10209		ctsio->residual = 0;
10210		ctsio->kern_data_len = alloc_len;
10211		ctsio->kern_total_len = alloc_len;
10212	}
10213	ctsio->kern_data_resid = 0;
10214	ctsio->kern_rel_offset = 0;
10215	ctsio->kern_sg_entries = 0;
10216
10217	/*
10218	 * The control device is always connected.  The disk device, on the
10219	 * other hand, may not be online all the time.  Need to change this
10220	 * to figure out whether the disk device is actually online or not.
10221	 */
10222	if (lun != NULL)
10223		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10224				  lun->be_lun->lun_type;
10225	else
10226		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10227
10228	lbp_ptr->page_code = SVPD_LBP;
10229	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10230	lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10231	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10232		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10233		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10234		lbp_ptr->prov_type = SVPD_LBP_THIN;
10235	}
10236
10237	ctl_set_success(ctsio);
10238	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10239	ctsio->be_move_done = ctl_config_move_done;
10240	ctl_datamove((union ctl_io *)ctsio);
10241	return (CTL_RETVAL_COMPLETE);
10242}
10243
10244/*
10245 * INQUIRY with the EVPD bit set.
10246 */
10247static int
10248ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10249{
10250	struct ctl_lun *lun;
10251	struct scsi_inquiry *cdb;
10252	int alloc_len, retval;
10253
10254	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10255	cdb = (struct scsi_inquiry *)ctsio->cdb;
10256	alloc_len = scsi_2btoul(cdb->length);
10257
10258	switch (cdb->page_code) {
10259	case SVPD_SUPPORTED_PAGES:
10260		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10261		break;
10262	case SVPD_UNIT_SERIAL_NUMBER:
10263		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10264		break;
10265	case SVPD_DEVICE_ID:
10266		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10267		break;
10268	case SVPD_EXTENDED_INQUIRY_DATA:
10269		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10270		break;
10271	case SVPD_MODE_PAGE_POLICY:
10272		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10273		break;
10274	case SVPD_SCSI_PORTS:
10275		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10276		break;
10277	case SVPD_SCSI_TPC:
10278		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10279		break;
10280	case SVPD_BLOCK_LIMITS:
10281		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10282			goto err;
10283		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10284		break;
10285	case SVPD_BDC:
10286		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10287			goto err;
10288		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10289		break;
10290	case SVPD_LBP:
10291		if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10292			goto err;
10293		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10294		break;
10295	default:
10296err:
10297		ctl_set_invalid_field(ctsio,
10298				      /*sks_valid*/ 1,
10299				      /*command*/ 1,
10300				      /*field*/ 2,
10301				      /*bit_valid*/ 0,
10302				      /*bit*/ 0);
10303		ctl_done((union ctl_io *)ctsio);
10304		retval = CTL_RETVAL_COMPLETE;
10305		break;
10306	}
10307
10308	return (retval);
10309}
10310
10311/*
10312 * Standard INQUIRY data.
10313 */
10314static int
10315ctl_inquiry_std(struct ctl_scsiio *ctsio)
10316{
10317	struct scsi_inquiry_data *inq_ptr;
10318	struct scsi_inquiry *cdb;
10319	struct ctl_softc *softc;
10320	struct ctl_lun *lun;
10321	char *val;
10322	uint32_t alloc_len, data_len;
10323	ctl_port_type port_type;
10324
10325	softc = control_softc;
10326
10327	/*
10328	 * Figure out whether we're talking to a Fibre Channel port or not.
10329	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10330	 * SCSI front ends.
10331	 */
10332	port_type = softc->ctl_ports[
10333	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10334	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10335		port_type = CTL_PORT_SCSI;
10336
10337	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10338	cdb = (struct scsi_inquiry *)ctsio->cdb;
10339	alloc_len = scsi_2btoul(cdb->length);
10340
10341	/*
10342	 * We malloc the full inquiry data size here and fill it
10343	 * in.  If the user only asks for less, we'll give him
10344	 * that much.
10345	 */
10346	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10347	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10348	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10349	ctsio->kern_sg_entries = 0;
10350	ctsio->kern_data_resid = 0;
10351	ctsio->kern_rel_offset = 0;
10352
10353	if (data_len < alloc_len) {
10354		ctsio->residual = alloc_len - data_len;
10355		ctsio->kern_data_len = data_len;
10356		ctsio->kern_total_len = data_len;
10357	} else {
10358		ctsio->residual = 0;
10359		ctsio->kern_data_len = alloc_len;
10360		ctsio->kern_total_len = alloc_len;
10361	}
10362
10363	/*
10364	 * If we have a LUN configured, report it as connected.  Otherwise,
10365	 * report that it is offline or no device is supported, depending
10366	 * on the value of inquiry_pq_no_lun.
10367	 *
10368	 * According to the spec (SPC-4 r34), the peripheral qualifier
10369	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10370	 *
10371	 * "A peripheral device having the specified peripheral device type
10372	 * is not connected to this logical unit. However, the device
10373	 * server is capable of supporting the specified peripheral device
10374	 * type on this logical unit."
10375	 *
10376	 * According to the same spec, the peripheral qualifier
10377	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10378	 *
10379	 * "The device server is not capable of supporting a peripheral
10380	 * device on this logical unit. For this peripheral qualifier the
10381	 * peripheral device type shall be set to 1Fh. All other peripheral
10382	 * device type values are reserved for this peripheral qualifier."
10383	 *
10384	 * Given the text, it would seem that we probably want to report that
10385	 * the LUN is offline here.  There is no LUN connected, but we can
10386	 * support a LUN at the given LUN number.
10387	 *
10388	 * In the real world, though, it sounds like things are a little
10389	 * different:
10390	 *
10391	 * - Linux, when presented with a LUN with the offline peripheral
10392	 *   qualifier, will create an sg driver instance for it.  So when
10393	 *   you attach it to CTL, you wind up with a ton of sg driver
10394	 *   instances.  (One for every LUN that Linux bothered to probe.)
10395	 *   Linux does this despite the fact that it issues a REPORT LUNs
10396	 *   to LUN 0 to get the inventory of supported LUNs.
10397	 *
10398	 * - There is other anecdotal evidence (from Emulex folks) about
10399	 *   arrays that use the offline peripheral qualifier for LUNs that
10400	 *   are on the "passive" path in an active/passive array.
10401	 *
10402	 * So the solution is provide a hopefully reasonable default
10403	 * (return bad/no LUN) and allow the user to change the behavior
10404	 * with a tunable/sysctl variable.
10405	 */
10406	if (lun != NULL)
10407		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10408				  lun->be_lun->lun_type;
10409	else if (softc->inquiry_pq_no_lun == 0)
10410		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10411	else
10412		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10413
10414	/* RMB in byte 2 is 0 */
10415	inq_ptr->version = SCSI_REV_SPC4;
10416
10417	/*
10418	 * According to SAM-3, even if a device only supports a single
10419	 * level of LUN addressing, it should still set the HISUP bit:
10420	 *
10421	 * 4.9.1 Logical unit numbers overview
10422	 *
10423	 * All logical unit number formats described in this standard are
10424	 * hierarchical in structure even when only a single level in that
10425	 * hierarchy is used. The HISUP bit shall be set to one in the
10426	 * standard INQUIRY data (see SPC-2) when any logical unit number
10427	 * format described in this standard is used.  Non-hierarchical
10428	 * formats are outside the scope of this standard.
10429	 *
10430	 * Therefore we set the HiSup bit here.
10431	 *
10432	 * The reponse format is 2, per SPC-3.
10433	 */
10434	inq_ptr->response_format = SID_HiSup | 2;
10435
10436	inq_ptr->additional_length = data_len -
10437	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10438	CTL_DEBUG_PRINT(("additional_length = %d\n",
10439			 inq_ptr->additional_length));
10440
10441	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10442	/* 16 bit addressing */
10443	if (port_type == CTL_PORT_SCSI)
10444		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10445	/* XXX set the SID_MultiP bit here if we're actually going to
10446	   respond on multiple ports */
10447	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10448
10449	/* 16 bit data bus, synchronous transfers */
10450	if (port_type == CTL_PORT_SCSI)
10451		inq_ptr->flags = SID_WBus16 | SID_Sync;
10452	/*
10453	 * XXX KDM do we want to support tagged queueing on the control
10454	 * device at all?
10455	 */
10456	if ((lun == NULL)
10457	 || (lun->be_lun->lun_type != T_PROCESSOR))
10458		inq_ptr->flags |= SID_CmdQue;
10459	/*
10460	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10461	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10462	 * name and 4 bytes for the revision.
10463	 */
10464	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10465	    "vendor")) == NULL) {
10466		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10467	} else {
10468		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10469		strncpy(inq_ptr->vendor, val,
10470		    min(sizeof(inq_ptr->vendor), strlen(val)));
10471	}
10472	if (lun == NULL) {
10473		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10474		    sizeof(inq_ptr->product));
10475	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10476		switch (lun->be_lun->lun_type) {
10477		case T_DIRECT:
10478			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10479			    sizeof(inq_ptr->product));
10480			break;
10481		case T_PROCESSOR:
10482			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10483			    sizeof(inq_ptr->product));
10484			break;
10485		default:
10486			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10487			    sizeof(inq_ptr->product));
10488			break;
10489		}
10490	} else {
10491		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10492		strncpy(inq_ptr->product, val,
10493		    min(sizeof(inq_ptr->product), strlen(val)));
10494	}
10495
10496	/*
10497	 * XXX make this a macro somewhere so it automatically gets
10498	 * incremented when we make changes.
10499	 */
10500	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10501	    "revision")) == NULL) {
10502		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10503	} else {
10504		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10505		strncpy(inq_ptr->revision, val,
10506		    min(sizeof(inq_ptr->revision), strlen(val)));
10507	}
10508
10509	/*
10510	 * For parallel SCSI, we support double transition and single
10511	 * transition clocking.  We also support QAS (Quick Arbitration
10512	 * and Selection) and Information Unit transfers on both the
10513	 * control and array devices.
10514	 */
10515	if (port_type == CTL_PORT_SCSI)
10516		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10517				    SID_SPI_IUS;
10518
10519	/* SAM-5 (no version claimed) */
10520	scsi_ulto2b(0x00A0, inq_ptr->version1);
10521	/* SPC-4 (no version claimed) */
10522	scsi_ulto2b(0x0460, inq_ptr->version2);
10523	if (port_type == CTL_PORT_FC) {
10524		/* FCP-2 ANSI INCITS.350:2003 */
10525		scsi_ulto2b(0x0917, inq_ptr->version3);
10526	} else if (port_type == CTL_PORT_SCSI) {
10527		/* SPI-4 ANSI INCITS.362:200x */
10528		scsi_ulto2b(0x0B56, inq_ptr->version3);
10529	} else if (port_type == CTL_PORT_ISCSI) {
10530		/* iSCSI (no version claimed) */
10531		scsi_ulto2b(0x0960, inq_ptr->version3);
10532	} else if (port_type == CTL_PORT_SAS) {
10533		/* SAS (no version claimed) */
10534		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10535	}
10536
10537	if (lun == NULL) {
10538		/* SBC-4 (no version claimed) */
10539		scsi_ulto2b(0x0600, inq_ptr->version4);
10540	} else {
10541		switch (lun->be_lun->lun_type) {
10542		case T_DIRECT:
10543			/* SBC-4 (no version claimed) */
10544			scsi_ulto2b(0x0600, inq_ptr->version4);
10545			break;
10546		case T_PROCESSOR:
10547		default:
10548			break;
10549		}
10550	}
10551
10552	ctl_set_success(ctsio);
10553	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10554	ctsio->be_move_done = ctl_config_move_done;
10555	ctl_datamove((union ctl_io *)ctsio);
10556	return (CTL_RETVAL_COMPLETE);
10557}
10558
10559int
10560ctl_inquiry(struct ctl_scsiio *ctsio)
10561{
10562	struct scsi_inquiry *cdb;
10563	int retval;
10564
10565	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10566
10567	cdb = (struct scsi_inquiry *)ctsio->cdb;
10568	if (cdb->byte2 & SI_EVPD)
10569		retval = ctl_inquiry_evpd(ctsio);
10570	else if (cdb->page_code == 0)
10571		retval = ctl_inquiry_std(ctsio);
10572	else {
10573		ctl_set_invalid_field(ctsio,
10574				      /*sks_valid*/ 1,
10575				      /*command*/ 1,
10576				      /*field*/ 2,
10577				      /*bit_valid*/ 0,
10578				      /*bit*/ 0);
10579		ctl_done((union ctl_io *)ctsio);
10580		return (CTL_RETVAL_COMPLETE);
10581	}
10582
10583	return (retval);
10584}
10585
10586/*
10587 * For known CDB types, parse the LBA and length.
10588 */
10589static int
10590ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10591{
10592	if (io->io_hdr.io_type != CTL_IO_SCSI)
10593		return (1);
10594
10595	switch (io->scsiio.cdb[0]) {
10596	case COMPARE_AND_WRITE: {
10597		struct scsi_compare_and_write *cdb;
10598
10599		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10600
10601		*lba = scsi_8btou64(cdb->addr);
10602		*len = cdb->length;
10603		break;
10604	}
10605	case READ_6:
10606	case WRITE_6: {
10607		struct scsi_rw_6 *cdb;
10608
10609		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10610
10611		*lba = scsi_3btoul(cdb->addr);
10612		/* only 5 bits are valid in the most significant address byte */
10613		*lba &= 0x1fffff;
10614		*len = cdb->length;
10615		break;
10616	}
10617	case READ_10:
10618	case WRITE_10: {
10619		struct scsi_rw_10 *cdb;
10620
10621		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10622
10623		*lba = scsi_4btoul(cdb->addr);
10624		*len = scsi_2btoul(cdb->length);
10625		break;
10626	}
10627	case WRITE_VERIFY_10: {
10628		struct scsi_write_verify_10 *cdb;
10629
10630		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10631
10632		*lba = scsi_4btoul(cdb->addr);
10633		*len = scsi_2btoul(cdb->length);
10634		break;
10635	}
10636	case READ_12:
10637	case WRITE_12: {
10638		struct scsi_rw_12 *cdb;
10639
10640		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10641
10642		*lba = scsi_4btoul(cdb->addr);
10643		*len = scsi_4btoul(cdb->length);
10644		break;
10645	}
10646	case WRITE_VERIFY_12: {
10647		struct scsi_write_verify_12 *cdb;
10648
10649		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10650
10651		*lba = scsi_4btoul(cdb->addr);
10652		*len = scsi_4btoul(cdb->length);
10653		break;
10654	}
10655	case READ_16:
10656	case WRITE_16:
10657	case WRITE_ATOMIC_16: {
10658		struct scsi_rw_16 *cdb;
10659
10660		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10661
10662		*lba = scsi_8btou64(cdb->addr);
10663		*len = scsi_4btoul(cdb->length);
10664		break;
10665	}
10666	case WRITE_VERIFY_16: {
10667		struct scsi_write_verify_16 *cdb;
10668
10669		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10670
10671		*lba = scsi_8btou64(cdb->addr);
10672		*len = scsi_4btoul(cdb->length);
10673		break;
10674	}
10675	case WRITE_SAME_10: {
10676		struct scsi_write_same_10 *cdb;
10677
10678		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10679
10680		*lba = scsi_4btoul(cdb->addr);
10681		*len = scsi_2btoul(cdb->length);
10682		break;
10683	}
10684	case WRITE_SAME_16: {
10685		struct scsi_write_same_16 *cdb;
10686
10687		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10688
10689		*lba = scsi_8btou64(cdb->addr);
10690		*len = scsi_4btoul(cdb->length);
10691		break;
10692	}
10693	case VERIFY_10: {
10694		struct scsi_verify_10 *cdb;
10695
10696		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10697
10698		*lba = scsi_4btoul(cdb->addr);
10699		*len = scsi_2btoul(cdb->length);
10700		break;
10701	}
10702	case VERIFY_12: {
10703		struct scsi_verify_12 *cdb;
10704
10705		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10706
10707		*lba = scsi_4btoul(cdb->addr);
10708		*len = scsi_4btoul(cdb->length);
10709		break;
10710	}
10711	case VERIFY_16: {
10712		struct scsi_verify_16 *cdb;
10713
10714		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10715
10716		*lba = scsi_8btou64(cdb->addr);
10717		*len = scsi_4btoul(cdb->length);
10718		break;
10719	}
10720	case UNMAP: {
10721		*lba = 0;
10722		*len = UINT64_MAX;
10723		break;
10724	}
10725	case SERVICE_ACTION_IN: {	/* GET LBA STATUS */
10726		struct scsi_get_lba_status *cdb;
10727
10728		cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10729		*lba = scsi_8btou64(cdb->addr);
10730		*len = UINT32_MAX;
10731		break;
10732	}
10733	default:
10734		return (1);
10735		break; /* NOTREACHED */
10736	}
10737
10738	return (0);
10739}
10740
10741static ctl_action
10742ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10743    bool seq)
10744{
10745	uint64_t endlba1, endlba2;
10746
10747	endlba1 = lba1 + len1 - (seq ? 0 : 1);
10748	endlba2 = lba2 + len2 - 1;
10749
10750	if ((endlba1 < lba2) || (endlba2 < lba1))
10751		return (CTL_ACTION_PASS);
10752	else
10753		return (CTL_ACTION_BLOCK);
10754}
10755
10756static int
10757ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10758{
10759	struct ctl_ptr_len_flags *ptrlen;
10760	struct scsi_unmap_desc *buf, *end, *range;
10761	uint64_t lba;
10762	uint32_t len;
10763
10764	/* If not UNMAP -- go other way. */
10765	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10766	    io->scsiio.cdb[0] != UNMAP)
10767		return (CTL_ACTION_ERROR);
10768
10769	/* If UNMAP without data -- block and wait for data. */
10770	ptrlen = (struct ctl_ptr_len_flags *)
10771	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10772	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10773	    ptrlen->ptr == NULL)
10774		return (CTL_ACTION_BLOCK);
10775
10776	/* UNMAP with data -- check for collision. */
10777	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10778	end = buf + ptrlen->len / sizeof(*buf);
10779	for (range = buf; range < end; range++) {
10780		lba = scsi_8btou64(range->lba);
10781		len = scsi_4btoul(range->length);
10782		if ((lba < lba2 + len2) && (lba + len > lba2))
10783			return (CTL_ACTION_BLOCK);
10784	}
10785	return (CTL_ACTION_PASS);
10786}
10787
10788static ctl_action
10789ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10790{
10791	uint64_t lba1, lba2;
10792	uint64_t len1, len2;
10793	int retval;
10794
10795	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10796		return (CTL_ACTION_ERROR);
10797
10798	retval = ctl_extent_check_unmap(io1, lba2, len2);
10799	if (retval != CTL_ACTION_ERROR)
10800		return (retval);
10801
10802	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10803		return (CTL_ACTION_ERROR);
10804
10805	return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10806}
10807
10808static ctl_action
10809ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10810{
10811	uint64_t lba1, lba2;
10812	uint64_t len1, len2;
10813
10814	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10815		return (CTL_ACTION_ERROR);
10816	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10817		return (CTL_ACTION_ERROR);
10818
10819	if (lba1 + len1 == lba2)
10820		return (CTL_ACTION_BLOCK);
10821	return (CTL_ACTION_PASS);
10822}
10823
10824static ctl_action
10825ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10826    union ctl_io *ooa_io)
10827{
10828	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10829	ctl_serialize_action *serialize_row;
10830
10831	/*
10832	 * The initiator attempted multiple untagged commands at the same
10833	 * time.  Can't do that.
10834	 */
10835	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10836	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10837	 && ((pending_io->io_hdr.nexus.targ_port ==
10838	      ooa_io->io_hdr.nexus.targ_port)
10839	  && (pending_io->io_hdr.nexus.initid.id ==
10840	      ooa_io->io_hdr.nexus.initid.id))
10841	 && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10842	      CTL_FLAG_STATUS_SENT)) == 0))
10843		return (CTL_ACTION_OVERLAP);
10844
10845	/*
10846	 * The initiator attempted to send multiple tagged commands with
10847	 * the same ID.  (It's fine if different initiators have the same
10848	 * tag ID.)
10849	 *
10850	 * Even if all of those conditions are true, we don't kill the I/O
10851	 * if the command ahead of us has been aborted.  We won't end up
10852	 * sending it to the FETD, and it's perfectly legal to resend a
10853	 * command with the same tag number as long as the previous
10854	 * instance of this tag number has been aborted somehow.
10855	 */
10856	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10857	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10858	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10859	 && ((pending_io->io_hdr.nexus.targ_port ==
10860	      ooa_io->io_hdr.nexus.targ_port)
10861	  && (pending_io->io_hdr.nexus.initid.id ==
10862	      ooa_io->io_hdr.nexus.initid.id))
10863	 && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10864	      CTL_FLAG_STATUS_SENT)) == 0))
10865		return (CTL_ACTION_OVERLAP_TAG);
10866
10867	/*
10868	 * If we get a head of queue tag, SAM-3 says that we should
10869	 * immediately execute it.
10870	 *
10871	 * What happens if this command would normally block for some other
10872	 * reason?  e.g. a request sense with a head of queue tag
10873	 * immediately after a write.  Normally that would block, but this
10874	 * will result in its getting executed immediately...
10875	 *
10876	 * We currently return "pass" instead of "skip", so we'll end up
10877	 * going through the rest of the queue to check for overlapped tags.
10878	 *
10879	 * XXX KDM check for other types of blockage first??
10880	 */
10881	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10882		return (CTL_ACTION_PASS);
10883
10884	/*
10885	 * Ordered tags have to block until all items ahead of them
10886	 * have completed.  If we get called with an ordered tag, we always
10887	 * block, if something else is ahead of us in the queue.
10888	 */
10889	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10890		return (CTL_ACTION_BLOCK);
10891
10892	/*
10893	 * Simple tags get blocked until all head of queue and ordered tags
10894	 * ahead of them have completed.  I'm lumping untagged commands in
10895	 * with simple tags here.  XXX KDM is that the right thing to do?
10896	 */
10897	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10898	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10899	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10900	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10901		return (CTL_ACTION_BLOCK);
10902
10903	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10904	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10905
10906	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10907
10908	switch (serialize_row[pending_entry->seridx]) {
10909	case CTL_SER_BLOCK:
10910		return (CTL_ACTION_BLOCK);
10911	case CTL_SER_EXTENT:
10912		return (ctl_extent_check(ooa_io, pending_io,
10913		    (lun->serseq == CTL_LUN_SERSEQ_ON)));
10914	case CTL_SER_EXTENTOPT:
10915		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10916		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10917			return (ctl_extent_check(ooa_io, pending_io,
10918			    (lun->serseq == CTL_LUN_SERSEQ_ON)));
10919		return (CTL_ACTION_PASS);
10920	case CTL_SER_EXTENTSEQ:
10921		if (lun->serseq != CTL_LUN_SERSEQ_OFF)
10922			return (ctl_extent_check_seq(ooa_io, pending_io));
10923		return (CTL_ACTION_PASS);
10924	case CTL_SER_PASS:
10925		return (CTL_ACTION_PASS);
10926	case CTL_SER_BLOCKOPT:
10927		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10928		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10929			return (CTL_ACTION_BLOCK);
10930		return (CTL_ACTION_PASS);
10931	case CTL_SER_SKIP:
10932		return (CTL_ACTION_SKIP);
10933	default:
10934		panic("invalid serialization value %d",
10935		      serialize_row[pending_entry->seridx]);
10936	}
10937
10938	return (CTL_ACTION_ERROR);
10939}
10940
10941/*
10942 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10943 * Assumptions:
10944 * - pending_io is generally either incoming, or on the blocked queue
10945 * - starting I/O is the I/O we want to start the check with.
10946 */
10947static ctl_action
10948ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10949	      union ctl_io *starting_io)
10950{
10951	union ctl_io *ooa_io;
10952	ctl_action action;
10953
10954	mtx_assert(&lun->lun_lock, MA_OWNED);
10955
10956	/*
10957	 * Run back along the OOA queue, starting with the current
10958	 * blocked I/O and going through every I/O before it on the
10959	 * queue.  If starting_io is NULL, we'll just end up returning
10960	 * CTL_ACTION_PASS.
10961	 */
10962	for (ooa_io = starting_io; ooa_io != NULL;
10963	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10964	     ooa_links)){
10965
10966		/*
10967		 * This routine just checks to see whether
10968		 * cur_blocked is blocked by ooa_io, which is ahead
10969		 * of it in the queue.  It doesn't queue/dequeue
10970		 * cur_blocked.
10971		 */
10972		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10973		switch (action) {
10974		case CTL_ACTION_BLOCK:
10975		case CTL_ACTION_OVERLAP:
10976		case CTL_ACTION_OVERLAP_TAG:
10977		case CTL_ACTION_SKIP:
10978		case CTL_ACTION_ERROR:
10979			return (action);
10980			break; /* NOTREACHED */
10981		case CTL_ACTION_PASS:
10982			break;
10983		default:
10984			panic("invalid action %d", action);
10985			break;  /* NOTREACHED */
10986		}
10987	}
10988
10989	return (CTL_ACTION_PASS);
10990}
10991
10992/*
10993 * Assumptions:
10994 * - An I/O has just completed, and has been removed from the per-LUN OOA
10995 *   queue, so some items on the blocked queue may now be unblocked.
10996 */
10997static int
10998ctl_check_blocked(struct ctl_lun *lun)
10999{
11000	union ctl_io *cur_blocked, *next_blocked;
11001
11002	mtx_assert(&lun->lun_lock, MA_OWNED);
11003
11004	/*
11005	 * Run forward from the head of the blocked queue, checking each
11006	 * entry against the I/Os prior to it on the OOA queue to see if
11007	 * there is still any blockage.
11008	 *
11009	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11010	 * with our removing a variable on it while it is traversing the
11011	 * list.
11012	 */
11013	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11014	     cur_blocked != NULL; cur_blocked = next_blocked) {
11015		union ctl_io *prev_ooa;
11016		ctl_action action;
11017
11018		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11019							  blocked_links);
11020
11021		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11022						      ctl_ooaq, ooa_links);
11023
11024		/*
11025		 * If cur_blocked happens to be the first item in the OOA
11026		 * queue now, prev_ooa will be NULL, and the action
11027		 * returned will just be CTL_ACTION_PASS.
11028		 */
11029		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11030
11031		switch (action) {
11032		case CTL_ACTION_BLOCK:
11033			/* Nothing to do here, still blocked */
11034			break;
11035		case CTL_ACTION_OVERLAP:
11036		case CTL_ACTION_OVERLAP_TAG:
11037			/*
11038			 * This shouldn't happen!  In theory we've already
11039			 * checked this command for overlap...
11040			 */
11041			break;
11042		case CTL_ACTION_PASS:
11043		case CTL_ACTION_SKIP: {
11044			const struct ctl_cmd_entry *entry;
11045			int isc_retval;
11046
11047			/*
11048			 * The skip case shouldn't happen, this transaction
11049			 * should have never made it onto the blocked queue.
11050			 */
11051			/*
11052			 * This I/O is no longer blocked, we can remove it
11053			 * from the blocked queue.  Since this is a TAILQ
11054			 * (doubly linked list), we can do O(1) removals
11055			 * from any place on the list.
11056			 */
11057			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11058				     blocked_links);
11059			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11060
11061			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11062				/*
11063				 * Need to send IO back to original side to
11064				 * run
11065				 */
11066				union ctl_ha_msg msg_info;
11067
11068				msg_info.hdr.original_sc =
11069					cur_blocked->io_hdr.original_sc;
11070				msg_info.hdr.serializing_sc = cur_blocked;
11071				msg_info.hdr.msg_type = CTL_MSG_R2R;
11072				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11073				     &msg_info, sizeof(msg_info), 0)) >
11074				     CTL_HA_STATUS_SUCCESS) {
11075					printf("CTL:Check Blocked error from "
11076					       "ctl_ha_msg_send %d\n",
11077					       isc_retval);
11078				}
11079				break;
11080			}
11081			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11082
11083			/*
11084			 * Check this I/O for LUN state changes that may
11085			 * have happened while this command was blocked.
11086			 * The LUN state may have been changed by a command
11087			 * ahead of us in the queue, so we need to re-check
11088			 * for any states that can be caused by SCSI
11089			 * commands.
11090			 */
11091			if (ctl_scsiio_lun_check(lun, entry,
11092						 &cur_blocked->scsiio) == 0) {
11093				cur_blocked->io_hdr.flags |=
11094				                      CTL_FLAG_IS_WAS_ON_RTR;
11095				ctl_enqueue_rtr(cur_blocked);
11096			} else
11097				ctl_done(cur_blocked);
11098			break;
11099		}
11100		default:
11101			/*
11102			 * This probably shouldn't happen -- we shouldn't
11103			 * get CTL_ACTION_ERROR, or anything else.
11104			 */
11105			break;
11106		}
11107	}
11108
11109	return (CTL_RETVAL_COMPLETE);
11110}
11111
11112/*
11113 * This routine (with one exception) checks LUN flags that can be set by
11114 * commands ahead of us in the OOA queue.  These flags have to be checked
11115 * when a command initially comes in, and when we pull a command off the
11116 * blocked queue and are preparing to execute it.  The reason we have to
11117 * check these flags for commands on the blocked queue is that the LUN
11118 * state may have been changed by a command ahead of us while we're on the
11119 * blocked queue.
11120 *
11121 * Ordering is somewhat important with these checks, so please pay
11122 * careful attention to the placement of any new checks.
11123 */
11124static int
11125ctl_scsiio_lun_check(struct ctl_lun *lun,
11126    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11127{
11128	struct ctl_softc *softc = lun->ctl_softc;
11129	int retval;
11130	uint32_t residx;
11131
11132	retval = 0;
11133
11134	mtx_assert(&lun->lun_lock, MA_OWNED);
11135
11136	/*
11137	 * If this shelf is a secondary shelf controller, we have to reject
11138	 * any media access commands.
11139	 */
11140	if ((softc->flags & CTL_FLAG_ACTIVE_SHELF) == 0 &&
11141	    (entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0) {
11142		ctl_set_lun_standby(ctsio);
11143		retval = 1;
11144		goto bailout;
11145	}
11146
11147	if (entry->pattern & CTL_LUN_PAT_WRITE) {
11148		if (lun->flags & CTL_LUN_READONLY) {
11149			ctl_set_sense(ctsio, /*current_error*/ 1,
11150			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11151			    /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
11152			retval = 1;
11153			goto bailout;
11154		}
11155		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11156		    .eca_and_aen & SCP_SWP) != 0) {
11157			ctl_set_sense(ctsio, /*current_error*/ 1,
11158			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11159			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11160			retval = 1;
11161			goto bailout;
11162		}
11163	}
11164
11165	/*
11166	 * Check for a reservation conflict.  If this command isn't allowed
11167	 * even on reserved LUNs, and if this initiator isn't the one who
11168	 * reserved us, reject the command with a reservation conflict.
11169	 */
11170	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11171	if ((lun->flags & CTL_LUN_RESERVED)
11172	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11173		if (lun->res_idx != residx) {
11174			ctl_set_reservation_conflict(ctsio);
11175			retval = 1;
11176			goto bailout;
11177		}
11178	}
11179
11180	if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11181	    (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11182		/* No reservation or command is allowed. */;
11183	} else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11184	    (lun->res_type == SPR_TYPE_WR_EX ||
11185	     lun->res_type == SPR_TYPE_WR_EX_RO ||
11186	     lun->res_type == SPR_TYPE_WR_EX_AR)) {
11187		/* The command is allowed for Write Exclusive resv. */;
11188	} else {
11189		/*
11190		 * if we aren't registered or it's a res holder type
11191		 * reservation and this isn't the res holder then set a
11192		 * conflict.
11193		 */
11194		if (ctl_get_prkey(lun, residx) == 0
11195		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11196			ctl_set_reservation_conflict(ctsio);
11197			retval = 1;
11198			goto bailout;
11199		}
11200
11201	}
11202
11203	if ((lun->flags & CTL_LUN_OFFLINE)
11204	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11205		ctl_set_lun_not_ready(ctsio);
11206		retval = 1;
11207		goto bailout;
11208	}
11209
11210	/*
11211	 * If the LUN is stopped, see if this particular command is allowed
11212	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11213	 */
11214	if ((lun->flags & CTL_LUN_STOPPED)
11215	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11216		/* "Logical unit not ready, initializing cmd. required" */
11217		ctl_set_lun_stopped(ctsio);
11218		retval = 1;
11219		goto bailout;
11220	}
11221
11222	if ((lun->flags & CTL_LUN_INOPERABLE)
11223	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11224		/* "Medium format corrupted" */
11225		ctl_set_medium_format_corrupted(ctsio);
11226		retval = 1;
11227		goto bailout;
11228	}
11229
11230bailout:
11231	return (retval);
11232
11233}
11234
11235static void
11236ctl_failover_io(union ctl_io *io, int have_lock)
11237{
11238	ctl_set_busy(&io->scsiio);
11239	ctl_done(io);
11240}
11241
11242#ifdef notyet
11243static void
11244ctl_failover(void)
11245{
11246	struct ctl_lun *lun;
11247	struct ctl_softc *softc;
11248	union ctl_io *next_io, *pending_io;
11249	union ctl_io *io;
11250	int lun_idx;
11251
11252	softc = control_softc;
11253
11254	mtx_lock(&softc->ctl_lock);
11255	/*
11256	 * Remove any cmds from the other SC from the rtr queue.  These
11257	 * will obviously only be for LUNs for which we're the primary.
11258	 * We can't send status or get/send data for these commands.
11259	 * Since they haven't been executed yet, we can just remove them.
11260	 * We'll either abort them or delete them below, depending on
11261	 * which HA mode we're in.
11262	 */
11263#ifdef notyet
11264	mtx_lock(&softc->queue_lock);
11265	for (io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue);
11266	     io != NULL; io = next_io) {
11267		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11268		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11269			STAILQ_REMOVE(&softc->rtr_queue, &io->io_hdr,
11270				      ctl_io_hdr, links);
11271	}
11272	mtx_unlock(&softc->queue_lock);
11273#endif
11274
11275	for (lun_idx=0; lun_idx < softc->num_luns; lun_idx++) {
11276		lun = softc->ctl_luns[lun_idx];
11277		if (lun==NULL)
11278			continue;
11279
11280		/*
11281		 * Processor LUNs are primary on both sides.
11282		 * XXX will this always be true?
11283		 */
11284		if (lun->be_lun->lun_type == T_PROCESSOR)
11285			continue;
11286
11287		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11288		 && (softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11289			printf("FAILOVER: primary lun %d\n", lun_idx);
11290		        /*
11291			 * Remove all commands from the other SC. First from the
11292			 * blocked queue then from the ooa queue. Once we have
11293			 * removed them. Call ctl_check_blocked to see if there
11294			 * is anything that can run.
11295			 */
11296			for (io = (union ctl_io *)TAILQ_FIRST(
11297			     &lun->blocked_queue); io != NULL; io = next_io) {
11298
11299		        	next_io = (union ctl_io *)TAILQ_NEXT(
11300				    &io->io_hdr, blocked_links);
11301
11302				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11303					TAILQ_REMOVE(&lun->blocked_queue,
11304						     &io->io_hdr,blocked_links);
11305					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11306					TAILQ_REMOVE(&lun->ooa_queue,
11307						     &io->io_hdr, ooa_links);
11308
11309					ctl_free_io(io);
11310				}
11311			}
11312
11313			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11314	     		     io != NULL; io = next_io) {
11315
11316		        	next_io = (union ctl_io *)TAILQ_NEXT(
11317				    &io->io_hdr, ooa_links);
11318
11319				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11320
11321					TAILQ_REMOVE(&lun->ooa_queue,
11322						&io->io_hdr,
11323					     	ooa_links);
11324
11325					ctl_free_io(io);
11326				}
11327			}
11328			ctl_check_blocked(lun);
11329		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11330			&& (softc->ha_mode == CTL_HA_MODE_XFER)) {
11331
11332			printf("FAILOVER: primary lun %d\n", lun_idx);
11333			/*
11334			 * Abort all commands from the other SC.  We can't
11335			 * send status back for them now.  These should get
11336			 * cleaned up when they are completed or come out
11337			 * for a datamove operation.
11338			 */
11339			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11340	     		     io != NULL; io = next_io) {
11341		        	next_io = (union ctl_io *)TAILQ_NEXT(
11342					&io->io_hdr, ooa_links);
11343
11344				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11345					io->io_hdr.flags |= CTL_FLAG_ABORT;
11346			}
11347		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11348			&& (softc->ha_mode == CTL_HA_MODE_XFER)) {
11349
11350			printf("FAILOVER: secondary lun %d\n", lun_idx);
11351
11352			lun->flags |= CTL_LUN_PRIMARY_SC;
11353
11354			/*
11355			 * We send all I/O that was sent to this controller
11356			 * and redirected to the other side back with
11357			 * busy status, and have the initiator retry it.
11358			 * Figuring out how much data has been transferred,
11359			 * etc. and picking up where we left off would be
11360			 * very tricky.
11361			 *
11362			 * XXX KDM need to remove I/O from the blocked
11363			 * queue as well!
11364			 */
11365			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11366			     &lun->ooa_queue); pending_io != NULL;
11367			     pending_io = next_io) {
11368
11369				next_io =  (union ctl_io *)TAILQ_NEXT(
11370					&pending_io->io_hdr, ooa_links);
11371
11372				pending_io->io_hdr.flags &=
11373					~CTL_FLAG_SENT_2OTHER_SC;
11374
11375				if (pending_io->io_hdr.flags &
11376				    CTL_FLAG_IO_ACTIVE) {
11377					pending_io->io_hdr.flags |=
11378						CTL_FLAG_FAILOVER;
11379				} else {
11380					ctl_set_busy(&pending_io->scsiio);
11381					ctl_done(pending_io);
11382				}
11383			}
11384
11385			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
11386		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11387			&& (softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11388			printf("FAILOVER: secondary lun %d\n", lun_idx);
11389			/*
11390			 * if the first io on the OOA is not on the RtR queue
11391			 * add it.
11392			 */
11393			lun->flags |= CTL_LUN_PRIMARY_SC;
11394
11395			pending_io = (union ctl_io *)TAILQ_FIRST(
11396			    &lun->ooa_queue);
11397			if (pending_io==NULL) {
11398				printf("Nothing on OOA queue\n");
11399				continue;
11400			}
11401
11402			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11403			if ((pending_io->io_hdr.flags &
11404			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11405				pending_io->io_hdr.flags |=
11406				    CTL_FLAG_IS_WAS_ON_RTR;
11407				ctl_enqueue_rtr(pending_io);
11408			}
11409#if 0
11410			else
11411			{
11412				printf("Tag 0x%04x is running\n",
11413				      pending_io->scsiio.tag_num);
11414			}
11415#endif
11416
11417			next_io = (union ctl_io *)TAILQ_NEXT(
11418			    &pending_io->io_hdr, ooa_links);
11419			for (pending_io=next_io; pending_io != NULL;
11420			     pending_io = next_io) {
11421				pending_io->io_hdr.flags &=
11422				    ~CTL_FLAG_SENT_2OTHER_SC;
11423				next_io = (union ctl_io *)TAILQ_NEXT(
11424					&pending_io->io_hdr, ooa_links);
11425				if (pending_io->io_hdr.flags &
11426				    CTL_FLAG_IS_WAS_ON_RTR) {
11427#if 0
11428				        printf("Tag 0x%04x is running\n",
11429				      		pending_io->scsiio.tag_num);
11430#endif
11431					continue;
11432				}
11433
11434				switch (ctl_check_ooa(lun, pending_io,
11435			            (union ctl_io *)TAILQ_PREV(
11436				    &pending_io->io_hdr, ctl_ooaq,
11437				    ooa_links))) {
11438
11439				case CTL_ACTION_BLOCK:
11440					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11441							  &pending_io->io_hdr,
11442							  blocked_links);
11443					pending_io->io_hdr.flags |=
11444					    CTL_FLAG_BLOCKED;
11445					break;
11446				case CTL_ACTION_PASS:
11447				case CTL_ACTION_SKIP:
11448					pending_io->io_hdr.flags |=
11449					    CTL_FLAG_IS_WAS_ON_RTR;
11450					ctl_enqueue_rtr(pending_io);
11451					break;
11452				case CTL_ACTION_OVERLAP:
11453					ctl_set_overlapped_cmd(
11454					    (struct ctl_scsiio *)pending_io);
11455					ctl_done(pending_io);
11456					break;
11457				case CTL_ACTION_OVERLAP_TAG:
11458					ctl_set_overlapped_tag(
11459					    (struct ctl_scsiio *)pending_io,
11460					    pending_io->scsiio.tag_num & 0xff);
11461					ctl_done(pending_io);
11462					break;
11463				case CTL_ACTION_ERROR:
11464				default:
11465					ctl_set_internal_failure(
11466						(struct ctl_scsiio *)pending_io,
11467						0,  // sks_valid
11468						0); //retry count
11469					ctl_done(pending_io);
11470					break;
11471				}
11472			}
11473
11474			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
11475		} else {
11476			panic("Unhandled HA mode failover, LUN flags = %#x, "
11477			      "ha_mode = #%x", lun->flags, softc->ha_mode);
11478		}
11479	}
11480	ctl_pause_rtr = 0;
11481	mtx_unlock(&softc->ctl_lock);
11482}
11483#endif
11484
11485static void
11486ctl_clear_ua(struct ctl_softc *ctl_softc, uint32_t initidx,
11487	     ctl_ua_type ua_type)
11488{
11489	struct ctl_lun *lun;
11490	ctl_ua_type *pu;
11491
11492	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
11493
11494	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
11495		mtx_lock(&lun->lun_lock);
11496		pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
11497		if (pu != NULL)
11498			pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua_type;
11499		mtx_unlock(&lun->lun_lock);
11500	}
11501}
11502
11503static int
11504ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11505{
11506	struct ctl_lun *lun;
11507	const struct ctl_cmd_entry *entry;
11508	uint32_t initidx, targ_lun;
11509	int retval;
11510
11511	retval = 0;
11512
11513	lun = NULL;
11514
11515	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11516	if ((targ_lun < CTL_MAX_LUNS)
11517	 && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11518		/*
11519		 * If the LUN is invalid, pretend that it doesn't exist.
11520		 * It will go away as soon as all pending I/O has been
11521		 * completed.
11522		 */
11523		mtx_lock(&lun->lun_lock);
11524		if (lun->flags & CTL_LUN_DISABLED) {
11525			mtx_unlock(&lun->lun_lock);
11526			lun = NULL;
11527			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11528			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11529		} else {
11530			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11531			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11532				lun->be_lun;
11533			if (lun->be_lun->lun_type == T_PROCESSOR) {
11534				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11535			}
11536
11537			/*
11538			 * Every I/O goes into the OOA queue for a
11539			 * particular LUN, and stays there until completion.
11540			 */
11541#ifdef CTL_TIME_IO
11542			if (TAILQ_EMPTY(&lun->ooa_queue)) {
11543				lun->idle_time += getsbinuptime() -
11544				    lun->last_busy;
11545			}
11546#endif
11547			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11548			    ooa_links);
11549		}
11550	} else {
11551		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11552		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11553	}
11554
11555	/* Get command entry and return error if it is unsuppotyed. */
11556	entry = ctl_validate_command(ctsio);
11557	if (entry == NULL) {
11558		if (lun)
11559			mtx_unlock(&lun->lun_lock);
11560		return (retval);
11561	}
11562
11563	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11564	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11565
11566	/*
11567	 * Check to see whether we can send this command to LUNs that don't
11568	 * exist.  This should pretty much only be the case for inquiry
11569	 * and request sense.  Further checks, below, really require having
11570	 * a LUN, so we can't really check the command anymore.  Just put
11571	 * it on the rtr queue.
11572	 */
11573	if (lun == NULL) {
11574		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11575			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11576			ctl_enqueue_rtr((union ctl_io *)ctsio);
11577			return (retval);
11578		}
11579
11580		ctl_set_unsupported_lun(ctsio);
11581		ctl_done((union ctl_io *)ctsio);
11582		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11583		return (retval);
11584	} else {
11585		/*
11586		 * Make sure we support this particular command on this LUN.
11587		 * e.g., we don't support writes to the control LUN.
11588		 */
11589		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11590			mtx_unlock(&lun->lun_lock);
11591			ctl_set_invalid_opcode(ctsio);
11592			ctl_done((union ctl_io *)ctsio);
11593			return (retval);
11594		}
11595	}
11596
11597	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11598
11599#ifdef CTL_WITH_CA
11600	/*
11601	 * If we've got a request sense, it'll clear the contingent
11602	 * allegiance condition.  Otherwise, if we have a CA condition for
11603	 * this initiator, clear it, because it sent down a command other
11604	 * than request sense.
11605	 */
11606	if ((ctsio->cdb[0] != REQUEST_SENSE)
11607	 && (ctl_is_set(lun->have_ca, initidx)))
11608		ctl_clear_mask(lun->have_ca, initidx);
11609#endif
11610
11611	/*
11612	 * If the command has this flag set, it handles its own unit
11613	 * attention reporting, we shouldn't do anything.  Otherwise we
11614	 * check for any pending unit attentions, and send them back to the
11615	 * initiator.  We only do this when a command initially comes in,
11616	 * not when we pull it off the blocked queue.
11617	 *
11618	 * According to SAM-3, section 5.3.2, the order that things get
11619	 * presented back to the host is basically unit attentions caused
11620	 * by some sort of reset event, busy status, reservation conflicts
11621	 * or task set full, and finally any other status.
11622	 *
11623	 * One issue here is that some of the unit attentions we report
11624	 * don't fall into the "reset" category (e.g. "reported luns data
11625	 * has changed").  So reporting it here, before the reservation
11626	 * check, may be technically wrong.  I guess the only thing to do
11627	 * would be to check for and report the reset events here, and then
11628	 * check for the other unit attention types after we check for a
11629	 * reservation conflict.
11630	 *
11631	 * XXX KDM need to fix this
11632	 */
11633	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11634		ctl_ua_type ua_type;
11635		scsi_sense_data_type sense_format;
11636
11637		if (lun->flags & CTL_LUN_SENSE_DESC)
11638			sense_format = SSD_TYPE_DESC;
11639		else
11640			sense_format = SSD_TYPE_FIXED;
11641
11642		ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11643		    sense_format);
11644		if (ua_type != CTL_UA_NONE) {
11645			mtx_unlock(&lun->lun_lock);
11646			ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11647			ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11648			ctsio->sense_len = SSD_FULL_SIZE;
11649			ctl_done((union ctl_io *)ctsio);
11650			return (retval);
11651		}
11652	}
11653
11654
11655	if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11656		mtx_unlock(&lun->lun_lock);
11657		ctl_done((union ctl_io *)ctsio);
11658		return (retval);
11659	}
11660
11661	/*
11662	 * XXX CHD this is where we want to send IO to other side if
11663	 * this LUN is secondary on this SC. We will need to make a copy
11664	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11665	 * the copy we send as FROM_OTHER.
11666	 * We also need to stuff the address of the original IO so we can
11667	 * find it easily. Something similar will need be done on the other
11668	 * side so when we are done we can find the copy.
11669	 */
11670	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11671		union ctl_ha_msg msg_info;
11672		int isc_retval;
11673
11674		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11675
11676		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11677		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11678#if 0
11679		printf("1. ctsio %p\n", ctsio);
11680#endif
11681		msg_info.hdr.serializing_sc = NULL;
11682		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11683		msg_info.scsi.tag_num = ctsio->tag_num;
11684		msg_info.scsi.tag_type = ctsio->tag_type;
11685		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11686
11687		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11688
11689		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11690		    (void *)&msg_info, sizeof(msg_info), 0)) >
11691		    CTL_HA_STATUS_SUCCESS) {
11692			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11693			       isc_retval);
11694			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11695		} else {
11696#if 0
11697			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11698#endif
11699		}
11700
11701		/*
11702		 * XXX KDM this I/O is off the incoming queue, but hasn't
11703		 * been inserted on any other queue.  We may need to come
11704		 * up with a holding queue while we wait for serialization
11705		 * so that we have an idea of what we're waiting for from
11706		 * the other side.
11707		 */
11708		mtx_unlock(&lun->lun_lock);
11709		return (retval);
11710	}
11711
11712	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11713			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11714			      ctl_ooaq, ooa_links))) {
11715	case CTL_ACTION_BLOCK:
11716		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11717		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11718				  blocked_links);
11719		mtx_unlock(&lun->lun_lock);
11720		return (retval);
11721	case CTL_ACTION_PASS:
11722	case CTL_ACTION_SKIP:
11723		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11724		mtx_unlock(&lun->lun_lock);
11725		ctl_enqueue_rtr((union ctl_io *)ctsio);
11726		break;
11727	case CTL_ACTION_OVERLAP:
11728		mtx_unlock(&lun->lun_lock);
11729		ctl_set_overlapped_cmd(ctsio);
11730		ctl_done((union ctl_io *)ctsio);
11731		break;
11732	case CTL_ACTION_OVERLAP_TAG:
11733		mtx_unlock(&lun->lun_lock);
11734		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11735		ctl_done((union ctl_io *)ctsio);
11736		break;
11737	case CTL_ACTION_ERROR:
11738	default:
11739		mtx_unlock(&lun->lun_lock);
11740		ctl_set_internal_failure(ctsio,
11741					 /*sks_valid*/ 0,
11742					 /*retry_count*/ 0);
11743		ctl_done((union ctl_io *)ctsio);
11744		break;
11745	}
11746	return (retval);
11747}
11748
11749const struct ctl_cmd_entry *
11750ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11751{
11752	const struct ctl_cmd_entry *entry;
11753	int service_action;
11754
11755	entry = &ctl_cmd_table[ctsio->cdb[0]];
11756	if (sa)
11757		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11758	if (entry->flags & CTL_CMD_FLAG_SA5) {
11759		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11760		entry = &((const struct ctl_cmd_entry *)
11761		    entry->execute)[service_action];
11762	}
11763	return (entry);
11764}
11765
11766const struct ctl_cmd_entry *
11767ctl_validate_command(struct ctl_scsiio *ctsio)
11768{
11769	const struct ctl_cmd_entry *entry;
11770	int i, sa;
11771	uint8_t diff;
11772
11773	entry = ctl_get_cmd_entry(ctsio, &sa);
11774	if (entry->execute == NULL) {
11775		if (sa)
11776			ctl_set_invalid_field(ctsio,
11777					      /*sks_valid*/ 1,
11778					      /*command*/ 1,
11779					      /*field*/ 1,
11780					      /*bit_valid*/ 1,
11781					      /*bit*/ 4);
11782		else
11783			ctl_set_invalid_opcode(ctsio);
11784		ctl_done((union ctl_io *)ctsio);
11785		return (NULL);
11786	}
11787	KASSERT(entry->length > 0,
11788	    ("Not defined length for command 0x%02x/0x%02x",
11789	     ctsio->cdb[0], ctsio->cdb[1]));
11790	for (i = 1; i < entry->length; i++) {
11791		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11792		if (diff == 0)
11793			continue;
11794		ctl_set_invalid_field(ctsio,
11795				      /*sks_valid*/ 1,
11796				      /*command*/ 1,
11797				      /*field*/ i,
11798				      /*bit_valid*/ 1,
11799				      /*bit*/ fls(diff) - 1);
11800		ctl_done((union ctl_io *)ctsio);
11801		return (NULL);
11802	}
11803	return (entry);
11804}
11805
11806static int
11807ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11808{
11809
11810	switch (lun_type) {
11811	case T_PROCESSOR:
11812		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11813		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11814			return (0);
11815		break;
11816	case T_DIRECT:
11817		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11818		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11819			return (0);
11820		break;
11821	default:
11822		return (0);
11823	}
11824	return (1);
11825}
11826
11827static int
11828ctl_scsiio(struct ctl_scsiio *ctsio)
11829{
11830	int retval;
11831	const struct ctl_cmd_entry *entry;
11832
11833	retval = CTL_RETVAL_COMPLETE;
11834
11835	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11836
11837	entry = ctl_get_cmd_entry(ctsio, NULL);
11838
11839	/*
11840	 * If this I/O has been aborted, just send it straight to
11841	 * ctl_done() without executing it.
11842	 */
11843	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11844		ctl_done((union ctl_io *)ctsio);
11845		goto bailout;
11846	}
11847
11848	/*
11849	 * All the checks should have been handled by ctl_scsiio_precheck().
11850	 * We should be clear now to just execute the I/O.
11851	 */
11852	retval = entry->execute(ctsio);
11853
11854bailout:
11855	return (retval);
11856}
11857
11858/*
11859 * Since we only implement one target right now, a bus reset simply resets
11860 * our single target.
11861 */
11862static int
11863ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11864{
11865	return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11866}
11867
11868static int
11869ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11870		 ctl_ua_type ua_type)
11871{
11872	struct ctl_lun *lun;
11873	int retval;
11874
11875	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11876		union ctl_ha_msg msg_info;
11877
11878		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11879		msg_info.hdr.nexus = io->io_hdr.nexus;
11880		if (ua_type==CTL_UA_TARG_RESET)
11881			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11882		else
11883			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11884		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11885		msg_info.hdr.original_sc = NULL;
11886		msg_info.hdr.serializing_sc = NULL;
11887		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11888		    (void *)&msg_info, sizeof(msg_info), 0)) {
11889		}
11890	}
11891	retval = 0;
11892
11893	mtx_lock(&softc->ctl_lock);
11894	STAILQ_FOREACH(lun, &softc->lun_list, links)
11895		retval += ctl_lun_reset(lun, io, ua_type);
11896	mtx_unlock(&softc->ctl_lock);
11897
11898	return (retval);
11899}
11900
11901/*
11902 * The LUN should always be set.  The I/O is optional, and is used to
11903 * distinguish between I/Os sent by this initiator, and by other
11904 * initiators.  We set unit attention for initiators other than this one.
11905 * SAM-3 is vague on this point.  It does say that a unit attention should
11906 * be established for other initiators when a LUN is reset (see section
11907 * 5.7.3), but it doesn't specifically say that the unit attention should
11908 * be established for this particular initiator when a LUN is reset.  Here
11909 * is the relevant text, from SAM-3 rev 8:
11910 *
11911 * 5.7.2 When a SCSI initiator port aborts its own tasks
11912 *
11913 * When a SCSI initiator port causes its own task(s) to be aborted, no
11914 * notification that the task(s) have been aborted shall be returned to
11915 * the SCSI initiator port other than the completion response for the
11916 * command or task management function action that caused the task(s) to
11917 * be aborted and notification(s) associated with related effects of the
11918 * action (e.g., a reset unit attention condition).
11919 *
11920 * XXX KDM for now, we're setting unit attention for all initiators.
11921 */
11922static int
11923ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11924{
11925	union ctl_io *xio;
11926#if 0
11927	uint32_t initidx;
11928#endif
11929#ifdef CTL_WITH_CA
11930	int i;
11931#endif
11932
11933	mtx_lock(&lun->lun_lock);
11934	/*
11935	 * Run through the OOA queue and abort each I/O.
11936	 */
11937	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11938	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11939		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11940	}
11941
11942	/*
11943	 * This version sets unit attention for every
11944	 */
11945#if 0
11946	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11947	ctl_est_ua_all(lun, initidx, ua_type);
11948#else
11949	ctl_est_ua_all(lun, -1, ua_type);
11950#endif
11951
11952	/*
11953	 * A reset (any kind, really) clears reservations established with
11954	 * RESERVE/RELEASE.  It does not clear reservations established
11955	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11956	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11957	 * reservations made with the RESERVE/RELEASE commands, because
11958	 * those commands are obsolete in SPC-3.
11959	 */
11960	lun->flags &= ~CTL_LUN_RESERVED;
11961
11962#ifdef CTL_WITH_CA
11963	for (i = 0; i < CTL_MAX_INITIATORS; i++)
11964		ctl_clear_mask(lun->have_ca, i);
11965#endif
11966	mtx_unlock(&lun->lun_lock);
11967
11968	return (0);
11969}
11970
11971static void
11972ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11973    int other_sc)
11974{
11975	union ctl_io *xio;
11976
11977	mtx_assert(&lun->lun_lock, MA_OWNED);
11978
11979	/*
11980	 * Run through the OOA queue and attempt to find the given I/O.
11981	 * The target port, initiator ID, tag type and tag number have to
11982	 * match the values that we got from the initiator.  If we have an
11983	 * untagged command to abort, simply abort the first untagged command
11984	 * we come to.  We only allow one untagged command at a time of course.
11985	 */
11986	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11987	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11988
11989		if ((targ_port == UINT32_MAX ||
11990		     targ_port == xio->io_hdr.nexus.targ_port) &&
11991		    (init_id == UINT32_MAX ||
11992		     init_id == xio->io_hdr.nexus.initid.id)) {
11993			if (targ_port != xio->io_hdr.nexus.targ_port ||
11994			    init_id != xio->io_hdr.nexus.initid.id)
11995				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11996			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11997			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11998				union ctl_ha_msg msg_info;
11999
12000				msg_info.hdr.nexus = xio->io_hdr.nexus;
12001				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12002				msg_info.task.tag_num = xio->scsiio.tag_num;
12003				msg_info.task.tag_type = xio->scsiio.tag_type;
12004				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12005				msg_info.hdr.original_sc = NULL;
12006				msg_info.hdr.serializing_sc = NULL;
12007				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12008				    (void *)&msg_info, sizeof(msg_info), 0);
12009			}
12010		}
12011	}
12012}
12013
12014static int
12015ctl_abort_task_set(union ctl_io *io)
12016{
12017	struct ctl_softc *softc = control_softc;
12018	struct ctl_lun *lun;
12019	uint32_t targ_lun;
12020
12021	/*
12022	 * Look up the LUN.
12023	 */
12024	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12025	mtx_lock(&softc->ctl_lock);
12026	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12027		lun = softc->ctl_luns[targ_lun];
12028	else {
12029		mtx_unlock(&softc->ctl_lock);
12030		return (1);
12031	}
12032
12033	mtx_lock(&lun->lun_lock);
12034	mtx_unlock(&softc->ctl_lock);
12035	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12036		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12037		    io->io_hdr.nexus.initid.id,
12038		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12039	} else { /* CTL_TASK_CLEAR_TASK_SET */
12040		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12041		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12042	}
12043	mtx_unlock(&lun->lun_lock);
12044	return (0);
12045}
12046
12047static int
12048ctl_i_t_nexus_reset(union ctl_io *io)
12049{
12050	struct ctl_softc *softc = control_softc;
12051	struct ctl_lun *lun;
12052	uint32_t initidx, residx;
12053
12054	initidx = ctl_get_initindex(&io->io_hdr.nexus);
12055	residx = ctl_get_resindex(&io->io_hdr.nexus);
12056	mtx_lock(&softc->ctl_lock);
12057	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12058		mtx_lock(&lun->lun_lock);
12059		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12060		    io->io_hdr.nexus.initid.id,
12061		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12062#ifdef CTL_WITH_CA
12063		ctl_clear_mask(lun->have_ca, initidx);
12064#endif
12065		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
12066			lun->flags &= ~CTL_LUN_RESERVED;
12067		ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
12068		mtx_unlock(&lun->lun_lock);
12069	}
12070	mtx_unlock(&softc->ctl_lock);
12071	return (0);
12072}
12073
12074static int
12075ctl_abort_task(union ctl_io *io)
12076{
12077	union ctl_io *xio;
12078	struct ctl_lun *lun;
12079	struct ctl_softc *softc;
12080#if 0
12081	struct sbuf sb;
12082	char printbuf[128];
12083#endif
12084	int found;
12085	uint32_t targ_lun;
12086
12087	softc = control_softc;
12088	found = 0;
12089
12090	/*
12091	 * Look up the LUN.
12092	 */
12093	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12094	mtx_lock(&softc->ctl_lock);
12095	if ((targ_lun < CTL_MAX_LUNS)
12096	 && (softc->ctl_luns[targ_lun] != NULL))
12097		lun = softc->ctl_luns[targ_lun];
12098	else {
12099		mtx_unlock(&softc->ctl_lock);
12100		return (1);
12101	}
12102
12103#if 0
12104	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12105	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12106#endif
12107
12108	mtx_lock(&lun->lun_lock);
12109	mtx_unlock(&softc->ctl_lock);
12110	/*
12111	 * Run through the OOA queue and attempt to find the given I/O.
12112	 * The target port, initiator ID, tag type and tag number have to
12113	 * match the values that we got from the initiator.  If we have an
12114	 * untagged command to abort, simply abort the first untagged command
12115	 * we come to.  We only allow one untagged command at a time of course.
12116	 */
12117	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12118	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12119#if 0
12120		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12121
12122		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12123			    lun->lun, xio->scsiio.tag_num,
12124			    xio->scsiio.tag_type,
12125			    (xio->io_hdr.blocked_links.tqe_prev
12126			    == NULL) ? "" : " BLOCKED",
12127			    (xio->io_hdr.flags &
12128			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12129			    (xio->io_hdr.flags &
12130			    CTL_FLAG_ABORT) ? " ABORT" : "",
12131			    (xio->io_hdr.flags &
12132			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12133		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12134		sbuf_finish(&sb);
12135		printf("%s\n", sbuf_data(&sb));
12136#endif
12137
12138		if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12139		 || (xio->io_hdr.nexus.initid.id != io->io_hdr.nexus.initid.id)
12140		 || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12141			continue;
12142
12143		/*
12144		 * If the abort says that the task is untagged, the
12145		 * task in the queue must be untagged.  Otherwise,
12146		 * we just check to see whether the tag numbers
12147		 * match.  This is because the QLogic firmware
12148		 * doesn't pass back the tag type in an abort
12149		 * request.
12150		 */
12151#if 0
12152		if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12153		  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12154		 || (xio->scsiio.tag_num == io->taskio.tag_num))
12155#endif
12156		/*
12157		 * XXX KDM we've got problems with FC, because it
12158		 * doesn't send down a tag type with aborts.  So we
12159		 * can only really go by the tag number...
12160		 * This may cause problems with parallel SCSI.
12161		 * Need to figure that out!!
12162		 */
12163		if (xio->scsiio.tag_num == io->taskio.tag_num) {
12164			xio->io_hdr.flags |= CTL_FLAG_ABORT;
12165			found = 1;
12166			if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12167			    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12168				union ctl_ha_msg msg_info;
12169
12170				io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
12171				msg_info.hdr.nexus = io->io_hdr.nexus;
12172				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12173				msg_info.task.tag_num = io->taskio.tag_num;
12174				msg_info.task.tag_type = io->taskio.tag_type;
12175				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12176				msg_info.hdr.original_sc = NULL;
12177				msg_info.hdr.serializing_sc = NULL;
12178#if 0
12179				printf("Sent Abort to other side\n");
12180#endif
12181				if (ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12182				    (void *)&msg_info, sizeof(msg_info), 0) !=
12183				    CTL_HA_STATUS_SUCCESS) {
12184				}
12185			}
12186#if 0
12187			printf("ctl_abort_task: found I/O to abort\n");
12188#endif
12189		}
12190	}
12191	mtx_unlock(&lun->lun_lock);
12192
12193	if (found == 0) {
12194		/*
12195		 * This isn't really an error.  It's entirely possible for
12196		 * the abort and command completion to cross on the wire.
12197		 * This is more of an informative/diagnostic error.
12198		 */
12199#if 0
12200		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12201		       "%d:%d:%d:%d tag %d type %d\n",
12202		       io->io_hdr.nexus.initid.id,
12203		       io->io_hdr.nexus.targ_port,
12204		       io->io_hdr.nexus.targ_target.id,
12205		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12206		       io->taskio.tag_type);
12207#endif
12208	}
12209	return (0);
12210}
12211
12212static void
12213ctl_run_task(union ctl_io *io)
12214{
12215	struct ctl_softc *softc = control_softc;
12216	int retval = 1;
12217	const char *task_desc;
12218
12219	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12220
12221	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12222	    ("ctl_run_task: Unextected io_type %d\n",
12223	     io->io_hdr.io_type));
12224
12225	task_desc = ctl_scsi_task_string(&io->taskio);
12226	if (task_desc != NULL) {
12227#ifdef NEEDTOPORT
12228		csevent_log(CSC_CTL | CSC_SHELF_SW |
12229			    CTL_TASK_REPORT,
12230			    csevent_LogType_Trace,
12231			    csevent_Severity_Information,
12232			    csevent_AlertLevel_Green,
12233			    csevent_FRU_Firmware,
12234			    csevent_FRU_Unknown,
12235			    "CTL: received task: %s",task_desc);
12236#endif
12237	} else {
12238#ifdef NEEDTOPORT
12239		csevent_log(CSC_CTL | CSC_SHELF_SW |
12240			    CTL_TASK_REPORT,
12241			    csevent_LogType_Trace,
12242			    csevent_Severity_Information,
12243			    csevent_AlertLevel_Green,
12244			    csevent_FRU_Firmware,
12245			    csevent_FRU_Unknown,
12246			    "CTL: received unknown task "
12247			    "type: %d (%#x)",
12248			    io->taskio.task_action,
12249			    io->taskio.task_action);
12250#endif
12251	}
12252	switch (io->taskio.task_action) {
12253	case CTL_TASK_ABORT_TASK:
12254		retval = ctl_abort_task(io);
12255		break;
12256	case CTL_TASK_ABORT_TASK_SET:
12257	case CTL_TASK_CLEAR_TASK_SET:
12258		retval = ctl_abort_task_set(io);
12259		break;
12260	case CTL_TASK_CLEAR_ACA:
12261		break;
12262	case CTL_TASK_I_T_NEXUS_RESET:
12263		retval = ctl_i_t_nexus_reset(io);
12264		break;
12265	case CTL_TASK_LUN_RESET: {
12266		struct ctl_lun *lun;
12267		uint32_t targ_lun;
12268
12269		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12270		mtx_lock(&softc->ctl_lock);
12271		if ((targ_lun < CTL_MAX_LUNS)
12272		 && (softc->ctl_luns[targ_lun] != NULL))
12273			lun = softc->ctl_luns[targ_lun];
12274		else {
12275			mtx_unlock(&softc->ctl_lock);
12276			retval = 1;
12277			break;
12278		}
12279
12280		if (!(io->io_hdr.flags &
12281		    CTL_FLAG_FROM_OTHER_SC)) {
12282			union ctl_ha_msg msg_info;
12283
12284			io->io_hdr.flags |=
12285				CTL_FLAG_SENT_2OTHER_SC;
12286			msg_info.hdr.msg_type =
12287				CTL_MSG_MANAGE_TASKS;
12288			msg_info.hdr.nexus = io->io_hdr.nexus;
12289			msg_info.task.task_action =
12290				CTL_TASK_LUN_RESET;
12291			msg_info.hdr.original_sc = NULL;
12292			msg_info.hdr.serializing_sc = NULL;
12293			if (CTL_HA_STATUS_SUCCESS !=
12294			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12295			    (void *)&msg_info,
12296			    sizeof(msg_info), 0)) {
12297			}
12298		}
12299
12300		retval = ctl_lun_reset(lun, io,
12301				       CTL_UA_LUN_RESET);
12302		mtx_unlock(&softc->ctl_lock);
12303		break;
12304	}
12305	case CTL_TASK_TARGET_RESET:
12306		retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
12307		break;
12308	case CTL_TASK_BUS_RESET:
12309		retval = ctl_bus_reset(softc, io);
12310		break;
12311	case CTL_TASK_PORT_LOGIN:
12312		break;
12313	case CTL_TASK_PORT_LOGOUT:
12314		break;
12315	default:
12316		printf("ctl_run_task: got unknown task management event %d\n",
12317		       io->taskio.task_action);
12318		break;
12319	}
12320	if (retval == 0)
12321		io->io_hdr.status = CTL_SUCCESS;
12322	else
12323		io->io_hdr.status = CTL_ERROR;
12324	ctl_done(io);
12325}
12326
12327/*
12328 * For HA operation.  Handle commands that come in from the other
12329 * controller.
12330 */
12331static void
12332ctl_handle_isc(union ctl_io *io)
12333{
12334	int free_io;
12335	struct ctl_lun *lun;
12336	struct ctl_softc *softc;
12337	uint32_t targ_lun;
12338
12339	softc = control_softc;
12340
12341	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12342	lun = softc->ctl_luns[targ_lun];
12343
12344	switch (io->io_hdr.msg_type) {
12345	case CTL_MSG_SERIALIZE:
12346		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12347		break;
12348	case CTL_MSG_R2R: {
12349		const struct ctl_cmd_entry *entry;
12350
12351		/*
12352		 * This is only used in SER_ONLY mode.
12353		 */
12354		free_io = 0;
12355		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12356		mtx_lock(&lun->lun_lock);
12357		if (ctl_scsiio_lun_check(lun,
12358		    entry, (struct ctl_scsiio *)io) != 0) {
12359			mtx_unlock(&lun->lun_lock);
12360			ctl_done(io);
12361			break;
12362		}
12363		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12364		mtx_unlock(&lun->lun_lock);
12365		ctl_enqueue_rtr(io);
12366		break;
12367	}
12368	case CTL_MSG_FINISH_IO:
12369		if (softc->ha_mode == CTL_HA_MODE_XFER) {
12370			free_io = 0;
12371			ctl_done(io);
12372		} else {
12373			free_io = 1;
12374			mtx_lock(&lun->lun_lock);
12375			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12376				     ooa_links);
12377			ctl_check_blocked(lun);
12378			mtx_unlock(&lun->lun_lock);
12379		}
12380		break;
12381	case CTL_MSG_PERS_ACTION:
12382		ctl_hndl_per_res_out_on_other_sc(
12383			(union ctl_ha_msg *)&io->presio.pr_msg);
12384		free_io = 1;
12385		break;
12386	case CTL_MSG_BAD_JUJU:
12387		free_io = 0;
12388		ctl_done(io);
12389		break;
12390	case CTL_MSG_DATAMOVE:
12391		/* Only used in XFER mode */
12392		free_io = 0;
12393		ctl_datamove_remote(io);
12394		break;
12395	case CTL_MSG_DATAMOVE_DONE:
12396		/* Only used in XFER mode */
12397		free_io = 0;
12398		io->scsiio.be_move_done(io);
12399		break;
12400	default:
12401		free_io = 1;
12402		printf("%s: Invalid message type %d\n",
12403		       __func__, io->io_hdr.msg_type);
12404		break;
12405	}
12406	if (free_io)
12407		ctl_free_io(io);
12408
12409}
12410
12411
12412/*
12413 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12414 * there is no match.
12415 */
12416static ctl_lun_error_pattern
12417ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12418{
12419	const struct ctl_cmd_entry *entry;
12420	ctl_lun_error_pattern filtered_pattern, pattern;
12421
12422	pattern = desc->error_pattern;
12423
12424	/*
12425	 * XXX KDM we need more data passed into this function to match a
12426	 * custom pattern, and we actually need to implement custom pattern
12427	 * matching.
12428	 */
12429	if (pattern & CTL_LUN_PAT_CMD)
12430		return (CTL_LUN_PAT_CMD);
12431
12432	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12433		return (CTL_LUN_PAT_ANY);
12434
12435	entry = ctl_get_cmd_entry(ctsio, NULL);
12436
12437	filtered_pattern = entry->pattern & pattern;
12438
12439	/*
12440	 * If the user requested specific flags in the pattern (e.g.
12441	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12442	 * flags.
12443	 *
12444	 * If the user did not specify any flags, it doesn't matter whether
12445	 * or not the command supports the flags.
12446	 */
12447	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12448	     (pattern & ~CTL_LUN_PAT_MASK))
12449		return (CTL_LUN_PAT_NONE);
12450
12451	/*
12452	 * If the user asked for a range check, see if the requested LBA
12453	 * range overlaps with this command's LBA range.
12454	 */
12455	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12456		uint64_t lba1;
12457		uint64_t len1;
12458		ctl_action action;
12459		int retval;
12460
12461		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12462		if (retval != 0)
12463			return (CTL_LUN_PAT_NONE);
12464
12465		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12466					      desc->lba_range.len, FALSE);
12467		/*
12468		 * A "pass" means that the LBA ranges don't overlap, so
12469		 * this doesn't match the user's range criteria.
12470		 */
12471		if (action == CTL_ACTION_PASS)
12472			return (CTL_LUN_PAT_NONE);
12473	}
12474
12475	return (filtered_pattern);
12476}
12477
12478static void
12479ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12480{
12481	struct ctl_error_desc *desc, *desc2;
12482
12483	mtx_assert(&lun->lun_lock, MA_OWNED);
12484
12485	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12486		ctl_lun_error_pattern pattern;
12487		/*
12488		 * Check to see whether this particular command matches
12489		 * the pattern in the descriptor.
12490		 */
12491		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12492		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12493			continue;
12494
12495		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12496		case CTL_LUN_INJ_ABORTED:
12497			ctl_set_aborted(&io->scsiio);
12498			break;
12499		case CTL_LUN_INJ_MEDIUM_ERR:
12500			ctl_set_medium_error(&io->scsiio);
12501			break;
12502		case CTL_LUN_INJ_UA:
12503			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12504			 * OCCURRED */
12505			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12506			break;
12507		case CTL_LUN_INJ_CUSTOM:
12508			/*
12509			 * We're assuming the user knows what he is doing.
12510			 * Just copy the sense information without doing
12511			 * checks.
12512			 */
12513			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12514			      MIN(sizeof(desc->custom_sense),
12515				  sizeof(io->scsiio.sense_data)));
12516			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12517			io->scsiio.sense_len = SSD_FULL_SIZE;
12518			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12519			break;
12520		case CTL_LUN_INJ_NONE:
12521		default:
12522			/*
12523			 * If this is an error injection type we don't know
12524			 * about, clear the continuous flag (if it is set)
12525			 * so it will get deleted below.
12526			 */
12527			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12528			break;
12529		}
12530		/*
12531		 * By default, each error injection action is a one-shot
12532		 */
12533		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12534			continue;
12535
12536		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12537
12538		free(desc, M_CTL);
12539	}
12540}
12541
12542#ifdef CTL_IO_DELAY
12543static void
12544ctl_datamove_timer_wakeup(void *arg)
12545{
12546	union ctl_io *io;
12547
12548	io = (union ctl_io *)arg;
12549
12550	ctl_datamove(io);
12551}
12552#endif /* CTL_IO_DELAY */
12553
12554void
12555ctl_datamove(union ctl_io *io)
12556{
12557	void (*fe_datamove)(union ctl_io *io);
12558
12559	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12560
12561	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12562
12563#ifdef CTL_TIME_IO
12564	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12565		char str[256];
12566		char path_str[64];
12567		struct sbuf sb;
12568
12569		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12570		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12571
12572		sbuf_cat(&sb, path_str);
12573		switch (io->io_hdr.io_type) {
12574		case CTL_IO_SCSI:
12575			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12576			sbuf_printf(&sb, "\n");
12577			sbuf_cat(&sb, path_str);
12578			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12579				    io->scsiio.tag_num, io->scsiio.tag_type);
12580			break;
12581		case CTL_IO_TASK:
12582			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12583				    "Tag Type: %d\n", io->taskio.task_action,
12584				    io->taskio.tag_num, io->taskio.tag_type);
12585			break;
12586		default:
12587			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12588			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12589			break;
12590		}
12591		sbuf_cat(&sb, path_str);
12592		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12593			    (intmax_t)time_uptime - io->io_hdr.start_time);
12594		sbuf_finish(&sb);
12595		printf("%s", sbuf_data(&sb));
12596	}
12597#endif /* CTL_TIME_IO */
12598
12599#ifdef CTL_IO_DELAY
12600	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12601		struct ctl_lun *lun;
12602
12603		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12604
12605		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12606	} else {
12607		struct ctl_lun *lun;
12608
12609		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12610		if ((lun != NULL)
12611		 && (lun->delay_info.datamove_delay > 0)) {
12612			struct callout *callout;
12613
12614			callout = (struct callout *)&io->io_hdr.timer_bytes;
12615			callout_init(callout, /*mpsafe*/ 1);
12616			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12617			callout_reset(callout,
12618				      lun->delay_info.datamove_delay * hz,
12619				      ctl_datamove_timer_wakeup, io);
12620			if (lun->delay_info.datamove_type ==
12621			    CTL_DELAY_TYPE_ONESHOT)
12622				lun->delay_info.datamove_delay = 0;
12623			return;
12624		}
12625	}
12626#endif
12627
12628	/*
12629	 * This command has been aborted.  Set the port status, so we fail
12630	 * the data move.
12631	 */
12632	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12633		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12634		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12635		       io->io_hdr.nexus.targ_port,
12636		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12637		       io->io_hdr.nexus.targ_lun);
12638		io->io_hdr.port_status = 31337;
12639		/*
12640		 * Note that the backend, in this case, will get the
12641		 * callback in its context.  In other cases it may get
12642		 * called in the frontend's interrupt thread context.
12643		 */
12644		io->scsiio.be_move_done(io);
12645		return;
12646	}
12647
12648	/* Don't confuse frontend with zero length data move. */
12649	if (io->scsiio.kern_data_len == 0) {
12650		io->scsiio.be_move_done(io);
12651		return;
12652	}
12653
12654	/*
12655	 * If we're in XFER mode and this I/O is from the other shelf
12656	 * controller, we need to send the DMA to the other side to
12657	 * actually transfer the data to/from the host.  In serialize only
12658	 * mode the transfer happens below CTL and ctl_datamove() is only
12659	 * called on the machine that originally received the I/O.
12660	 */
12661	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12662	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12663		union ctl_ha_msg msg;
12664		uint32_t sg_entries_sent;
12665		int do_sg_copy;
12666		int i;
12667
12668		memset(&msg, 0, sizeof(msg));
12669		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12670		msg.hdr.original_sc = io->io_hdr.original_sc;
12671		msg.hdr.serializing_sc = io;
12672		msg.hdr.nexus = io->io_hdr.nexus;
12673		msg.dt.flags = io->io_hdr.flags;
12674		/*
12675		 * We convert everything into a S/G list here.  We can't
12676		 * pass by reference, only by value between controllers.
12677		 * So we can't pass a pointer to the S/G list, only as many
12678		 * S/G entries as we can fit in here.  If it's possible for
12679		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12680		 * then we need to break this up into multiple transfers.
12681		 */
12682		if (io->scsiio.kern_sg_entries == 0) {
12683			msg.dt.kern_sg_entries = 1;
12684			/*
12685			 * If this is in cached memory, flush the cache
12686			 * before we send the DMA request to the other
12687			 * controller.  We want to do this in either the
12688			 * read or the write case.  The read case is
12689			 * straightforward.  In the write case, we want to
12690			 * make sure nothing is in the local cache that
12691			 * could overwrite the DMAed data.
12692			 */
12693			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12694				/*
12695				 * XXX KDM use bus_dmamap_sync() here.
12696				 */
12697			}
12698
12699			/*
12700			 * Convert to a physical address if this is a
12701			 * virtual address.
12702			 */
12703			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12704				msg.dt.sg_list[0].addr =
12705					io->scsiio.kern_data_ptr;
12706			} else {
12707				/*
12708				 * XXX KDM use busdma here!
12709				 */
12710#if 0
12711				msg.dt.sg_list[0].addr = (void *)
12712					vtophys(io->scsiio.kern_data_ptr);
12713#endif
12714			}
12715
12716			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12717			do_sg_copy = 0;
12718		} else {
12719			struct ctl_sg_entry *sgl;
12720
12721			do_sg_copy = 1;
12722			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12723			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12724			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12725				/*
12726				 * XXX KDM use bus_dmamap_sync() here.
12727				 */
12728			}
12729		}
12730
12731		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12732		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12733		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12734		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12735		msg.dt.sg_sequence = 0;
12736
12737		/*
12738		 * Loop until we've sent all of the S/G entries.  On the
12739		 * other end, we'll recompose these S/G entries into one
12740		 * contiguous list before passing it to the
12741		 */
12742		for (sg_entries_sent = 0; sg_entries_sent <
12743		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12744			msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list)/
12745				sizeof(msg.dt.sg_list[0])),
12746				msg.dt.kern_sg_entries - sg_entries_sent);
12747
12748			if (do_sg_copy != 0) {
12749				struct ctl_sg_entry *sgl;
12750				int j;
12751
12752				sgl = (struct ctl_sg_entry *)
12753					io->scsiio.kern_data_ptr;
12754				/*
12755				 * If this is in cached memory, flush the cache
12756				 * before we send the DMA request to the other
12757				 * controller.  We want to do this in either
12758				 * the * read or the write case.  The read
12759				 * case is straightforward.  In the write
12760				 * case, we want to make sure nothing is
12761				 * in the local cache that could overwrite
12762				 * the DMAed data.
12763				 */
12764
12765				for (i = sg_entries_sent, j = 0;
12766				     i < msg.dt.cur_sg_entries; i++, j++) {
12767					if ((io->io_hdr.flags &
12768					     CTL_FLAG_NO_DATASYNC) == 0) {
12769						/*
12770						 * XXX KDM use bus_dmamap_sync()
12771						 */
12772					}
12773					if ((io->io_hdr.flags &
12774					     CTL_FLAG_BUS_ADDR) == 0) {
12775						/*
12776						 * XXX KDM use busdma.
12777						 */
12778#if 0
12779						msg.dt.sg_list[j].addr =(void *)
12780						       vtophys(sgl[i].addr);
12781#endif
12782					} else {
12783						msg.dt.sg_list[j].addr =
12784							sgl[i].addr;
12785					}
12786					msg.dt.sg_list[j].len = sgl[i].len;
12787				}
12788			}
12789
12790			sg_entries_sent += msg.dt.cur_sg_entries;
12791			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12792				msg.dt.sg_last = 1;
12793			else
12794				msg.dt.sg_last = 0;
12795
12796			/*
12797			 * XXX KDM drop and reacquire the lock here?
12798			 */
12799			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12800			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12801				/*
12802				 * XXX do something here.
12803				 */
12804			}
12805
12806			msg.dt.sent_sg_entries = sg_entries_sent;
12807		}
12808		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12809		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12810			ctl_failover_io(io, /*have_lock*/ 0);
12811
12812	} else {
12813
12814		/*
12815		 * Lookup the fe_datamove() function for this particular
12816		 * front end.
12817		 */
12818		fe_datamove =
12819		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12820
12821		fe_datamove(io);
12822	}
12823}
12824
12825static void
12826ctl_send_datamove_done(union ctl_io *io, int have_lock)
12827{
12828	union ctl_ha_msg msg;
12829	int isc_status;
12830
12831	memset(&msg, 0, sizeof(msg));
12832
12833	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12834	msg.hdr.original_sc = io;
12835	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12836	msg.hdr.nexus = io->io_hdr.nexus;
12837	msg.hdr.status = io->io_hdr.status;
12838	msg.scsi.tag_num = io->scsiio.tag_num;
12839	msg.scsi.tag_type = io->scsiio.tag_type;
12840	msg.scsi.scsi_status = io->scsiio.scsi_status;
12841	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12842	       sizeof(io->scsiio.sense_data));
12843	msg.scsi.sense_len = io->scsiio.sense_len;
12844	msg.scsi.sense_residual = io->scsiio.sense_residual;
12845	msg.scsi.fetd_status = io->io_hdr.port_status;
12846	msg.scsi.residual = io->scsiio.residual;
12847	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12848
12849	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12850		ctl_failover_io(io, /*have_lock*/ have_lock);
12851		return;
12852	}
12853
12854	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12855	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12856		/* XXX do something if this fails */
12857	}
12858
12859}
12860
12861/*
12862 * The DMA to the remote side is done, now we need to tell the other side
12863 * we're done so it can continue with its data movement.
12864 */
12865static void
12866ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12867{
12868	union ctl_io *io;
12869
12870	io = rq->context;
12871
12872	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12873		printf("%s: ISC DMA write failed with error %d", __func__,
12874		       rq->ret);
12875		ctl_set_internal_failure(&io->scsiio,
12876					 /*sks_valid*/ 1,
12877					 /*retry_count*/ rq->ret);
12878	}
12879
12880	ctl_dt_req_free(rq);
12881
12882	/*
12883	 * In this case, we had to malloc the memory locally.  Free it.
12884	 */
12885	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12886		int i;
12887		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12888			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12889	}
12890	/*
12891	 * The data is in local and remote memory, so now we need to send
12892	 * status (good or back) back to the other side.
12893	 */
12894	ctl_send_datamove_done(io, /*have_lock*/ 0);
12895}
12896
12897/*
12898 * We've moved the data from the host/controller into local memory.  Now we
12899 * need to push it over to the remote controller's memory.
12900 */
12901static int
12902ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12903{
12904	int retval;
12905
12906	retval = 0;
12907
12908	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12909					  ctl_datamove_remote_write_cb);
12910
12911	return (retval);
12912}
12913
12914static void
12915ctl_datamove_remote_write(union ctl_io *io)
12916{
12917	int retval;
12918	void (*fe_datamove)(union ctl_io *io);
12919
12920	/*
12921	 * - Get the data from the host/HBA into local memory.
12922	 * - DMA memory from the local controller to the remote controller.
12923	 * - Send status back to the remote controller.
12924	 */
12925
12926	retval = ctl_datamove_remote_sgl_setup(io);
12927	if (retval != 0)
12928		return;
12929
12930	/* Switch the pointer over so the FETD knows what to do */
12931	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12932
12933	/*
12934	 * Use a custom move done callback, since we need to send completion
12935	 * back to the other controller, not to the backend on this side.
12936	 */
12937	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12938
12939	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12940
12941	fe_datamove(io);
12942
12943	return;
12944
12945}
12946
12947static int
12948ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12949{
12950#if 0
12951	char str[256];
12952	char path_str[64];
12953	struct sbuf sb;
12954#endif
12955
12956	/*
12957	 * In this case, we had to malloc the memory locally.  Free it.
12958	 */
12959	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12960		int i;
12961		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12962			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12963	}
12964
12965#if 0
12966	scsi_path_string(io, path_str, sizeof(path_str));
12967	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12968	sbuf_cat(&sb, path_str);
12969	scsi_command_string(&io->scsiio, NULL, &sb);
12970	sbuf_printf(&sb, "\n");
12971	sbuf_cat(&sb, path_str);
12972	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12973		    io->scsiio.tag_num, io->scsiio.tag_type);
12974	sbuf_cat(&sb, path_str);
12975	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12976		    io->io_hdr.flags, io->io_hdr.status);
12977	sbuf_finish(&sb);
12978	printk("%s", sbuf_data(&sb));
12979#endif
12980
12981
12982	/*
12983	 * The read is done, now we need to send status (good or bad) back
12984	 * to the other side.
12985	 */
12986	ctl_send_datamove_done(io, /*have_lock*/ 0);
12987
12988	return (0);
12989}
12990
12991static void
12992ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12993{
12994	union ctl_io *io;
12995	void (*fe_datamove)(union ctl_io *io);
12996
12997	io = rq->context;
12998
12999	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13000		printf("%s: ISC DMA read failed with error %d", __func__,
13001		       rq->ret);
13002		ctl_set_internal_failure(&io->scsiio,
13003					 /*sks_valid*/ 1,
13004					 /*retry_count*/ rq->ret);
13005	}
13006
13007	ctl_dt_req_free(rq);
13008
13009	/* Switch the pointer over so the FETD knows what to do */
13010	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13011
13012	/*
13013	 * Use a custom move done callback, since we need to send completion
13014	 * back to the other controller, not to the backend on this side.
13015	 */
13016	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13017
13018	/* XXX KDM add checks like the ones in ctl_datamove? */
13019
13020	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13021
13022	fe_datamove(io);
13023}
13024
13025static int
13026ctl_datamove_remote_sgl_setup(union ctl_io *io)
13027{
13028	struct ctl_sg_entry *local_sglist, *remote_sglist;
13029	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13030	struct ctl_softc *softc;
13031	int retval;
13032	int i;
13033
13034	retval = 0;
13035	softc = control_softc;
13036
13037	local_sglist = io->io_hdr.local_sglist;
13038	local_dma_sglist = io->io_hdr.local_dma_sglist;
13039	remote_sglist = io->io_hdr.remote_sglist;
13040	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13041
13042	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13043		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13044			local_sglist[i].len = remote_sglist[i].len;
13045
13046			/*
13047			 * XXX Detect the situation where the RS-level I/O
13048			 * redirector on the other side has already read the
13049			 * data off of the AOR RS on this side, and
13050			 * transferred it to remote (mirror) memory on the
13051			 * other side.  Since we already have the data in
13052			 * memory here, we just need to use it.
13053			 *
13054			 * XXX KDM this can probably be removed once we
13055			 * get the cache device code in and take the
13056			 * current AOR implementation out.
13057			 */
13058#ifdef NEEDTOPORT
13059			if ((remote_sglist[i].addr >=
13060			     (void *)vtophys(softc->mirr->addr))
13061			 && (remote_sglist[i].addr <
13062			     ((void *)vtophys(softc->mirr->addr) +
13063			     CacheMirrorOffset))) {
13064				local_sglist[i].addr = remote_sglist[i].addr -
13065					CacheMirrorOffset;
13066				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13067				     CTL_FLAG_DATA_IN)
13068					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13069			} else {
13070				local_sglist[i].addr = remote_sglist[i].addr +
13071					CacheMirrorOffset;
13072			}
13073#endif
13074#if 0
13075			printf("%s: local %p, remote %p, len %d\n",
13076			       __func__, local_sglist[i].addr,
13077			       remote_sglist[i].addr, local_sglist[i].len);
13078#endif
13079		}
13080	} else {
13081		uint32_t len_to_go;
13082
13083		/*
13084		 * In this case, we don't have automatically allocated
13085		 * memory for this I/O on this controller.  This typically
13086		 * happens with internal CTL I/O -- e.g. inquiry, mode
13087		 * sense, etc.  Anything coming from RAIDCore will have
13088		 * a mirror area available.
13089		 */
13090		len_to_go = io->scsiio.kern_data_len;
13091
13092		/*
13093		 * Clear the no datasync flag, we have to use malloced
13094		 * buffers.
13095		 */
13096		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13097
13098		/*
13099		 * The difficult thing here is that the size of the various
13100		 * S/G segments may be different than the size from the
13101		 * remote controller.  That'll make it harder when DMAing
13102		 * the data back to the other side.
13103		 */
13104		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13105		     sizeof(io->io_hdr.remote_sglist[0])) &&
13106		     (len_to_go > 0); i++) {
13107			local_sglist[i].len = MIN(len_to_go, 131072);
13108			CTL_SIZE_8B(local_dma_sglist[i].len,
13109				    local_sglist[i].len);
13110			local_sglist[i].addr =
13111				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13112
13113			local_dma_sglist[i].addr = local_sglist[i].addr;
13114
13115			if (local_sglist[i].addr == NULL) {
13116				int j;
13117
13118				printf("malloc failed for %zd bytes!",
13119				       local_dma_sglist[i].len);
13120				for (j = 0; j < i; j++) {
13121					free(local_sglist[j].addr, M_CTL);
13122				}
13123				ctl_set_internal_failure(&io->scsiio,
13124							 /*sks_valid*/ 1,
13125							 /*retry_count*/ 4857);
13126				retval = 1;
13127				goto bailout_error;
13128
13129			}
13130			/* XXX KDM do we need a sync here? */
13131
13132			len_to_go -= local_sglist[i].len;
13133		}
13134		/*
13135		 * Reset the number of S/G entries accordingly.  The
13136		 * original number of S/G entries is available in
13137		 * rem_sg_entries.
13138		 */
13139		io->scsiio.kern_sg_entries = i;
13140
13141#if 0
13142		printf("%s: kern_sg_entries = %d\n", __func__,
13143		       io->scsiio.kern_sg_entries);
13144		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13145			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13146			       local_sglist[i].addr, local_sglist[i].len,
13147			       local_dma_sglist[i].len);
13148#endif
13149	}
13150
13151
13152	return (retval);
13153
13154bailout_error:
13155
13156	ctl_send_datamove_done(io, /*have_lock*/ 0);
13157
13158	return (retval);
13159}
13160
13161static int
13162ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13163			 ctl_ha_dt_cb callback)
13164{
13165	struct ctl_ha_dt_req *rq;
13166	struct ctl_sg_entry *remote_sglist, *local_sglist;
13167	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13168	uint32_t local_used, remote_used, total_used;
13169	int retval;
13170	int i, j;
13171
13172	retval = 0;
13173
13174	rq = ctl_dt_req_alloc();
13175
13176	/*
13177	 * If we failed to allocate the request, and if the DMA didn't fail
13178	 * anyway, set busy status.  This is just a resource allocation
13179	 * failure.
13180	 */
13181	if ((rq == NULL)
13182	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13183		ctl_set_busy(&io->scsiio);
13184
13185	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13186
13187		if (rq != NULL)
13188			ctl_dt_req_free(rq);
13189
13190		/*
13191		 * The data move failed.  We need to return status back
13192		 * to the other controller.  No point in trying to DMA
13193		 * data to the remote controller.
13194		 */
13195
13196		ctl_send_datamove_done(io, /*have_lock*/ 0);
13197
13198		retval = 1;
13199
13200		goto bailout;
13201	}
13202
13203	local_sglist = io->io_hdr.local_sglist;
13204	local_dma_sglist = io->io_hdr.local_dma_sglist;
13205	remote_sglist = io->io_hdr.remote_sglist;
13206	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13207	local_used = 0;
13208	remote_used = 0;
13209	total_used = 0;
13210
13211	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13212		rq->ret = CTL_HA_STATUS_SUCCESS;
13213		rq->context = io;
13214		callback(rq);
13215		goto bailout;
13216	}
13217
13218	/*
13219	 * Pull/push the data over the wire from/to the other controller.
13220	 * This takes into account the possibility that the local and
13221	 * remote sglists may not be identical in terms of the size of
13222	 * the elements and the number of elements.
13223	 *
13224	 * One fundamental assumption here is that the length allocated for
13225	 * both the local and remote sglists is identical.  Otherwise, we've
13226	 * essentially got a coding error of some sort.
13227	 */
13228	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13229		int isc_ret;
13230		uint32_t cur_len, dma_length;
13231		uint8_t *tmp_ptr;
13232
13233		rq->id = CTL_HA_DATA_CTL;
13234		rq->command = command;
13235		rq->context = io;
13236
13237		/*
13238		 * Both pointers should be aligned.  But it is possible
13239		 * that the allocation length is not.  They should both
13240		 * also have enough slack left over at the end, though,
13241		 * to round up to the next 8 byte boundary.
13242		 */
13243		cur_len = MIN(local_sglist[i].len - local_used,
13244			      remote_sglist[j].len - remote_used);
13245
13246		/*
13247		 * In this case, we have a size issue and need to decrease
13248		 * the size, except in the case where we actually have less
13249		 * than 8 bytes left.  In that case, we need to increase
13250		 * the DMA length to get the last bit.
13251		 */
13252		if ((cur_len & 0x7) != 0) {
13253			if (cur_len > 0x7) {
13254				cur_len = cur_len - (cur_len & 0x7);
13255				dma_length = cur_len;
13256			} else {
13257				CTL_SIZE_8B(dma_length, cur_len);
13258			}
13259
13260		} else
13261			dma_length = cur_len;
13262
13263		/*
13264		 * If we had to allocate memory for this I/O, instead of using
13265		 * the non-cached mirror memory, we'll need to flush the cache
13266		 * before trying to DMA to the other controller.
13267		 *
13268		 * We could end up doing this multiple times for the same
13269		 * segment if we have a larger local segment than remote
13270		 * segment.  That shouldn't be an issue.
13271		 */
13272		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13273			/*
13274			 * XXX KDM use bus_dmamap_sync() here.
13275			 */
13276		}
13277
13278		rq->size = dma_length;
13279
13280		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13281		tmp_ptr += local_used;
13282
13283		/* Use physical addresses when talking to ISC hardware */
13284		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13285			/* XXX KDM use busdma */
13286#if 0
13287			rq->local = vtophys(tmp_ptr);
13288#endif
13289		} else
13290			rq->local = tmp_ptr;
13291
13292		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13293		tmp_ptr += remote_used;
13294		rq->remote = tmp_ptr;
13295
13296		rq->callback = NULL;
13297
13298		local_used += cur_len;
13299		if (local_used >= local_sglist[i].len) {
13300			i++;
13301			local_used = 0;
13302		}
13303
13304		remote_used += cur_len;
13305		if (remote_used >= remote_sglist[j].len) {
13306			j++;
13307			remote_used = 0;
13308		}
13309		total_used += cur_len;
13310
13311		if (total_used >= io->scsiio.kern_data_len)
13312			rq->callback = callback;
13313
13314		if ((rq->size & 0x7) != 0) {
13315			printf("%s: warning: size %d is not on 8b boundary\n",
13316			       __func__, rq->size);
13317		}
13318		if (((uintptr_t)rq->local & 0x7) != 0) {
13319			printf("%s: warning: local %p not on 8b boundary\n",
13320			       __func__, rq->local);
13321		}
13322		if (((uintptr_t)rq->remote & 0x7) != 0) {
13323			printf("%s: warning: remote %p not on 8b boundary\n",
13324			       __func__, rq->local);
13325		}
13326#if 0
13327		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13328		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13329		       rq->local, rq->remote, rq->size);
13330#endif
13331
13332		isc_ret = ctl_dt_single(rq);
13333		if (isc_ret == CTL_HA_STATUS_WAIT)
13334			continue;
13335
13336		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13337			rq->ret = CTL_HA_STATUS_SUCCESS;
13338		} else {
13339			rq->ret = isc_ret;
13340		}
13341		callback(rq);
13342		goto bailout;
13343	}
13344
13345bailout:
13346	return (retval);
13347
13348}
13349
13350static void
13351ctl_datamove_remote_read(union ctl_io *io)
13352{
13353	int retval;
13354	int i;
13355
13356	/*
13357	 * This will send an error to the other controller in the case of a
13358	 * failure.
13359	 */
13360	retval = ctl_datamove_remote_sgl_setup(io);
13361	if (retval != 0)
13362		return;
13363
13364	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13365					  ctl_datamove_remote_read_cb);
13366	if ((retval != 0)
13367	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13368		/*
13369		 * Make sure we free memory if there was an error..  The
13370		 * ctl_datamove_remote_xfer() function will send the
13371		 * datamove done message, or call the callback with an
13372		 * error if there is a problem.
13373		 */
13374		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13375			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13376	}
13377
13378	return;
13379}
13380
13381/*
13382 * Process a datamove request from the other controller.  This is used for
13383 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13384 * first.  Once that is complete, the data gets DMAed into the remote
13385 * controller's memory.  For reads, we DMA from the remote controller's
13386 * memory into our memory first, and then move it out to the FETD.
13387 */
13388static void
13389ctl_datamove_remote(union ctl_io *io)
13390{
13391	struct ctl_softc *softc;
13392
13393	softc = control_softc;
13394
13395	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13396
13397	/*
13398	 * Note that we look for an aborted I/O here, but don't do some of
13399	 * the other checks that ctl_datamove() normally does.
13400	 * We don't need to run the datamove delay code, since that should
13401	 * have been done if need be on the other controller.
13402	 */
13403	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13404		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13405		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13406		       io->io_hdr.nexus.targ_port,
13407		       io->io_hdr.nexus.targ_target.id,
13408		       io->io_hdr.nexus.targ_lun);
13409		io->io_hdr.port_status = 31338;
13410		ctl_send_datamove_done(io, /*have_lock*/ 0);
13411		return;
13412	}
13413
13414	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13415		ctl_datamove_remote_write(io);
13416	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13417		ctl_datamove_remote_read(io);
13418	} else {
13419		union ctl_ha_msg msg;
13420		struct scsi_sense_data *sense;
13421		uint8_t sks[3];
13422		int retry_count;
13423
13424		memset(&msg, 0, sizeof(msg));
13425
13426		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13427		msg.hdr.status = CTL_SCSI_ERROR;
13428		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13429
13430		retry_count = 4243;
13431
13432		sense = &msg.scsi.sense_data;
13433		sks[0] = SSD_SCS_VALID;
13434		sks[1] = (retry_count >> 8) & 0xff;
13435		sks[2] = retry_count & 0xff;
13436
13437		/* "Internal target failure" */
13438		scsi_set_sense_data(sense,
13439				    /*sense_format*/ SSD_TYPE_NONE,
13440				    /*current_error*/ 1,
13441				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13442				    /*asc*/ 0x44,
13443				    /*ascq*/ 0x00,
13444				    /*type*/ SSD_ELEM_SKS,
13445				    /*size*/ sizeof(sks),
13446				    /*data*/ sks,
13447				    SSD_ELEM_NONE);
13448
13449		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13450		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13451			ctl_failover_io(io, /*have_lock*/ 1);
13452			return;
13453		}
13454
13455		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13456		    CTL_HA_STATUS_SUCCESS) {
13457			/* XXX KDM what to do if this fails? */
13458		}
13459		return;
13460	}
13461
13462}
13463
13464static int
13465ctl_process_done(union ctl_io *io)
13466{
13467	struct ctl_lun *lun;
13468	struct ctl_softc *softc = control_softc;
13469	void (*fe_done)(union ctl_io *io);
13470	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13471
13472	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13473
13474	fe_done = softc->ctl_ports[targ_port]->fe_done;
13475
13476#ifdef CTL_TIME_IO
13477	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13478		char str[256];
13479		char path_str[64];
13480		struct sbuf sb;
13481
13482		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13483		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13484
13485		sbuf_cat(&sb, path_str);
13486		switch (io->io_hdr.io_type) {
13487		case CTL_IO_SCSI:
13488			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13489			sbuf_printf(&sb, "\n");
13490			sbuf_cat(&sb, path_str);
13491			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13492				    io->scsiio.tag_num, io->scsiio.tag_type);
13493			break;
13494		case CTL_IO_TASK:
13495			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13496				    "Tag Type: %d\n", io->taskio.task_action,
13497				    io->taskio.tag_num, io->taskio.tag_type);
13498			break;
13499		default:
13500			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13501			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13502			break;
13503		}
13504		sbuf_cat(&sb, path_str);
13505		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13506			    (intmax_t)time_uptime - io->io_hdr.start_time);
13507		sbuf_finish(&sb);
13508		printf("%s", sbuf_data(&sb));
13509	}
13510#endif /* CTL_TIME_IO */
13511
13512	switch (io->io_hdr.io_type) {
13513	case CTL_IO_SCSI:
13514		break;
13515	case CTL_IO_TASK:
13516		if (ctl_debug & CTL_DEBUG_INFO)
13517			ctl_io_error_print(io, NULL);
13518		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13519			ctl_free_io(io);
13520		else
13521			fe_done(io);
13522		return (CTL_RETVAL_COMPLETE);
13523	default:
13524		panic("ctl_process_done: invalid io type %d\n",
13525		      io->io_hdr.io_type);
13526		break; /* NOTREACHED */
13527	}
13528
13529	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13530	if (lun == NULL) {
13531		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13532				 io->io_hdr.nexus.targ_mapped_lun));
13533		goto bailout;
13534	}
13535
13536	mtx_lock(&lun->lun_lock);
13537
13538	/*
13539	 * Check to see if we have any errors to inject here.  We only
13540	 * inject errors for commands that don't already have errors set.
13541	 */
13542	if ((STAILQ_FIRST(&lun->error_list) != NULL) &&
13543	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13544	    ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13545		ctl_inject_error(lun, io);
13546
13547	/*
13548	 * XXX KDM how do we treat commands that aren't completed
13549	 * successfully?
13550	 *
13551	 * XXX KDM should we also track I/O latency?
13552	 */
13553	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13554	    io->io_hdr.io_type == CTL_IO_SCSI) {
13555#ifdef CTL_TIME_IO
13556		struct bintime cur_bt;
13557#endif
13558		int type;
13559
13560		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13561		    CTL_FLAG_DATA_IN)
13562			type = CTL_STATS_READ;
13563		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13564		    CTL_FLAG_DATA_OUT)
13565			type = CTL_STATS_WRITE;
13566		else
13567			type = CTL_STATS_NO_IO;
13568
13569		lun->stats.ports[targ_port].bytes[type] +=
13570		    io->scsiio.kern_total_len;
13571		lun->stats.ports[targ_port].operations[type]++;
13572#ifdef CTL_TIME_IO
13573		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13574		   &io->io_hdr.dma_bt);
13575		lun->stats.ports[targ_port].num_dmas[type] +=
13576		    io->io_hdr.num_dmas;
13577		getbintime(&cur_bt);
13578		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13579		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13580#endif
13581	}
13582
13583	/*
13584	 * Remove this from the OOA queue.
13585	 */
13586	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13587#ifdef CTL_TIME_IO
13588	if (TAILQ_EMPTY(&lun->ooa_queue))
13589		lun->last_busy = getsbinuptime();
13590#endif
13591
13592	/*
13593	 * Run through the blocked queue on this LUN and see if anything
13594	 * has become unblocked, now that this transaction is done.
13595	 */
13596	ctl_check_blocked(lun);
13597
13598	/*
13599	 * If the LUN has been invalidated, free it if there is nothing
13600	 * left on its OOA queue.
13601	 */
13602	if ((lun->flags & CTL_LUN_INVALID)
13603	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13604		mtx_unlock(&lun->lun_lock);
13605		mtx_lock(&softc->ctl_lock);
13606		ctl_free_lun(lun);
13607		mtx_unlock(&softc->ctl_lock);
13608	} else
13609		mtx_unlock(&lun->lun_lock);
13610
13611bailout:
13612
13613	/*
13614	 * If this command has been aborted, make sure we set the status
13615	 * properly.  The FETD is responsible for freeing the I/O and doing
13616	 * whatever it needs to do to clean up its state.
13617	 */
13618	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13619		ctl_set_task_aborted(&io->scsiio);
13620
13621	/*
13622	 * If enabled, print command error status.
13623	 */
13624	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13625	    (ctl_debug & CTL_DEBUG_INFO) != 0)
13626		ctl_io_error_print(io, NULL);
13627
13628	/*
13629	 * Tell the FETD or the other shelf controller we're done with this
13630	 * command.  Note that only SCSI commands get to this point.  Task
13631	 * management commands are completed above.
13632	 *
13633	 * We only send status to the other controller if we're in XFER
13634	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13635	 * received the I/O (from CTL's perspective), and so the status is
13636	 * generated there.
13637	 *
13638	 * XXX KDM if we hold the lock here, we could cause a deadlock
13639	 * if the frontend comes back in in this context to queue
13640	 * something.
13641	 */
13642	if ((softc->ha_mode == CTL_HA_MODE_XFER)
13643	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13644		union ctl_ha_msg msg;
13645
13646		memset(&msg, 0, sizeof(msg));
13647		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13648		msg.hdr.original_sc = io->io_hdr.original_sc;
13649		msg.hdr.nexus = io->io_hdr.nexus;
13650		msg.hdr.status = io->io_hdr.status;
13651		msg.scsi.scsi_status = io->scsiio.scsi_status;
13652		msg.scsi.tag_num = io->scsiio.tag_num;
13653		msg.scsi.tag_type = io->scsiio.tag_type;
13654		msg.scsi.sense_len = io->scsiio.sense_len;
13655		msg.scsi.sense_residual = io->scsiio.sense_residual;
13656		msg.scsi.residual = io->scsiio.residual;
13657		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13658		       sizeof(io->scsiio.sense_data));
13659		/*
13660		 * We copy this whether or not this is an I/O-related
13661		 * command.  Otherwise, we'd have to go and check to see
13662		 * whether it's a read/write command, and it really isn't
13663		 * worth it.
13664		 */
13665		memcpy(&msg.scsi.lbalen,
13666		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13667		       sizeof(msg.scsi.lbalen));
13668
13669		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13670				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13671			/* XXX do something here */
13672		}
13673
13674		ctl_free_io(io);
13675	} else
13676		fe_done(io);
13677
13678	return (CTL_RETVAL_COMPLETE);
13679}
13680
13681#ifdef CTL_WITH_CA
13682/*
13683 * Front end should call this if it doesn't do autosense.  When the request
13684 * sense comes back in from the initiator, we'll dequeue this and send it.
13685 */
13686int
13687ctl_queue_sense(union ctl_io *io)
13688{
13689	struct ctl_lun *lun;
13690	struct ctl_port *port;
13691	struct ctl_softc *softc;
13692	uint32_t initidx, targ_lun;
13693
13694	softc = control_softc;
13695
13696	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13697
13698	/*
13699	 * LUN lookup will likely move to the ctl_work_thread() once we
13700	 * have our new queueing infrastructure (that doesn't put things on
13701	 * a per-LUN queue initially).  That is so that we can handle
13702	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13703	 * can't deal with that right now.
13704	 */
13705	mtx_lock(&softc->ctl_lock);
13706
13707	/*
13708	 * If we don't have a LUN for this, just toss the sense
13709	 * information.
13710	 */
13711	port = ctl_io_port(&ctsio->io_hdr);
13712	targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13713	if ((targ_lun < CTL_MAX_LUNS)
13714	 && (softc->ctl_luns[targ_lun] != NULL))
13715		lun = softc->ctl_luns[targ_lun];
13716	else
13717		goto bailout;
13718
13719	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13720
13721	mtx_lock(&lun->lun_lock);
13722	/*
13723	 * Already have CA set for this LUN...toss the sense information.
13724	 */
13725	if (ctl_is_set(lun->have_ca, initidx)) {
13726		mtx_unlock(&lun->lun_lock);
13727		goto bailout;
13728	}
13729
13730	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13731	       MIN(sizeof(lun->pending_sense[initidx]),
13732	       sizeof(io->scsiio.sense_data)));
13733	ctl_set_mask(lun->have_ca, initidx);
13734	mtx_unlock(&lun->lun_lock);
13735
13736bailout:
13737	mtx_unlock(&softc->ctl_lock);
13738
13739	ctl_free_io(io);
13740
13741	return (CTL_RETVAL_COMPLETE);
13742}
13743#endif
13744
13745/*
13746 * Primary command inlet from frontend ports.  All SCSI and task I/O
13747 * requests must go through this function.
13748 */
13749int
13750ctl_queue(union ctl_io *io)
13751{
13752	struct ctl_port *port;
13753
13754	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13755
13756#ifdef CTL_TIME_IO
13757	io->io_hdr.start_time = time_uptime;
13758	getbintime(&io->io_hdr.start_bt);
13759#endif /* CTL_TIME_IO */
13760
13761	/* Map FE-specific LUN ID into global one. */
13762	port = ctl_io_port(&io->io_hdr);
13763	io->io_hdr.nexus.targ_mapped_lun =
13764	    ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13765
13766	switch (io->io_hdr.io_type) {
13767	case CTL_IO_SCSI:
13768	case CTL_IO_TASK:
13769		if (ctl_debug & CTL_DEBUG_CDB)
13770			ctl_io_print(io);
13771		ctl_enqueue_incoming(io);
13772		break;
13773	default:
13774		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13775		return (EINVAL);
13776	}
13777
13778	return (CTL_RETVAL_COMPLETE);
13779}
13780
13781#ifdef CTL_IO_DELAY
13782static void
13783ctl_done_timer_wakeup(void *arg)
13784{
13785	union ctl_io *io;
13786
13787	io = (union ctl_io *)arg;
13788	ctl_done(io);
13789}
13790#endif /* CTL_IO_DELAY */
13791
13792void
13793ctl_done(union ctl_io *io)
13794{
13795
13796	/*
13797	 * Enable this to catch duplicate completion issues.
13798	 */
13799#if 0
13800	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13801		printf("%s: type %d msg %d cdb %x iptl: "
13802		       "%d:%d:%d:%d tag 0x%04x "
13803		       "flag %#x status %x\n",
13804			__func__,
13805			io->io_hdr.io_type,
13806			io->io_hdr.msg_type,
13807			io->scsiio.cdb[0],
13808			io->io_hdr.nexus.initid.id,
13809			io->io_hdr.nexus.targ_port,
13810			io->io_hdr.nexus.targ_target.id,
13811			io->io_hdr.nexus.targ_lun,
13812			(io->io_hdr.io_type ==
13813			CTL_IO_TASK) ?
13814			io->taskio.tag_num :
13815			io->scsiio.tag_num,
13816		        io->io_hdr.flags,
13817			io->io_hdr.status);
13818	} else
13819		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13820#endif
13821
13822	/*
13823	 * This is an internal copy of an I/O, and should not go through
13824	 * the normal done processing logic.
13825	 */
13826	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13827		return;
13828
13829	/*
13830	 * We need to send a msg to the serializing shelf to finish the IO
13831	 * as well.  We don't send a finish message to the other shelf if
13832	 * this is a task management command.  Task management commands
13833	 * aren't serialized in the OOA queue, but rather just executed on
13834	 * both shelf controllers for commands that originated on that
13835	 * controller.
13836	 */
13837	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13838	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13839		union ctl_ha_msg msg_io;
13840
13841		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13842		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13843		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13844		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13845		}
13846		/* continue on to finish IO */
13847	}
13848#ifdef CTL_IO_DELAY
13849	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13850		struct ctl_lun *lun;
13851
13852		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13853
13854		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13855	} else {
13856		struct ctl_lun *lun;
13857
13858		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13859
13860		if ((lun != NULL)
13861		 && (lun->delay_info.done_delay > 0)) {
13862			struct callout *callout;
13863
13864			callout = (struct callout *)&io->io_hdr.timer_bytes;
13865			callout_init(callout, /*mpsafe*/ 1);
13866			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13867			callout_reset(callout,
13868				      lun->delay_info.done_delay * hz,
13869				      ctl_done_timer_wakeup, io);
13870			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13871				lun->delay_info.done_delay = 0;
13872			return;
13873		}
13874	}
13875#endif /* CTL_IO_DELAY */
13876
13877	ctl_enqueue_done(io);
13878}
13879
13880int
13881ctl_isc(struct ctl_scsiio *ctsio)
13882{
13883	struct ctl_lun *lun;
13884	int retval;
13885
13886	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13887
13888	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13889
13890	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13891
13892	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13893
13894	return (retval);
13895}
13896
13897
13898static void
13899ctl_work_thread(void *arg)
13900{
13901	struct ctl_thread *thr = (struct ctl_thread *)arg;
13902	struct ctl_softc *softc = thr->ctl_softc;
13903	union ctl_io *io;
13904	int retval;
13905
13906	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13907
13908	for (;;) {
13909		retval = 0;
13910
13911		/*
13912		 * We handle the queues in this order:
13913		 * - ISC
13914		 * - done queue (to free up resources, unblock other commands)
13915		 * - RtR queue
13916		 * - incoming queue
13917		 *
13918		 * If those queues are empty, we break out of the loop and
13919		 * go to sleep.
13920		 */
13921		mtx_lock(&thr->queue_lock);
13922		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13923		if (io != NULL) {
13924			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13925			mtx_unlock(&thr->queue_lock);
13926			ctl_handle_isc(io);
13927			continue;
13928		}
13929		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13930		if (io != NULL) {
13931			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13932			/* clear any blocked commands, call fe_done */
13933			mtx_unlock(&thr->queue_lock);
13934			retval = ctl_process_done(io);
13935			continue;
13936		}
13937		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13938		if (io != NULL) {
13939			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13940			mtx_unlock(&thr->queue_lock);
13941			if (io->io_hdr.io_type == CTL_IO_TASK)
13942				ctl_run_task(io);
13943			else
13944				ctl_scsiio_precheck(softc, &io->scsiio);
13945			continue;
13946		}
13947		if (!ctl_pause_rtr) {
13948			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13949			if (io != NULL) {
13950				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13951				mtx_unlock(&thr->queue_lock);
13952				retval = ctl_scsiio(&io->scsiio);
13953				if (retval != CTL_RETVAL_COMPLETE)
13954					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13955				continue;
13956			}
13957		}
13958
13959		/* Sleep until we have something to do. */
13960		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13961	}
13962}
13963
13964static void
13965ctl_lun_thread(void *arg)
13966{
13967	struct ctl_softc *softc = (struct ctl_softc *)arg;
13968	struct ctl_be_lun *be_lun;
13969	int retval;
13970
13971	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13972
13973	for (;;) {
13974		retval = 0;
13975		mtx_lock(&softc->ctl_lock);
13976		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13977		if (be_lun != NULL) {
13978			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13979			mtx_unlock(&softc->ctl_lock);
13980			ctl_create_lun(be_lun);
13981			continue;
13982		}
13983
13984		/* Sleep until we have something to do. */
13985		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13986		    PDROP | PRIBIO, "-", 0);
13987	}
13988}
13989
13990static void
13991ctl_thresh_thread(void *arg)
13992{
13993	struct ctl_softc *softc = (struct ctl_softc *)arg;
13994	struct ctl_lun *lun;
13995	struct ctl_be_lun *be_lun;
13996	struct scsi_da_rw_recovery_page *rwpage;
13997	struct ctl_logical_block_provisioning_page *page;
13998	const char *attr;
13999	uint64_t thres, val;
14000	int i, e;
14001
14002	CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
14003
14004	for (;;) {
14005		mtx_lock(&softc->ctl_lock);
14006		STAILQ_FOREACH(lun, &softc->lun_list, links) {
14007			be_lun = lun->be_lun;
14008			if ((lun->flags & CTL_LUN_DISABLED) ||
14009			    (lun->flags & CTL_LUN_OFFLINE) ||
14010			    lun->backend->lun_attr == NULL)
14011				continue;
14012			rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
14013			if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
14014				continue;
14015			e = 0;
14016			page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
14017			for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
14018				if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
14019					continue;
14020				thres = scsi_4btoul(page->descr[i].count);
14021				thres <<= CTL_LBP_EXPONENT;
14022				switch (page->descr[i].resource) {
14023				case 0x01:
14024					attr = "blocksavail";
14025					break;
14026				case 0x02:
14027					attr = "blocksused";
14028					break;
14029				case 0xf1:
14030					attr = "poolblocksavail";
14031					break;
14032				case 0xf2:
14033					attr = "poolblocksused";
14034					break;
14035				default:
14036					continue;
14037				}
14038				mtx_unlock(&softc->ctl_lock); // XXX
14039				val = lun->backend->lun_attr(
14040				    lun->be_lun->be_lun, attr);
14041				mtx_lock(&softc->ctl_lock);
14042				if (val == UINT64_MAX)
14043					continue;
14044				if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
14045				    == SLBPPD_ARMING_INC)
14046					e |= (val >= thres);
14047				else
14048					e |= (val <= thres);
14049			}
14050			mtx_lock(&lun->lun_lock);
14051			if (e) {
14052				if (lun->lasttpt == 0 ||
14053				    time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
14054					lun->lasttpt = time_uptime;
14055					ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
14056				}
14057			} else {
14058				lun->lasttpt = 0;
14059				ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
14060			}
14061			mtx_unlock(&lun->lun_lock);
14062		}
14063		mtx_unlock(&softc->ctl_lock);
14064		pause("-", CTL_LBP_PERIOD * hz);
14065	}
14066}
14067
14068static void
14069ctl_enqueue_incoming(union ctl_io *io)
14070{
14071	struct ctl_softc *softc = control_softc;
14072	struct ctl_thread *thr;
14073	u_int idx;
14074
14075	idx = (io->io_hdr.nexus.targ_port * 127 +
14076	       io->io_hdr.nexus.initid.id) % worker_threads;
14077	thr = &softc->threads[idx];
14078	mtx_lock(&thr->queue_lock);
14079	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14080	mtx_unlock(&thr->queue_lock);
14081	wakeup(thr);
14082}
14083
14084static void
14085ctl_enqueue_rtr(union ctl_io *io)
14086{
14087	struct ctl_softc *softc = control_softc;
14088	struct ctl_thread *thr;
14089
14090	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14091	mtx_lock(&thr->queue_lock);
14092	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14093	mtx_unlock(&thr->queue_lock);
14094	wakeup(thr);
14095}
14096
14097static void
14098ctl_enqueue_done(union ctl_io *io)
14099{
14100	struct ctl_softc *softc = control_softc;
14101	struct ctl_thread *thr;
14102
14103	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14104	mtx_lock(&thr->queue_lock);
14105	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14106	mtx_unlock(&thr->queue_lock);
14107	wakeup(thr);
14108}
14109
14110#ifdef notyet
14111static void
14112ctl_enqueue_isc(union ctl_io *io)
14113{
14114	struct ctl_softc *softc = control_softc;
14115	struct ctl_thread *thr;
14116
14117	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14118	mtx_lock(&thr->queue_lock);
14119	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14120	mtx_unlock(&thr->queue_lock);
14121	wakeup(thr);
14122}
14123
14124/* Initialization and failover */
14125
14126void
14127ctl_init_isc_msg(void)
14128{
14129	printf("CTL: Still calling this thing\n");
14130}
14131
14132/*
14133 * Init component
14134 * 	Initializes component into configuration defined by bootMode
14135 *	(see hasc-sv.c)
14136 *  	returns hasc_Status:
14137 * 		OK
14138 *		ERROR - fatal error
14139 */
14140static ctl_ha_comp_status
14141ctl_isc_init(struct ctl_ha_component *c)
14142{
14143	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14144
14145	c->status = ret;
14146	return ret;
14147}
14148
14149/* Start component
14150 * 	Starts component in state requested. If component starts successfully,
14151 *	it must set its own state to the requestrd state
14152 *	When requested state is HASC_STATE_HA, the component may refine it
14153 * 	by adding _SLAVE or _MASTER flags.
14154 *	Currently allowed state transitions are:
14155 *	UNKNOWN->HA		- initial startup
14156 *	UNKNOWN->SINGLE - initial startup when no parter detected
14157 *	HA->SINGLE		- failover
14158 * returns ctl_ha_comp_status:
14159 * 		OK	- component successfully started in requested state
14160 *		FAILED  - could not start the requested state, failover may
14161 * 			  be possible
14162 *		ERROR	- fatal error detected, no future startup possible
14163 */
14164static ctl_ha_comp_status
14165ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14166{
14167	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14168
14169	printf("%s: go\n", __func__);
14170
14171	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14172	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14173		control_softc->is_single = 0;
14174		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14175		    != CTL_HA_STATUS_SUCCESS) {
14176			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14177			ret = CTL_HA_COMP_STATUS_ERROR;
14178		}
14179	} else if (CTL_HA_STATE_IS_HA(c->state)
14180		&& CTL_HA_STATE_IS_SINGLE(state)){
14181		// HA->SINGLE transition
14182	        ctl_failover();
14183		control_softc->is_single = 1;
14184	} else {
14185		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14186		       c->state, state);
14187		ret = CTL_HA_COMP_STATUS_ERROR;
14188	}
14189	if (CTL_HA_STATE_IS_SINGLE(state))
14190		control_softc->is_single = 1;
14191
14192	c->state = state;
14193	c->status = ret;
14194	return ret;
14195}
14196
14197/*
14198 * Quiesce component
14199 * The component must clear any error conditions (set status to OK) and
14200 * prepare itself to another Start call
14201 * returns ctl_ha_comp_status:
14202 * 	OK
14203 *	ERROR
14204 */
14205static ctl_ha_comp_status
14206ctl_isc_quiesce(struct ctl_ha_component *c)
14207{
14208	int ret = CTL_HA_COMP_STATUS_OK;
14209
14210	ctl_pause_rtr = 1;
14211	c->status = ret;
14212	return ret;
14213}
14214
14215struct ctl_ha_component ctl_ha_component_ctlisc =
14216{
14217	.name = "CTL ISC",
14218	.state = CTL_HA_STATE_UNKNOWN,
14219	.init = ctl_isc_init,
14220	.start = ctl_isc_start,
14221	.quiesce = ctl_isc_quiesce
14222};
14223#endif
14224
14225/*
14226 *  vim: ts=8
14227 */
14228