ctl.c revision 271949
1/*-
2 * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3 * Copyright (c) 2012 The FreeBSD Foundation
4 * All rights reserved.
5 *
6 * Portions of this software were developed by Edward Tomasz Napierala
7 * under sponsorship from the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions, and the following disclaimer,
14 *    without modification.
15 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16 *    substantially similar to the "NO WARRANTY" disclaimer below
17 *    ("Disclaimer") and any redistribution must be conditioned upon
18 *    including a substantially similar Disclaimer requirement for further
19 *    binary redistribution.
20 *
21 * NO WARRANTY
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGES.
33 *
34 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
35 */
36/*
37 * CAM Target Layer, a SCSI device emulation subsystem.
38 *
39 * Author: Ken Merry <ken@FreeBSD.org>
40 */
41
42#define _CTL_C
43
44#include <sys/cdefs.h>
45__FBSDID("$FreeBSD: head/sys/cam/ctl/ctl.c 271949 2014-09-22 09:22:58Z mav $");
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/kernel.h>
50#include <sys/types.h>
51#include <sys/kthread.h>
52#include <sys/bio.h>
53#include <sys/fcntl.h>
54#include <sys/lock.h>
55#include <sys/module.h>
56#include <sys/mutex.h>
57#include <sys/condvar.h>
58#include <sys/malloc.h>
59#include <sys/conf.h>
60#include <sys/ioccom.h>
61#include <sys/queue.h>
62#include <sys/sbuf.h>
63#include <sys/smp.h>
64#include <sys/endian.h>
65#include <sys/sysctl.h>
66
67#include <cam/cam.h>
68#include <cam/scsi/scsi_all.h>
69#include <cam/scsi/scsi_da.h>
70#include <cam/ctl/ctl_io.h>
71#include <cam/ctl/ctl.h>
72#include <cam/ctl/ctl_frontend.h>
73#include <cam/ctl/ctl_frontend_internal.h>
74#include <cam/ctl/ctl_util.h>
75#include <cam/ctl/ctl_backend.h>
76#include <cam/ctl/ctl_ioctl.h>
77#include <cam/ctl/ctl_ha.h>
78#include <cam/ctl/ctl_private.h>
79#include <cam/ctl/ctl_debug.h>
80#include <cam/ctl/ctl_scsi_all.h>
81#include <cam/ctl/ctl_error.h>
82
83struct ctl_softc *control_softc = NULL;
84
85/*
86 * Size and alignment macros needed for Copan-specific HA hardware.  These
87 * can go away when the HA code is re-written, and uses busdma for any
88 * hardware.
89 */
90#define	CTL_ALIGN_8B(target, source, type)				\
91	if (((uint32_t)source & 0x7) != 0)				\
92		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
93	else								\
94		target = (type)source;
95
96#define	CTL_SIZE_8B(target, size)					\
97	if ((size & 0x7) != 0)						\
98		target = size + (0x8 - (size & 0x7));			\
99	else								\
100		target = size;
101
102#define CTL_ALIGN_8B_MARGIN	16
103
104/*
105 * Template mode pages.
106 */
107
108/*
109 * Note that these are default values only.  The actual values will be
110 * filled in when the user does a mode sense.
111 */
112static struct copan_power_subpage power_page_default = {
113	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
114	/*subpage*/ PWR_SUBPAGE_CODE,
115	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
116			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
117	/*page_version*/ PWR_VERSION,
118	/* total_luns */ 26,
119	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
120	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
121		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
122		      0, 0, 0, 0, 0, 0}
123};
124
125static struct copan_power_subpage power_page_changeable = {
126	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
127	/*subpage*/ PWR_SUBPAGE_CODE,
128	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
129			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
130	/*page_version*/ 0,
131	/* total_luns */ 0,
132	/* max_active_luns*/ 0,
133	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
134		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
135		      0, 0, 0, 0, 0, 0}
136};
137
138static struct copan_aps_subpage aps_page_default = {
139	APS_PAGE_CODE | SMPH_SPF, //page_code
140	APS_SUBPAGE_CODE, //subpage
141	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
142	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
143	APS_VERSION, //page_version
144	0, //lock_active
145	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
146	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
147	0, 0, 0, 0, 0} //reserved
148};
149
150static struct copan_aps_subpage aps_page_changeable = {
151	APS_PAGE_CODE | SMPH_SPF, //page_code
152	APS_SUBPAGE_CODE, //subpage
153	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
154	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
155	0, //page_version
156	0, //lock_active
157	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
158	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
159	0, 0, 0, 0, 0} //reserved
160};
161
162static struct copan_debugconf_subpage debugconf_page_default = {
163	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
164	DBGCNF_SUBPAGE_CODE,		/* subpage */
165	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
166	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
167	DBGCNF_VERSION,			/* page_version */
168	{CTL_TIME_IO_DEFAULT_SECS>>8,
169	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
170};
171
172static struct copan_debugconf_subpage debugconf_page_changeable = {
173	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
174	DBGCNF_SUBPAGE_CODE,		/* subpage */
175	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
176	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
177	0,				/* page_version */
178	{0xff,0xff},			/* ctl_time_io_secs */
179};
180
181static struct scsi_format_page format_page_default = {
182	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
183	/*page_length*/sizeof(struct scsi_format_page) - 2,
184	/*tracks_per_zone*/ {0, 0},
185	/*alt_sectors_per_zone*/ {0, 0},
186	/*alt_tracks_per_zone*/ {0, 0},
187	/*alt_tracks_per_lun*/ {0, 0},
188	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
189			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
190	/*bytes_per_sector*/ {0, 0},
191	/*interleave*/ {0, 0},
192	/*track_skew*/ {0, 0},
193	/*cylinder_skew*/ {0, 0},
194	/*flags*/ SFP_HSEC,
195	/*reserved*/ {0, 0, 0}
196};
197
198static struct scsi_format_page format_page_changeable = {
199	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
200	/*page_length*/sizeof(struct scsi_format_page) - 2,
201	/*tracks_per_zone*/ {0, 0},
202	/*alt_sectors_per_zone*/ {0, 0},
203	/*alt_tracks_per_zone*/ {0, 0},
204	/*alt_tracks_per_lun*/ {0, 0},
205	/*sectors_per_track*/ {0, 0},
206	/*bytes_per_sector*/ {0, 0},
207	/*interleave*/ {0, 0},
208	/*track_skew*/ {0, 0},
209	/*cylinder_skew*/ {0, 0},
210	/*flags*/ 0,
211	/*reserved*/ {0, 0, 0}
212};
213
214static struct scsi_rigid_disk_page rigid_disk_page_default = {
215	/*page_code*/SMS_RIGID_DISK_PAGE,
216	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
217	/*cylinders*/ {0, 0, 0},
218	/*heads*/ CTL_DEFAULT_HEADS,
219	/*start_write_precomp*/ {0, 0, 0},
220	/*start_reduced_current*/ {0, 0, 0},
221	/*step_rate*/ {0, 0},
222	/*landing_zone_cylinder*/ {0, 0, 0},
223	/*rpl*/ SRDP_RPL_DISABLED,
224	/*rotational_offset*/ 0,
225	/*reserved1*/ 0,
226	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
227			   CTL_DEFAULT_ROTATION_RATE & 0xff},
228	/*reserved2*/ {0, 0}
229};
230
231static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
232	/*page_code*/SMS_RIGID_DISK_PAGE,
233	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
234	/*cylinders*/ {0, 0, 0},
235	/*heads*/ 0,
236	/*start_write_precomp*/ {0, 0, 0},
237	/*start_reduced_current*/ {0, 0, 0},
238	/*step_rate*/ {0, 0},
239	/*landing_zone_cylinder*/ {0, 0, 0},
240	/*rpl*/ 0,
241	/*rotational_offset*/ 0,
242	/*reserved1*/ 0,
243	/*rotation_rate*/ {0, 0},
244	/*reserved2*/ {0, 0}
245};
246
247static struct scsi_caching_page caching_page_default = {
248	/*page_code*/SMS_CACHING_PAGE,
249	/*page_length*/sizeof(struct scsi_caching_page) - 2,
250	/*flags1*/ SCP_DISC | SCP_WCE,
251	/*ret_priority*/ 0,
252	/*disable_pf_transfer_len*/ {0xff, 0xff},
253	/*min_prefetch*/ {0, 0},
254	/*max_prefetch*/ {0xff, 0xff},
255	/*max_pf_ceiling*/ {0xff, 0xff},
256	/*flags2*/ 0,
257	/*cache_segments*/ 0,
258	/*cache_seg_size*/ {0, 0},
259	/*reserved*/ 0,
260	/*non_cache_seg_size*/ {0, 0, 0}
261};
262
263static struct scsi_caching_page caching_page_changeable = {
264	/*page_code*/SMS_CACHING_PAGE,
265	/*page_length*/sizeof(struct scsi_caching_page) - 2,
266	/*flags1*/ SCP_WCE | SCP_RCD,
267	/*ret_priority*/ 0,
268	/*disable_pf_transfer_len*/ {0, 0},
269	/*min_prefetch*/ {0, 0},
270	/*max_prefetch*/ {0, 0},
271	/*max_pf_ceiling*/ {0, 0},
272	/*flags2*/ 0,
273	/*cache_segments*/ 0,
274	/*cache_seg_size*/ {0, 0},
275	/*reserved*/ 0,
276	/*non_cache_seg_size*/ {0, 0, 0}
277};
278
279static struct scsi_control_page control_page_default = {
280	/*page_code*/SMS_CONTROL_MODE_PAGE,
281	/*page_length*/sizeof(struct scsi_control_page) - 2,
282	/*rlec*/0,
283	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
284	/*eca_and_aen*/0,
285	/*flags4*/SCP_TAS,
286	/*aen_holdoff_period*/{0, 0},
287	/*busy_timeout_period*/{0, 0},
288	/*extended_selftest_completion_time*/{0, 0}
289};
290
291static struct scsi_control_page control_page_changeable = {
292	/*page_code*/SMS_CONTROL_MODE_PAGE,
293	/*page_length*/sizeof(struct scsi_control_page) - 2,
294	/*rlec*/SCP_DSENSE,
295	/*queue_flags*/SCP_QUEUE_ALG_MASK,
296	/*eca_and_aen*/0,
297	/*flags4*/0,
298	/*aen_holdoff_period*/{0, 0},
299	/*busy_timeout_period*/{0, 0},
300	/*extended_selftest_completion_time*/{0, 0}
301};
302
303
304/*
305 * XXX KDM move these into the softc.
306 */
307static int rcv_sync_msg;
308static int persis_offset;
309static uint8_t ctl_pause_rtr;
310static int     ctl_is_single = 1;
311static int     index_to_aps_page;
312
313SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
314static int worker_threads = -1;
315SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
316    &worker_threads, 1, "Number of worker threads");
317static int verbose = 0;
318SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
319    &verbose, 0, "Show SCSI errors returned to initiator");
320
321/*
322 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
323 * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
324 * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
325 * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
326 */
327#define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
328
329static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
330				  int param);
331static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
332static int ctl_init(void);
333void ctl_shutdown(void);
334static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
335static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
336static void ctl_ioctl_online(void *arg);
337static void ctl_ioctl_offline(void *arg);
338static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
339static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
340static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
341static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
342static int ctl_ioctl_submit_wait(union ctl_io *io);
343static void ctl_ioctl_datamove(union ctl_io *io);
344static void ctl_ioctl_done(union ctl_io *io);
345static void ctl_ioctl_hard_startstop_callback(void *arg,
346					      struct cfi_metatask *metatask);
347static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
348static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
349			      struct ctl_ooa *ooa_hdr,
350			      struct ctl_ooa_entry *kern_entries);
351static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
352		     struct thread *td);
353static uint32_t ctl_map_lun(int port_num, uint32_t lun);
354static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
355#ifdef unused
356static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
357				   uint32_t targ_target, uint32_t targ_lun,
358				   int can_wait);
359static void ctl_kfree_io(union ctl_io *io);
360#endif /* unused */
361static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
362			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
363static int ctl_free_lun(struct ctl_lun *lun);
364static void ctl_create_lun(struct ctl_be_lun *be_lun);
365/**
366static void ctl_failover_change_pages(struct ctl_softc *softc,
367				      struct ctl_scsiio *ctsio, int master);
368**/
369
370static int ctl_do_mode_select(union ctl_io *io);
371static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
372			   uint64_t res_key, uint64_t sa_res_key,
373			   uint8_t type, uint32_t residx,
374			   struct ctl_scsiio *ctsio,
375			   struct scsi_per_res_out *cdb,
376			   struct scsi_per_res_out_parms* param);
377static void ctl_pro_preempt_other(struct ctl_lun *lun,
378				  union ctl_ha_msg *msg);
379static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
380static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
381static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
382static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
383static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
384static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
385static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
386					 int alloc_len);
387static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
388					 int alloc_len);
389static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
390static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
391static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
392static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
393static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
394static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
395static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
396    union ctl_io *pending_io, union ctl_io *ooa_io);
397static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
398				union ctl_io *starting_io);
399static int ctl_check_blocked(struct ctl_lun *lun);
400static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
401				struct ctl_lun *lun,
402				const struct ctl_cmd_entry *entry,
403				struct ctl_scsiio *ctsio);
404//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
405static void ctl_failover(void);
406static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
407			       struct ctl_scsiio *ctsio);
408static int ctl_scsiio(struct ctl_scsiio *ctsio);
409
410static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
411static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
412			    ctl_ua_type ua_type);
413static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
414			 ctl_ua_type ua_type);
415static int ctl_abort_task(union ctl_io *io);
416static int ctl_abort_task_set(union ctl_io *io);
417static int ctl_i_t_nexus_reset(union ctl_io *io);
418static void ctl_run_task(union ctl_io *io);
419#ifdef CTL_IO_DELAY
420static void ctl_datamove_timer_wakeup(void *arg);
421static void ctl_done_timer_wakeup(void *arg);
422#endif /* CTL_IO_DELAY */
423
424static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
425static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
426static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
427static void ctl_datamove_remote_write(union ctl_io *io);
428static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
429static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
430static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
431static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
432				    ctl_ha_dt_cb callback);
433static void ctl_datamove_remote_read(union ctl_io *io);
434static void ctl_datamove_remote(union ctl_io *io);
435static int ctl_process_done(union ctl_io *io);
436static void ctl_lun_thread(void *arg);
437static void ctl_work_thread(void *arg);
438static void ctl_enqueue_incoming(union ctl_io *io);
439static void ctl_enqueue_rtr(union ctl_io *io);
440static void ctl_enqueue_done(union ctl_io *io);
441static void ctl_enqueue_isc(union ctl_io *io);
442static const struct ctl_cmd_entry *
443    ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
444static const struct ctl_cmd_entry *
445    ctl_validate_command(struct ctl_scsiio *ctsio);
446static int ctl_cmd_applicable(uint8_t lun_type,
447    const struct ctl_cmd_entry *entry);
448
449/*
450 * Load the serialization table.  This isn't very pretty, but is probably
451 * the easiest way to do it.
452 */
453#include "ctl_ser_table.c"
454
455/*
456 * We only need to define open, close and ioctl routines for this driver.
457 */
458static struct cdevsw ctl_cdevsw = {
459	.d_version =	D_VERSION,
460	.d_flags =	0,
461	.d_open =	ctl_open,
462	.d_close =	ctl_close,
463	.d_ioctl =	ctl_ioctl,
464	.d_name =	"ctl",
465};
466
467
468MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
469MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
470
471static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
472
473static moduledata_t ctl_moduledata = {
474	"ctl",
475	ctl_module_event_handler,
476	NULL
477};
478
479DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
480MODULE_VERSION(ctl, 1);
481
482static struct ctl_frontend ioctl_frontend =
483{
484	.name = "ioctl",
485};
486
487static void
488ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
489			    union ctl_ha_msg *msg_info)
490{
491	struct ctl_scsiio *ctsio;
492
493	if (msg_info->hdr.original_sc == NULL) {
494		printf("%s: original_sc == NULL!\n", __func__);
495		/* XXX KDM now what? */
496		return;
497	}
498
499	ctsio = &msg_info->hdr.original_sc->scsiio;
500	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
501	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
502	ctsio->io_hdr.status = msg_info->hdr.status;
503	ctsio->scsi_status = msg_info->scsi.scsi_status;
504	ctsio->sense_len = msg_info->scsi.sense_len;
505	ctsio->sense_residual = msg_info->scsi.sense_residual;
506	ctsio->residual = msg_info->scsi.residual;
507	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
508	       sizeof(ctsio->sense_data));
509	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
510	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
511	ctl_enqueue_isc((union ctl_io *)ctsio);
512}
513
514static void
515ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
516				union ctl_ha_msg *msg_info)
517{
518	struct ctl_scsiio *ctsio;
519
520	if (msg_info->hdr.serializing_sc == NULL) {
521		printf("%s: serializing_sc == NULL!\n", __func__);
522		/* XXX KDM now what? */
523		return;
524	}
525
526	ctsio = &msg_info->hdr.serializing_sc->scsiio;
527#if 0
528	/*
529	 * Attempt to catch the situation where an I/O has
530	 * been freed, and we're using it again.
531	 */
532	if (ctsio->io_hdr.io_type == 0xff) {
533		union ctl_io *tmp_io;
534		tmp_io = (union ctl_io *)ctsio;
535		printf("%s: %p use after free!\n", __func__,
536		       ctsio);
537		printf("%s: type %d msg %d cdb %x iptl: "
538		       "%d:%d:%d:%d tag 0x%04x "
539		       "flag %#x status %x\n",
540			__func__,
541			tmp_io->io_hdr.io_type,
542			tmp_io->io_hdr.msg_type,
543			tmp_io->scsiio.cdb[0],
544			tmp_io->io_hdr.nexus.initid.id,
545			tmp_io->io_hdr.nexus.targ_port,
546			tmp_io->io_hdr.nexus.targ_target.id,
547			tmp_io->io_hdr.nexus.targ_lun,
548			(tmp_io->io_hdr.io_type ==
549			CTL_IO_TASK) ?
550			tmp_io->taskio.tag_num :
551			tmp_io->scsiio.tag_num,
552		        tmp_io->io_hdr.flags,
553			tmp_io->io_hdr.status);
554	}
555#endif
556	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
557	ctl_enqueue_isc((union ctl_io *)ctsio);
558}
559
560/*
561 * ISC (Inter Shelf Communication) event handler.  Events from the HA
562 * subsystem come in here.
563 */
564static void
565ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
566{
567	struct ctl_softc *ctl_softc;
568	union ctl_io *io;
569	struct ctl_prio *presio;
570	ctl_ha_status isc_status;
571
572	ctl_softc = control_softc;
573	io = NULL;
574
575
576#if 0
577	printf("CTL: Isc Msg event %d\n", event);
578#endif
579	if (event == CTL_HA_EVT_MSG_RECV) {
580		union ctl_ha_msg msg_info;
581
582		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
583					     sizeof(msg_info), /*wait*/ 0);
584#if 0
585		printf("CTL: msg_type %d\n", msg_info.msg_type);
586#endif
587		if (isc_status != 0) {
588			printf("Error receiving message, status = %d\n",
589			       isc_status);
590			return;
591		}
592
593		switch (msg_info.hdr.msg_type) {
594		case CTL_MSG_SERIALIZE:
595#if 0
596			printf("Serialize\n");
597#endif
598			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
599			if (io == NULL) {
600				printf("ctl_isc_event_handler: can't allocate "
601				       "ctl_io!\n");
602				/* Bad Juju */
603				/* Need to set busy and send msg back */
604				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
605				msg_info.hdr.status = CTL_SCSI_ERROR;
606				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
607				msg_info.scsi.sense_len = 0;
608			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
609				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
610				}
611				goto bailout;
612			}
613			ctl_zero_io(io);
614			// populate ctsio from msg_info
615			io->io_hdr.io_type = CTL_IO_SCSI;
616			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
617			io->io_hdr.original_sc = msg_info.hdr.original_sc;
618#if 0
619			printf("pOrig %x\n", (int)msg_info.original_sc);
620#endif
621			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
622					    CTL_FLAG_IO_ACTIVE;
623			/*
624			 * If we're in serialization-only mode, we don't
625			 * want to go through full done processing.  Thus
626			 * the COPY flag.
627			 *
628			 * XXX KDM add another flag that is more specific.
629			 */
630			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
631				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
632			io->io_hdr.nexus = msg_info.hdr.nexus;
633#if 0
634			printf("targ %d, port %d, iid %d, lun %d\n",
635			       io->io_hdr.nexus.targ_target.id,
636			       io->io_hdr.nexus.targ_port,
637			       io->io_hdr.nexus.initid.id,
638			       io->io_hdr.nexus.targ_lun);
639#endif
640			io->scsiio.tag_num = msg_info.scsi.tag_num;
641			io->scsiio.tag_type = msg_info.scsi.tag_type;
642			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
643			       CTL_MAX_CDBLEN);
644			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
645				const struct ctl_cmd_entry *entry;
646
647				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
648				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
649				io->io_hdr.flags |=
650					entry->flags & CTL_FLAG_DATA_MASK;
651			}
652			ctl_enqueue_isc(io);
653			break;
654
655		/* Performed on the Originating SC, XFER mode only */
656		case CTL_MSG_DATAMOVE: {
657			struct ctl_sg_entry *sgl;
658			int i, j;
659
660			io = msg_info.hdr.original_sc;
661			if (io == NULL) {
662				printf("%s: original_sc == NULL!\n", __func__);
663				/* XXX KDM do something here */
664				break;
665			}
666			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
667			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
668			/*
669			 * Keep track of this, we need to send it back over
670			 * when the datamove is complete.
671			 */
672			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
673
674			if (msg_info.dt.sg_sequence == 0) {
675				/*
676				 * XXX KDM we use the preallocated S/G list
677				 * here, but we'll need to change this to
678				 * dynamic allocation if we need larger S/G
679				 * lists.
680				 */
681				if (msg_info.dt.kern_sg_entries >
682				    sizeof(io->io_hdr.remote_sglist) /
683				    sizeof(io->io_hdr.remote_sglist[0])) {
684					printf("%s: number of S/G entries "
685					    "needed %u > allocated num %zd\n",
686					    __func__,
687					    msg_info.dt.kern_sg_entries,
688					    sizeof(io->io_hdr.remote_sglist)/
689					    sizeof(io->io_hdr.remote_sglist[0]));
690
691					/*
692					 * XXX KDM send a message back to
693					 * the other side to shut down the
694					 * DMA.  The error will come back
695					 * through via the normal channel.
696					 */
697					break;
698				}
699				sgl = io->io_hdr.remote_sglist;
700				memset(sgl, 0,
701				       sizeof(io->io_hdr.remote_sglist));
702
703				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
704
705				io->scsiio.kern_sg_entries =
706					msg_info.dt.kern_sg_entries;
707				io->scsiio.rem_sg_entries =
708					msg_info.dt.kern_sg_entries;
709				io->scsiio.kern_data_len =
710					msg_info.dt.kern_data_len;
711				io->scsiio.kern_total_len =
712					msg_info.dt.kern_total_len;
713				io->scsiio.kern_data_resid =
714					msg_info.dt.kern_data_resid;
715				io->scsiio.kern_rel_offset =
716					msg_info.dt.kern_rel_offset;
717				/*
718				 * Clear out per-DMA flags.
719				 */
720				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
721				/*
722				 * Add per-DMA flags that are set for this
723				 * particular DMA request.
724				 */
725				io->io_hdr.flags |= msg_info.dt.flags &
726						    CTL_FLAG_RDMA_MASK;
727			} else
728				sgl = (struct ctl_sg_entry *)
729					io->scsiio.kern_data_ptr;
730
731			for (i = msg_info.dt.sent_sg_entries, j = 0;
732			     i < (msg_info.dt.sent_sg_entries +
733			     msg_info.dt.cur_sg_entries); i++, j++) {
734				sgl[i].addr = msg_info.dt.sg_list[j].addr;
735				sgl[i].len = msg_info.dt.sg_list[j].len;
736
737#if 0
738				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
739				       __func__,
740				       msg_info.dt.sg_list[j].addr,
741				       msg_info.dt.sg_list[j].len,
742				       sgl[i].addr, sgl[i].len, j, i);
743#endif
744			}
745#if 0
746			memcpy(&sgl[msg_info.dt.sent_sg_entries],
747			       msg_info.dt.sg_list,
748			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
749#endif
750
751			/*
752			 * If this is the last piece of the I/O, we've got
753			 * the full S/G list.  Queue processing in the thread.
754			 * Otherwise wait for the next piece.
755			 */
756			if (msg_info.dt.sg_last != 0)
757				ctl_enqueue_isc(io);
758			break;
759		}
760		/* Performed on the Serializing (primary) SC, XFER mode only */
761		case CTL_MSG_DATAMOVE_DONE: {
762			if (msg_info.hdr.serializing_sc == NULL) {
763				printf("%s: serializing_sc == NULL!\n",
764				       __func__);
765				/* XXX KDM now what? */
766				break;
767			}
768			/*
769			 * We grab the sense information here in case
770			 * there was a failure, so we can return status
771			 * back to the initiator.
772			 */
773			io = msg_info.hdr.serializing_sc;
774			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
775			io->io_hdr.status = msg_info.hdr.status;
776			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
777			io->scsiio.sense_len = msg_info.scsi.sense_len;
778			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
779			io->io_hdr.port_status = msg_info.scsi.fetd_status;
780			io->scsiio.residual = msg_info.scsi.residual;
781			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
782			       sizeof(io->scsiio.sense_data));
783			ctl_enqueue_isc(io);
784			break;
785		}
786
787		/* Preformed on Originating SC, SER_ONLY mode */
788		case CTL_MSG_R2R:
789			io = msg_info.hdr.original_sc;
790			if (io == NULL) {
791				printf("%s: Major Bummer\n", __func__);
792				return;
793			} else {
794#if 0
795				printf("pOrig %x\n",(int) ctsio);
796#endif
797			}
798			io->io_hdr.msg_type = CTL_MSG_R2R;
799			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
800			ctl_enqueue_isc(io);
801			break;
802
803		/*
804		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
805		 * mode.
806		 * Performed on the Originating (i.e. secondary) SC in XFER
807		 * mode
808		 */
809		case CTL_MSG_FINISH_IO:
810			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
811				ctl_isc_handler_finish_xfer(ctl_softc,
812							    &msg_info);
813			else
814				ctl_isc_handler_finish_ser_only(ctl_softc,
815								&msg_info);
816			break;
817
818		/* Preformed on Originating SC */
819		case CTL_MSG_BAD_JUJU:
820			io = msg_info.hdr.original_sc;
821			if (io == NULL) {
822				printf("%s: Bad JUJU!, original_sc is NULL!\n",
823				       __func__);
824				break;
825			}
826			ctl_copy_sense_data(&msg_info, io);
827			/*
828			 * IO should have already been cleaned up on other
829			 * SC so clear this flag so we won't send a message
830			 * back to finish the IO there.
831			 */
832			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
833			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
834
835			/* io = msg_info.hdr.serializing_sc; */
836			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
837			ctl_enqueue_isc(io);
838			break;
839
840		/* Handle resets sent from the other side */
841		case CTL_MSG_MANAGE_TASKS: {
842			struct ctl_taskio *taskio;
843			taskio = (struct ctl_taskio *)ctl_alloc_io(
844				(void *)ctl_softc->othersc_pool);
845			if (taskio == NULL) {
846				printf("ctl_isc_event_handler: can't allocate "
847				       "ctl_io!\n");
848				/* Bad Juju */
849				/* should I just call the proper reset func
850				   here??? */
851				goto bailout;
852			}
853			ctl_zero_io((union ctl_io *)taskio);
854			taskio->io_hdr.io_type = CTL_IO_TASK;
855			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
856			taskio->io_hdr.nexus = msg_info.hdr.nexus;
857			taskio->task_action = msg_info.task.task_action;
858			taskio->tag_num = msg_info.task.tag_num;
859			taskio->tag_type = msg_info.task.tag_type;
860#ifdef CTL_TIME_IO
861			taskio->io_hdr.start_time = time_uptime;
862			getbintime(&taskio->io_hdr.start_bt);
863#if 0
864			cs_prof_gettime(&taskio->io_hdr.start_ticks);
865#endif
866#endif /* CTL_TIME_IO */
867			ctl_run_task((union ctl_io *)taskio);
868			break;
869		}
870		/* Persistent Reserve action which needs attention */
871		case CTL_MSG_PERS_ACTION:
872			presio = (struct ctl_prio *)ctl_alloc_io(
873				(void *)ctl_softc->othersc_pool);
874			if (presio == NULL) {
875				printf("ctl_isc_event_handler: can't allocate "
876				       "ctl_io!\n");
877				/* Bad Juju */
878				/* Need to set busy and send msg back */
879				goto bailout;
880			}
881			ctl_zero_io((union ctl_io *)presio);
882			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
883			presio->pr_msg = msg_info.pr;
884			ctl_enqueue_isc((union ctl_io *)presio);
885			break;
886		case CTL_MSG_SYNC_FE:
887			rcv_sync_msg = 1;
888			break;
889		case CTL_MSG_APS_LOCK: {
890			// It's quicker to execute this then to
891			// queue it.
892			struct ctl_lun *lun;
893			struct ctl_page_index *page_index;
894			struct copan_aps_subpage *current_sp;
895			uint32_t targ_lun;
896
897			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
898			lun = ctl_softc->ctl_luns[targ_lun];
899			mtx_lock(&lun->lun_lock);
900			page_index = &lun->mode_pages.index[index_to_aps_page];
901			current_sp = (struct copan_aps_subpage *)
902				     (page_index->page_data +
903				     (page_index->page_len * CTL_PAGE_CURRENT));
904
905			current_sp->lock_active = msg_info.aps.lock_flag;
906			mtx_unlock(&lun->lun_lock);
907		        break;
908		}
909		default:
910		        printf("How did I get here?\n");
911		}
912	} else if (event == CTL_HA_EVT_MSG_SENT) {
913		if (param != CTL_HA_STATUS_SUCCESS) {
914			printf("Bad status from ctl_ha_msg_send status %d\n",
915			       param);
916		}
917		return;
918	} else if (event == CTL_HA_EVT_DISCONNECT) {
919		printf("CTL: Got a disconnect from Isc\n");
920		return;
921	} else {
922		printf("ctl_isc_event_handler: Unknown event %d\n", event);
923		return;
924	}
925
926bailout:
927	return;
928}
929
930static void
931ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
932{
933	struct scsi_sense_data *sense;
934
935	sense = &dest->scsiio.sense_data;
936	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
937	dest->scsiio.scsi_status = src->scsi.scsi_status;
938	dest->scsiio.sense_len = src->scsi.sense_len;
939	dest->io_hdr.status = src->hdr.status;
940}
941
942static int
943ctl_init(void)
944{
945	struct ctl_softc *softc;
946	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
947	struct ctl_port *port;
948        uint8_t sc_id =0;
949	int i, error, retval;
950	//int isc_retval;
951
952	retval = 0;
953	ctl_pause_rtr = 0;
954        rcv_sync_msg = 0;
955
956	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
957			       M_WAITOK | M_ZERO);
958	softc = control_softc;
959
960	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
961			      "cam/ctl");
962
963	softc->dev->si_drv1 = softc;
964
965	/*
966	 * By default, return a "bad LUN" peripheral qualifier for unknown
967	 * LUNs.  The user can override this default using the tunable or
968	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
969	 */
970	softc->inquiry_pq_no_lun = 1;
971	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
972			  &softc->inquiry_pq_no_lun);
973	sysctl_ctx_init(&softc->sysctl_ctx);
974	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
975		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
976		CTLFLAG_RD, 0, "CAM Target Layer");
977
978	if (softc->sysctl_tree == NULL) {
979		printf("%s: unable to allocate sysctl tree\n", __func__);
980		destroy_dev(softc->dev);
981		free(control_softc, M_DEVBUF);
982		control_softc = NULL;
983		return (ENOMEM);
984	}
985
986	SYSCTL_ADD_INT(&softc->sysctl_ctx,
987		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
988		       "inquiry_pq_no_lun", CTLFLAG_RW,
989		       &softc->inquiry_pq_no_lun, 0,
990		       "Report no lun possible for invalid LUNs");
991
992	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
993	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
994	softc->open_count = 0;
995
996	/*
997	 * Default to actually sending a SYNCHRONIZE CACHE command down to
998	 * the drive.
999	 */
1000	softc->flags = CTL_FLAG_REAL_SYNC;
1001
1002	/*
1003	 * In Copan's HA scheme, the "master" and "slave" roles are
1004	 * figured out through the slot the controller is in.  Although it
1005	 * is an active/active system, someone has to be in charge.
1006 	 */
1007#ifdef NEEDTOPORT
1008        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1009#endif
1010
1011        if (sc_id == 0) {
1012		softc->flags |= CTL_FLAG_MASTER_SHELF;
1013		persis_offset = 0;
1014	} else
1015		persis_offset = CTL_MAX_INITIATORS;
1016
1017	/*
1018	 * XXX KDM need to figure out where we want to get our target ID
1019	 * and WWID.  Is it different on each port?
1020	 */
1021	softc->target.id = 0;
1022	softc->target.wwid[0] = 0x12345678;
1023	softc->target.wwid[1] = 0x87654321;
1024	STAILQ_INIT(&softc->lun_list);
1025	STAILQ_INIT(&softc->pending_lun_queue);
1026	STAILQ_INIT(&softc->fe_list);
1027	STAILQ_INIT(&softc->port_list);
1028	STAILQ_INIT(&softc->be_list);
1029	STAILQ_INIT(&softc->io_pools);
1030	ctl_tpc_init(softc);
1031
1032	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1033			    &internal_pool)!= 0){
1034		printf("ctl: can't allocate %d entry internal pool, "
1035		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1036		return (ENOMEM);
1037	}
1038
1039	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1040			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1041		printf("ctl: can't allocate %d entry emergency pool, "
1042		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1043		ctl_pool_free(internal_pool);
1044		return (ENOMEM);
1045	}
1046
1047	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1048	                    &other_pool) != 0)
1049	{
1050		printf("ctl: can't allocate %d entry other SC pool, "
1051		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1052		ctl_pool_free(internal_pool);
1053		ctl_pool_free(emergency_pool);
1054		return (ENOMEM);
1055	}
1056
1057	softc->internal_pool = internal_pool;
1058	softc->emergency_pool = emergency_pool;
1059	softc->othersc_pool = other_pool;
1060
1061	if (worker_threads <= 0)
1062		worker_threads = max(1, mp_ncpus / 4);
1063	if (worker_threads > CTL_MAX_THREADS)
1064		worker_threads = CTL_MAX_THREADS;
1065
1066	for (i = 0; i < worker_threads; i++) {
1067		struct ctl_thread *thr = &softc->threads[i];
1068
1069		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1070		thr->ctl_softc = softc;
1071		STAILQ_INIT(&thr->incoming_queue);
1072		STAILQ_INIT(&thr->rtr_queue);
1073		STAILQ_INIT(&thr->done_queue);
1074		STAILQ_INIT(&thr->isc_queue);
1075
1076		error = kproc_kthread_add(ctl_work_thread, thr,
1077		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1078		if (error != 0) {
1079			printf("error creating CTL work thread!\n");
1080			ctl_pool_free(internal_pool);
1081			ctl_pool_free(emergency_pool);
1082			ctl_pool_free(other_pool);
1083			return (error);
1084		}
1085	}
1086	error = kproc_kthread_add(ctl_lun_thread, softc,
1087	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1088	if (error != 0) {
1089		printf("error creating CTL lun thread!\n");
1090		ctl_pool_free(internal_pool);
1091		ctl_pool_free(emergency_pool);
1092		ctl_pool_free(other_pool);
1093		return (error);
1094	}
1095	if (bootverbose)
1096		printf("ctl: CAM Target Layer loaded\n");
1097
1098	/*
1099	 * Initialize the ioctl front end.
1100	 */
1101	ctl_frontend_register(&ioctl_frontend);
1102	port = &softc->ioctl_info.port;
1103	port->frontend = &ioctl_frontend;
1104	sprintf(softc->ioctl_info.port_name, "ioctl");
1105	port->port_type = CTL_PORT_IOCTL;
1106	port->num_requested_ctl_io = 100;
1107	port->port_name = softc->ioctl_info.port_name;
1108	port->port_online = ctl_ioctl_online;
1109	port->port_offline = ctl_ioctl_offline;
1110	port->onoff_arg = &softc->ioctl_info;
1111	port->lun_enable = ctl_ioctl_lun_enable;
1112	port->lun_disable = ctl_ioctl_lun_disable;
1113	port->targ_lun_arg = &softc->ioctl_info;
1114	port->fe_datamove = ctl_ioctl_datamove;
1115	port->fe_done = ctl_ioctl_done;
1116	port->max_targets = 15;
1117	port->max_target_id = 15;
1118
1119	if (ctl_port_register(&softc->ioctl_info.port,
1120	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1121		printf("ctl: ioctl front end registration failed, will "
1122		       "continue anyway\n");
1123	}
1124
1125#ifdef CTL_IO_DELAY
1126	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1127		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1128		       sizeof(struct callout), CTL_TIMER_BYTES);
1129		return (EINVAL);
1130	}
1131#endif /* CTL_IO_DELAY */
1132
1133	return (0);
1134}
1135
1136void
1137ctl_shutdown(void)
1138{
1139	struct ctl_softc *softc;
1140	struct ctl_lun *lun, *next_lun;
1141	struct ctl_io_pool *pool;
1142
1143	softc = (struct ctl_softc *)control_softc;
1144
1145	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1146		printf("ctl: ioctl front end deregistration failed\n");
1147
1148	mtx_lock(&softc->ctl_lock);
1149
1150	/*
1151	 * Free up each LUN.
1152	 */
1153	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1154		next_lun = STAILQ_NEXT(lun, links);
1155		ctl_free_lun(lun);
1156	}
1157
1158	mtx_unlock(&softc->ctl_lock);
1159
1160	ctl_frontend_deregister(&ioctl_frontend);
1161
1162	/*
1163	 * This will rip the rug out from under any FETDs or anyone else
1164	 * that has a pool allocated.  Since we increment our module
1165	 * refcount any time someone outside the main CTL module allocates
1166	 * a pool, we shouldn't have any problems here.  The user won't be
1167	 * able to unload the CTL module until client modules have
1168	 * successfully unloaded.
1169	 */
1170	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1171		ctl_pool_free(pool);
1172
1173#if 0
1174	ctl_shutdown_thread(softc->work_thread);
1175	mtx_destroy(&softc->queue_lock);
1176#endif
1177
1178	ctl_tpc_shutdown(softc);
1179	mtx_destroy(&softc->pool_lock);
1180	mtx_destroy(&softc->ctl_lock);
1181
1182	destroy_dev(softc->dev);
1183
1184	sysctl_ctx_free(&softc->sysctl_ctx);
1185
1186	free(control_softc, M_DEVBUF);
1187	control_softc = NULL;
1188
1189	if (bootverbose)
1190		printf("ctl: CAM Target Layer unloaded\n");
1191}
1192
1193static int
1194ctl_module_event_handler(module_t mod, int what, void *arg)
1195{
1196
1197	switch (what) {
1198	case MOD_LOAD:
1199		return (ctl_init());
1200	case MOD_UNLOAD:
1201		return (EBUSY);
1202	default:
1203		return (EOPNOTSUPP);
1204	}
1205}
1206
1207/*
1208 * XXX KDM should we do some access checks here?  Bump a reference count to
1209 * prevent a CTL module from being unloaded while someone has it open?
1210 */
1211static int
1212ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1213{
1214	return (0);
1215}
1216
1217static int
1218ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1219{
1220	return (0);
1221}
1222
1223int
1224ctl_port_enable(ctl_port_type port_type)
1225{
1226	struct ctl_softc *softc;
1227	struct ctl_port *port;
1228
1229	if (ctl_is_single == 0) {
1230		union ctl_ha_msg msg_info;
1231		int isc_retval;
1232
1233#if 0
1234		printf("%s: HA mode, synchronizing frontend enable\n",
1235		        __func__);
1236#endif
1237		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1238	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1239		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1240			printf("Sync msg send error retval %d\n", isc_retval);
1241		}
1242		if (!rcv_sync_msg) {
1243			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1244			        sizeof(msg_info), 1);
1245		}
1246#if 0
1247        	printf("CTL:Frontend Enable\n");
1248	} else {
1249		printf("%s: single mode, skipping frontend synchronization\n",
1250		        __func__);
1251#endif
1252	}
1253
1254	softc = control_softc;
1255
1256	STAILQ_FOREACH(port, &softc->port_list, links) {
1257		if (port_type & port->port_type)
1258		{
1259#if 0
1260			printf("port %d\n", port->targ_port);
1261#endif
1262			ctl_port_online(port);
1263		}
1264	}
1265
1266	return (0);
1267}
1268
1269int
1270ctl_port_disable(ctl_port_type port_type)
1271{
1272	struct ctl_softc *softc;
1273	struct ctl_port *port;
1274
1275	softc = control_softc;
1276
1277	STAILQ_FOREACH(port, &softc->port_list, links) {
1278		if (port_type & port->port_type)
1279			ctl_port_offline(port);
1280	}
1281
1282	return (0);
1283}
1284
1285/*
1286 * Returns 0 for success, 1 for failure.
1287 * Currently the only failure mode is if there aren't enough entries
1288 * allocated.  So, in case of a failure, look at num_entries_dropped,
1289 * reallocate and try again.
1290 */
1291int
1292ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1293	      int *num_entries_filled, int *num_entries_dropped,
1294	      ctl_port_type port_type, int no_virtual)
1295{
1296	struct ctl_softc *softc;
1297	struct ctl_port *port;
1298	int entries_dropped, entries_filled;
1299	int retval;
1300	int i;
1301
1302	softc = control_softc;
1303
1304	retval = 0;
1305	entries_filled = 0;
1306	entries_dropped = 0;
1307
1308	i = 0;
1309	mtx_lock(&softc->ctl_lock);
1310	STAILQ_FOREACH(port, &softc->port_list, links) {
1311		struct ctl_port_entry *entry;
1312
1313		if ((port->port_type & port_type) == 0)
1314			continue;
1315
1316		if ((no_virtual != 0)
1317		 && (port->virtual_port != 0))
1318			continue;
1319
1320		if (entries_filled >= num_entries_alloced) {
1321			entries_dropped++;
1322			continue;
1323		}
1324		entry = &entries[i];
1325
1326		entry->port_type = port->port_type;
1327		strlcpy(entry->port_name, port->port_name,
1328			sizeof(entry->port_name));
1329		entry->physical_port = port->physical_port;
1330		entry->virtual_port = port->virtual_port;
1331		entry->wwnn = port->wwnn;
1332		entry->wwpn = port->wwpn;
1333
1334		i++;
1335		entries_filled++;
1336	}
1337
1338	mtx_unlock(&softc->ctl_lock);
1339
1340	if (entries_dropped > 0)
1341		retval = 1;
1342
1343	*num_entries_dropped = entries_dropped;
1344	*num_entries_filled = entries_filled;
1345
1346	return (retval);
1347}
1348
1349static void
1350ctl_ioctl_online(void *arg)
1351{
1352	struct ctl_ioctl_info *ioctl_info;
1353
1354	ioctl_info = (struct ctl_ioctl_info *)arg;
1355
1356	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1357}
1358
1359static void
1360ctl_ioctl_offline(void *arg)
1361{
1362	struct ctl_ioctl_info *ioctl_info;
1363
1364	ioctl_info = (struct ctl_ioctl_info *)arg;
1365
1366	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1367}
1368
1369/*
1370 * Remove an initiator by port number and initiator ID.
1371 * Returns 0 for success, -1 for failure.
1372 */
1373int
1374ctl_remove_initiator(struct ctl_port *port, int iid)
1375{
1376	struct ctl_softc *softc = control_softc;
1377
1378	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1379
1380	if (iid > CTL_MAX_INIT_PER_PORT) {
1381		printf("%s: initiator ID %u > maximun %u!\n",
1382		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1383		return (-1);
1384	}
1385
1386	mtx_lock(&softc->ctl_lock);
1387	port->wwpn_iid[iid].in_use--;
1388	port->wwpn_iid[iid].last_use = time_uptime;
1389	mtx_unlock(&softc->ctl_lock);
1390
1391	return (0);
1392}
1393
1394/*
1395 * Add an initiator to the initiator map.
1396 * Returns iid for success, < 0 for failure.
1397 */
1398int
1399ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1400{
1401	struct ctl_softc *softc = control_softc;
1402	time_t best_time;
1403	int i, best;
1404
1405	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1406
1407	if (iid >= CTL_MAX_INIT_PER_PORT) {
1408		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1409		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1410		free(name, M_CTL);
1411		return (-1);
1412	}
1413
1414	mtx_lock(&softc->ctl_lock);
1415
1416	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1417		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1418			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1419				iid = i;
1420				break;
1421			}
1422			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1423			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1424				iid = i;
1425				break;
1426			}
1427		}
1428	}
1429
1430	if (iid < 0) {
1431		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1432			if (port->wwpn_iid[i].in_use == 0 &&
1433			    port->wwpn_iid[i].wwpn == 0 &&
1434			    port->wwpn_iid[i].name == NULL) {
1435				iid = i;
1436				break;
1437			}
1438		}
1439	}
1440
1441	if (iid < 0) {
1442		best = -1;
1443		best_time = INT32_MAX;
1444		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1445			if (port->wwpn_iid[i].in_use == 0) {
1446				if (port->wwpn_iid[i].last_use < best_time) {
1447					best = i;
1448					best_time = port->wwpn_iid[i].last_use;
1449				}
1450			}
1451		}
1452		iid = best;
1453	}
1454
1455	if (iid < 0) {
1456		mtx_unlock(&softc->ctl_lock);
1457		free(name, M_CTL);
1458		return (-2);
1459	}
1460
1461	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1462		/*
1463		 * This is not an error yet.
1464		 */
1465		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1466#if 0
1467			printf("%s: port %d iid %u WWPN %#jx arrived"
1468			    " again\n", __func__, port->targ_port,
1469			    iid, (uintmax_t)wwpn);
1470#endif
1471			goto take;
1472		}
1473		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1474		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1475#if 0
1476			printf("%s: port %d iid %u name '%s' arrived"
1477			    " again\n", __func__, port->targ_port,
1478			    iid, name);
1479#endif
1480			goto take;
1481		}
1482
1483		/*
1484		 * This is an error, but what do we do about it?  The
1485		 * driver is telling us we have a new WWPN for this
1486		 * initiator ID, so we pretty much need to use it.
1487		 */
1488		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1489		    " but WWPN %#jx '%s' is still at that address\n",
1490		    __func__, port->targ_port, iid, wwpn, name,
1491		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1492		    port->wwpn_iid[iid].name);
1493
1494		/*
1495		 * XXX KDM clear have_ca and ua_pending on each LUN for
1496		 * this initiator.
1497		 */
1498	}
1499take:
1500	free(port->wwpn_iid[iid].name, M_CTL);
1501	port->wwpn_iid[iid].name = name;
1502	port->wwpn_iid[iid].wwpn = wwpn;
1503	port->wwpn_iid[iid].in_use++;
1504	mtx_unlock(&softc->ctl_lock);
1505
1506	return (iid);
1507}
1508
1509static int
1510ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1511{
1512	int len;
1513
1514	switch (port->port_type) {
1515	case CTL_PORT_FC:
1516	{
1517		struct scsi_transportid_fcp *id =
1518		    (struct scsi_transportid_fcp *)buf;
1519		if (port->wwpn_iid[iid].wwpn == 0)
1520			return (0);
1521		memset(id, 0, sizeof(*id));
1522		id->format_protocol = SCSI_PROTO_FC;
1523		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1524		return (sizeof(*id));
1525	}
1526	case CTL_PORT_ISCSI:
1527	{
1528		struct scsi_transportid_iscsi_port *id =
1529		    (struct scsi_transportid_iscsi_port *)buf;
1530		if (port->wwpn_iid[iid].name == NULL)
1531			return (0);
1532		memset(id, 0, 256);
1533		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1534		    SCSI_PROTO_ISCSI;
1535		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1536		len = roundup2(min(len, 252), 4);
1537		scsi_ulto2b(len, id->additional_length);
1538		return (sizeof(*id) + len);
1539	}
1540	case CTL_PORT_SAS:
1541	{
1542		struct scsi_transportid_sas *id =
1543		    (struct scsi_transportid_sas *)buf;
1544		if (port->wwpn_iid[iid].wwpn == 0)
1545			return (0);
1546		memset(id, 0, sizeof(*id));
1547		id->format_protocol = SCSI_PROTO_SAS;
1548		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1549		return (sizeof(*id));
1550	}
1551	default:
1552	{
1553		struct scsi_transportid_spi *id =
1554		    (struct scsi_transportid_spi *)buf;
1555		memset(id, 0, sizeof(*id));
1556		id->format_protocol = SCSI_PROTO_SPI;
1557		scsi_ulto2b(iid, id->scsi_addr);
1558		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1559		return (sizeof(*id));
1560	}
1561	}
1562}
1563
1564static int
1565ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1566{
1567	return (0);
1568}
1569
1570static int
1571ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1572{
1573	return (0);
1574}
1575
1576/*
1577 * Data movement routine for the CTL ioctl frontend port.
1578 */
1579static int
1580ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1581{
1582	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1583	struct ctl_sg_entry ext_entry, kern_entry;
1584	int ext_sglen, ext_sg_entries, kern_sg_entries;
1585	int ext_sg_start, ext_offset;
1586	int len_to_copy, len_copied;
1587	int kern_watermark, ext_watermark;
1588	int ext_sglist_malloced;
1589	int i, j;
1590
1591	ext_sglist_malloced = 0;
1592	ext_sg_start = 0;
1593	ext_offset = 0;
1594
1595	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1596
1597	/*
1598	 * If this flag is set, fake the data transfer.
1599	 */
1600	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1601		ctsio->ext_data_filled = ctsio->ext_data_len;
1602		goto bailout;
1603	}
1604
1605	/*
1606	 * To simplify things here, if we have a single buffer, stick it in
1607	 * a S/G entry and just make it a single entry S/G list.
1608	 */
1609	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1610		int len_seen;
1611
1612		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1613
1614		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1615							   M_WAITOK);
1616		ext_sglist_malloced = 1;
1617		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1618				   ext_sglen) != 0) {
1619			ctl_set_internal_failure(ctsio,
1620						 /*sks_valid*/ 0,
1621						 /*retry_count*/ 0);
1622			goto bailout;
1623		}
1624		ext_sg_entries = ctsio->ext_sg_entries;
1625		len_seen = 0;
1626		for (i = 0; i < ext_sg_entries; i++) {
1627			if ((len_seen + ext_sglist[i].len) >=
1628			     ctsio->ext_data_filled) {
1629				ext_sg_start = i;
1630				ext_offset = ctsio->ext_data_filled - len_seen;
1631				break;
1632			}
1633			len_seen += ext_sglist[i].len;
1634		}
1635	} else {
1636		ext_sglist = &ext_entry;
1637		ext_sglist->addr = ctsio->ext_data_ptr;
1638		ext_sglist->len = ctsio->ext_data_len;
1639		ext_sg_entries = 1;
1640		ext_sg_start = 0;
1641		ext_offset = ctsio->ext_data_filled;
1642	}
1643
1644	if (ctsio->kern_sg_entries > 0) {
1645		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1646		kern_sg_entries = ctsio->kern_sg_entries;
1647	} else {
1648		kern_sglist = &kern_entry;
1649		kern_sglist->addr = ctsio->kern_data_ptr;
1650		kern_sglist->len = ctsio->kern_data_len;
1651		kern_sg_entries = 1;
1652	}
1653
1654
1655	kern_watermark = 0;
1656	ext_watermark = ext_offset;
1657	len_copied = 0;
1658	for (i = ext_sg_start, j = 0;
1659	     i < ext_sg_entries && j < kern_sg_entries;) {
1660		uint8_t *ext_ptr, *kern_ptr;
1661
1662		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1663				      kern_sglist[j].len - kern_watermark);
1664
1665		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1666		ext_ptr = ext_ptr + ext_watermark;
1667		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1668			/*
1669			 * XXX KDM fix this!
1670			 */
1671			panic("need to implement bus address support");
1672#if 0
1673			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1674#endif
1675		} else
1676			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1677		kern_ptr = kern_ptr + kern_watermark;
1678
1679		kern_watermark += len_to_copy;
1680		ext_watermark += len_to_copy;
1681
1682		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1683		     CTL_FLAG_DATA_IN) {
1684			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1685					 "bytes to user\n", len_to_copy));
1686			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1687					 "to %p\n", kern_ptr, ext_ptr));
1688			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1689				ctl_set_internal_failure(ctsio,
1690							 /*sks_valid*/ 0,
1691							 /*retry_count*/ 0);
1692				goto bailout;
1693			}
1694		} else {
1695			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1696					 "bytes from user\n", len_to_copy));
1697			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1698					 "to %p\n", ext_ptr, kern_ptr));
1699			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1700				ctl_set_internal_failure(ctsio,
1701							 /*sks_valid*/ 0,
1702							 /*retry_count*/0);
1703				goto bailout;
1704			}
1705		}
1706
1707		len_copied += len_to_copy;
1708
1709		if (ext_sglist[i].len == ext_watermark) {
1710			i++;
1711			ext_watermark = 0;
1712		}
1713
1714		if (kern_sglist[j].len == kern_watermark) {
1715			j++;
1716			kern_watermark = 0;
1717		}
1718	}
1719
1720	ctsio->ext_data_filled += len_copied;
1721
1722	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1723			 "kern_sg_entries: %d\n", ext_sg_entries,
1724			 kern_sg_entries));
1725	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1726			 "kern_data_len = %d\n", ctsio->ext_data_len,
1727			 ctsio->kern_data_len));
1728
1729
1730	/* XXX KDM set residual?? */
1731bailout:
1732
1733	if (ext_sglist_malloced != 0)
1734		free(ext_sglist, M_CTL);
1735
1736	return (CTL_RETVAL_COMPLETE);
1737}
1738
1739/*
1740 * Serialize a command that went down the "wrong" side, and so was sent to
1741 * this controller for execution.  The logic is a little different than the
1742 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1743 * sent back to the other side, but in the success case, we execute the
1744 * command on this side (XFER mode) or tell the other side to execute it
1745 * (SER_ONLY mode).
1746 */
1747static int
1748ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1749{
1750	struct ctl_softc *ctl_softc;
1751	union ctl_ha_msg msg_info;
1752	struct ctl_lun *lun;
1753	int retval = 0;
1754	uint32_t targ_lun;
1755
1756	ctl_softc = control_softc;
1757
1758	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1759	lun = ctl_softc->ctl_luns[targ_lun];
1760	if (lun==NULL)
1761	{
1762		/*
1763		 * Why isn't LUN defined? The other side wouldn't
1764		 * send a cmd if the LUN is undefined.
1765		 */
1766		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1767
1768		/* "Logical unit not supported" */
1769		ctl_set_sense_data(&msg_info.scsi.sense_data,
1770				   lun,
1771				   /*sense_format*/SSD_TYPE_NONE,
1772				   /*current_error*/ 1,
1773				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1774				   /*asc*/ 0x25,
1775				   /*ascq*/ 0x00,
1776				   SSD_ELEM_NONE);
1777
1778		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1779		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1780		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1781		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1782		msg_info.hdr.serializing_sc = NULL;
1783		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1784	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1785				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1786		}
1787		return(1);
1788
1789	}
1790
1791	mtx_lock(&lun->lun_lock);
1792    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1793
1794	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1795		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1796		 ooa_links))) {
1797	case CTL_ACTION_BLOCK:
1798		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1799		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1800				  blocked_links);
1801		break;
1802	case CTL_ACTION_PASS:
1803	case CTL_ACTION_SKIP:
1804		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1805			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1806			ctl_enqueue_rtr((union ctl_io *)ctsio);
1807		} else {
1808
1809			/* send msg back to other side */
1810			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1811			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1812			msg_info.hdr.msg_type = CTL_MSG_R2R;
1813#if 0
1814			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1815#endif
1816		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1817			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1818			}
1819		}
1820		break;
1821	case CTL_ACTION_OVERLAP:
1822		/* OVERLAPPED COMMANDS ATTEMPTED */
1823		ctl_set_sense_data(&msg_info.scsi.sense_data,
1824				   lun,
1825				   /*sense_format*/SSD_TYPE_NONE,
1826				   /*current_error*/ 1,
1827				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1828				   /*asc*/ 0x4E,
1829				   /*ascq*/ 0x00,
1830				   SSD_ELEM_NONE);
1831
1832		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1833		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1834		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1835		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1836		msg_info.hdr.serializing_sc = NULL;
1837		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1838#if 0
1839		printf("BAD JUJU:Major Bummer Overlap\n");
1840#endif
1841		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1842		retval = 1;
1843		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1844		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1845		}
1846		break;
1847	case CTL_ACTION_OVERLAP_TAG:
1848		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1849		ctl_set_sense_data(&msg_info.scsi.sense_data,
1850				   lun,
1851				   /*sense_format*/SSD_TYPE_NONE,
1852				   /*current_error*/ 1,
1853				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1854				   /*asc*/ 0x4D,
1855				   /*ascq*/ ctsio->tag_num & 0xff,
1856				   SSD_ELEM_NONE);
1857
1858		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1859		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1860		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1861		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1862		msg_info.hdr.serializing_sc = NULL;
1863		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1864#if 0
1865		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1866#endif
1867		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1868		retval = 1;
1869		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1870		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1871		}
1872		break;
1873	case CTL_ACTION_ERROR:
1874	default:
1875		/* "Internal target failure" */
1876		ctl_set_sense_data(&msg_info.scsi.sense_data,
1877				   lun,
1878				   /*sense_format*/SSD_TYPE_NONE,
1879				   /*current_error*/ 1,
1880				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1881				   /*asc*/ 0x44,
1882				   /*ascq*/ 0x00,
1883				   SSD_ELEM_NONE);
1884
1885		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1886		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1887		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1888		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1889		msg_info.hdr.serializing_sc = NULL;
1890		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1891#if 0
1892		printf("BAD JUJU:Major Bummer HW Error\n");
1893#endif
1894		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1895		retval = 1;
1896		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1897		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1898		}
1899		break;
1900	}
1901	mtx_unlock(&lun->lun_lock);
1902	return (retval);
1903}
1904
1905static int
1906ctl_ioctl_submit_wait(union ctl_io *io)
1907{
1908	struct ctl_fe_ioctl_params params;
1909	ctl_fe_ioctl_state last_state;
1910	int done, retval;
1911
1912	retval = 0;
1913
1914	bzero(&params, sizeof(params));
1915
1916	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1917	cv_init(&params.sem, "ctlioccv");
1918	params.state = CTL_IOCTL_INPROG;
1919	last_state = params.state;
1920
1921	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1922
1923	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1924
1925	/* This shouldn't happen */
1926	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1927		return (retval);
1928
1929	done = 0;
1930
1931	do {
1932		mtx_lock(&params.ioctl_mtx);
1933		/*
1934		 * Check the state here, and don't sleep if the state has
1935		 * already changed (i.e. wakeup has already occured, but we
1936		 * weren't waiting yet).
1937		 */
1938		if (params.state == last_state) {
1939			/* XXX KDM cv_wait_sig instead? */
1940			cv_wait(&params.sem, &params.ioctl_mtx);
1941		}
1942		last_state = params.state;
1943
1944		switch (params.state) {
1945		case CTL_IOCTL_INPROG:
1946			/* Why did we wake up? */
1947			/* XXX KDM error here? */
1948			mtx_unlock(&params.ioctl_mtx);
1949			break;
1950		case CTL_IOCTL_DATAMOVE:
1951			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1952
1953			/*
1954			 * change last_state back to INPROG to avoid
1955			 * deadlock on subsequent data moves.
1956			 */
1957			params.state = last_state = CTL_IOCTL_INPROG;
1958
1959			mtx_unlock(&params.ioctl_mtx);
1960			ctl_ioctl_do_datamove(&io->scsiio);
1961			/*
1962			 * Note that in some cases, most notably writes,
1963			 * this will queue the I/O and call us back later.
1964			 * In other cases, generally reads, this routine
1965			 * will immediately call back and wake us up,
1966			 * probably using our own context.
1967			 */
1968			io->scsiio.be_move_done(io);
1969			break;
1970		case CTL_IOCTL_DONE:
1971			mtx_unlock(&params.ioctl_mtx);
1972			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1973			done = 1;
1974			break;
1975		default:
1976			mtx_unlock(&params.ioctl_mtx);
1977			/* XXX KDM error here? */
1978			break;
1979		}
1980	} while (done == 0);
1981
1982	mtx_destroy(&params.ioctl_mtx);
1983	cv_destroy(&params.sem);
1984
1985	return (CTL_RETVAL_COMPLETE);
1986}
1987
1988static void
1989ctl_ioctl_datamove(union ctl_io *io)
1990{
1991	struct ctl_fe_ioctl_params *params;
1992
1993	params = (struct ctl_fe_ioctl_params *)
1994		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1995
1996	mtx_lock(&params->ioctl_mtx);
1997	params->state = CTL_IOCTL_DATAMOVE;
1998	cv_broadcast(&params->sem);
1999	mtx_unlock(&params->ioctl_mtx);
2000}
2001
2002static void
2003ctl_ioctl_done(union ctl_io *io)
2004{
2005	struct ctl_fe_ioctl_params *params;
2006
2007	params = (struct ctl_fe_ioctl_params *)
2008		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2009
2010	mtx_lock(&params->ioctl_mtx);
2011	params->state = CTL_IOCTL_DONE;
2012	cv_broadcast(&params->sem);
2013	mtx_unlock(&params->ioctl_mtx);
2014}
2015
2016static void
2017ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2018{
2019	struct ctl_fe_ioctl_startstop_info *sd_info;
2020
2021	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2022
2023	sd_info->hs_info.status = metatask->status;
2024	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2025	sd_info->hs_info.luns_complete =
2026		metatask->taskinfo.startstop.luns_complete;
2027	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2028
2029	cv_broadcast(&sd_info->sem);
2030}
2031
2032static void
2033ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2034{
2035	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2036
2037	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2038
2039	mtx_lock(fe_bbr_info->lock);
2040	fe_bbr_info->bbr_info->status = metatask->status;
2041	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2042	fe_bbr_info->wakeup_done = 1;
2043	mtx_unlock(fe_bbr_info->lock);
2044
2045	cv_broadcast(&fe_bbr_info->sem);
2046}
2047
2048/*
2049 * Returns 0 for success, errno for failure.
2050 */
2051static int
2052ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2053		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2054{
2055	union ctl_io *io;
2056	int retval;
2057
2058	retval = 0;
2059
2060	mtx_lock(&lun->lun_lock);
2061	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2062	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2063	     ooa_links)) {
2064		struct ctl_ooa_entry *entry;
2065
2066		/*
2067		 * If we've got more than we can fit, just count the
2068		 * remaining entries.
2069		 */
2070		if (*cur_fill_num >= ooa_hdr->alloc_num)
2071			continue;
2072
2073		entry = &kern_entries[*cur_fill_num];
2074
2075		entry->tag_num = io->scsiio.tag_num;
2076		entry->lun_num = lun->lun;
2077#ifdef CTL_TIME_IO
2078		entry->start_bt = io->io_hdr.start_bt;
2079#endif
2080		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2081		entry->cdb_len = io->scsiio.cdb_len;
2082		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2083			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2084
2085		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2086			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2087
2088		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2089			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2090
2091		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2092			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2093
2094		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2095			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2096	}
2097	mtx_unlock(&lun->lun_lock);
2098
2099	return (retval);
2100}
2101
2102static void *
2103ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2104		 size_t error_str_len)
2105{
2106	void *kptr;
2107
2108	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2109
2110	if (copyin(user_addr, kptr, len) != 0) {
2111		snprintf(error_str, error_str_len, "Error copying %d bytes "
2112			 "from user address %p to kernel address %p", len,
2113			 user_addr, kptr);
2114		free(kptr, M_CTL);
2115		return (NULL);
2116	}
2117
2118	return (kptr);
2119}
2120
2121static void
2122ctl_free_args(int num_args, struct ctl_be_arg *args)
2123{
2124	int i;
2125
2126	if (args == NULL)
2127		return;
2128
2129	for (i = 0; i < num_args; i++) {
2130		free(args[i].kname, M_CTL);
2131		free(args[i].kvalue, M_CTL);
2132	}
2133
2134	free(args, M_CTL);
2135}
2136
2137static struct ctl_be_arg *
2138ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2139		char *error_str, size_t error_str_len)
2140{
2141	struct ctl_be_arg *args;
2142	int i;
2143
2144	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2145				error_str, error_str_len);
2146
2147	if (args == NULL)
2148		goto bailout;
2149
2150	for (i = 0; i < num_args; i++) {
2151		args[i].kname = NULL;
2152		args[i].kvalue = NULL;
2153	}
2154
2155	for (i = 0; i < num_args; i++) {
2156		uint8_t *tmpptr;
2157
2158		args[i].kname = ctl_copyin_alloc(args[i].name,
2159			args[i].namelen, error_str, error_str_len);
2160		if (args[i].kname == NULL)
2161			goto bailout;
2162
2163		if (args[i].kname[args[i].namelen - 1] != '\0') {
2164			snprintf(error_str, error_str_len, "Argument %d "
2165				 "name is not NUL-terminated", i);
2166			goto bailout;
2167		}
2168
2169		if (args[i].flags & CTL_BEARG_RD) {
2170			tmpptr = ctl_copyin_alloc(args[i].value,
2171				args[i].vallen, error_str, error_str_len);
2172			if (tmpptr == NULL)
2173				goto bailout;
2174			if ((args[i].flags & CTL_BEARG_ASCII)
2175			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2176				snprintf(error_str, error_str_len, "Argument "
2177				    "%d value is not NUL-terminated", i);
2178				goto bailout;
2179			}
2180			args[i].kvalue = tmpptr;
2181		} else {
2182			args[i].kvalue = malloc(args[i].vallen,
2183			    M_CTL, M_WAITOK | M_ZERO);
2184		}
2185	}
2186
2187	return (args);
2188bailout:
2189
2190	ctl_free_args(num_args, args);
2191
2192	return (NULL);
2193}
2194
2195static void
2196ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2197{
2198	int i;
2199
2200	for (i = 0; i < num_args; i++) {
2201		if (args[i].flags & CTL_BEARG_WR)
2202			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2203	}
2204}
2205
2206/*
2207 * Escape characters that are illegal or not recommended in XML.
2208 */
2209int
2210ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2211{
2212	int retval;
2213
2214	retval = 0;
2215
2216	for (; *str; str++) {
2217		switch (*str) {
2218		case '&':
2219			retval = sbuf_printf(sb, "&amp;");
2220			break;
2221		case '>':
2222			retval = sbuf_printf(sb, "&gt;");
2223			break;
2224		case '<':
2225			retval = sbuf_printf(sb, "&lt;");
2226			break;
2227		default:
2228			retval = sbuf_putc(sb, *str);
2229			break;
2230		}
2231
2232		if (retval != 0)
2233			break;
2234
2235	}
2236
2237	return (retval);
2238}
2239
2240static int
2241ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2242	  struct thread *td)
2243{
2244	struct ctl_softc *softc;
2245	int retval;
2246
2247	softc = control_softc;
2248
2249	retval = 0;
2250
2251	switch (cmd) {
2252	case CTL_IO: {
2253		union ctl_io *io;
2254		void *pool_tmp;
2255
2256		/*
2257		 * If we haven't been "enabled", don't allow any SCSI I/O
2258		 * to this FETD.
2259		 */
2260		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2261			retval = EPERM;
2262			break;
2263		}
2264
2265		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2266		if (io == NULL) {
2267			printf("ctl_ioctl: can't allocate ctl_io!\n");
2268			retval = ENOSPC;
2269			break;
2270		}
2271
2272		/*
2273		 * Need to save the pool reference so it doesn't get
2274		 * spammed by the user's ctl_io.
2275		 */
2276		pool_tmp = io->io_hdr.pool;
2277
2278		memcpy(io, (void *)addr, sizeof(*io));
2279
2280		io->io_hdr.pool = pool_tmp;
2281		/*
2282		 * No status yet, so make sure the status is set properly.
2283		 */
2284		io->io_hdr.status = CTL_STATUS_NONE;
2285
2286		/*
2287		 * The user sets the initiator ID, target and LUN IDs.
2288		 */
2289		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2290		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2291		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2292		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2293			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2294
2295		retval = ctl_ioctl_submit_wait(io);
2296
2297		if (retval != 0) {
2298			ctl_free_io(io);
2299			break;
2300		}
2301
2302		memcpy((void *)addr, io, sizeof(*io));
2303
2304		/* return this to our pool */
2305		ctl_free_io(io);
2306
2307		break;
2308	}
2309	case CTL_ENABLE_PORT:
2310	case CTL_DISABLE_PORT:
2311	case CTL_SET_PORT_WWNS: {
2312		struct ctl_port *port;
2313		struct ctl_port_entry *entry;
2314
2315		entry = (struct ctl_port_entry *)addr;
2316
2317		mtx_lock(&softc->ctl_lock);
2318		STAILQ_FOREACH(port, &softc->port_list, links) {
2319			int action, done;
2320
2321			action = 0;
2322			done = 0;
2323
2324			if ((entry->port_type == CTL_PORT_NONE)
2325			 && (entry->targ_port == port->targ_port)) {
2326				/*
2327				 * If the user only wants to enable or
2328				 * disable or set WWNs on a specific port,
2329				 * do the operation and we're done.
2330				 */
2331				action = 1;
2332				done = 1;
2333			} else if (entry->port_type & port->port_type) {
2334				/*
2335				 * Compare the user's type mask with the
2336				 * particular frontend type to see if we
2337				 * have a match.
2338				 */
2339				action = 1;
2340				done = 0;
2341
2342				/*
2343				 * Make sure the user isn't trying to set
2344				 * WWNs on multiple ports at the same time.
2345				 */
2346				if (cmd == CTL_SET_PORT_WWNS) {
2347					printf("%s: Can't set WWNs on "
2348					       "multiple ports\n", __func__);
2349					retval = EINVAL;
2350					break;
2351				}
2352			}
2353			if (action != 0) {
2354				/*
2355				 * XXX KDM we have to drop the lock here,
2356				 * because the online/offline operations
2357				 * can potentially block.  We need to
2358				 * reference count the frontends so they
2359				 * can't go away,
2360				 */
2361				mtx_unlock(&softc->ctl_lock);
2362
2363				if (cmd == CTL_ENABLE_PORT) {
2364					struct ctl_lun *lun;
2365
2366					STAILQ_FOREACH(lun, &softc->lun_list,
2367						       links) {
2368						port->lun_enable(port->targ_lun_arg,
2369						    lun->target,
2370						    lun->lun);
2371					}
2372
2373					ctl_port_online(port);
2374				} else if (cmd == CTL_DISABLE_PORT) {
2375					struct ctl_lun *lun;
2376
2377					ctl_port_offline(port);
2378
2379					STAILQ_FOREACH(lun, &softc->lun_list,
2380						       links) {
2381						port->lun_disable(
2382						    port->targ_lun_arg,
2383						    lun->target,
2384						    lun->lun);
2385					}
2386				}
2387
2388				mtx_lock(&softc->ctl_lock);
2389
2390				if (cmd == CTL_SET_PORT_WWNS)
2391					ctl_port_set_wwns(port,
2392					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2393					    1 : 0, entry->wwnn,
2394					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2395					    1 : 0, entry->wwpn);
2396			}
2397			if (done != 0)
2398				break;
2399		}
2400		mtx_unlock(&softc->ctl_lock);
2401		break;
2402	}
2403	case CTL_GET_PORT_LIST: {
2404		struct ctl_port *port;
2405		struct ctl_port_list *list;
2406		int i;
2407
2408		list = (struct ctl_port_list *)addr;
2409
2410		if (list->alloc_len != (list->alloc_num *
2411		    sizeof(struct ctl_port_entry))) {
2412			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2413			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2414			       "%zu\n", __func__, list->alloc_len,
2415			       list->alloc_num, sizeof(struct ctl_port_entry));
2416			retval = EINVAL;
2417			break;
2418		}
2419		list->fill_len = 0;
2420		list->fill_num = 0;
2421		list->dropped_num = 0;
2422		i = 0;
2423		mtx_lock(&softc->ctl_lock);
2424		STAILQ_FOREACH(port, &softc->port_list, links) {
2425			struct ctl_port_entry entry, *list_entry;
2426
2427			if (list->fill_num >= list->alloc_num) {
2428				list->dropped_num++;
2429				continue;
2430			}
2431
2432			entry.port_type = port->port_type;
2433			strlcpy(entry.port_name, port->port_name,
2434				sizeof(entry.port_name));
2435			entry.targ_port = port->targ_port;
2436			entry.physical_port = port->physical_port;
2437			entry.virtual_port = port->virtual_port;
2438			entry.wwnn = port->wwnn;
2439			entry.wwpn = port->wwpn;
2440			if (port->status & CTL_PORT_STATUS_ONLINE)
2441				entry.online = 1;
2442			else
2443				entry.online = 0;
2444
2445			list_entry = &list->entries[i];
2446
2447			retval = copyout(&entry, list_entry, sizeof(entry));
2448			if (retval != 0) {
2449				printf("%s: CTL_GET_PORT_LIST: copyout "
2450				       "returned %d\n", __func__, retval);
2451				break;
2452			}
2453			i++;
2454			list->fill_num++;
2455			list->fill_len += sizeof(entry);
2456		}
2457		mtx_unlock(&softc->ctl_lock);
2458
2459		/*
2460		 * If this is non-zero, we had a copyout fault, so there's
2461		 * probably no point in attempting to set the status inside
2462		 * the structure.
2463		 */
2464		if (retval != 0)
2465			break;
2466
2467		if (list->dropped_num > 0)
2468			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2469		else
2470			list->status = CTL_PORT_LIST_OK;
2471		break;
2472	}
2473	case CTL_DUMP_OOA: {
2474		struct ctl_lun *lun;
2475		union ctl_io *io;
2476		char printbuf[128];
2477		struct sbuf sb;
2478
2479		mtx_lock(&softc->ctl_lock);
2480		printf("Dumping OOA queues:\n");
2481		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2482			mtx_lock(&lun->lun_lock);
2483			for (io = (union ctl_io *)TAILQ_FIRST(
2484			     &lun->ooa_queue); io != NULL;
2485			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2486			     ooa_links)) {
2487				sbuf_new(&sb, printbuf, sizeof(printbuf),
2488					 SBUF_FIXEDLEN);
2489				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2490					    (intmax_t)lun->lun,
2491					    io->scsiio.tag_num,
2492					    (io->io_hdr.flags &
2493					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2494					    (io->io_hdr.flags &
2495					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2496					    (io->io_hdr.flags &
2497					    CTL_FLAG_ABORT) ? " ABORT" : "",
2498			                    (io->io_hdr.flags &
2499		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2500				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2501				sbuf_finish(&sb);
2502				printf("%s\n", sbuf_data(&sb));
2503			}
2504			mtx_unlock(&lun->lun_lock);
2505		}
2506		printf("OOA queues dump done\n");
2507		mtx_unlock(&softc->ctl_lock);
2508		break;
2509	}
2510	case CTL_GET_OOA: {
2511		struct ctl_lun *lun;
2512		struct ctl_ooa *ooa_hdr;
2513		struct ctl_ooa_entry *entries;
2514		uint32_t cur_fill_num;
2515
2516		ooa_hdr = (struct ctl_ooa *)addr;
2517
2518		if ((ooa_hdr->alloc_len == 0)
2519		 || (ooa_hdr->alloc_num == 0)) {
2520			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2521			       "must be non-zero\n", __func__,
2522			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2523			retval = EINVAL;
2524			break;
2525		}
2526
2527		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2528		    sizeof(struct ctl_ooa_entry))) {
2529			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2530			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2531			       __func__, ooa_hdr->alloc_len,
2532			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2533			retval = EINVAL;
2534			break;
2535		}
2536
2537		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2538		if (entries == NULL) {
2539			printf("%s: could not allocate %d bytes for OOA "
2540			       "dump\n", __func__, ooa_hdr->alloc_len);
2541			retval = ENOMEM;
2542			break;
2543		}
2544
2545		mtx_lock(&softc->ctl_lock);
2546		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2547		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2548		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2549			mtx_unlock(&softc->ctl_lock);
2550			free(entries, M_CTL);
2551			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2552			       __func__, (uintmax_t)ooa_hdr->lun_num);
2553			retval = EINVAL;
2554			break;
2555		}
2556
2557		cur_fill_num = 0;
2558
2559		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2560			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2561				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2562					ooa_hdr, entries);
2563				if (retval != 0)
2564					break;
2565			}
2566			if (retval != 0) {
2567				mtx_unlock(&softc->ctl_lock);
2568				free(entries, M_CTL);
2569				break;
2570			}
2571		} else {
2572			lun = softc->ctl_luns[ooa_hdr->lun_num];
2573
2574			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2575						    entries);
2576		}
2577		mtx_unlock(&softc->ctl_lock);
2578
2579		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2580		ooa_hdr->fill_len = ooa_hdr->fill_num *
2581			sizeof(struct ctl_ooa_entry);
2582		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2583		if (retval != 0) {
2584			printf("%s: error copying out %d bytes for OOA dump\n",
2585			       __func__, ooa_hdr->fill_len);
2586		}
2587
2588		getbintime(&ooa_hdr->cur_bt);
2589
2590		if (cur_fill_num > ooa_hdr->alloc_num) {
2591			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2592			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2593		} else {
2594			ooa_hdr->dropped_num = 0;
2595			ooa_hdr->status = CTL_OOA_OK;
2596		}
2597
2598		free(entries, M_CTL);
2599		break;
2600	}
2601	case CTL_CHECK_OOA: {
2602		union ctl_io *io;
2603		struct ctl_lun *lun;
2604		struct ctl_ooa_info *ooa_info;
2605
2606
2607		ooa_info = (struct ctl_ooa_info *)addr;
2608
2609		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2610			ooa_info->status = CTL_OOA_INVALID_LUN;
2611			break;
2612		}
2613		mtx_lock(&softc->ctl_lock);
2614		lun = softc->ctl_luns[ooa_info->lun_id];
2615		if (lun == NULL) {
2616			mtx_unlock(&softc->ctl_lock);
2617			ooa_info->status = CTL_OOA_INVALID_LUN;
2618			break;
2619		}
2620		mtx_lock(&lun->lun_lock);
2621		mtx_unlock(&softc->ctl_lock);
2622		ooa_info->num_entries = 0;
2623		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2624		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2625		     &io->io_hdr, ooa_links)) {
2626			ooa_info->num_entries++;
2627		}
2628		mtx_unlock(&lun->lun_lock);
2629
2630		ooa_info->status = CTL_OOA_SUCCESS;
2631
2632		break;
2633	}
2634	case CTL_HARD_START:
2635	case CTL_HARD_STOP: {
2636		struct ctl_fe_ioctl_startstop_info ss_info;
2637		struct cfi_metatask *metatask;
2638		struct mtx hs_mtx;
2639
2640		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2641
2642		cv_init(&ss_info.sem, "hard start/stop cv" );
2643
2644		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2645		if (metatask == NULL) {
2646			retval = ENOMEM;
2647			mtx_destroy(&hs_mtx);
2648			break;
2649		}
2650
2651		if (cmd == CTL_HARD_START)
2652			metatask->tasktype = CFI_TASK_STARTUP;
2653		else
2654			metatask->tasktype = CFI_TASK_SHUTDOWN;
2655
2656		metatask->callback = ctl_ioctl_hard_startstop_callback;
2657		metatask->callback_arg = &ss_info;
2658
2659		cfi_action(metatask);
2660
2661		/* Wait for the callback */
2662		mtx_lock(&hs_mtx);
2663		cv_wait_sig(&ss_info.sem, &hs_mtx);
2664		mtx_unlock(&hs_mtx);
2665
2666		/*
2667		 * All information has been copied from the metatask by the
2668		 * time cv_broadcast() is called, so we free the metatask here.
2669		 */
2670		cfi_free_metatask(metatask);
2671
2672		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2673
2674		mtx_destroy(&hs_mtx);
2675		break;
2676	}
2677	case CTL_BBRREAD: {
2678		struct ctl_bbrread_info *bbr_info;
2679		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2680		struct mtx bbr_mtx;
2681		struct cfi_metatask *metatask;
2682
2683		bbr_info = (struct ctl_bbrread_info *)addr;
2684
2685		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2686
2687		bzero(&bbr_mtx, sizeof(bbr_mtx));
2688		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2689
2690		fe_bbr_info.bbr_info = bbr_info;
2691		fe_bbr_info.lock = &bbr_mtx;
2692
2693		cv_init(&fe_bbr_info.sem, "BBR read cv");
2694		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2695
2696		if (metatask == NULL) {
2697			mtx_destroy(&bbr_mtx);
2698			cv_destroy(&fe_bbr_info.sem);
2699			retval = ENOMEM;
2700			break;
2701		}
2702		metatask->tasktype = CFI_TASK_BBRREAD;
2703		metatask->callback = ctl_ioctl_bbrread_callback;
2704		metatask->callback_arg = &fe_bbr_info;
2705		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2706		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2707		metatask->taskinfo.bbrread.len = bbr_info->len;
2708
2709		cfi_action(metatask);
2710
2711		mtx_lock(&bbr_mtx);
2712		while (fe_bbr_info.wakeup_done == 0)
2713			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2714		mtx_unlock(&bbr_mtx);
2715
2716		bbr_info->status = metatask->status;
2717		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2718		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2719		memcpy(&bbr_info->sense_data,
2720		       &metatask->taskinfo.bbrread.sense_data,
2721		       ctl_min(sizeof(bbr_info->sense_data),
2722			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2723
2724		cfi_free_metatask(metatask);
2725
2726		mtx_destroy(&bbr_mtx);
2727		cv_destroy(&fe_bbr_info.sem);
2728
2729		break;
2730	}
2731	case CTL_DELAY_IO: {
2732		struct ctl_io_delay_info *delay_info;
2733#ifdef CTL_IO_DELAY
2734		struct ctl_lun *lun;
2735#endif /* CTL_IO_DELAY */
2736
2737		delay_info = (struct ctl_io_delay_info *)addr;
2738
2739#ifdef CTL_IO_DELAY
2740		mtx_lock(&softc->ctl_lock);
2741
2742		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2743		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2744			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2745		} else {
2746			lun = softc->ctl_luns[delay_info->lun_id];
2747			mtx_lock(&lun->lun_lock);
2748
2749			delay_info->status = CTL_DELAY_STATUS_OK;
2750
2751			switch (delay_info->delay_type) {
2752			case CTL_DELAY_TYPE_CONT:
2753				break;
2754			case CTL_DELAY_TYPE_ONESHOT:
2755				break;
2756			default:
2757				delay_info->status =
2758					CTL_DELAY_STATUS_INVALID_TYPE;
2759				break;
2760			}
2761
2762			switch (delay_info->delay_loc) {
2763			case CTL_DELAY_LOC_DATAMOVE:
2764				lun->delay_info.datamove_type =
2765					delay_info->delay_type;
2766				lun->delay_info.datamove_delay =
2767					delay_info->delay_secs;
2768				break;
2769			case CTL_DELAY_LOC_DONE:
2770				lun->delay_info.done_type =
2771					delay_info->delay_type;
2772				lun->delay_info.done_delay =
2773					delay_info->delay_secs;
2774				break;
2775			default:
2776				delay_info->status =
2777					CTL_DELAY_STATUS_INVALID_LOC;
2778				break;
2779			}
2780			mtx_unlock(&lun->lun_lock);
2781		}
2782
2783		mtx_unlock(&softc->ctl_lock);
2784#else
2785		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2786#endif /* CTL_IO_DELAY */
2787		break;
2788	}
2789	case CTL_REALSYNC_SET: {
2790		int *syncstate;
2791
2792		syncstate = (int *)addr;
2793
2794		mtx_lock(&softc->ctl_lock);
2795		switch (*syncstate) {
2796		case 0:
2797			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2798			break;
2799		case 1:
2800			softc->flags |= CTL_FLAG_REAL_SYNC;
2801			break;
2802		default:
2803			retval = EINVAL;
2804			break;
2805		}
2806		mtx_unlock(&softc->ctl_lock);
2807		break;
2808	}
2809	case CTL_REALSYNC_GET: {
2810		int *syncstate;
2811
2812		syncstate = (int*)addr;
2813
2814		mtx_lock(&softc->ctl_lock);
2815		if (softc->flags & CTL_FLAG_REAL_SYNC)
2816			*syncstate = 1;
2817		else
2818			*syncstate = 0;
2819		mtx_unlock(&softc->ctl_lock);
2820
2821		break;
2822	}
2823	case CTL_SETSYNC:
2824	case CTL_GETSYNC: {
2825		struct ctl_sync_info *sync_info;
2826		struct ctl_lun *lun;
2827
2828		sync_info = (struct ctl_sync_info *)addr;
2829
2830		mtx_lock(&softc->ctl_lock);
2831		lun = softc->ctl_luns[sync_info->lun_id];
2832		if (lun == NULL) {
2833			mtx_unlock(&softc->ctl_lock);
2834			sync_info->status = CTL_GS_SYNC_NO_LUN;
2835		}
2836		/*
2837		 * Get or set the sync interval.  We're not bounds checking
2838		 * in the set case, hopefully the user won't do something
2839		 * silly.
2840		 */
2841		mtx_lock(&lun->lun_lock);
2842		mtx_unlock(&softc->ctl_lock);
2843		if (cmd == CTL_GETSYNC)
2844			sync_info->sync_interval = lun->sync_interval;
2845		else
2846			lun->sync_interval = sync_info->sync_interval;
2847		mtx_unlock(&lun->lun_lock);
2848
2849		sync_info->status = CTL_GS_SYNC_OK;
2850
2851		break;
2852	}
2853	case CTL_GETSTATS: {
2854		struct ctl_stats *stats;
2855		struct ctl_lun *lun;
2856		int i;
2857
2858		stats = (struct ctl_stats *)addr;
2859
2860		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2861		     stats->alloc_len) {
2862			stats->status = CTL_SS_NEED_MORE_SPACE;
2863			stats->num_luns = softc->num_luns;
2864			break;
2865		}
2866		/*
2867		 * XXX KDM no locking here.  If the LUN list changes,
2868		 * things can blow up.
2869		 */
2870		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2871		     i++, lun = STAILQ_NEXT(lun, links)) {
2872			retval = copyout(&lun->stats, &stats->lun_stats[i],
2873					 sizeof(lun->stats));
2874			if (retval != 0)
2875				break;
2876		}
2877		stats->num_luns = softc->num_luns;
2878		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2879				 softc->num_luns;
2880		stats->status = CTL_SS_OK;
2881#ifdef CTL_TIME_IO
2882		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2883#else
2884		stats->flags = CTL_STATS_FLAG_NONE;
2885#endif
2886		getnanouptime(&stats->timestamp);
2887		break;
2888	}
2889	case CTL_ERROR_INJECT: {
2890		struct ctl_error_desc *err_desc, *new_err_desc;
2891		struct ctl_lun *lun;
2892
2893		err_desc = (struct ctl_error_desc *)addr;
2894
2895		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2896				      M_WAITOK | M_ZERO);
2897		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2898
2899		mtx_lock(&softc->ctl_lock);
2900		lun = softc->ctl_luns[err_desc->lun_id];
2901		if (lun == NULL) {
2902			mtx_unlock(&softc->ctl_lock);
2903			free(new_err_desc, M_CTL);
2904			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2905			       __func__, (uintmax_t)err_desc->lun_id);
2906			retval = EINVAL;
2907			break;
2908		}
2909		mtx_lock(&lun->lun_lock);
2910		mtx_unlock(&softc->ctl_lock);
2911
2912		/*
2913		 * We could do some checking here to verify the validity
2914		 * of the request, but given the complexity of error
2915		 * injection requests, the checking logic would be fairly
2916		 * complex.
2917		 *
2918		 * For now, if the request is invalid, it just won't get
2919		 * executed and might get deleted.
2920		 */
2921		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2922
2923		/*
2924		 * XXX KDM check to make sure the serial number is unique,
2925		 * in case we somehow manage to wrap.  That shouldn't
2926		 * happen for a very long time, but it's the right thing to
2927		 * do.
2928		 */
2929		new_err_desc->serial = lun->error_serial;
2930		err_desc->serial = lun->error_serial;
2931		lun->error_serial++;
2932
2933		mtx_unlock(&lun->lun_lock);
2934		break;
2935	}
2936	case CTL_ERROR_INJECT_DELETE: {
2937		struct ctl_error_desc *delete_desc, *desc, *desc2;
2938		struct ctl_lun *lun;
2939		int delete_done;
2940
2941		delete_desc = (struct ctl_error_desc *)addr;
2942		delete_done = 0;
2943
2944		mtx_lock(&softc->ctl_lock);
2945		lun = softc->ctl_luns[delete_desc->lun_id];
2946		if (lun == NULL) {
2947			mtx_unlock(&softc->ctl_lock);
2948			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2949			       __func__, (uintmax_t)delete_desc->lun_id);
2950			retval = EINVAL;
2951			break;
2952		}
2953		mtx_lock(&lun->lun_lock);
2954		mtx_unlock(&softc->ctl_lock);
2955		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2956			if (desc->serial != delete_desc->serial)
2957				continue;
2958
2959			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2960				      links);
2961			free(desc, M_CTL);
2962			delete_done = 1;
2963		}
2964		mtx_unlock(&lun->lun_lock);
2965		if (delete_done == 0) {
2966			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2967			       "error serial %ju on LUN %u\n", __func__,
2968			       delete_desc->serial, delete_desc->lun_id);
2969			retval = EINVAL;
2970			break;
2971		}
2972		break;
2973	}
2974	case CTL_DUMP_STRUCTS: {
2975		int i, j, k, idx;
2976		struct ctl_port *port;
2977		struct ctl_frontend *fe;
2978
2979		mtx_lock(&softc->ctl_lock);
2980		printf("CTL Persistent Reservation information start:\n");
2981		for (i = 0; i < CTL_MAX_LUNS; i++) {
2982			struct ctl_lun *lun;
2983
2984			lun = softc->ctl_luns[i];
2985
2986			if ((lun == NULL)
2987			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2988				continue;
2989
2990			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2991				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2992					idx = j * CTL_MAX_INIT_PER_PORT + k;
2993					if (lun->per_res[idx].registered == 0)
2994						continue;
2995					printf("  LUN %d port %d iid %d key "
2996					       "%#jx\n", i, j, k,
2997					       (uintmax_t)scsi_8btou64(
2998					       lun->per_res[idx].res_key.key));
2999				}
3000			}
3001		}
3002		printf("CTL Persistent Reservation information end\n");
3003		printf("CTL Ports:\n");
3004		STAILQ_FOREACH(port, &softc->port_list, links) {
3005			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3006			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3007			       port->frontend->name, port->port_type,
3008			       port->physical_port, port->virtual_port,
3009			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3010			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3011				if (port->wwpn_iid[j].in_use == 0 &&
3012				    port->wwpn_iid[j].wwpn == 0 &&
3013				    port->wwpn_iid[j].name == NULL)
3014					continue;
3015
3016				printf("    iid %u use %d WWPN %#jx '%s'\n",
3017				    j, port->wwpn_iid[j].in_use,
3018				    (uintmax_t)port->wwpn_iid[j].wwpn,
3019				    port->wwpn_iid[j].name);
3020			}
3021		}
3022		printf("CTL Port information end\n");
3023		mtx_unlock(&softc->ctl_lock);
3024		/*
3025		 * XXX KDM calling this without a lock.  We'd likely want
3026		 * to drop the lock before calling the frontend's dump
3027		 * routine anyway.
3028		 */
3029		printf("CTL Frontends:\n");
3030		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3031			printf("  Frontend '%s'\n", fe->name);
3032			if (fe->fe_dump != NULL)
3033				fe->fe_dump();
3034		}
3035		printf("CTL Frontend information end\n");
3036		break;
3037	}
3038	case CTL_LUN_REQ: {
3039		struct ctl_lun_req *lun_req;
3040		struct ctl_backend_driver *backend;
3041
3042		lun_req = (struct ctl_lun_req *)addr;
3043
3044		backend = ctl_backend_find(lun_req->backend);
3045		if (backend == NULL) {
3046			lun_req->status = CTL_LUN_ERROR;
3047			snprintf(lun_req->error_str,
3048				 sizeof(lun_req->error_str),
3049				 "Backend \"%s\" not found.",
3050				 lun_req->backend);
3051			break;
3052		}
3053		if (lun_req->num_be_args > 0) {
3054			lun_req->kern_be_args = ctl_copyin_args(
3055				lun_req->num_be_args,
3056				lun_req->be_args,
3057				lun_req->error_str,
3058				sizeof(lun_req->error_str));
3059			if (lun_req->kern_be_args == NULL) {
3060				lun_req->status = CTL_LUN_ERROR;
3061				break;
3062			}
3063		}
3064
3065		retval = backend->ioctl(dev, cmd, addr, flag, td);
3066
3067		if (lun_req->num_be_args > 0) {
3068			ctl_copyout_args(lun_req->num_be_args,
3069				      lun_req->kern_be_args);
3070			ctl_free_args(lun_req->num_be_args,
3071				      lun_req->kern_be_args);
3072		}
3073		break;
3074	}
3075	case CTL_LUN_LIST: {
3076		struct sbuf *sb;
3077		struct ctl_lun *lun;
3078		struct ctl_lun_list *list;
3079		struct ctl_option *opt;
3080
3081		list = (struct ctl_lun_list *)addr;
3082
3083		/*
3084		 * Allocate a fixed length sbuf here, based on the length
3085		 * of the user's buffer.  We could allocate an auto-extending
3086		 * buffer, and then tell the user how much larger our
3087		 * amount of data is than his buffer, but that presents
3088		 * some problems:
3089		 *
3090		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3091		 *     we can't hold a lock while calling them with an
3092		 *     auto-extending buffer.
3093 		 *
3094		 * 2.  There is not currently a LUN reference counting
3095		 *     mechanism, outside of outstanding transactions on
3096		 *     the LUN's OOA queue.  So a LUN could go away on us
3097		 *     while we're getting the LUN number, backend-specific
3098		 *     information, etc.  Thus, given the way things
3099		 *     currently work, we need to hold the CTL lock while
3100		 *     grabbing LUN information.
3101		 *
3102		 * So, from the user's standpoint, the best thing to do is
3103		 * allocate what he thinks is a reasonable buffer length,
3104		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3105		 * double the buffer length and try again.  (And repeat
3106		 * that until he succeeds.)
3107		 */
3108		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3109		if (sb == NULL) {
3110			list->status = CTL_LUN_LIST_ERROR;
3111			snprintf(list->error_str, sizeof(list->error_str),
3112				 "Unable to allocate %d bytes for LUN list",
3113				 list->alloc_len);
3114			break;
3115		}
3116
3117		sbuf_printf(sb, "<ctllunlist>\n");
3118
3119		mtx_lock(&softc->ctl_lock);
3120		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3121			mtx_lock(&lun->lun_lock);
3122			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3123					     (uintmax_t)lun->lun);
3124
3125			/*
3126			 * Bail out as soon as we see that we've overfilled
3127			 * the buffer.
3128			 */
3129			if (retval != 0)
3130				break;
3131
3132			retval = sbuf_printf(sb, "\t<backend_type>%s"
3133					     "</backend_type>\n",
3134					     (lun->backend == NULL) ?  "none" :
3135					     lun->backend->name);
3136
3137			if (retval != 0)
3138				break;
3139
3140			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3141					     lun->be_lun->lun_type);
3142
3143			if (retval != 0)
3144				break;
3145
3146			if (lun->backend == NULL) {
3147				retval = sbuf_printf(sb, "</lun>\n");
3148				if (retval != 0)
3149					break;
3150				continue;
3151			}
3152
3153			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3154					     (lun->be_lun->maxlba > 0) ?
3155					     lun->be_lun->maxlba + 1 : 0);
3156
3157			if (retval != 0)
3158				break;
3159
3160			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3161					     lun->be_lun->blocksize);
3162
3163			if (retval != 0)
3164				break;
3165
3166			retval = sbuf_printf(sb, "\t<serial_number>");
3167
3168			if (retval != 0)
3169				break;
3170
3171			retval = ctl_sbuf_printf_esc(sb,
3172						     lun->be_lun->serial_num);
3173
3174			if (retval != 0)
3175				break;
3176
3177			retval = sbuf_printf(sb, "</serial_number>\n");
3178
3179			if (retval != 0)
3180				break;
3181
3182			retval = sbuf_printf(sb, "\t<device_id>");
3183
3184			if (retval != 0)
3185				break;
3186
3187			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3188
3189			if (retval != 0)
3190				break;
3191
3192			retval = sbuf_printf(sb, "</device_id>\n");
3193
3194			if (retval != 0)
3195				break;
3196
3197			if (lun->backend->lun_info != NULL) {
3198				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3199				if (retval != 0)
3200					break;
3201			}
3202			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3203				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3204				    opt->name, opt->value, opt->name);
3205				if (retval != 0)
3206					break;
3207			}
3208
3209			retval = sbuf_printf(sb, "</lun>\n");
3210
3211			if (retval != 0)
3212				break;
3213			mtx_unlock(&lun->lun_lock);
3214		}
3215		if (lun != NULL)
3216			mtx_unlock(&lun->lun_lock);
3217		mtx_unlock(&softc->ctl_lock);
3218
3219		if ((retval != 0)
3220		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3221			retval = 0;
3222			sbuf_delete(sb);
3223			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3224			snprintf(list->error_str, sizeof(list->error_str),
3225				 "Out of space, %d bytes is too small",
3226				 list->alloc_len);
3227			break;
3228		}
3229
3230		sbuf_finish(sb);
3231
3232		retval = copyout(sbuf_data(sb), list->lun_xml,
3233				 sbuf_len(sb) + 1);
3234
3235		list->fill_len = sbuf_len(sb) + 1;
3236		list->status = CTL_LUN_LIST_OK;
3237		sbuf_delete(sb);
3238		break;
3239	}
3240	case CTL_ISCSI: {
3241		struct ctl_iscsi *ci;
3242		struct ctl_frontend *fe;
3243
3244		ci = (struct ctl_iscsi *)addr;
3245
3246		fe = ctl_frontend_find("iscsi");
3247		if (fe == NULL) {
3248			ci->status = CTL_ISCSI_ERROR;
3249			snprintf(ci->error_str, sizeof(ci->error_str),
3250			    "Frontend \"iscsi\" not found.");
3251			break;
3252		}
3253
3254		retval = fe->ioctl(dev, cmd, addr, flag, td);
3255		break;
3256	}
3257	case CTL_PORT_REQ: {
3258		struct ctl_req *req;
3259		struct ctl_frontend *fe;
3260
3261		req = (struct ctl_req *)addr;
3262
3263		fe = ctl_frontend_find(req->driver);
3264		if (fe == NULL) {
3265			req->status = CTL_LUN_ERROR;
3266			snprintf(req->error_str, sizeof(req->error_str),
3267			    "Frontend \"%s\" not found.", req->driver);
3268			break;
3269		}
3270		if (req->num_args > 0) {
3271			req->kern_args = ctl_copyin_args(req->num_args,
3272			    req->args, req->error_str, sizeof(req->error_str));
3273			if (req->kern_args == NULL) {
3274				req->status = CTL_LUN_ERROR;
3275				break;
3276			}
3277		}
3278
3279		retval = fe->ioctl(dev, cmd, addr, flag, td);
3280
3281		if (req->num_args > 0) {
3282			ctl_copyout_args(req->num_args, req->kern_args);
3283			ctl_free_args(req->num_args, req->kern_args);
3284		}
3285		break;
3286	}
3287	case CTL_PORT_LIST: {
3288		struct sbuf *sb;
3289		struct ctl_port *port;
3290		struct ctl_lun_list *list;
3291		struct ctl_option *opt;
3292
3293		list = (struct ctl_lun_list *)addr;
3294
3295		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3296		if (sb == NULL) {
3297			list->status = CTL_LUN_LIST_ERROR;
3298			snprintf(list->error_str, sizeof(list->error_str),
3299				 "Unable to allocate %d bytes for LUN list",
3300				 list->alloc_len);
3301			break;
3302		}
3303
3304		sbuf_printf(sb, "<ctlportlist>\n");
3305
3306		mtx_lock(&softc->ctl_lock);
3307		STAILQ_FOREACH(port, &softc->port_list, links) {
3308			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3309					     (uintmax_t)port->targ_port);
3310
3311			/*
3312			 * Bail out as soon as we see that we've overfilled
3313			 * the buffer.
3314			 */
3315			if (retval != 0)
3316				break;
3317
3318			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3319			    "</frontend_type>\n", port->frontend->name);
3320			if (retval != 0)
3321				break;
3322
3323			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3324					     port->port_type);
3325			if (retval != 0)
3326				break;
3327
3328			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3329			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3330			if (retval != 0)
3331				break;
3332
3333			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3334			    port->port_name);
3335			if (retval != 0)
3336				break;
3337
3338			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3339			    port->physical_port);
3340			if (retval != 0)
3341				break;
3342
3343			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3344			    port->virtual_port);
3345			if (retval != 0)
3346				break;
3347
3348			retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3349			    (uintmax_t)port->wwnn);
3350			if (retval != 0)
3351				break;
3352
3353			retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3354			    (uintmax_t)port->wwpn);
3355			if (retval != 0)
3356				break;
3357
3358			if (port->port_info != NULL) {
3359				retval = port->port_info(port->onoff_arg, sb);
3360				if (retval != 0)
3361					break;
3362			}
3363			STAILQ_FOREACH(opt, &port->options, links) {
3364				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3365				    opt->name, opt->value, opt->name);
3366				if (retval != 0)
3367					break;
3368			}
3369
3370			retval = sbuf_printf(sb, "</targ_port>\n");
3371			if (retval != 0)
3372				break;
3373		}
3374		mtx_unlock(&softc->ctl_lock);
3375
3376		if ((retval != 0)
3377		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3378			retval = 0;
3379			sbuf_delete(sb);
3380			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3381			snprintf(list->error_str, sizeof(list->error_str),
3382				 "Out of space, %d bytes is too small",
3383				 list->alloc_len);
3384			break;
3385		}
3386
3387		sbuf_finish(sb);
3388
3389		retval = copyout(sbuf_data(sb), list->lun_xml,
3390				 sbuf_len(sb) + 1);
3391
3392		list->fill_len = sbuf_len(sb) + 1;
3393		list->status = CTL_LUN_LIST_OK;
3394		sbuf_delete(sb);
3395		break;
3396	}
3397	default: {
3398		/* XXX KDM should we fix this? */
3399#if 0
3400		struct ctl_backend_driver *backend;
3401		unsigned int type;
3402		int found;
3403
3404		found = 0;
3405
3406		/*
3407		 * We encode the backend type as the ioctl type for backend
3408		 * ioctls.  So parse it out here, and then search for a
3409		 * backend of this type.
3410		 */
3411		type = _IOC_TYPE(cmd);
3412
3413		STAILQ_FOREACH(backend, &softc->be_list, links) {
3414			if (backend->type == type) {
3415				found = 1;
3416				break;
3417			}
3418		}
3419		if (found == 0) {
3420			printf("ctl: unknown ioctl command %#lx or backend "
3421			       "%d\n", cmd, type);
3422			retval = EINVAL;
3423			break;
3424		}
3425		retval = backend->ioctl(dev, cmd, addr, flag, td);
3426#endif
3427		retval = ENOTTY;
3428		break;
3429	}
3430	}
3431	return (retval);
3432}
3433
3434uint32_t
3435ctl_get_initindex(struct ctl_nexus *nexus)
3436{
3437	if (nexus->targ_port < CTL_MAX_PORTS)
3438		return (nexus->initid.id +
3439			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3440	else
3441		return (nexus->initid.id +
3442		       ((nexus->targ_port - CTL_MAX_PORTS) *
3443			CTL_MAX_INIT_PER_PORT));
3444}
3445
3446uint32_t
3447ctl_get_resindex(struct ctl_nexus *nexus)
3448{
3449	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3450}
3451
3452uint32_t
3453ctl_port_idx(int port_num)
3454{
3455	if (port_num < CTL_MAX_PORTS)
3456		return(port_num);
3457	else
3458		return(port_num - CTL_MAX_PORTS);
3459}
3460
3461static uint32_t
3462ctl_map_lun(int port_num, uint32_t lun_id)
3463{
3464	struct ctl_port *port;
3465
3466	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3467	if (port == NULL)
3468		return (UINT32_MAX);
3469	if (port->lun_map == NULL)
3470		return (lun_id);
3471	return (port->lun_map(port->targ_lun_arg, lun_id));
3472}
3473
3474static uint32_t
3475ctl_map_lun_back(int port_num, uint32_t lun_id)
3476{
3477	struct ctl_port *port;
3478	uint32_t i;
3479
3480	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3481	if (port->lun_map == NULL)
3482		return (lun_id);
3483	for (i = 0; i < CTL_MAX_LUNS; i++) {
3484		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3485			return (i);
3486	}
3487	return (UINT32_MAX);
3488}
3489
3490/*
3491 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3492 * that are a power of 2.
3493 */
3494int
3495ctl_ffz(uint32_t *mask, uint32_t size)
3496{
3497	uint32_t num_chunks, num_pieces;
3498	int i, j;
3499
3500	num_chunks = (size >> 5);
3501	if (num_chunks == 0)
3502		num_chunks++;
3503	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3504
3505	for (i = 0; i < num_chunks; i++) {
3506		for (j = 0; j < num_pieces; j++) {
3507			if ((mask[i] & (1 << j)) == 0)
3508				return ((i << 5) + j);
3509		}
3510	}
3511
3512	return (-1);
3513}
3514
3515int
3516ctl_set_mask(uint32_t *mask, uint32_t bit)
3517{
3518	uint32_t chunk, piece;
3519
3520	chunk = bit >> 5;
3521	piece = bit % (sizeof(uint32_t) * 8);
3522
3523	if ((mask[chunk] & (1 << piece)) != 0)
3524		return (-1);
3525	else
3526		mask[chunk] |= (1 << piece);
3527
3528	return (0);
3529}
3530
3531int
3532ctl_clear_mask(uint32_t *mask, uint32_t bit)
3533{
3534	uint32_t chunk, piece;
3535
3536	chunk = bit >> 5;
3537	piece = bit % (sizeof(uint32_t) * 8);
3538
3539	if ((mask[chunk] & (1 << piece)) == 0)
3540		return (-1);
3541	else
3542		mask[chunk] &= ~(1 << piece);
3543
3544	return (0);
3545}
3546
3547int
3548ctl_is_set(uint32_t *mask, uint32_t bit)
3549{
3550	uint32_t chunk, piece;
3551
3552	chunk = bit >> 5;
3553	piece = bit % (sizeof(uint32_t) * 8);
3554
3555	if ((mask[chunk] & (1 << piece)) == 0)
3556		return (0);
3557	else
3558		return (1);
3559}
3560
3561#ifdef unused
3562/*
3563 * The bus, target and lun are optional, they can be filled in later.
3564 * can_wait is used to determine whether we can wait on the malloc or not.
3565 */
3566union ctl_io*
3567ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3568	      uint32_t targ_lun, int can_wait)
3569{
3570	union ctl_io *io;
3571
3572	if (can_wait)
3573		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3574	else
3575		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3576
3577	if (io != NULL) {
3578		io->io_hdr.io_type = io_type;
3579		io->io_hdr.targ_port = targ_port;
3580		/*
3581		 * XXX KDM this needs to change/go away.  We need to move
3582		 * to a preallocated pool of ctl_scsiio structures.
3583		 */
3584		io->io_hdr.nexus.targ_target.id = targ_target;
3585		io->io_hdr.nexus.targ_lun = targ_lun;
3586	}
3587
3588	return (io);
3589}
3590
3591void
3592ctl_kfree_io(union ctl_io *io)
3593{
3594	free(io, M_CTL);
3595}
3596#endif /* unused */
3597
3598/*
3599 * ctl_softc, pool_type, total_ctl_io are passed in.
3600 * npool is passed out.
3601 */
3602int
3603ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3604		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3605{
3606	uint32_t i;
3607	union ctl_io *cur_io, *next_io;
3608	struct ctl_io_pool *pool;
3609	int retval;
3610
3611	retval = 0;
3612
3613	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3614					    M_NOWAIT | M_ZERO);
3615	if (pool == NULL) {
3616		retval = ENOMEM;
3617		goto bailout;
3618	}
3619
3620	pool->type = pool_type;
3621	pool->ctl_softc = ctl_softc;
3622
3623	mtx_lock(&ctl_softc->pool_lock);
3624	pool->id = ctl_softc->cur_pool_id++;
3625	mtx_unlock(&ctl_softc->pool_lock);
3626
3627	pool->flags = CTL_POOL_FLAG_NONE;
3628	pool->refcount = 1;		/* Reference for validity. */
3629	STAILQ_INIT(&pool->free_queue);
3630
3631	/*
3632	 * XXX KDM other options here:
3633	 * - allocate a page at a time
3634	 * - allocate one big chunk of memory.
3635	 * Page allocation might work well, but would take a little more
3636	 * tracking.
3637	 */
3638	for (i = 0; i < total_ctl_io; i++) {
3639		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3640						M_NOWAIT);
3641		if (cur_io == NULL) {
3642			retval = ENOMEM;
3643			break;
3644		}
3645		cur_io->io_hdr.pool = pool;
3646		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3647		pool->total_ctl_io++;
3648		pool->free_ctl_io++;
3649	}
3650
3651	if (retval != 0) {
3652		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3653		     cur_io != NULL; cur_io = next_io) {
3654			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3655							      links);
3656			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3657				      ctl_io_hdr, links);
3658			free(cur_io, M_CTLIO);
3659		}
3660
3661		free(pool, M_CTL);
3662		goto bailout;
3663	}
3664	mtx_lock(&ctl_softc->pool_lock);
3665	ctl_softc->num_pools++;
3666	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3667	/*
3668	 * Increment our usage count if this is an external consumer, so we
3669	 * can't get unloaded until the external consumer (most likely a
3670	 * FETD) unloads and frees his pool.
3671	 *
3672	 * XXX KDM will this increment the caller's module use count, or
3673	 * mine?
3674	 */
3675#if 0
3676	if ((pool_type != CTL_POOL_EMERGENCY)
3677	 && (pool_type != CTL_POOL_INTERNAL)
3678	 && (pool_type != CTL_POOL_4OTHERSC))
3679		MOD_INC_USE_COUNT;
3680#endif
3681
3682	mtx_unlock(&ctl_softc->pool_lock);
3683
3684	*npool = pool;
3685
3686bailout:
3687
3688	return (retval);
3689}
3690
3691static int
3692ctl_pool_acquire(struct ctl_io_pool *pool)
3693{
3694
3695	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3696
3697	if (pool->flags & CTL_POOL_FLAG_INVALID)
3698		return (EINVAL);
3699
3700	pool->refcount++;
3701
3702	return (0);
3703}
3704
3705static void
3706ctl_pool_release(struct ctl_io_pool *pool)
3707{
3708	struct ctl_softc *ctl_softc = pool->ctl_softc;
3709	union ctl_io *io;
3710
3711	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3712
3713	if (--pool->refcount != 0)
3714		return;
3715
3716	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3717		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3718			      links);
3719		free(io, M_CTLIO);
3720	}
3721
3722	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3723	ctl_softc->num_pools--;
3724
3725	/*
3726	 * XXX KDM will this decrement the caller's usage count or mine?
3727	 */
3728#if 0
3729	if ((pool->type != CTL_POOL_EMERGENCY)
3730	 && (pool->type != CTL_POOL_INTERNAL)
3731	 && (pool->type != CTL_POOL_4OTHERSC))
3732		MOD_DEC_USE_COUNT;
3733#endif
3734
3735	free(pool, M_CTL);
3736}
3737
3738void
3739ctl_pool_free(struct ctl_io_pool *pool)
3740{
3741	struct ctl_softc *ctl_softc;
3742
3743	if (pool == NULL)
3744		return;
3745
3746	ctl_softc = pool->ctl_softc;
3747	mtx_lock(&ctl_softc->pool_lock);
3748	pool->flags |= CTL_POOL_FLAG_INVALID;
3749	ctl_pool_release(pool);
3750	mtx_unlock(&ctl_softc->pool_lock);
3751}
3752
3753/*
3754 * This routine does not block (except for spinlocks of course).
3755 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3756 * possible.
3757 */
3758union ctl_io *
3759ctl_alloc_io(void *pool_ref)
3760{
3761	union ctl_io *io;
3762	struct ctl_softc *ctl_softc;
3763	struct ctl_io_pool *pool, *npool;
3764	struct ctl_io_pool *emergency_pool;
3765
3766	pool = (struct ctl_io_pool *)pool_ref;
3767
3768	if (pool == NULL) {
3769		printf("%s: pool is NULL\n", __func__);
3770		return (NULL);
3771	}
3772
3773	emergency_pool = NULL;
3774
3775	ctl_softc = pool->ctl_softc;
3776
3777	mtx_lock(&ctl_softc->pool_lock);
3778	/*
3779	 * First, try to get the io structure from the user's pool.
3780	 */
3781	if (ctl_pool_acquire(pool) == 0) {
3782		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3783		if (io != NULL) {
3784			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3785			pool->total_allocated++;
3786			pool->free_ctl_io--;
3787			mtx_unlock(&ctl_softc->pool_lock);
3788			return (io);
3789		} else
3790			ctl_pool_release(pool);
3791	}
3792	/*
3793	 * If he doesn't have any io structures left, search for an
3794	 * emergency pool and grab one from there.
3795	 */
3796	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3797		if (npool->type != CTL_POOL_EMERGENCY)
3798			continue;
3799
3800		if (ctl_pool_acquire(npool) != 0)
3801			continue;
3802
3803		emergency_pool = npool;
3804
3805		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3806		if (io != NULL) {
3807			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3808			npool->total_allocated++;
3809			npool->free_ctl_io--;
3810			mtx_unlock(&ctl_softc->pool_lock);
3811			return (io);
3812		} else
3813			ctl_pool_release(npool);
3814	}
3815
3816	/* Drop the spinlock before we malloc */
3817	mtx_unlock(&ctl_softc->pool_lock);
3818
3819	/*
3820	 * The emergency pool (if it exists) didn't have one, so try an
3821	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3822	 */
3823	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3824	if (io != NULL) {
3825		/*
3826		 * If the emergency pool exists but is empty, add this
3827		 * ctl_io to its list when it gets freed.
3828		 */
3829		if (emergency_pool != NULL) {
3830			mtx_lock(&ctl_softc->pool_lock);
3831			if (ctl_pool_acquire(emergency_pool) == 0) {
3832				io->io_hdr.pool = emergency_pool;
3833				emergency_pool->total_ctl_io++;
3834				/*
3835				 * Need to bump this, otherwise
3836				 * total_allocated and total_freed won't
3837				 * match when we no longer have anything
3838				 * outstanding.
3839				 */
3840				emergency_pool->total_allocated++;
3841			}
3842			mtx_unlock(&ctl_softc->pool_lock);
3843		} else
3844			io->io_hdr.pool = NULL;
3845	}
3846
3847	return (io);
3848}
3849
3850void
3851ctl_free_io(union ctl_io *io)
3852{
3853	if (io == NULL)
3854		return;
3855
3856	/*
3857	 * If this ctl_io has a pool, return it to that pool.
3858	 */
3859	if (io->io_hdr.pool != NULL) {
3860		struct ctl_io_pool *pool;
3861
3862		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3863		mtx_lock(&pool->ctl_softc->pool_lock);
3864		io->io_hdr.io_type = 0xff;
3865		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3866		pool->total_freed++;
3867		pool->free_ctl_io++;
3868		ctl_pool_release(pool);
3869		mtx_unlock(&pool->ctl_softc->pool_lock);
3870	} else {
3871		/*
3872		 * Otherwise, just free it.  We probably malloced it and
3873		 * the emergency pool wasn't available.
3874		 */
3875		free(io, M_CTLIO);
3876	}
3877
3878}
3879
3880void
3881ctl_zero_io(union ctl_io *io)
3882{
3883	void *pool_ref;
3884
3885	if (io == NULL)
3886		return;
3887
3888	/*
3889	 * May need to preserve linked list pointers at some point too.
3890	 */
3891	pool_ref = io->io_hdr.pool;
3892
3893	memset(io, 0, sizeof(*io));
3894
3895	io->io_hdr.pool = pool_ref;
3896}
3897
3898/*
3899 * This routine is currently used for internal copies of ctl_ios that need
3900 * to persist for some reason after we've already returned status to the
3901 * FETD.  (Thus the flag set.)
3902 *
3903 * XXX XXX
3904 * Note that this makes a blind copy of all fields in the ctl_io, except
3905 * for the pool reference.  This includes any memory that has been
3906 * allocated!  That memory will no longer be valid after done has been
3907 * called, so this would be VERY DANGEROUS for command that actually does
3908 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3909 * start and stop commands, which don't transfer any data, so this is not a
3910 * problem.  If it is used for anything else, the caller would also need to
3911 * allocate data buffer space and this routine would need to be modified to
3912 * copy the data buffer(s) as well.
3913 */
3914void
3915ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3916{
3917	void *pool_ref;
3918
3919	if ((src == NULL)
3920	 || (dest == NULL))
3921		return;
3922
3923	/*
3924	 * May need to preserve linked list pointers at some point too.
3925	 */
3926	pool_ref = dest->io_hdr.pool;
3927
3928	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3929
3930	dest->io_hdr.pool = pool_ref;
3931	/*
3932	 * We need to know that this is an internal copy, and doesn't need
3933	 * to get passed back to the FETD that allocated it.
3934	 */
3935	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3936}
3937
3938#ifdef NEEDTOPORT
3939static void
3940ctl_update_power_subpage(struct copan_power_subpage *page)
3941{
3942	int num_luns, num_partitions, config_type;
3943	struct ctl_softc *softc;
3944	cs_BOOL_t aor_present, shelf_50pct_power;
3945	cs_raidset_personality_t rs_type;
3946	int max_active_luns;
3947
3948	softc = control_softc;
3949
3950	/* subtract out the processor LUN */
3951	num_luns = softc->num_luns - 1;
3952	/*
3953	 * Default to 7 LUNs active, which was the only number we allowed
3954	 * in the past.
3955	 */
3956	max_active_luns = 7;
3957
3958	num_partitions = config_GetRsPartitionInfo();
3959	config_type = config_GetConfigType();
3960	shelf_50pct_power = config_GetShelfPowerMode();
3961	aor_present = config_IsAorRsPresent();
3962
3963	rs_type = ddb_GetRsRaidType(1);
3964	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3965	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3966		EPRINT(0, "Unsupported RS type %d!", rs_type);
3967	}
3968
3969
3970	page->total_luns = num_luns;
3971
3972	switch (config_type) {
3973	case 40:
3974		/*
3975		 * In a 40 drive configuration, it doesn't matter what DC
3976		 * cards we have, whether we have AOR enabled or not,
3977		 * partitioning or not, or what type of RAIDset we have.
3978		 * In that scenario, we can power up every LUN we present
3979		 * to the user.
3980		 */
3981		max_active_luns = num_luns;
3982
3983		break;
3984	case 64:
3985		if (shelf_50pct_power == CS_FALSE) {
3986			/* 25% power */
3987			if (aor_present == CS_TRUE) {
3988				if (rs_type ==
3989				     CS_RAIDSET_PERSONALITY_RAID5) {
3990					max_active_luns = 7;
3991				} else if (rs_type ==
3992					 CS_RAIDSET_PERSONALITY_RAID1){
3993					max_active_luns = 14;
3994				} else {
3995					/* XXX KDM now what?? */
3996				}
3997			} else {
3998				if (rs_type ==
3999				     CS_RAIDSET_PERSONALITY_RAID5) {
4000					max_active_luns = 8;
4001				} else if (rs_type ==
4002					 CS_RAIDSET_PERSONALITY_RAID1){
4003					max_active_luns = 16;
4004				} else {
4005					/* XXX KDM now what?? */
4006				}
4007			}
4008		} else {
4009			/* 50% power */
4010			/*
4011			 * With 50% power in a 64 drive configuration, we
4012			 * can power all LUNs we present.
4013			 */
4014			max_active_luns = num_luns;
4015		}
4016		break;
4017	case 112:
4018		if (shelf_50pct_power == CS_FALSE) {
4019			/* 25% power */
4020			if (aor_present == CS_TRUE) {
4021				if (rs_type ==
4022				     CS_RAIDSET_PERSONALITY_RAID5) {
4023					max_active_luns = 7;
4024				} else if (rs_type ==
4025					 CS_RAIDSET_PERSONALITY_RAID1){
4026					max_active_luns = 14;
4027				} else {
4028					/* XXX KDM now what?? */
4029				}
4030			} else {
4031				if (rs_type ==
4032				     CS_RAIDSET_PERSONALITY_RAID5) {
4033					max_active_luns = 8;
4034				} else if (rs_type ==
4035					 CS_RAIDSET_PERSONALITY_RAID1){
4036					max_active_luns = 16;
4037				} else {
4038					/* XXX KDM now what?? */
4039				}
4040			}
4041		} else {
4042			/* 50% power */
4043			if (aor_present == CS_TRUE) {
4044				if (rs_type ==
4045				     CS_RAIDSET_PERSONALITY_RAID5) {
4046					max_active_luns = 14;
4047				} else if (rs_type ==
4048					 CS_RAIDSET_PERSONALITY_RAID1){
4049					/*
4050					 * We're assuming here that disk
4051					 * caching is enabled, and so we're
4052					 * able to power up half of each
4053					 * LUN, and cache all writes.
4054					 */
4055					max_active_luns = num_luns;
4056				} else {
4057					/* XXX KDM now what?? */
4058				}
4059			} else {
4060				if (rs_type ==
4061				     CS_RAIDSET_PERSONALITY_RAID5) {
4062					max_active_luns = 15;
4063				} else if (rs_type ==
4064					 CS_RAIDSET_PERSONALITY_RAID1){
4065					max_active_luns = 30;
4066				} else {
4067					/* XXX KDM now what?? */
4068				}
4069			}
4070		}
4071		break;
4072	default:
4073		/*
4074		 * In this case, we have an unknown configuration, so we
4075		 * just use the default from above.
4076		 */
4077		break;
4078	}
4079
4080	page->max_active_luns = max_active_luns;
4081#if 0
4082	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4083	       page->total_luns, page->max_active_luns);
4084#endif
4085}
4086#endif /* NEEDTOPORT */
4087
4088/*
4089 * This routine could be used in the future to load default and/or saved
4090 * mode page parameters for a particuar lun.
4091 */
4092static int
4093ctl_init_page_index(struct ctl_lun *lun)
4094{
4095	int i;
4096	struct ctl_page_index *page_index;
4097	struct ctl_softc *softc;
4098	const char *value;
4099
4100	memcpy(&lun->mode_pages.index, page_index_template,
4101	       sizeof(page_index_template));
4102
4103	softc = lun->ctl_softc;
4104
4105	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4106
4107		page_index = &lun->mode_pages.index[i];
4108		/*
4109		 * If this is a disk-only mode page, there's no point in
4110		 * setting it up.  For some pages, we have to have some
4111		 * basic information about the disk in order to calculate the
4112		 * mode page data.
4113		 */
4114		if ((lun->be_lun->lun_type != T_DIRECT)
4115		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4116			continue;
4117
4118		switch (page_index->page_code & SMPH_PC_MASK) {
4119		case SMS_FORMAT_DEVICE_PAGE: {
4120			struct scsi_format_page *format_page;
4121
4122			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4123				panic("subpage is incorrect!");
4124
4125			/*
4126			 * Sectors per track are set above.  Bytes per
4127			 * sector need to be set here on a per-LUN basis.
4128			 */
4129			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4130			       &format_page_default,
4131			       sizeof(format_page_default));
4132			memcpy(&lun->mode_pages.format_page[
4133			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4134			       sizeof(format_page_changeable));
4135			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4136			       &format_page_default,
4137			       sizeof(format_page_default));
4138			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4139			       &format_page_default,
4140			       sizeof(format_page_default));
4141
4142			format_page = &lun->mode_pages.format_page[
4143				CTL_PAGE_CURRENT];
4144			scsi_ulto2b(lun->be_lun->blocksize,
4145				    format_page->bytes_per_sector);
4146
4147			format_page = &lun->mode_pages.format_page[
4148				CTL_PAGE_DEFAULT];
4149			scsi_ulto2b(lun->be_lun->blocksize,
4150				    format_page->bytes_per_sector);
4151
4152			format_page = &lun->mode_pages.format_page[
4153				CTL_PAGE_SAVED];
4154			scsi_ulto2b(lun->be_lun->blocksize,
4155				    format_page->bytes_per_sector);
4156
4157			page_index->page_data =
4158				(uint8_t *)lun->mode_pages.format_page;
4159			break;
4160		}
4161		case SMS_RIGID_DISK_PAGE: {
4162			struct scsi_rigid_disk_page *rigid_disk_page;
4163			uint32_t sectors_per_cylinder;
4164			uint64_t cylinders;
4165#ifndef	__XSCALE__
4166			int shift;
4167#endif /* !__XSCALE__ */
4168
4169			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4170				panic("invalid subpage value %d",
4171				      page_index->subpage);
4172
4173			/*
4174			 * Rotation rate and sectors per track are set
4175			 * above.  We calculate the cylinders here based on
4176			 * capacity.  Due to the number of heads and
4177			 * sectors per track we're using, smaller arrays
4178			 * may turn out to have 0 cylinders.  Linux and
4179			 * FreeBSD don't pay attention to these mode pages
4180			 * to figure out capacity, but Solaris does.  It
4181			 * seems to deal with 0 cylinders just fine, and
4182			 * works out a fake geometry based on the capacity.
4183			 */
4184			memcpy(&lun->mode_pages.rigid_disk_page[
4185			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4186			       sizeof(rigid_disk_page_default));
4187			memcpy(&lun->mode_pages.rigid_disk_page[
4188			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4189			       sizeof(rigid_disk_page_changeable));
4190			memcpy(&lun->mode_pages.rigid_disk_page[
4191			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4192			       sizeof(rigid_disk_page_default));
4193			memcpy(&lun->mode_pages.rigid_disk_page[
4194			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4195			       sizeof(rigid_disk_page_default));
4196
4197			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4198				CTL_DEFAULT_HEADS;
4199
4200			/*
4201			 * The divide method here will be more accurate,
4202			 * probably, but results in floating point being
4203			 * used in the kernel on i386 (__udivdi3()).  On the
4204			 * XScale, though, __udivdi3() is implemented in
4205			 * software.
4206			 *
4207			 * The shift method for cylinder calculation is
4208			 * accurate if sectors_per_cylinder is a power of
4209			 * 2.  Otherwise it might be slightly off -- you
4210			 * might have a bit of a truncation problem.
4211			 */
4212#ifdef	__XSCALE__
4213			cylinders = (lun->be_lun->maxlba + 1) /
4214				sectors_per_cylinder;
4215#else
4216			for (shift = 31; shift > 0; shift--) {
4217				if (sectors_per_cylinder & (1 << shift))
4218					break;
4219			}
4220			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4221#endif
4222
4223			/*
4224			 * We've basically got 3 bytes, or 24 bits for the
4225			 * cylinder size in the mode page.  If we're over,
4226			 * just round down to 2^24.
4227			 */
4228			if (cylinders > 0xffffff)
4229				cylinders = 0xffffff;
4230
4231			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4232				CTL_PAGE_CURRENT];
4233			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4234
4235			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4236				CTL_PAGE_DEFAULT];
4237			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4238
4239			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4240				CTL_PAGE_SAVED];
4241			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4242
4243			page_index->page_data =
4244				(uint8_t *)lun->mode_pages.rigid_disk_page;
4245			break;
4246		}
4247		case SMS_CACHING_PAGE: {
4248			struct scsi_caching_page *caching_page;
4249
4250			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4251				panic("invalid subpage value %d",
4252				      page_index->subpage);
4253			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4254			       &caching_page_default,
4255			       sizeof(caching_page_default));
4256			memcpy(&lun->mode_pages.caching_page[
4257			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4258			       sizeof(caching_page_changeable));
4259			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4260			       &caching_page_default,
4261			       sizeof(caching_page_default));
4262			caching_page = &lun->mode_pages.caching_page[
4263			    CTL_PAGE_SAVED];
4264			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4265			if (value != NULL && strcmp(value, "off") == 0)
4266				caching_page->flags1 &= ~SCP_WCE;
4267			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4268			if (value != NULL && strcmp(value, "off") == 0)
4269				caching_page->flags1 |= SCP_RCD;
4270			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4271			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4272			       sizeof(caching_page_default));
4273			page_index->page_data =
4274				(uint8_t *)lun->mode_pages.caching_page;
4275			break;
4276		}
4277		case SMS_CONTROL_MODE_PAGE: {
4278			struct scsi_control_page *control_page;
4279
4280			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4281				panic("invalid subpage value %d",
4282				      page_index->subpage);
4283
4284			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4285			       &control_page_default,
4286			       sizeof(control_page_default));
4287			memcpy(&lun->mode_pages.control_page[
4288			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4289			       sizeof(control_page_changeable));
4290			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4291			       &control_page_default,
4292			       sizeof(control_page_default));
4293			control_page = &lun->mode_pages.control_page[
4294			    CTL_PAGE_SAVED];
4295			value = ctl_get_opt(&lun->be_lun->options, "reordering");
4296			if (value != NULL && strcmp(value, "unrestricted") == 0) {
4297				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4298				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4299			}
4300			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4301			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4302			       sizeof(control_page_default));
4303			page_index->page_data =
4304				(uint8_t *)lun->mode_pages.control_page;
4305			break;
4306
4307		}
4308		case SMS_VENDOR_SPECIFIC_PAGE:{
4309			switch (page_index->subpage) {
4310			case PWR_SUBPAGE_CODE: {
4311				struct copan_power_subpage *current_page,
4312							   *saved_page;
4313
4314				memcpy(&lun->mode_pages.power_subpage[
4315				       CTL_PAGE_CURRENT],
4316				       &power_page_default,
4317				       sizeof(power_page_default));
4318				memcpy(&lun->mode_pages.power_subpage[
4319				       CTL_PAGE_CHANGEABLE],
4320				       &power_page_changeable,
4321				       sizeof(power_page_changeable));
4322				memcpy(&lun->mode_pages.power_subpage[
4323				       CTL_PAGE_DEFAULT],
4324				       &power_page_default,
4325				       sizeof(power_page_default));
4326				memcpy(&lun->mode_pages.power_subpage[
4327				       CTL_PAGE_SAVED],
4328				       &power_page_default,
4329				       sizeof(power_page_default));
4330				page_index->page_data =
4331				    (uint8_t *)lun->mode_pages.power_subpage;
4332
4333				current_page = (struct copan_power_subpage *)
4334					(page_index->page_data +
4335					 (page_index->page_len *
4336					  CTL_PAGE_CURRENT));
4337			        saved_page = (struct copan_power_subpage *)
4338				        (page_index->page_data +
4339					 (page_index->page_len *
4340					  CTL_PAGE_SAVED));
4341				break;
4342			}
4343			case APS_SUBPAGE_CODE: {
4344				struct copan_aps_subpage *current_page,
4345							 *saved_page;
4346
4347				// This gets set multiple times but
4348				// it should always be the same. It's
4349				// only done during init so who cares.
4350				index_to_aps_page = i;
4351
4352				memcpy(&lun->mode_pages.aps_subpage[
4353				       CTL_PAGE_CURRENT],
4354				       &aps_page_default,
4355				       sizeof(aps_page_default));
4356				memcpy(&lun->mode_pages.aps_subpage[
4357				       CTL_PAGE_CHANGEABLE],
4358				       &aps_page_changeable,
4359				       sizeof(aps_page_changeable));
4360				memcpy(&lun->mode_pages.aps_subpage[
4361				       CTL_PAGE_DEFAULT],
4362				       &aps_page_default,
4363				       sizeof(aps_page_default));
4364				memcpy(&lun->mode_pages.aps_subpage[
4365				       CTL_PAGE_SAVED],
4366				       &aps_page_default,
4367				       sizeof(aps_page_default));
4368				page_index->page_data =
4369					(uint8_t *)lun->mode_pages.aps_subpage;
4370
4371				current_page = (struct copan_aps_subpage *)
4372					(page_index->page_data +
4373					 (page_index->page_len *
4374					  CTL_PAGE_CURRENT));
4375				saved_page = (struct copan_aps_subpage *)
4376					(page_index->page_data +
4377					 (page_index->page_len *
4378					  CTL_PAGE_SAVED));
4379				break;
4380			}
4381			case DBGCNF_SUBPAGE_CODE: {
4382				struct copan_debugconf_subpage *current_page,
4383							       *saved_page;
4384
4385				memcpy(&lun->mode_pages.debugconf_subpage[
4386				       CTL_PAGE_CURRENT],
4387				       &debugconf_page_default,
4388				       sizeof(debugconf_page_default));
4389				memcpy(&lun->mode_pages.debugconf_subpage[
4390				       CTL_PAGE_CHANGEABLE],
4391				       &debugconf_page_changeable,
4392				       sizeof(debugconf_page_changeable));
4393				memcpy(&lun->mode_pages.debugconf_subpage[
4394				       CTL_PAGE_DEFAULT],
4395				       &debugconf_page_default,
4396				       sizeof(debugconf_page_default));
4397				memcpy(&lun->mode_pages.debugconf_subpage[
4398				       CTL_PAGE_SAVED],
4399				       &debugconf_page_default,
4400				       sizeof(debugconf_page_default));
4401				page_index->page_data =
4402					(uint8_t *)lun->mode_pages.debugconf_subpage;
4403
4404				current_page = (struct copan_debugconf_subpage *)
4405					(page_index->page_data +
4406					 (page_index->page_len *
4407					  CTL_PAGE_CURRENT));
4408				saved_page = (struct copan_debugconf_subpage *)
4409					(page_index->page_data +
4410					 (page_index->page_len *
4411					  CTL_PAGE_SAVED));
4412				break;
4413			}
4414			default:
4415				panic("invalid subpage value %d",
4416				      page_index->subpage);
4417				break;
4418			}
4419   			break;
4420		}
4421		default:
4422			panic("invalid page value %d",
4423			      page_index->page_code & SMPH_PC_MASK);
4424			break;
4425    	}
4426	}
4427
4428	return (CTL_RETVAL_COMPLETE);
4429}
4430
4431/*
4432 * LUN allocation.
4433 *
4434 * Requirements:
4435 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4436 *   wants us to allocate the LUN and he can block.
4437 * - ctl_softc is always set
4438 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4439 *
4440 * Returns 0 for success, non-zero (errno) for failure.
4441 */
4442static int
4443ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4444	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4445{
4446	struct ctl_lun *nlun, *lun;
4447	struct ctl_port *port;
4448	struct scsi_vpd_id_descriptor *desc;
4449	struct scsi_vpd_id_t10 *t10id;
4450	const char *eui, *naa, *scsiname, *vendor;
4451	int lun_number, i, lun_malloced;
4452	int devidlen, idlen1, idlen2 = 0, len;
4453
4454	if (be_lun == NULL)
4455		return (EINVAL);
4456
4457	/*
4458	 * We currently only support Direct Access or Processor LUN types.
4459	 */
4460	switch (be_lun->lun_type) {
4461	case T_DIRECT:
4462		break;
4463	case T_PROCESSOR:
4464		break;
4465	case T_SEQUENTIAL:
4466	case T_CHANGER:
4467	default:
4468		be_lun->lun_config_status(be_lun->be_lun,
4469					  CTL_LUN_CONFIG_FAILURE);
4470		break;
4471	}
4472	if (ctl_lun == NULL) {
4473		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4474		lun_malloced = 1;
4475	} else {
4476		lun_malloced = 0;
4477		lun = ctl_lun;
4478	}
4479
4480	memset(lun, 0, sizeof(*lun));
4481	if (lun_malloced)
4482		lun->flags = CTL_LUN_MALLOCED;
4483
4484	/* Generate LUN ID. */
4485	devidlen = max(CTL_DEVID_MIN_LEN,
4486	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4487	idlen1 = sizeof(*t10id) + devidlen;
4488	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4489	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4490	if (scsiname != NULL) {
4491		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4492		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4493	}
4494	eui = ctl_get_opt(&be_lun->options, "eui");
4495	if (eui != NULL) {
4496		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4497	}
4498	naa = ctl_get_opt(&be_lun->options, "naa");
4499	if (naa != NULL) {
4500		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4501	}
4502	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4503	    M_CTL, M_WAITOK | M_ZERO);
4504	lun->lun_devid->len = len;
4505	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4506	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4507	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4508	desc->length = idlen1;
4509	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4510	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4511	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4512		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4513	} else {
4514		strncpy(t10id->vendor, vendor,
4515		    min(sizeof(t10id->vendor), strlen(vendor)));
4516	}
4517	strncpy((char *)t10id->vendor_spec_id,
4518	    (char *)be_lun->device_id, devidlen);
4519	if (scsiname != NULL) {
4520		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4521		    desc->length);
4522		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4523		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4524		    SVPD_ID_TYPE_SCSI_NAME;
4525		desc->length = idlen2;
4526		strlcpy(desc->identifier, scsiname, idlen2);
4527	}
4528	if (eui != NULL) {
4529		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4530		    desc->length);
4531		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4532		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4533		    SVPD_ID_TYPE_EUI64;
4534		desc->length = 8;
4535		scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4536	}
4537	if (naa != NULL) {
4538		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4539		    desc->length);
4540		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4541		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4542		    SVPD_ID_TYPE_NAA;
4543		desc->length = 8;
4544		scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4545	}
4546
4547	mtx_lock(&ctl_softc->ctl_lock);
4548	/*
4549	 * See if the caller requested a particular LUN number.  If so, see
4550	 * if it is available.  Otherwise, allocate the first available LUN.
4551	 */
4552	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4553		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4554		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4555			mtx_unlock(&ctl_softc->ctl_lock);
4556			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4557				printf("ctl: requested LUN ID %d is higher "
4558				       "than CTL_MAX_LUNS - 1 (%d)\n",
4559				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4560			} else {
4561				/*
4562				 * XXX KDM return an error, or just assign
4563				 * another LUN ID in this case??
4564				 */
4565				printf("ctl: requested LUN ID %d is already "
4566				       "in use\n", be_lun->req_lun_id);
4567			}
4568			if (lun->flags & CTL_LUN_MALLOCED)
4569				free(lun, M_CTL);
4570			be_lun->lun_config_status(be_lun->be_lun,
4571						  CTL_LUN_CONFIG_FAILURE);
4572			return (ENOSPC);
4573		}
4574		lun_number = be_lun->req_lun_id;
4575	} else {
4576		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4577		if (lun_number == -1) {
4578			mtx_unlock(&ctl_softc->ctl_lock);
4579			printf("ctl: can't allocate LUN on target %ju, out of "
4580			       "LUNs\n", (uintmax_t)target_id.id);
4581			if (lun->flags & CTL_LUN_MALLOCED)
4582				free(lun, M_CTL);
4583			be_lun->lun_config_status(be_lun->be_lun,
4584						  CTL_LUN_CONFIG_FAILURE);
4585			return (ENOSPC);
4586		}
4587	}
4588	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4589
4590	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4591	lun->target = target_id;
4592	lun->lun = lun_number;
4593	lun->be_lun = be_lun;
4594	/*
4595	 * The processor LUN is always enabled.  Disk LUNs come on line
4596	 * disabled, and must be enabled by the backend.
4597	 */
4598	lun->flags |= CTL_LUN_DISABLED;
4599	lun->backend = be_lun->be;
4600	be_lun->ctl_lun = lun;
4601	be_lun->lun_id = lun_number;
4602	atomic_add_int(&be_lun->be->num_luns, 1);
4603	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4604		lun->flags |= CTL_LUN_STOPPED;
4605
4606	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4607		lun->flags |= CTL_LUN_INOPERABLE;
4608
4609	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4610		lun->flags |= CTL_LUN_PRIMARY_SC;
4611
4612	lun->ctl_softc = ctl_softc;
4613	TAILQ_INIT(&lun->ooa_queue);
4614	TAILQ_INIT(&lun->blocked_queue);
4615	STAILQ_INIT(&lun->error_list);
4616	ctl_tpc_lun_init(lun);
4617
4618	/*
4619	 * Initialize the mode page index.
4620	 */
4621	ctl_init_page_index(lun);
4622
4623	/*
4624	 * Set the poweron UA for all initiators on this LUN only.
4625	 */
4626	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4627		lun->pending_ua[i] = CTL_UA_POWERON;
4628
4629	/*
4630	 * Now, before we insert this lun on the lun list, set the lun
4631	 * inventory changed UA for all other luns.
4632	 */
4633	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4634		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4635			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4636		}
4637	}
4638
4639	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4640
4641	ctl_softc->ctl_luns[lun_number] = lun;
4642
4643	ctl_softc->num_luns++;
4644
4645	/* Setup statistics gathering */
4646	lun->stats.device_type = be_lun->lun_type;
4647	lun->stats.lun_number = lun_number;
4648	if (lun->stats.device_type == T_DIRECT)
4649		lun->stats.blocksize = be_lun->blocksize;
4650	else
4651		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4652	for (i = 0;i < CTL_MAX_PORTS;i++)
4653		lun->stats.ports[i].targ_port = i;
4654
4655	mtx_unlock(&ctl_softc->ctl_lock);
4656
4657	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4658
4659	/*
4660	 * Run through each registered FETD and bring it online if it isn't
4661	 * already.  Enable the target ID if it hasn't been enabled, and
4662	 * enable this particular LUN.
4663	 */
4664	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4665		int retval;
4666
4667		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4668		if (retval != 0) {
4669			printf("ctl_alloc_lun: FETD %s port %d returned error "
4670			       "%d for lun_enable on target %ju lun %d\n",
4671			       port->port_name, port->targ_port, retval,
4672			       (uintmax_t)target_id.id, lun_number);
4673		} else
4674			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4675	}
4676	return (0);
4677}
4678
4679/*
4680 * Delete a LUN.
4681 * Assumptions:
4682 * - LUN has already been marked invalid and any pending I/O has been taken
4683 *   care of.
4684 */
4685static int
4686ctl_free_lun(struct ctl_lun *lun)
4687{
4688	struct ctl_softc *softc;
4689#if 0
4690	struct ctl_port *port;
4691#endif
4692	struct ctl_lun *nlun;
4693	int i;
4694
4695	softc = lun->ctl_softc;
4696
4697	mtx_assert(&softc->ctl_lock, MA_OWNED);
4698
4699	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4700
4701	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4702
4703	softc->ctl_luns[lun->lun] = NULL;
4704
4705	if (!TAILQ_EMPTY(&lun->ooa_queue))
4706		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4707
4708	softc->num_luns--;
4709
4710	/*
4711	 * XXX KDM this scheme only works for a single target/multiple LUN
4712	 * setup.  It needs to be revamped for a multiple target scheme.
4713	 *
4714	 * XXX KDM this results in port->lun_disable() getting called twice,
4715	 * once when ctl_disable_lun() is called, and a second time here.
4716	 * We really need to re-think the LUN disable semantics.  There
4717	 * should probably be several steps/levels to LUN removal:
4718	 *  - disable
4719	 *  - invalidate
4720	 *  - free
4721 	 *
4722	 * Right now we only have a disable method when communicating to
4723	 * the front end ports, at least for individual LUNs.
4724	 */
4725#if 0
4726	STAILQ_FOREACH(port, &softc->port_list, links) {
4727		int retval;
4728
4729		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4730					 lun->lun);
4731		if (retval != 0) {
4732			printf("ctl_free_lun: FETD %s port %d returned error "
4733			       "%d for lun_disable on target %ju lun %jd\n",
4734			       port->port_name, port->targ_port, retval,
4735			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4736		}
4737
4738		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4739			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4740
4741			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4742			if (retval != 0) {
4743				printf("ctl_free_lun: FETD %s port %d "
4744				       "returned error %d for targ_disable on "
4745				       "target %ju\n", port->port_name,
4746				       port->targ_port, retval,
4747				       (uintmax_t)lun->target.id);
4748			} else
4749				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4750
4751			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4752				continue;
4753
4754#if 0
4755			port->port_offline(port->onoff_arg);
4756			port->status &= ~CTL_PORT_STATUS_ONLINE;
4757#endif
4758		}
4759	}
4760#endif
4761
4762	/*
4763	 * Tell the backend to free resources, if this LUN has a backend.
4764	 */
4765	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4766	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4767
4768	ctl_tpc_lun_shutdown(lun);
4769	mtx_destroy(&lun->lun_lock);
4770	free(lun->lun_devid, M_CTL);
4771	if (lun->flags & CTL_LUN_MALLOCED)
4772		free(lun, M_CTL);
4773
4774	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4775		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4776			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4777		}
4778	}
4779
4780	return (0);
4781}
4782
4783static void
4784ctl_create_lun(struct ctl_be_lun *be_lun)
4785{
4786	struct ctl_softc *ctl_softc;
4787
4788	ctl_softc = control_softc;
4789
4790	/*
4791	 * ctl_alloc_lun() should handle all potential failure cases.
4792	 */
4793	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4794}
4795
4796int
4797ctl_add_lun(struct ctl_be_lun *be_lun)
4798{
4799	struct ctl_softc *ctl_softc = control_softc;
4800
4801	mtx_lock(&ctl_softc->ctl_lock);
4802	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4803	mtx_unlock(&ctl_softc->ctl_lock);
4804	wakeup(&ctl_softc->pending_lun_queue);
4805
4806	return (0);
4807}
4808
4809int
4810ctl_enable_lun(struct ctl_be_lun *be_lun)
4811{
4812	struct ctl_softc *ctl_softc;
4813	struct ctl_port *port, *nport;
4814	struct ctl_lun *lun;
4815	int retval;
4816
4817	ctl_softc = control_softc;
4818
4819	lun = (struct ctl_lun *)be_lun->ctl_lun;
4820
4821	mtx_lock(&ctl_softc->ctl_lock);
4822	mtx_lock(&lun->lun_lock);
4823	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4824		/*
4825		 * eh?  Why did we get called if the LUN is already
4826		 * enabled?
4827		 */
4828		mtx_unlock(&lun->lun_lock);
4829		mtx_unlock(&ctl_softc->ctl_lock);
4830		return (0);
4831	}
4832	lun->flags &= ~CTL_LUN_DISABLED;
4833	mtx_unlock(&lun->lun_lock);
4834
4835	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4836		nport = STAILQ_NEXT(port, links);
4837
4838		/*
4839		 * Drop the lock while we call the FETD's enable routine.
4840		 * This can lead to a callback into CTL (at least in the
4841		 * case of the internal initiator frontend.
4842		 */
4843		mtx_unlock(&ctl_softc->ctl_lock);
4844		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4845		mtx_lock(&ctl_softc->ctl_lock);
4846		if (retval != 0) {
4847			printf("%s: FETD %s port %d returned error "
4848			       "%d for lun_enable on target %ju lun %jd\n",
4849			       __func__, port->port_name, port->targ_port, retval,
4850			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4851		}
4852#if 0
4853		 else {
4854            /* NOTE:  TODO:  why does lun enable affect port status? */
4855			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4856		}
4857#endif
4858	}
4859
4860	mtx_unlock(&ctl_softc->ctl_lock);
4861
4862	return (0);
4863}
4864
4865int
4866ctl_disable_lun(struct ctl_be_lun *be_lun)
4867{
4868	struct ctl_softc *ctl_softc;
4869	struct ctl_port *port;
4870	struct ctl_lun *lun;
4871	int retval;
4872
4873	ctl_softc = control_softc;
4874
4875	lun = (struct ctl_lun *)be_lun->ctl_lun;
4876
4877	mtx_lock(&ctl_softc->ctl_lock);
4878	mtx_lock(&lun->lun_lock);
4879	if (lun->flags & CTL_LUN_DISABLED) {
4880		mtx_unlock(&lun->lun_lock);
4881		mtx_unlock(&ctl_softc->ctl_lock);
4882		return (0);
4883	}
4884	lun->flags |= CTL_LUN_DISABLED;
4885	mtx_unlock(&lun->lun_lock);
4886
4887	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4888		mtx_unlock(&ctl_softc->ctl_lock);
4889		/*
4890		 * Drop the lock before we call the frontend's disable
4891		 * routine, to avoid lock order reversals.
4892		 *
4893		 * XXX KDM what happens if the frontend list changes while
4894		 * we're traversing it?  It's unlikely, but should be handled.
4895		 */
4896		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4897					 lun->lun);
4898		mtx_lock(&ctl_softc->ctl_lock);
4899		if (retval != 0) {
4900			printf("ctl_alloc_lun: FETD %s port %d returned error "
4901			       "%d for lun_disable on target %ju lun %jd\n",
4902			       port->port_name, port->targ_port, retval,
4903			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4904		}
4905	}
4906
4907	mtx_unlock(&ctl_softc->ctl_lock);
4908
4909	return (0);
4910}
4911
4912int
4913ctl_start_lun(struct ctl_be_lun *be_lun)
4914{
4915	struct ctl_softc *ctl_softc;
4916	struct ctl_lun *lun;
4917
4918	ctl_softc = control_softc;
4919
4920	lun = (struct ctl_lun *)be_lun->ctl_lun;
4921
4922	mtx_lock(&lun->lun_lock);
4923	lun->flags &= ~CTL_LUN_STOPPED;
4924	mtx_unlock(&lun->lun_lock);
4925
4926	return (0);
4927}
4928
4929int
4930ctl_stop_lun(struct ctl_be_lun *be_lun)
4931{
4932	struct ctl_softc *ctl_softc;
4933	struct ctl_lun *lun;
4934
4935	ctl_softc = control_softc;
4936
4937	lun = (struct ctl_lun *)be_lun->ctl_lun;
4938
4939	mtx_lock(&lun->lun_lock);
4940	lun->flags |= CTL_LUN_STOPPED;
4941	mtx_unlock(&lun->lun_lock);
4942
4943	return (0);
4944}
4945
4946int
4947ctl_lun_offline(struct ctl_be_lun *be_lun)
4948{
4949	struct ctl_softc *ctl_softc;
4950	struct ctl_lun *lun;
4951
4952	ctl_softc = control_softc;
4953
4954	lun = (struct ctl_lun *)be_lun->ctl_lun;
4955
4956	mtx_lock(&lun->lun_lock);
4957	lun->flags |= CTL_LUN_OFFLINE;
4958	mtx_unlock(&lun->lun_lock);
4959
4960	return (0);
4961}
4962
4963int
4964ctl_lun_online(struct ctl_be_lun *be_lun)
4965{
4966	struct ctl_softc *ctl_softc;
4967	struct ctl_lun *lun;
4968
4969	ctl_softc = control_softc;
4970
4971	lun = (struct ctl_lun *)be_lun->ctl_lun;
4972
4973	mtx_lock(&lun->lun_lock);
4974	lun->flags &= ~CTL_LUN_OFFLINE;
4975	mtx_unlock(&lun->lun_lock);
4976
4977	return (0);
4978}
4979
4980int
4981ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4982{
4983	struct ctl_softc *ctl_softc;
4984	struct ctl_lun *lun;
4985
4986	ctl_softc = control_softc;
4987
4988	lun = (struct ctl_lun *)be_lun->ctl_lun;
4989
4990	mtx_lock(&lun->lun_lock);
4991
4992	/*
4993	 * The LUN needs to be disabled before it can be marked invalid.
4994	 */
4995	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4996		mtx_unlock(&lun->lun_lock);
4997		return (-1);
4998	}
4999	/*
5000	 * Mark the LUN invalid.
5001	 */
5002	lun->flags |= CTL_LUN_INVALID;
5003
5004	/*
5005	 * If there is nothing in the OOA queue, go ahead and free the LUN.
5006	 * If we have something in the OOA queue, we'll free it when the
5007	 * last I/O completes.
5008	 */
5009	if (TAILQ_EMPTY(&lun->ooa_queue)) {
5010		mtx_unlock(&lun->lun_lock);
5011		mtx_lock(&ctl_softc->ctl_lock);
5012		ctl_free_lun(lun);
5013		mtx_unlock(&ctl_softc->ctl_lock);
5014	} else
5015		mtx_unlock(&lun->lun_lock);
5016
5017	return (0);
5018}
5019
5020int
5021ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5022{
5023	struct ctl_softc *ctl_softc;
5024	struct ctl_lun *lun;
5025
5026	ctl_softc = control_softc;
5027	lun = (struct ctl_lun *)be_lun->ctl_lun;
5028
5029	mtx_lock(&lun->lun_lock);
5030	lun->flags |= CTL_LUN_INOPERABLE;
5031	mtx_unlock(&lun->lun_lock);
5032
5033	return (0);
5034}
5035
5036int
5037ctl_lun_operable(struct ctl_be_lun *be_lun)
5038{
5039	struct ctl_softc *ctl_softc;
5040	struct ctl_lun *lun;
5041
5042	ctl_softc = control_softc;
5043	lun = (struct ctl_lun *)be_lun->ctl_lun;
5044
5045	mtx_lock(&lun->lun_lock);
5046	lun->flags &= ~CTL_LUN_INOPERABLE;
5047	mtx_unlock(&lun->lun_lock);
5048
5049	return (0);
5050}
5051
5052int
5053ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5054		   int lock)
5055{
5056	struct ctl_softc *softc;
5057	struct ctl_lun *lun;
5058	struct copan_aps_subpage *current_sp;
5059	struct ctl_page_index *page_index;
5060	int i;
5061
5062	softc = control_softc;
5063
5064	mtx_lock(&softc->ctl_lock);
5065
5066	lun = (struct ctl_lun *)be_lun->ctl_lun;
5067	mtx_lock(&lun->lun_lock);
5068
5069	page_index = NULL;
5070	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5071		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5072		     APS_PAGE_CODE)
5073			continue;
5074
5075		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5076			continue;
5077		page_index = &lun->mode_pages.index[i];
5078	}
5079
5080	if (page_index == NULL) {
5081		mtx_unlock(&lun->lun_lock);
5082		mtx_unlock(&softc->ctl_lock);
5083		printf("%s: APS subpage not found for lun %ju!\n", __func__,
5084		       (uintmax_t)lun->lun);
5085		return (1);
5086	}
5087#if 0
5088	if ((softc->aps_locked_lun != 0)
5089	 && (softc->aps_locked_lun != lun->lun)) {
5090		printf("%s: attempt to lock LUN %llu when %llu is already "
5091		       "locked\n");
5092		mtx_unlock(&lun->lun_lock);
5093		mtx_unlock(&softc->ctl_lock);
5094		return (1);
5095	}
5096#endif
5097
5098	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5099		(page_index->page_len * CTL_PAGE_CURRENT));
5100
5101	if (lock != 0) {
5102		current_sp->lock_active = APS_LOCK_ACTIVE;
5103		softc->aps_locked_lun = lun->lun;
5104	} else {
5105		current_sp->lock_active = 0;
5106		softc->aps_locked_lun = 0;
5107	}
5108
5109
5110	/*
5111	 * If we're in HA mode, try to send the lock message to the other
5112	 * side.
5113	 */
5114	if (ctl_is_single == 0) {
5115		int isc_retval;
5116		union ctl_ha_msg lock_msg;
5117
5118		lock_msg.hdr.nexus = *nexus;
5119		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5120		if (lock != 0)
5121			lock_msg.aps.lock_flag = 1;
5122		else
5123			lock_msg.aps.lock_flag = 0;
5124		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5125					 sizeof(lock_msg), 0);
5126		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5127			printf("%s: APS (lock=%d) error returned from "
5128			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5129			mtx_unlock(&lun->lun_lock);
5130			mtx_unlock(&softc->ctl_lock);
5131			return (1);
5132		}
5133	}
5134
5135	mtx_unlock(&lun->lun_lock);
5136	mtx_unlock(&softc->ctl_lock);
5137
5138	return (0);
5139}
5140
5141void
5142ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5143{
5144	struct ctl_lun *lun;
5145	struct ctl_softc *softc;
5146	int i;
5147
5148	softc = control_softc;
5149
5150	lun = (struct ctl_lun *)be_lun->ctl_lun;
5151
5152	mtx_lock(&lun->lun_lock);
5153
5154	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5155		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5156
5157	mtx_unlock(&lun->lun_lock);
5158}
5159
5160/*
5161 * Backend "memory move is complete" callback for requests that never
5162 * make it down to say RAIDCore's configuration code.
5163 */
5164int
5165ctl_config_move_done(union ctl_io *io)
5166{
5167	int retval;
5168
5169	retval = CTL_RETVAL_COMPLETE;
5170
5171
5172	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5173	/*
5174	 * XXX KDM this shouldn't happen, but what if it does?
5175	 */
5176	if (io->io_hdr.io_type != CTL_IO_SCSI)
5177		panic("I/O type isn't CTL_IO_SCSI!");
5178
5179	if ((io->io_hdr.port_status == 0)
5180	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5181	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5182		io->io_hdr.status = CTL_SUCCESS;
5183	else if ((io->io_hdr.port_status != 0)
5184	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5185	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5186		/*
5187		 * For hardware error sense keys, the sense key
5188		 * specific value is defined to be a retry count,
5189		 * but we use it to pass back an internal FETD
5190		 * error code.  XXX KDM  Hopefully the FETD is only
5191		 * using 16 bits for an error code, since that's
5192		 * all the space we have in the sks field.
5193		 */
5194		ctl_set_internal_failure(&io->scsiio,
5195					 /*sks_valid*/ 1,
5196					 /*retry_count*/
5197					 io->io_hdr.port_status);
5198		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5199			free(io->scsiio.kern_data_ptr, M_CTL);
5200		ctl_done(io);
5201		goto bailout;
5202	}
5203
5204	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5205	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5206	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5207		/*
5208		 * XXX KDM just assuming a single pointer here, and not a
5209		 * S/G list.  If we start using S/G lists for config data,
5210		 * we'll need to know how to clean them up here as well.
5211		 */
5212		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5213			free(io->scsiio.kern_data_ptr, M_CTL);
5214		/* Hopefully the user has already set the status... */
5215		ctl_done(io);
5216	} else {
5217		/*
5218		 * XXX KDM now we need to continue data movement.  Some
5219		 * options:
5220		 * - call ctl_scsiio() again?  We don't do this for data
5221		 *   writes, because for those at least we know ahead of
5222		 *   time where the write will go and how long it is.  For
5223		 *   config writes, though, that information is largely
5224		 *   contained within the write itself, thus we need to
5225		 *   parse out the data again.
5226		 *
5227		 * - Call some other function once the data is in?
5228		 */
5229
5230		/*
5231		 * XXX KDM call ctl_scsiio() again for now, and check flag
5232		 * bits to see whether we're allocated or not.
5233		 */
5234		retval = ctl_scsiio(&io->scsiio);
5235	}
5236bailout:
5237	return (retval);
5238}
5239
5240/*
5241 * This gets called by a backend driver when it is done with a
5242 * data_submit method.
5243 */
5244void
5245ctl_data_submit_done(union ctl_io *io)
5246{
5247	/*
5248	 * If the IO_CONT flag is set, we need to call the supplied
5249	 * function to continue processing the I/O, instead of completing
5250	 * the I/O just yet.
5251	 *
5252	 * If there is an error, though, we don't want to keep processing.
5253	 * Instead, just send status back to the initiator.
5254	 */
5255	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5256	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5257	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5258	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5259		io->scsiio.io_cont(io);
5260		return;
5261	}
5262	ctl_done(io);
5263}
5264
5265/*
5266 * This gets called by a backend driver when it is done with a
5267 * configuration write.
5268 */
5269void
5270ctl_config_write_done(union ctl_io *io)
5271{
5272	uint8_t *buf;
5273
5274	/*
5275	 * If the IO_CONT flag is set, we need to call the supplied
5276	 * function to continue processing the I/O, instead of completing
5277	 * the I/O just yet.
5278	 *
5279	 * If there is an error, though, we don't want to keep processing.
5280	 * Instead, just send status back to the initiator.
5281	 */
5282	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5283	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5284	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5285		io->scsiio.io_cont(io);
5286		return;
5287	}
5288	/*
5289	 * Since a configuration write can be done for commands that actually
5290	 * have data allocated, like write buffer, and commands that have
5291	 * no data, like start/stop unit, we need to check here.
5292	 */
5293	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5294		buf = io->scsiio.kern_data_ptr;
5295	else
5296		buf = NULL;
5297	ctl_done(io);
5298	if (buf)
5299		free(buf, M_CTL);
5300}
5301
5302/*
5303 * SCSI release command.
5304 */
5305int
5306ctl_scsi_release(struct ctl_scsiio *ctsio)
5307{
5308	int length, longid, thirdparty_id, resv_id;
5309	struct ctl_softc *ctl_softc;
5310	struct ctl_lun *lun;
5311	uint32_t residx;
5312
5313	length = 0;
5314	resv_id = 0;
5315
5316	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5317
5318	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5319	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5320	ctl_softc = control_softc;
5321
5322	switch (ctsio->cdb[0]) {
5323	case RELEASE_10: {
5324		struct scsi_release_10 *cdb;
5325
5326		cdb = (struct scsi_release_10 *)ctsio->cdb;
5327
5328		if (cdb->byte2 & SR10_LONGID)
5329			longid = 1;
5330		else
5331			thirdparty_id = cdb->thirdparty_id;
5332
5333		resv_id = cdb->resv_id;
5334		length = scsi_2btoul(cdb->length);
5335		break;
5336	}
5337	}
5338
5339
5340	/*
5341	 * XXX KDM right now, we only support LUN reservation.  We don't
5342	 * support 3rd party reservations, or extent reservations, which
5343	 * might actually need the parameter list.  If we've gotten this
5344	 * far, we've got a LUN reservation.  Anything else got kicked out
5345	 * above.  So, according to SPC, ignore the length.
5346	 */
5347	length = 0;
5348
5349	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5350	 && (length > 0)) {
5351		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5352		ctsio->kern_data_len = length;
5353		ctsio->kern_total_len = length;
5354		ctsio->kern_data_resid = 0;
5355		ctsio->kern_rel_offset = 0;
5356		ctsio->kern_sg_entries = 0;
5357		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5358		ctsio->be_move_done = ctl_config_move_done;
5359		ctl_datamove((union ctl_io *)ctsio);
5360
5361		return (CTL_RETVAL_COMPLETE);
5362	}
5363
5364	if (length > 0)
5365		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5366
5367	mtx_lock(&lun->lun_lock);
5368
5369	/*
5370	 * According to SPC, it is not an error for an intiator to attempt
5371	 * to release a reservation on a LUN that isn't reserved, or that
5372	 * is reserved by another initiator.  The reservation can only be
5373	 * released, though, by the initiator who made it or by one of
5374	 * several reset type events.
5375	 */
5376	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5377			lun->flags &= ~CTL_LUN_RESERVED;
5378
5379	mtx_unlock(&lun->lun_lock);
5380
5381	ctsio->scsi_status = SCSI_STATUS_OK;
5382	ctsio->io_hdr.status = CTL_SUCCESS;
5383
5384	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5385		free(ctsio->kern_data_ptr, M_CTL);
5386		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5387	}
5388
5389	ctl_done((union ctl_io *)ctsio);
5390	return (CTL_RETVAL_COMPLETE);
5391}
5392
5393int
5394ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5395{
5396	int extent, thirdparty, longid;
5397	int resv_id, length;
5398	uint64_t thirdparty_id;
5399	struct ctl_softc *ctl_softc;
5400	struct ctl_lun *lun;
5401	uint32_t residx;
5402
5403	extent = 0;
5404	thirdparty = 0;
5405	longid = 0;
5406	resv_id = 0;
5407	length = 0;
5408	thirdparty_id = 0;
5409
5410	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5411
5412	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5413	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5414	ctl_softc = control_softc;
5415
5416	switch (ctsio->cdb[0]) {
5417	case RESERVE_10: {
5418		struct scsi_reserve_10 *cdb;
5419
5420		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5421
5422		if (cdb->byte2 & SR10_LONGID)
5423			longid = 1;
5424		else
5425			thirdparty_id = cdb->thirdparty_id;
5426
5427		resv_id = cdb->resv_id;
5428		length = scsi_2btoul(cdb->length);
5429		break;
5430	}
5431	}
5432
5433	/*
5434	 * XXX KDM right now, we only support LUN reservation.  We don't
5435	 * support 3rd party reservations, or extent reservations, which
5436	 * might actually need the parameter list.  If we've gotten this
5437	 * far, we've got a LUN reservation.  Anything else got kicked out
5438	 * above.  So, according to SPC, ignore the length.
5439	 */
5440	length = 0;
5441
5442	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5443	 && (length > 0)) {
5444		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5445		ctsio->kern_data_len = length;
5446		ctsio->kern_total_len = length;
5447		ctsio->kern_data_resid = 0;
5448		ctsio->kern_rel_offset = 0;
5449		ctsio->kern_sg_entries = 0;
5450		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5451		ctsio->be_move_done = ctl_config_move_done;
5452		ctl_datamove((union ctl_io *)ctsio);
5453
5454		return (CTL_RETVAL_COMPLETE);
5455	}
5456
5457	if (length > 0)
5458		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5459
5460	mtx_lock(&lun->lun_lock);
5461	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5462		ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5463		ctsio->io_hdr.status = CTL_SCSI_ERROR;
5464		goto bailout;
5465	}
5466
5467	lun->flags |= CTL_LUN_RESERVED;
5468	lun->res_idx = residx;
5469
5470	ctsio->scsi_status = SCSI_STATUS_OK;
5471	ctsio->io_hdr.status = CTL_SUCCESS;
5472
5473bailout:
5474	mtx_unlock(&lun->lun_lock);
5475
5476	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5477		free(ctsio->kern_data_ptr, M_CTL);
5478		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5479	}
5480
5481	ctl_done((union ctl_io *)ctsio);
5482	return (CTL_RETVAL_COMPLETE);
5483}
5484
5485int
5486ctl_start_stop(struct ctl_scsiio *ctsio)
5487{
5488	struct scsi_start_stop_unit *cdb;
5489	struct ctl_lun *lun;
5490	struct ctl_softc *ctl_softc;
5491	int retval;
5492
5493	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5494
5495	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5496	ctl_softc = control_softc;
5497	retval = 0;
5498
5499	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5500
5501	/*
5502	 * XXX KDM
5503	 * We don't support the immediate bit on a stop unit.  In order to
5504	 * do that, we would need to code up a way to know that a stop is
5505	 * pending, and hold off any new commands until it completes, one
5506	 * way or another.  Then we could accept or reject those commands
5507	 * depending on its status.  We would almost need to do the reverse
5508	 * of what we do below for an immediate start -- return the copy of
5509	 * the ctl_io to the FETD with status to send to the host (and to
5510	 * free the copy!) and then free the original I/O once the stop
5511	 * actually completes.  That way, the OOA queue mechanism can work
5512	 * to block commands that shouldn't proceed.  Another alternative
5513	 * would be to put the copy in the queue in place of the original,
5514	 * and return the original back to the caller.  That could be
5515	 * slightly safer..
5516	 */
5517	if ((cdb->byte2 & SSS_IMMED)
5518	 && ((cdb->how & SSS_START) == 0)) {
5519		ctl_set_invalid_field(ctsio,
5520				      /*sks_valid*/ 1,
5521				      /*command*/ 1,
5522				      /*field*/ 1,
5523				      /*bit_valid*/ 1,
5524				      /*bit*/ 0);
5525		ctl_done((union ctl_io *)ctsio);
5526		return (CTL_RETVAL_COMPLETE);
5527	}
5528
5529	if ((lun->flags & CTL_LUN_PR_RESERVED)
5530	 && ((cdb->how & SSS_START)==0)) {
5531		uint32_t residx;
5532
5533		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5534		if (!lun->per_res[residx].registered
5535		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5536
5537			ctl_set_reservation_conflict(ctsio);
5538			ctl_done((union ctl_io *)ctsio);
5539			return (CTL_RETVAL_COMPLETE);
5540		}
5541	}
5542
5543	/*
5544	 * If there is no backend on this device, we can't start or stop
5545	 * it.  In theory we shouldn't get any start/stop commands in the
5546	 * first place at this level if the LUN doesn't have a backend.
5547	 * That should get stopped by the command decode code.
5548	 */
5549	if (lun->backend == NULL) {
5550		ctl_set_invalid_opcode(ctsio);
5551		ctl_done((union ctl_io *)ctsio);
5552		return (CTL_RETVAL_COMPLETE);
5553	}
5554
5555	/*
5556	 * XXX KDM Copan-specific offline behavior.
5557	 * Figure out a reasonable way to port this?
5558	 */
5559#ifdef NEEDTOPORT
5560	mtx_lock(&lun->lun_lock);
5561
5562	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5563	 && (lun->flags & CTL_LUN_OFFLINE)) {
5564		/*
5565		 * If the LUN is offline, and the on/offline bit isn't set,
5566		 * reject the start or stop.  Otherwise, let it through.
5567		 */
5568		mtx_unlock(&lun->lun_lock);
5569		ctl_set_lun_not_ready(ctsio);
5570		ctl_done((union ctl_io *)ctsio);
5571	} else {
5572		mtx_unlock(&lun->lun_lock);
5573#endif /* NEEDTOPORT */
5574		/*
5575		 * This could be a start or a stop when we're online,
5576		 * or a stop/offline or start/online.  A start or stop when
5577		 * we're offline is covered in the case above.
5578		 */
5579		/*
5580		 * In the non-immediate case, we send the request to
5581		 * the backend and return status to the user when
5582		 * it is done.
5583		 *
5584		 * In the immediate case, we allocate a new ctl_io
5585		 * to hold a copy of the request, and send that to
5586		 * the backend.  We then set good status on the
5587		 * user's request and return it immediately.
5588		 */
5589		if (cdb->byte2 & SSS_IMMED) {
5590			union ctl_io *new_io;
5591
5592			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5593			if (new_io == NULL) {
5594				ctl_set_busy(ctsio);
5595				ctl_done((union ctl_io *)ctsio);
5596			} else {
5597				ctl_copy_io((union ctl_io *)ctsio,
5598					    new_io);
5599				retval = lun->backend->config_write(new_io);
5600				ctl_set_success(ctsio);
5601				ctl_done((union ctl_io *)ctsio);
5602			}
5603		} else {
5604			retval = lun->backend->config_write(
5605				(union ctl_io *)ctsio);
5606		}
5607#ifdef NEEDTOPORT
5608	}
5609#endif
5610	return (retval);
5611}
5612
5613/*
5614 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5615 * we don't really do anything with the LBA and length fields if the user
5616 * passes them in.  Instead we'll just flush out the cache for the entire
5617 * LUN.
5618 */
5619int
5620ctl_sync_cache(struct ctl_scsiio *ctsio)
5621{
5622	struct ctl_lun *lun;
5623	struct ctl_softc *ctl_softc;
5624	uint64_t starting_lba;
5625	uint32_t block_count;
5626	int retval;
5627
5628	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5629
5630	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5631	ctl_softc = control_softc;
5632	retval = 0;
5633
5634	switch (ctsio->cdb[0]) {
5635	case SYNCHRONIZE_CACHE: {
5636		struct scsi_sync_cache *cdb;
5637		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5638
5639		starting_lba = scsi_4btoul(cdb->begin_lba);
5640		block_count = scsi_2btoul(cdb->lb_count);
5641		break;
5642	}
5643	case SYNCHRONIZE_CACHE_16: {
5644		struct scsi_sync_cache_16 *cdb;
5645		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5646
5647		starting_lba = scsi_8btou64(cdb->begin_lba);
5648		block_count = scsi_4btoul(cdb->lb_count);
5649		break;
5650	}
5651	default:
5652		ctl_set_invalid_opcode(ctsio);
5653		ctl_done((union ctl_io *)ctsio);
5654		goto bailout;
5655		break; /* NOTREACHED */
5656	}
5657
5658	/*
5659	 * We check the LBA and length, but don't do anything with them.
5660	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5661	 * get flushed.  This check will just help satisfy anyone who wants
5662	 * to see an error for an out of range LBA.
5663	 */
5664	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5665		ctl_set_lba_out_of_range(ctsio);
5666		ctl_done((union ctl_io *)ctsio);
5667		goto bailout;
5668	}
5669
5670	/*
5671	 * If this LUN has no backend, we can't flush the cache anyway.
5672	 */
5673	if (lun->backend == NULL) {
5674		ctl_set_invalid_opcode(ctsio);
5675		ctl_done((union ctl_io *)ctsio);
5676		goto bailout;
5677	}
5678
5679	/*
5680	 * Check to see whether we're configured to send the SYNCHRONIZE
5681	 * CACHE command directly to the back end.
5682	 */
5683	mtx_lock(&lun->lun_lock);
5684	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5685	 && (++(lun->sync_count) >= lun->sync_interval)) {
5686		lun->sync_count = 0;
5687		mtx_unlock(&lun->lun_lock);
5688		retval = lun->backend->config_write((union ctl_io *)ctsio);
5689	} else {
5690		mtx_unlock(&lun->lun_lock);
5691		ctl_set_success(ctsio);
5692		ctl_done((union ctl_io *)ctsio);
5693	}
5694
5695bailout:
5696
5697	return (retval);
5698}
5699
5700int
5701ctl_format(struct ctl_scsiio *ctsio)
5702{
5703	struct scsi_format *cdb;
5704	struct ctl_lun *lun;
5705	struct ctl_softc *ctl_softc;
5706	int length, defect_list_len;
5707
5708	CTL_DEBUG_PRINT(("ctl_format\n"));
5709
5710	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5711	ctl_softc = control_softc;
5712
5713	cdb = (struct scsi_format *)ctsio->cdb;
5714
5715	length = 0;
5716	if (cdb->byte2 & SF_FMTDATA) {
5717		if (cdb->byte2 & SF_LONGLIST)
5718			length = sizeof(struct scsi_format_header_long);
5719		else
5720			length = sizeof(struct scsi_format_header_short);
5721	}
5722
5723	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5724	 && (length > 0)) {
5725		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5726		ctsio->kern_data_len = length;
5727		ctsio->kern_total_len = length;
5728		ctsio->kern_data_resid = 0;
5729		ctsio->kern_rel_offset = 0;
5730		ctsio->kern_sg_entries = 0;
5731		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5732		ctsio->be_move_done = ctl_config_move_done;
5733		ctl_datamove((union ctl_io *)ctsio);
5734
5735		return (CTL_RETVAL_COMPLETE);
5736	}
5737
5738	defect_list_len = 0;
5739
5740	if (cdb->byte2 & SF_FMTDATA) {
5741		if (cdb->byte2 & SF_LONGLIST) {
5742			struct scsi_format_header_long *header;
5743
5744			header = (struct scsi_format_header_long *)
5745				ctsio->kern_data_ptr;
5746
5747			defect_list_len = scsi_4btoul(header->defect_list_len);
5748			if (defect_list_len != 0) {
5749				ctl_set_invalid_field(ctsio,
5750						      /*sks_valid*/ 1,
5751						      /*command*/ 0,
5752						      /*field*/ 2,
5753						      /*bit_valid*/ 0,
5754						      /*bit*/ 0);
5755				goto bailout;
5756			}
5757		} else {
5758			struct scsi_format_header_short *header;
5759
5760			header = (struct scsi_format_header_short *)
5761				ctsio->kern_data_ptr;
5762
5763			defect_list_len = scsi_2btoul(header->defect_list_len);
5764			if (defect_list_len != 0) {
5765				ctl_set_invalid_field(ctsio,
5766						      /*sks_valid*/ 1,
5767						      /*command*/ 0,
5768						      /*field*/ 2,
5769						      /*bit_valid*/ 0,
5770						      /*bit*/ 0);
5771				goto bailout;
5772			}
5773		}
5774	}
5775
5776	/*
5777	 * The format command will clear out the "Medium format corrupted"
5778	 * status if set by the configuration code.  That status is really
5779	 * just a way to notify the host that we have lost the media, and
5780	 * get them to issue a command that will basically make them think
5781	 * they're blowing away the media.
5782	 */
5783	mtx_lock(&lun->lun_lock);
5784	lun->flags &= ~CTL_LUN_INOPERABLE;
5785	mtx_unlock(&lun->lun_lock);
5786
5787	ctsio->scsi_status = SCSI_STATUS_OK;
5788	ctsio->io_hdr.status = CTL_SUCCESS;
5789bailout:
5790
5791	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5792		free(ctsio->kern_data_ptr, M_CTL);
5793		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5794	}
5795
5796	ctl_done((union ctl_io *)ctsio);
5797	return (CTL_RETVAL_COMPLETE);
5798}
5799
5800int
5801ctl_read_buffer(struct ctl_scsiio *ctsio)
5802{
5803	struct scsi_read_buffer *cdb;
5804	struct ctl_lun *lun;
5805	int buffer_offset, len;
5806	static uint8_t descr[4];
5807	static uint8_t echo_descr[4] = { 0 };
5808
5809	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5810
5811	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5812	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5813
5814	if (lun->flags & CTL_LUN_PR_RESERVED) {
5815		uint32_t residx;
5816
5817		/*
5818		 * XXX KDM need a lock here.
5819		 */
5820		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5821		if ((lun->res_type == SPR_TYPE_EX_AC
5822		  && residx != lun->pr_res_idx)
5823		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5824		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5825		  && !lun->per_res[residx].registered)) {
5826			ctl_set_reservation_conflict(ctsio);
5827			ctl_done((union ctl_io *)ctsio);
5828			return (CTL_RETVAL_COMPLETE);
5829	        }
5830	}
5831
5832	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5833	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5834	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5835		ctl_set_invalid_field(ctsio,
5836				      /*sks_valid*/ 1,
5837				      /*command*/ 1,
5838				      /*field*/ 1,
5839				      /*bit_valid*/ 1,
5840				      /*bit*/ 4);
5841		ctl_done((union ctl_io *)ctsio);
5842		return (CTL_RETVAL_COMPLETE);
5843	}
5844
5845	len = scsi_3btoul(cdb->length);
5846	buffer_offset = scsi_3btoul(cdb->offset);
5847
5848	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5849		ctl_set_invalid_field(ctsio,
5850				      /*sks_valid*/ 1,
5851				      /*command*/ 1,
5852				      /*field*/ 6,
5853				      /*bit_valid*/ 0,
5854				      /*bit*/ 0);
5855		ctl_done((union ctl_io *)ctsio);
5856		return (CTL_RETVAL_COMPLETE);
5857	}
5858
5859	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5860		descr[0] = 0;
5861		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5862		ctsio->kern_data_ptr = descr;
5863		len = min(len, sizeof(descr));
5864	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5865		ctsio->kern_data_ptr = echo_descr;
5866		len = min(len, sizeof(echo_descr));
5867	} else
5868		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5869	ctsio->kern_data_len = len;
5870	ctsio->kern_total_len = len;
5871	ctsio->kern_data_resid = 0;
5872	ctsio->kern_rel_offset = 0;
5873	ctsio->kern_sg_entries = 0;
5874	ctsio->be_move_done = ctl_config_move_done;
5875	ctl_datamove((union ctl_io *)ctsio);
5876
5877	return (CTL_RETVAL_COMPLETE);
5878}
5879
5880int
5881ctl_write_buffer(struct ctl_scsiio *ctsio)
5882{
5883	struct scsi_write_buffer *cdb;
5884	struct ctl_lun *lun;
5885	int buffer_offset, len;
5886
5887	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5888
5889	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5890	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5891
5892	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5893		ctl_set_invalid_field(ctsio,
5894				      /*sks_valid*/ 1,
5895				      /*command*/ 1,
5896				      /*field*/ 1,
5897				      /*bit_valid*/ 1,
5898				      /*bit*/ 4);
5899		ctl_done((union ctl_io *)ctsio);
5900		return (CTL_RETVAL_COMPLETE);
5901	}
5902
5903	len = scsi_3btoul(cdb->length);
5904	buffer_offset = scsi_3btoul(cdb->offset);
5905
5906	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5907		ctl_set_invalid_field(ctsio,
5908				      /*sks_valid*/ 1,
5909				      /*command*/ 1,
5910				      /*field*/ 6,
5911				      /*bit_valid*/ 0,
5912				      /*bit*/ 0);
5913		ctl_done((union ctl_io *)ctsio);
5914		return (CTL_RETVAL_COMPLETE);
5915	}
5916
5917	/*
5918	 * If we've got a kernel request that hasn't been malloced yet,
5919	 * malloc it and tell the caller the data buffer is here.
5920	 */
5921	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5922		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5923		ctsio->kern_data_len = len;
5924		ctsio->kern_total_len = len;
5925		ctsio->kern_data_resid = 0;
5926		ctsio->kern_rel_offset = 0;
5927		ctsio->kern_sg_entries = 0;
5928		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5929		ctsio->be_move_done = ctl_config_move_done;
5930		ctl_datamove((union ctl_io *)ctsio);
5931
5932		return (CTL_RETVAL_COMPLETE);
5933	}
5934
5935	ctl_done((union ctl_io *)ctsio);
5936
5937	return (CTL_RETVAL_COMPLETE);
5938}
5939
5940int
5941ctl_write_same(struct ctl_scsiio *ctsio)
5942{
5943	struct ctl_lun *lun;
5944	struct ctl_lba_len_flags *lbalen;
5945	uint64_t lba;
5946	uint32_t num_blocks;
5947	int len, retval;
5948	uint8_t byte2;
5949
5950	retval = CTL_RETVAL_COMPLETE;
5951
5952	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5953
5954	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5955
5956	switch (ctsio->cdb[0]) {
5957	case WRITE_SAME_10: {
5958		struct scsi_write_same_10 *cdb;
5959
5960		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5961
5962		lba = scsi_4btoul(cdb->addr);
5963		num_blocks = scsi_2btoul(cdb->length);
5964		byte2 = cdb->byte2;
5965		break;
5966	}
5967	case WRITE_SAME_16: {
5968		struct scsi_write_same_16 *cdb;
5969
5970		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5971
5972		lba = scsi_8btou64(cdb->addr);
5973		num_blocks = scsi_4btoul(cdb->length);
5974		byte2 = cdb->byte2;
5975		break;
5976	}
5977	default:
5978		/*
5979		 * We got a command we don't support.  This shouldn't
5980		 * happen, commands should be filtered out above us.
5981		 */
5982		ctl_set_invalid_opcode(ctsio);
5983		ctl_done((union ctl_io *)ctsio);
5984
5985		return (CTL_RETVAL_COMPLETE);
5986		break; /* NOTREACHED */
5987	}
5988
5989	/* NDOB flag can be used only together with UNMAP */
5990	if ((byte2 & (SWS_NDOB | SWS_UNMAP)) == SWS_NDOB) {
5991		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5992		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5993		ctl_done((union ctl_io *)ctsio);
5994		return (CTL_RETVAL_COMPLETE);
5995	}
5996
5997	/*
5998	 * The first check is to make sure we're in bounds, the second
5999	 * check is to catch wrap-around problems.  If the lba + num blocks
6000	 * is less than the lba, then we've wrapped around and the block
6001	 * range is invalid anyway.
6002	 */
6003	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6004	 || ((lba + num_blocks) < lba)) {
6005		ctl_set_lba_out_of_range(ctsio);
6006		ctl_done((union ctl_io *)ctsio);
6007		return (CTL_RETVAL_COMPLETE);
6008	}
6009
6010	/* Zero number of blocks means "to the last logical block" */
6011	if (num_blocks == 0) {
6012		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
6013			ctl_set_invalid_field(ctsio,
6014					      /*sks_valid*/ 0,
6015					      /*command*/ 1,
6016					      /*field*/ 0,
6017					      /*bit_valid*/ 0,
6018					      /*bit*/ 0);
6019			ctl_done((union ctl_io *)ctsio);
6020			return (CTL_RETVAL_COMPLETE);
6021		}
6022		num_blocks = (lun->be_lun->maxlba + 1) - lba;
6023	}
6024
6025	len = lun->be_lun->blocksize;
6026
6027	/*
6028	 * If we've got a kernel request that hasn't been malloced yet,
6029	 * malloc it and tell the caller the data buffer is here.
6030	 */
6031	if ((byte2 & SWS_NDOB) == 0 &&
6032	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6033		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6034		ctsio->kern_data_len = len;
6035		ctsio->kern_total_len = len;
6036		ctsio->kern_data_resid = 0;
6037		ctsio->kern_rel_offset = 0;
6038		ctsio->kern_sg_entries = 0;
6039		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6040		ctsio->be_move_done = ctl_config_move_done;
6041		ctl_datamove((union ctl_io *)ctsio);
6042
6043		return (CTL_RETVAL_COMPLETE);
6044	}
6045
6046	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6047	lbalen->lba = lba;
6048	lbalen->len = num_blocks;
6049	lbalen->flags = byte2;
6050	retval = lun->backend->config_write((union ctl_io *)ctsio);
6051
6052	return (retval);
6053}
6054
6055int
6056ctl_unmap(struct ctl_scsiio *ctsio)
6057{
6058	struct ctl_lun *lun;
6059	struct scsi_unmap *cdb;
6060	struct ctl_ptr_len_flags *ptrlen;
6061	struct scsi_unmap_header *hdr;
6062	struct scsi_unmap_desc *buf, *end, *endnz, *range;
6063	uint64_t lba;
6064	uint32_t num_blocks;
6065	int len, retval;
6066	uint8_t byte2;
6067
6068	retval = CTL_RETVAL_COMPLETE;
6069
6070	CTL_DEBUG_PRINT(("ctl_unmap\n"));
6071
6072	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6073	cdb = (struct scsi_unmap *)ctsio->cdb;
6074
6075	len = scsi_2btoul(cdb->length);
6076	byte2 = cdb->byte2;
6077
6078	/*
6079	 * If we've got a kernel request that hasn't been malloced yet,
6080	 * malloc it and tell the caller the data buffer is here.
6081	 */
6082	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6083		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6084		ctsio->kern_data_len = len;
6085		ctsio->kern_total_len = len;
6086		ctsio->kern_data_resid = 0;
6087		ctsio->kern_rel_offset = 0;
6088		ctsio->kern_sg_entries = 0;
6089		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6090		ctsio->be_move_done = ctl_config_move_done;
6091		ctl_datamove((union ctl_io *)ctsio);
6092
6093		return (CTL_RETVAL_COMPLETE);
6094	}
6095
6096	len = ctsio->kern_total_len - ctsio->kern_data_resid;
6097	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6098	if (len < sizeof (*hdr) ||
6099	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6100	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6101	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6102		ctl_set_invalid_field(ctsio,
6103				      /*sks_valid*/ 0,
6104				      /*command*/ 0,
6105				      /*field*/ 0,
6106				      /*bit_valid*/ 0,
6107				      /*bit*/ 0);
6108		ctl_done((union ctl_io *)ctsio);
6109		return (CTL_RETVAL_COMPLETE);
6110	}
6111	len = scsi_2btoul(hdr->desc_length);
6112	buf = (struct scsi_unmap_desc *)(hdr + 1);
6113	end = buf + len / sizeof(*buf);
6114
6115	endnz = buf;
6116	for (range = buf; range < end; range++) {
6117		lba = scsi_8btou64(range->lba);
6118		num_blocks = scsi_4btoul(range->length);
6119		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6120		 || ((lba + num_blocks) < lba)) {
6121			ctl_set_lba_out_of_range(ctsio);
6122			ctl_done((union ctl_io *)ctsio);
6123			return (CTL_RETVAL_COMPLETE);
6124		}
6125		if (num_blocks != 0)
6126			endnz = range + 1;
6127	}
6128
6129	/*
6130	 * Block backend can not handle zero last range.
6131	 * Filter it out and return if there is nothing left.
6132	 */
6133	len = (uint8_t *)endnz - (uint8_t *)buf;
6134	if (len == 0) {
6135		ctl_set_success(ctsio);
6136		ctl_done((union ctl_io *)ctsio);
6137		return (CTL_RETVAL_COMPLETE);
6138	}
6139
6140	mtx_lock(&lun->lun_lock);
6141	ptrlen = (struct ctl_ptr_len_flags *)
6142	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6143	ptrlen->ptr = (void *)buf;
6144	ptrlen->len = len;
6145	ptrlen->flags = byte2;
6146	ctl_check_blocked(lun);
6147	mtx_unlock(&lun->lun_lock);
6148
6149	retval = lun->backend->config_write((union ctl_io *)ctsio);
6150	return (retval);
6151}
6152
6153/*
6154 * Note that this function currently doesn't actually do anything inside
6155 * CTL to enforce things if the DQue bit is turned on.
6156 *
6157 * Also note that this function can't be used in the default case, because
6158 * the DQue bit isn't set in the changeable mask for the control mode page
6159 * anyway.  This is just here as an example for how to implement a page
6160 * handler, and a placeholder in case we want to allow the user to turn
6161 * tagged queueing on and off.
6162 *
6163 * The D_SENSE bit handling is functional, however, and will turn
6164 * descriptor sense on and off for a given LUN.
6165 */
6166int
6167ctl_control_page_handler(struct ctl_scsiio *ctsio,
6168			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6169{
6170	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6171	struct ctl_lun *lun;
6172	struct ctl_softc *softc;
6173	int set_ua;
6174	uint32_t initidx;
6175
6176	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6177	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6178	set_ua = 0;
6179
6180	user_cp = (struct scsi_control_page *)page_ptr;
6181	current_cp = (struct scsi_control_page *)
6182		(page_index->page_data + (page_index->page_len *
6183		CTL_PAGE_CURRENT));
6184	saved_cp = (struct scsi_control_page *)
6185		(page_index->page_data + (page_index->page_len *
6186		CTL_PAGE_SAVED));
6187
6188	softc = control_softc;
6189
6190	mtx_lock(&lun->lun_lock);
6191	if (((current_cp->rlec & SCP_DSENSE) == 0)
6192	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6193		/*
6194		 * Descriptor sense is currently turned off and the user
6195		 * wants to turn it on.
6196		 */
6197		current_cp->rlec |= SCP_DSENSE;
6198		saved_cp->rlec |= SCP_DSENSE;
6199		lun->flags |= CTL_LUN_SENSE_DESC;
6200		set_ua = 1;
6201	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6202		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6203		/*
6204		 * Descriptor sense is currently turned on, and the user
6205		 * wants to turn it off.
6206		 */
6207		current_cp->rlec &= ~SCP_DSENSE;
6208		saved_cp->rlec &= ~SCP_DSENSE;
6209		lun->flags &= ~CTL_LUN_SENSE_DESC;
6210		set_ua = 1;
6211	}
6212	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6213	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6214		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6215		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6216		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6217		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6218		set_ua = 1;
6219	}
6220	if (set_ua != 0) {
6221		int i;
6222		/*
6223		 * Let other initiators know that the mode
6224		 * parameters for this LUN have changed.
6225		 */
6226		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6227			if (i == initidx)
6228				continue;
6229
6230			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6231		}
6232	}
6233	mtx_unlock(&lun->lun_lock);
6234
6235	return (0);
6236}
6237
6238int
6239ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6240		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6241{
6242	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6243	struct ctl_lun *lun;
6244	int set_ua;
6245	uint32_t initidx;
6246
6247	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6248	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6249	set_ua = 0;
6250
6251	user_cp = (struct scsi_caching_page *)page_ptr;
6252	current_cp = (struct scsi_caching_page *)
6253		(page_index->page_data + (page_index->page_len *
6254		CTL_PAGE_CURRENT));
6255	saved_cp = (struct scsi_caching_page *)
6256		(page_index->page_data + (page_index->page_len *
6257		CTL_PAGE_SAVED));
6258
6259	mtx_lock(&lun->lun_lock);
6260	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6261	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6262		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6263		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6264		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6265		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6266		set_ua = 1;
6267	}
6268	if (set_ua != 0) {
6269		int i;
6270		/*
6271		 * Let other initiators know that the mode
6272		 * parameters for this LUN have changed.
6273		 */
6274		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6275			if (i == initidx)
6276				continue;
6277
6278			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6279		}
6280	}
6281	mtx_unlock(&lun->lun_lock);
6282
6283	return (0);
6284}
6285
6286int
6287ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6288		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6289{
6290	return (0);
6291}
6292
6293int
6294ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6295			   struct ctl_page_index *page_index, int pc)
6296{
6297	struct copan_power_subpage *page;
6298
6299	page = (struct copan_power_subpage *)page_index->page_data +
6300		(page_index->page_len * pc);
6301
6302	switch (pc) {
6303	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6304		/*
6305		 * We don't update the changable bits for this page.
6306		 */
6307		break;
6308	case SMS_PAGE_CTRL_CURRENT >> 6:
6309	case SMS_PAGE_CTRL_DEFAULT >> 6:
6310	case SMS_PAGE_CTRL_SAVED >> 6:
6311#ifdef NEEDTOPORT
6312		ctl_update_power_subpage(page);
6313#endif
6314		break;
6315	default:
6316#ifdef NEEDTOPORT
6317		EPRINT(0, "Invalid PC %d!!", pc);
6318#endif
6319		break;
6320	}
6321	return (0);
6322}
6323
6324
6325int
6326ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6327		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6328{
6329	struct copan_aps_subpage *user_sp;
6330	struct copan_aps_subpage *current_sp;
6331	union ctl_modepage_info *modepage_info;
6332	struct ctl_softc *softc;
6333	struct ctl_lun *lun;
6334	int retval;
6335
6336	retval = CTL_RETVAL_COMPLETE;
6337	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6338		     (page_index->page_len * CTL_PAGE_CURRENT));
6339	softc = control_softc;
6340	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6341
6342	user_sp = (struct copan_aps_subpage *)page_ptr;
6343
6344	modepage_info = (union ctl_modepage_info *)
6345		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6346
6347	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6348	modepage_info->header.subpage = page_index->subpage;
6349	modepage_info->aps.lock_active = user_sp->lock_active;
6350
6351	mtx_lock(&softc->ctl_lock);
6352
6353	/*
6354	 * If there is a request to lock the LUN and another LUN is locked
6355	 * this is an error. If the requested LUN is already locked ignore
6356	 * the request. If no LUN is locked attempt to lock it.
6357	 * if there is a request to unlock the LUN and the LUN is currently
6358	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6359	 * if another LUN is locked or no LUN is locked.
6360	 */
6361	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6362		if (softc->aps_locked_lun == lun->lun) {
6363			/*
6364			 * This LUN is already locked, so we're done.
6365			 */
6366			retval = CTL_RETVAL_COMPLETE;
6367		} else if (softc->aps_locked_lun == 0) {
6368			/*
6369			 * No one has the lock, pass the request to the
6370			 * backend.
6371			 */
6372			retval = lun->backend->config_write(
6373				(union ctl_io *)ctsio);
6374		} else {
6375			/*
6376			 * Someone else has the lock, throw out the request.
6377			 */
6378			ctl_set_already_locked(ctsio);
6379			free(ctsio->kern_data_ptr, M_CTL);
6380			ctl_done((union ctl_io *)ctsio);
6381
6382			/*
6383			 * Set the return value so that ctl_do_mode_select()
6384			 * won't try to complete the command.  We already
6385			 * completed it here.
6386			 */
6387			retval = CTL_RETVAL_ERROR;
6388		}
6389	} else if (softc->aps_locked_lun == lun->lun) {
6390		/*
6391		 * This LUN is locked, so pass the unlock request to the
6392		 * backend.
6393		 */
6394		retval = lun->backend->config_write((union ctl_io *)ctsio);
6395	}
6396	mtx_unlock(&softc->ctl_lock);
6397
6398	return (retval);
6399}
6400
6401int
6402ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6403				struct ctl_page_index *page_index,
6404				uint8_t *page_ptr)
6405{
6406	uint8_t *c;
6407	int i;
6408
6409	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6410	ctl_time_io_secs =
6411		(c[0] << 8) |
6412		(c[1] << 0) |
6413		0;
6414	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6415	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6416	printf("page data:");
6417	for (i=0; i<8; i++)
6418		printf(" %.2x",page_ptr[i]);
6419	printf("\n");
6420	return (0);
6421}
6422
6423int
6424ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6425			       struct ctl_page_index *page_index,
6426			       int pc)
6427{
6428	struct copan_debugconf_subpage *page;
6429
6430	page = (struct copan_debugconf_subpage *)page_index->page_data +
6431		(page_index->page_len * pc);
6432
6433	switch (pc) {
6434	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6435	case SMS_PAGE_CTRL_DEFAULT >> 6:
6436	case SMS_PAGE_CTRL_SAVED >> 6:
6437		/*
6438		 * We don't update the changable or default bits for this page.
6439		 */
6440		break;
6441	case SMS_PAGE_CTRL_CURRENT >> 6:
6442		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6443		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6444		break;
6445	default:
6446#ifdef NEEDTOPORT
6447		EPRINT(0, "Invalid PC %d!!", pc);
6448#endif /* NEEDTOPORT */
6449		break;
6450	}
6451	return (0);
6452}
6453
6454
6455static int
6456ctl_do_mode_select(union ctl_io *io)
6457{
6458	struct scsi_mode_page_header *page_header;
6459	struct ctl_page_index *page_index;
6460	struct ctl_scsiio *ctsio;
6461	int control_dev, page_len;
6462	int page_len_offset, page_len_size;
6463	union ctl_modepage_info *modepage_info;
6464	struct ctl_lun *lun;
6465	int *len_left, *len_used;
6466	int retval, i;
6467
6468	ctsio = &io->scsiio;
6469	page_index = NULL;
6470	page_len = 0;
6471	retval = CTL_RETVAL_COMPLETE;
6472
6473	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6474
6475	if (lun->be_lun->lun_type != T_DIRECT)
6476		control_dev = 1;
6477	else
6478		control_dev = 0;
6479
6480	modepage_info = (union ctl_modepage_info *)
6481		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6482	len_left = &modepage_info->header.len_left;
6483	len_used = &modepage_info->header.len_used;
6484
6485do_next_page:
6486
6487	page_header = (struct scsi_mode_page_header *)
6488		(ctsio->kern_data_ptr + *len_used);
6489
6490	if (*len_left == 0) {
6491		free(ctsio->kern_data_ptr, M_CTL);
6492		ctl_set_success(ctsio);
6493		ctl_done((union ctl_io *)ctsio);
6494		return (CTL_RETVAL_COMPLETE);
6495	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6496
6497		free(ctsio->kern_data_ptr, M_CTL);
6498		ctl_set_param_len_error(ctsio);
6499		ctl_done((union ctl_io *)ctsio);
6500		return (CTL_RETVAL_COMPLETE);
6501
6502	} else if ((page_header->page_code & SMPH_SPF)
6503		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6504
6505		free(ctsio->kern_data_ptr, M_CTL);
6506		ctl_set_param_len_error(ctsio);
6507		ctl_done((union ctl_io *)ctsio);
6508		return (CTL_RETVAL_COMPLETE);
6509	}
6510
6511
6512	/*
6513	 * XXX KDM should we do something with the block descriptor?
6514	 */
6515	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6516
6517		if ((control_dev != 0)
6518		 && (lun->mode_pages.index[i].page_flags &
6519		     CTL_PAGE_FLAG_DISK_ONLY))
6520			continue;
6521
6522		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6523		    (page_header->page_code & SMPH_PC_MASK))
6524			continue;
6525
6526		/*
6527		 * If neither page has a subpage code, then we've got a
6528		 * match.
6529		 */
6530		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6531		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6532			page_index = &lun->mode_pages.index[i];
6533			page_len = page_header->page_length;
6534			break;
6535		}
6536
6537		/*
6538		 * If both pages have subpages, then the subpage numbers
6539		 * have to match.
6540		 */
6541		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6542		  && (page_header->page_code & SMPH_SPF)) {
6543			struct scsi_mode_page_header_sp *sph;
6544
6545			sph = (struct scsi_mode_page_header_sp *)page_header;
6546
6547			if (lun->mode_pages.index[i].subpage ==
6548			    sph->subpage) {
6549				page_index = &lun->mode_pages.index[i];
6550				page_len = scsi_2btoul(sph->page_length);
6551				break;
6552			}
6553		}
6554	}
6555
6556	/*
6557	 * If we couldn't find the page, or if we don't have a mode select
6558	 * handler for it, send back an error to the user.
6559	 */
6560	if ((page_index == NULL)
6561	 || (page_index->select_handler == NULL)) {
6562		ctl_set_invalid_field(ctsio,
6563				      /*sks_valid*/ 1,
6564				      /*command*/ 0,
6565				      /*field*/ *len_used,
6566				      /*bit_valid*/ 0,
6567				      /*bit*/ 0);
6568		free(ctsio->kern_data_ptr, M_CTL);
6569		ctl_done((union ctl_io *)ctsio);
6570		return (CTL_RETVAL_COMPLETE);
6571	}
6572
6573	if (page_index->page_code & SMPH_SPF) {
6574		page_len_offset = 2;
6575		page_len_size = 2;
6576	} else {
6577		page_len_size = 1;
6578		page_len_offset = 1;
6579	}
6580
6581	/*
6582	 * If the length the initiator gives us isn't the one we specify in
6583	 * the mode page header, or if they didn't specify enough data in
6584	 * the CDB to avoid truncating this page, kick out the request.
6585	 */
6586	if ((page_len != (page_index->page_len - page_len_offset -
6587			  page_len_size))
6588	 || (*len_left < page_index->page_len)) {
6589
6590
6591		ctl_set_invalid_field(ctsio,
6592				      /*sks_valid*/ 1,
6593				      /*command*/ 0,
6594				      /*field*/ *len_used + page_len_offset,
6595				      /*bit_valid*/ 0,
6596				      /*bit*/ 0);
6597		free(ctsio->kern_data_ptr, M_CTL);
6598		ctl_done((union ctl_io *)ctsio);
6599		return (CTL_RETVAL_COMPLETE);
6600	}
6601
6602	/*
6603	 * Run through the mode page, checking to make sure that the bits
6604	 * the user changed are actually legal for him to change.
6605	 */
6606	for (i = 0; i < page_index->page_len; i++) {
6607		uint8_t *user_byte, *change_mask, *current_byte;
6608		int bad_bit;
6609		int j;
6610
6611		user_byte = (uint8_t *)page_header + i;
6612		change_mask = page_index->page_data +
6613			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6614		current_byte = page_index->page_data +
6615			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6616
6617		/*
6618		 * Check to see whether the user set any bits in this byte
6619		 * that he is not allowed to set.
6620		 */
6621		if ((*user_byte & ~(*change_mask)) ==
6622		    (*current_byte & ~(*change_mask)))
6623			continue;
6624
6625		/*
6626		 * Go through bit by bit to determine which one is illegal.
6627		 */
6628		bad_bit = 0;
6629		for (j = 7; j >= 0; j--) {
6630			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6631			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6632				bad_bit = i;
6633				break;
6634			}
6635		}
6636		ctl_set_invalid_field(ctsio,
6637				      /*sks_valid*/ 1,
6638				      /*command*/ 0,
6639				      /*field*/ *len_used + i,
6640				      /*bit_valid*/ 1,
6641				      /*bit*/ bad_bit);
6642		free(ctsio->kern_data_ptr, M_CTL);
6643		ctl_done((union ctl_io *)ctsio);
6644		return (CTL_RETVAL_COMPLETE);
6645	}
6646
6647	/*
6648	 * Decrement these before we call the page handler, since we may
6649	 * end up getting called back one way or another before the handler
6650	 * returns to this context.
6651	 */
6652	*len_left -= page_index->page_len;
6653	*len_used += page_index->page_len;
6654
6655	retval = page_index->select_handler(ctsio, page_index,
6656					    (uint8_t *)page_header);
6657
6658	/*
6659	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6660	 * wait until this queued command completes to finish processing
6661	 * the mode page.  If it returns anything other than
6662	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6663	 * already set the sense information, freed the data pointer, and
6664	 * completed the io for us.
6665	 */
6666	if (retval != CTL_RETVAL_COMPLETE)
6667		goto bailout_no_done;
6668
6669	/*
6670	 * If the initiator sent us more than one page, parse the next one.
6671	 */
6672	if (*len_left > 0)
6673		goto do_next_page;
6674
6675	ctl_set_success(ctsio);
6676	free(ctsio->kern_data_ptr, M_CTL);
6677	ctl_done((union ctl_io *)ctsio);
6678
6679bailout_no_done:
6680
6681	return (CTL_RETVAL_COMPLETE);
6682
6683}
6684
6685int
6686ctl_mode_select(struct ctl_scsiio *ctsio)
6687{
6688	int param_len, pf, sp;
6689	int header_size, bd_len;
6690	int len_left, len_used;
6691	struct ctl_page_index *page_index;
6692	struct ctl_lun *lun;
6693	int control_dev, page_len;
6694	union ctl_modepage_info *modepage_info;
6695	int retval;
6696
6697	pf = 0;
6698	sp = 0;
6699	page_len = 0;
6700	len_used = 0;
6701	len_left = 0;
6702	retval = 0;
6703	bd_len = 0;
6704	page_index = NULL;
6705
6706	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6707
6708	if (lun->be_lun->lun_type != T_DIRECT)
6709		control_dev = 1;
6710	else
6711		control_dev = 0;
6712
6713	switch (ctsio->cdb[0]) {
6714	case MODE_SELECT_6: {
6715		struct scsi_mode_select_6 *cdb;
6716
6717		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6718
6719		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6720		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6721
6722		param_len = cdb->length;
6723		header_size = sizeof(struct scsi_mode_header_6);
6724		break;
6725	}
6726	case MODE_SELECT_10: {
6727		struct scsi_mode_select_10 *cdb;
6728
6729		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6730
6731		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6732		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6733
6734		param_len = scsi_2btoul(cdb->length);
6735		header_size = sizeof(struct scsi_mode_header_10);
6736		break;
6737	}
6738	default:
6739		ctl_set_invalid_opcode(ctsio);
6740		ctl_done((union ctl_io *)ctsio);
6741		return (CTL_RETVAL_COMPLETE);
6742		break; /* NOTREACHED */
6743	}
6744
6745	/*
6746	 * From SPC-3:
6747	 * "A parameter list length of zero indicates that the Data-Out Buffer
6748	 * shall be empty. This condition shall not be considered as an error."
6749	 */
6750	if (param_len == 0) {
6751		ctl_set_success(ctsio);
6752		ctl_done((union ctl_io *)ctsio);
6753		return (CTL_RETVAL_COMPLETE);
6754	}
6755
6756	/*
6757	 * Since we'll hit this the first time through, prior to
6758	 * allocation, we don't need to free a data buffer here.
6759	 */
6760	if (param_len < header_size) {
6761		ctl_set_param_len_error(ctsio);
6762		ctl_done((union ctl_io *)ctsio);
6763		return (CTL_RETVAL_COMPLETE);
6764	}
6765
6766	/*
6767	 * Allocate the data buffer and grab the user's data.  In theory,
6768	 * we shouldn't have to sanity check the parameter list length here
6769	 * because the maximum size is 64K.  We should be able to malloc
6770	 * that much without too many problems.
6771	 */
6772	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6773		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6774		ctsio->kern_data_len = param_len;
6775		ctsio->kern_total_len = param_len;
6776		ctsio->kern_data_resid = 0;
6777		ctsio->kern_rel_offset = 0;
6778		ctsio->kern_sg_entries = 0;
6779		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6780		ctsio->be_move_done = ctl_config_move_done;
6781		ctl_datamove((union ctl_io *)ctsio);
6782
6783		return (CTL_RETVAL_COMPLETE);
6784	}
6785
6786	switch (ctsio->cdb[0]) {
6787	case MODE_SELECT_6: {
6788		struct scsi_mode_header_6 *mh6;
6789
6790		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6791		bd_len = mh6->blk_desc_len;
6792		break;
6793	}
6794	case MODE_SELECT_10: {
6795		struct scsi_mode_header_10 *mh10;
6796
6797		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6798		bd_len = scsi_2btoul(mh10->blk_desc_len);
6799		break;
6800	}
6801	default:
6802		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6803		break;
6804	}
6805
6806	if (param_len < (header_size + bd_len)) {
6807		free(ctsio->kern_data_ptr, M_CTL);
6808		ctl_set_param_len_error(ctsio);
6809		ctl_done((union ctl_io *)ctsio);
6810		return (CTL_RETVAL_COMPLETE);
6811	}
6812
6813	/*
6814	 * Set the IO_CONT flag, so that if this I/O gets passed to
6815	 * ctl_config_write_done(), it'll get passed back to
6816	 * ctl_do_mode_select() for further processing, or completion if
6817	 * we're all done.
6818	 */
6819	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6820	ctsio->io_cont = ctl_do_mode_select;
6821
6822	modepage_info = (union ctl_modepage_info *)
6823		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6824
6825	memset(modepage_info, 0, sizeof(*modepage_info));
6826
6827	len_left = param_len - header_size - bd_len;
6828	len_used = header_size + bd_len;
6829
6830	modepage_info->header.len_left = len_left;
6831	modepage_info->header.len_used = len_used;
6832
6833	return (ctl_do_mode_select((union ctl_io *)ctsio));
6834}
6835
6836int
6837ctl_mode_sense(struct ctl_scsiio *ctsio)
6838{
6839	struct ctl_lun *lun;
6840	int pc, page_code, dbd, llba, subpage;
6841	int alloc_len, page_len, header_len, total_len;
6842	struct scsi_mode_block_descr *block_desc;
6843	struct ctl_page_index *page_index;
6844	int control_dev;
6845
6846	dbd = 0;
6847	llba = 0;
6848	block_desc = NULL;
6849	page_index = NULL;
6850
6851	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6852
6853	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6854
6855	if (lun->be_lun->lun_type != T_DIRECT)
6856		control_dev = 1;
6857	else
6858		control_dev = 0;
6859
6860	if (lun->flags & CTL_LUN_PR_RESERVED) {
6861		uint32_t residx;
6862
6863		/*
6864		 * XXX KDM need a lock here.
6865		 */
6866		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6867		if ((lun->res_type == SPR_TYPE_EX_AC
6868		  && residx != lun->pr_res_idx)
6869		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6870		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6871		  && !lun->per_res[residx].registered)) {
6872			ctl_set_reservation_conflict(ctsio);
6873			ctl_done((union ctl_io *)ctsio);
6874			return (CTL_RETVAL_COMPLETE);
6875		}
6876	}
6877
6878	switch (ctsio->cdb[0]) {
6879	case MODE_SENSE_6: {
6880		struct scsi_mode_sense_6 *cdb;
6881
6882		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6883
6884		header_len = sizeof(struct scsi_mode_hdr_6);
6885		if (cdb->byte2 & SMS_DBD)
6886			dbd = 1;
6887		else
6888			header_len += sizeof(struct scsi_mode_block_descr);
6889
6890		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6891		page_code = cdb->page & SMS_PAGE_CODE;
6892		subpage = cdb->subpage;
6893		alloc_len = cdb->length;
6894		break;
6895	}
6896	case MODE_SENSE_10: {
6897		struct scsi_mode_sense_10 *cdb;
6898
6899		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6900
6901		header_len = sizeof(struct scsi_mode_hdr_10);
6902
6903		if (cdb->byte2 & SMS_DBD)
6904			dbd = 1;
6905		else
6906			header_len += sizeof(struct scsi_mode_block_descr);
6907		if (cdb->byte2 & SMS10_LLBAA)
6908			llba = 1;
6909		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6910		page_code = cdb->page & SMS_PAGE_CODE;
6911		subpage = cdb->subpage;
6912		alloc_len = scsi_2btoul(cdb->length);
6913		break;
6914	}
6915	default:
6916		ctl_set_invalid_opcode(ctsio);
6917		ctl_done((union ctl_io *)ctsio);
6918		return (CTL_RETVAL_COMPLETE);
6919		break; /* NOTREACHED */
6920	}
6921
6922	/*
6923	 * We have to make a first pass through to calculate the size of
6924	 * the pages that match the user's query.  Then we allocate enough
6925	 * memory to hold it, and actually copy the data into the buffer.
6926	 */
6927	switch (page_code) {
6928	case SMS_ALL_PAGES_PAGE: {
6929		int i;
6930
6931		page_len = 0;
6932
6933		/*
6934		 * At the moment, values other than 0 and 0xff here are
6935		 * reserved according to SPC-3.
6936		 */
6937		if ((subpage != SMS_SUBPAGE_PAGE_0)
6938		 && (subpage != SMS_SUBPAGE_ALL)) {
6939			ctl_set_invalid_field(ctsio,
6940					      /*sks_valid*/ 1,
6941					      /*command*/ 1,
6942					      /*field*/ 3,
6943					      /*bit_valid*/ 0,
6944					      /*bit*/ 0);
6945			ctl_done((union ctl_io *)ctsio);
6946			return (CTL_RETVAL_COMPLETE);
6947		}
6948
6949		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6950			if ((control_dev != 0)
6951			 && (lun->mode_pages.index[i].page_flags &
6952			     CTL_PAGE_FLAG_DISK_ONLY))
6953				continue;
6954
6955			/*
6956			 * We don't use this subpage if the user didn't
6957			 * request all subpages.
6958			 */
6959			if ((lun->mode_pages.index[i].subpage != 0)
6960			 && (subpage == SMS_SUBPAGE_PAGE_0))
6961				continue;
6962
6963#if 0
6964			printf("found page %#x len %d\n",
6965			       lun->mode_pages.index[i].page_code &
6966			       SMPH_PC_MASK,
6967			       lun->mode_pages.index[i].page_len);
6968#endif
6969			page_len += lun->mode_pages.index[i].page_len;
6970		}
6971		break;
6972	}
6973	default: {
6974		int i;
6975
6976		page_len = 0;
6977
6978		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6979			/* Look for the right page code */
6980			if ((lun->mode_pages.index[i].page_code &
6981			     SMPH_PC_MASK) != page_code)
6982				continue;
6983
6984			/* Look for the right subpage or the subpage wildcard*/
6985			if ((lun->mode_pages.index[i].subpage != subpage)
6986			 && (subpage != SMS_SUBPAGE_ALL))
6987				continue;
6988
6989			/* Make sure the page is supported for this dev type */
6990			if ((control_dev != 0)
6991			 && (lun->mode_pages.index[i].page_flags &
6992			     CTL_PAGE_FLAG_DISK_ONLY))
6993				continue;
6994
6995#if 0
6996			printf("found page %#x len %d\n",
6997			       lun->mode_pages.index[i].page_code &
6998			       SMPH_PC_MASK,
6999			       lun->mode_pages.index[i].page_len);
7000#endif
7001
7002			page_len += lun->mode_pages.index[i].page_len;
7003		}
7004
7005		if (page_len == 0) {
7006			ctl_set_invalid_field(ctsio,
7007					      /*sks_valid*/ 1,
7008					      /*command*/ 1,
7009					      /*field*/ 2,
7010					      /*bit_valid*/ 1,
7011					      /*bit*/ 5);
7012			ctl_done((union ctl_io *)ctsio);
7013			return (CTL_RETVAL_COMPLETE);
7014		}
7015		break;
7016	}
7017	}
7018
7019	total_len = header_len + page_len;
7020#if 0
7021	printf("header_len = %d, page_len = %d, total_len = %d\n",
7022	       header_len, page_len, total_len);
7023#endif
7024
7025	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7026	ctsio->kern_sg_entries = 0;
7027	ctsio->kern_data_resid = 0;
7028	ctsio->kern_rel_offset = 0;
7029	if (total_len < alloc_len) {
7030		ctsio->residual = alloc_len - total_len;
7031		ctsio->kern_data_len = total_len;
7032		ctsio->kern_total_len = total_len;
7033	} else {
7034		ctsio->residual = 0;
7035		ctsio->kern_data_len = alloc_len;
7036		ctsio->kern_total_len = alloc_len;
7037	}
7038
7039	switch (ctsio->cdb[0]) {
7040	case MODE_SENSE_6: {
7041		struct scsi_mode_hdr_6 *header;
7042
7043		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7044
7045		header->datalen = ctl_min(total_len - 1, 254);
7046		if (control_dev == 0)
7047			header->dev_specific = 0x10; /* DPOFUA */
7048		if (dbd)
7049			header->block_descr_len = 0;
7050		else
7051			header->block_descr_len =
7052				sizeof(struct scsi_mode_block_descr);
7053		block_desc = (struct scsi_mode_block_descr *)&header[1];
7054		break;
7055	}
7056	case MODE_SENSE_10: {
7057		struct scsi_mode_hdr_10 *header;
7058		int datalen;
7059
7060		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7061
7062		datalen = ctl_min(total_len - 2, 65533);
7063		scsi_ulto2b(datalen, header->datalen);
7064		if (control_dev == 0)
7065			header->dev_specific = 0x10; /* DPOFUA */
7066		if (dbd)
7067			scsi_ulto2b(0, header->block_descr_len);
7068		else
7069			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7070				    header->block_descr_len);
7071		block_desc = (struct scsi_mode_block_descr *)&header[1];
7072		break;
7073	}
7074	default:
7075		panic("invalid CDB type %#x", ctsio->cdb[0]);
7076		break; /* NOTREACHED */
7077	}
7078
7079	/*
7080	 * If we've got a disk, use its blocksize in the block
7081	 * descriptor.  Otherwise, just set it to 0.
7082	 */
7083	if (dbd == 0) {
7084		if (control_dev == 0)
7085			scsi_ulto3b(lun->be_lun->blocksize,
7086				    block_desc->block_len);
7087		else
7088			scsi_ulto3b(0, block_desc->block_len);
7089	}
7090
7091	switch (page_code) {
7092	case SMS_ALL_PAGES_PAGE: {
7093		int i, data_used;
7094
7095		data_used = header_len;
7096		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7097			struct ctl_page_index *page_index;
7098
7099			page_index = &lun->mode_pages.index[i];
7100
7101			if ((control_dev != 0)
7102			 && (page_index->page_flags &
7103			    CTL_PAGE_FLAG_DISK_ONLY))
7104				continue;
7105
7106			/*
7107			 * We don't use this subpage if the user didn't
7108			 * request all subpages.  We already checked (above)
7109			 * to make sure the user only specified a subpage
7110			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7111			 */
7112			if ((page_index->subpage != 0)
7113			 && (subpage == SMS_SUBPAGE_PAGE_0))
7114				continue;
7115
7116			/*
7117			 * Call the handler, if it exists, to update the
7118			 * page to the latest values.
7119			 */
7120			if (page_index->sense_handler != NULL)
7121				page_index->sense_handler(ctsio, page_index,pc);
7122
7123			memcpy(ctsio->kern_data_ptr + data_used,
7124			       page_index->page_data +
7125			       (page_index->page_len * pc),
7126			       page_index->page_len);
7127			data_used += page_index->page_len;
7128		}
7129		break;
7130	}
7131	default: {
7132		int i, data_used;
7133
7134		data_used = header_len;
7135
7136		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7137			struct ctl_page_index *page_index;
7138
7139			page_index = &lun->mode_pages.index[i];
7140
7141			/* Look for the right page code */
7142			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7143				continue;
7144
7145			/* Look for the right subpage or the subpage wildcard*/
7146			if ((page_index->subpage != subpage)
7147			 && (subpage != SMS_SUBPAGE_ALL))
7148				continue;
7149
7150			/* Make sure the page is supported for this dev type */
7151			if ((control_dev != 0)
7152			 && (page_index->page_flags &
7153			     CTL_PAGE_FLAG_DISK_ONLY))
7154				continue;
7155
7156			/*
7157			 * Call the handler, if it exists, to update the
7158			 * page to the latest values.
7159			 */
7160			if (page_index->sense_handler != NULL)
7161				page_index->sense_handler(ctsio, page_index,pc);
7162
7163			memcpy(ctsio->kern_data_ptr + data_used,
7164			       page_index->page_data +
7165			       (page_index->page_len * pc),
7166			       page_index->page_len);
7167			data_used += page_index->page_len;
7168		}
7169		break;
7170	}
7171	}
7172
7173	ctsio->scsi_status = SCSI_STATUS_OK;
7174
7175	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7176	ctsio->be_move_done = ctl_config_move_done;
7177	ctl_datamove((union ctl_io *)ctsio);
7178
7179	return (CTL_RETVAL_COMPLETE);
7180}
7181
7182int
7183ctl_read_capacity(struct ctl_scsiio *ctsio)
7184{
7185	struct scsi_read_capacity *cdb;
7186	struct scsi_read_capacity_data *data;
7187	struct ctl_lun *lun;
7188	uint32_t lba;
7189
7190	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7191
7192	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7193
7194	lba = scsi_4btoul(cdb->addr);
7195	if (((cdb->pmi & SRC_PMI) == 0)
7196	 && (lba != 0)) {
7197		ctl_set_invalid_field(/*ctsio*/ ctsio,
7198				      /*sks_valid*/ 1,
7199				      /*command*/ 1,
7200				      /*field*/ 2,
7201				      /*bit_valid*/ 0,
7202				      /*bit*/ 0);
7203		ctl_done((union ctl_io *)ctsio);
7204		return (CTL_RETVAL_COMPLETE);
7205	}
7206
7207	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7208
7209	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7210	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7211	ctsio->residual = 0;
7212	ctsio->kern_data_len = sizeof(*data);
7213	ctsio->kern_total_len = sizeof(*data);
7214	ctsio->kern_data_resid = 0;
7215	ctsio->kern_rel_offset = 0;
7216	ctsio->kern_sg_entries = 0;
7217
7218	/*
7219	 * If the maximum LBA is greater than 0xfffffffe, the user must
7220	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7221	 * serivce action set.
7222	 */
7223	if (lun->be_lun->maxlba > 0xfffffffe)
7224		scsi_ulto4b(0xffffffff, data->addr);
7225	else
7226		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7227
7228	/*
7229	 * XXX KDM this may not be 512 bytes...
7230	 */
7231	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7232
7233	ctsio->scsi_status = SCSI_STATUS_OK;
7234
7235	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7236	ctsio->be_move_done = ctl_config_move_done;
7237	ctl_datamove((union ctl_io *)ctsio);
7238
7239	return (CTL_RETVAL_COMPLETE);
7240}
7241
7242int
7243ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7244{
7245	struct scsi_read_capacity_16 *cdb;
7246	struct scsi_read_capacity_data_long *data;
7247	struct ctl_lun *lun;
7248	uint64_t lba;
7249	uint32_t alloc_len;
7250
7251	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7252
7253	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7254
7255	alloc_len = scsi_4btoul(cdb->alloc_len);
7256	lba = scsi_8btou64(cdb->addr);
7257
7258	if ((cdb->reladr & SRC16_PMI)
7259	 && (lba != 0)) {
7260		ctl_set_invalid_field(/*ctsio*/ ctsio,
7261				      /*sks_valid*/ 1,
7262				      /*command*/ 1,
7263				      /*field*/ 2,
7264				      /*bit_valid*/ 0,
7265				      /*bit*/ 0);
7266		ctl_done((union ctl_io *)ctsio);
7267		return (CTL_RETVAL_COMPLETE);
7268	}
7269
7270	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7271
7272	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7273	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7274
7275	if (sizeof(*data) < alloc_len) {
7276		ctsio->residual = alloc_len - sizeof(*data);
7277		ctsio->kern_data_len = sizeof(*data);
7278		ctsio->kern_total_len = sizeof(*data);
7279	} else {
7280		ctsio->residual = 0;
7281		ctsio->kern_data_len = alloc_len;
7282		ctsio->kern_total_len = alloc_len;
7283	}
7284	ctsio->kern_data_resid = 0;
7285	ctsio->kern_rel_offset = 0;
7286	ctsio->kern_sg_entries = 0;
7287
7288	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7289	/* XXX KDM this may not be 512 bytes... */
7290	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7291	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7292	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7293	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7294		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7295
7296	ctsio->scsi_status = SCSI_STATUS_OK;
7297
7298	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7299	ctsio->be_move_done = ctl_config_move_done;
7300	ctl_datamove((union ctl_io *)ctsio);
7301
7302	return (CTL_RETVAL_COMPLETE);
7303}
7304
7305int
7306ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7307{
7308	struct scsi_maintenance_in *cdb;
7309	int retval;
7310	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7311	int num_target_port_groups, num_target_ports, single;
7312	struct ctl_lun *lun;
7313	struct ctl_softc *softc;
7314	struct ctl_port *port;
7315	struct scsi_target_group_data *rtg_ptr;
7316	struct scsi_target_group_data_extended *rtg_ext_ptr;
7317	struct scsi_target_port_group_descriptor *tpg_desc;
7318
7319	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7320
7321	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7322	softc = control_softc;
7323	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7324
7325	retval = CTL_RETVAL_COMPLETE;
7326
7327	switch (cdb->byte2 & STG_PDF_MASK) {
7328	case STG_PDF_LENGTH:
7329		ext = 0;
7330		break;
7331	case STG_PDF_EXTENDED:
7332		ext = 1;
7333		break;
7334	default:
7335		ctl_set_invalid_field(/*ctsio*/ ctsio,
7336				      /*sks_valid*/ 1,
7337				      /*command*/ 1,
7338				      /*field*/ 2,
7339				      /*bit_valid*/ 1,
7340				      /*bit*/ 5);
7341		ctl_done((union ctl_io *)ctsio);
7342		return(retval);
7343	}
7344
7345	single = ctl_is_single;
7346	if (single)
7347		num_target_port_groups = 1;
7348	else
7349		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7350	num_target_ports = 0;
7351	mtx_lock(&softc->ctl_lock);
7352	STAILQ_FOREACH(port, &softc->port_list, links) {
7353		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7354			continue;
7355		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7356			continue;
7357		num_target_ports++;
7358	}
7359	mtx_unlock(&softc->ctl_lock);
7360
7361	if (ext)
7362		total_len = sizeof(struct scsi_target_group_data_extended);
7363	else
7364		total_len = sizeof(struct scsi_target_group_data);
7365	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7366		num_target_port_groups +
7367	    sizeof(struct scsi_target_port_descriptor) *
7368		num_target_ports * num_target_port_groups;
7369
7370	alloc_len = scsi_4btoul(cdb->length);
7371
7372	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7373
7374	ctsio->kern_sg_entries = 0;
7375
7376	if (total_len < alloc_len) {
7377		ctsio->residual = alloc_len - total_len;
7378		ctsio->kern_data_len = total_len;
7379		ctsio->kern_total_len = total_len;
7380	} else {
7381		ctsio->residual = 0;
7382		ctsio->kern_data_len = alloc_len;
7383		ctsio->kern_total_len = alloc_len;
7384	}
7385	ctsio->kern_data_resid = 0;
7386	ctsio->kern_rel_offset = 0;
7387
7388	if (ext) {
7389		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7390		    ctsio->kern_data_ptr;
7391		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7392		rtg_ext_ptr->format_type = 0x10;
7393		rtg_ext_ptr->implicit_transition_time = 0;
7394		tpg_desc = &rtg_ext_ptr->groups[0];
7395	} else {
7396		rtg_ptr = (struct scsi_target_group_data *)
7397		    ctsio->kern_data_ptr;
7398		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7399		tpg_desc = &rtg_ptr->groups[0];
7400	}
7401
7402	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7403	mtx_lock(&softc->ctl_lock);
7404	for (g = 0; g < num_target_port_groups; g++) {
7405		if (g == pg)
7406			tpg_desc->pref_state = TPG_PRIMARY |
7407			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7408		else
7409			tpg_desc->pref_state =
7410			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7411		tpg_desc->support = TPG_AO_SUP;
7412		if (!single)
7413			tpg_desc->support |= TPG_AN_SUP;
7414		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7415		tpg_desc->status = TPG_IMPLICIT;
7416		pc = 0;
7417		STAILQ_FOREACH(port, &softc->port_list, links) {
7418			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7419				continue;
7420			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7421			    CTL_MAX_LUNS)
7422				continue;
7423			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7424			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7425			    relative_target_port_identifier);
7426			pc++;
7427		}
7428		tpg_desc->target_port_count = pc;
7429		tpg_desc = (struct scsi_target_port_group_descriptor *)
7430		    &tpg_desc->descriptors[pc];
7431	}
7432	mtx_unlock(&softc->ctl_lock);
7433
7434	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7435	ctsio->be_move_done = ctl_config_move_done;
7436
7437	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7438			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7439			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7440			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7441			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7442
7443	ctl_datamove((union ctl_io *)ctsio);
7444	return(retval);
7445}
7446
7447int
7448ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7449{
7450	struct ctl_lun *lun;
7451	struct scsi_report_supported_opcodes *cdb;
7452	const struct ctl_cmd_entry *entry, *sentry;
7453	struct scsi_report_supported_opcodes_all *all;
7454	struct scsi_report_supported_opcodes_descr *descr;
7455	struct scsi_report_supported_opcodes_one *one;
7456	int retval;
7457	int alloc_len, total_len;
7458	int opcode, service_action, i, j, num;
7459
7460	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7461
7462	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7463	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7464
7465	retval = CTL_RETVAL_COMPLETE;
7466
7467	opcode = cdb->requested_opcode;
7468	service_action = scsi_2btoul(cdb->requested_service_action);
7469	switch (cdb->options & RSO_OPTIONS_MASK) {
7470	case RSO_OPTIONS_ALL:
7471		num = 0;
7472		for (i = 0; i < 256; i++) {
7473			entry = &ctl_cmd_table[i];
7474			if (entry->flags & CTL_CMD_FLAG_SA5) {
7475				for (j = 0; j < 32; j++) {
7476					sentry = &((const struct ctl_cmd_entry *)
7477					    entry->execute)[j];
7478					if (ctl_cmd_applicable(
7479					    lun->be_lun->lun_type, sentry))
7480						num++;
7481				}
7482			} else {
7483				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7484				    entry))
7485					num++;
7486			}
7487		}
7488		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7489		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7490		break;
7491	case RSO_OPTIONS_OC:
7492		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7493			ctl_set_invalid_field(/*ctsio*/ ctsio,
7494					      /*sks_valid*/ 1,
7495					      /*command*/ 1,
7496					      /*field*/ 2,
7497					      /*bit_valid*/ 1,
7498					      /*bit*/ 2);
7499			ctl_done((union ctl_io *)ctsio);
7500			return (CTL_RETVAL_COMPLETE);
7501		}
7502		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7503		break;
7504	case RSO_OPTIONS_OC_SA:
7505		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7506		    service_action >= 32) {
7507			ctl_set_invalid_field(/*ctsio*/ ctsio,
7508					      /*sks_valid*/ 1,
7509					      /*command*/ 1,
7510					      /*field*/ 2,
7511					      /*bit_valid*/ 1,
7512					      /*bit*/ 2);
7513			ctl_done((union ctl_io *)ctsio);
7514			return (CTL_RETVAL_COMPLETE);
7515		}
7516		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7517		break;
7518	default:
7519		ctl_set_invalid_field(/*ctsio*/ ctsio,
7520				      /*sks_valid*/ 1,
7521				      /*command*/ 1,
7522				      /*field*/ 2,
7523				      /*bit_valid*/ 1,
7524				      /*bit*/ 2);
7525		ctl_done((union ctl_io *)ctsio);
7526		return (CTL_RETVAL_COMPLETE);
7527	}
7528
7529	alloc_len = scsi_4btoul(cdb->length);
7530
7531	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7532
7533	ctsio->kern_sg_entries = 0;
7534
7535	if (total_len < alloc_len) {
7536		ctsio->residual = alloc_len - total_len;
7537		ctsio->kern_data_len = total_len;
7538		ctsio->kern_total_len = total_len;
7539	} else {
7540		ctsio->residual = 0;
7541		ctsio->kern_data_len = alloc_len;
7542		ctsio->kern_total_len = alloc_len;
7543	}
7544	ctsio->kern_data_resid = 0;
7545	ctsio->kern_rel_offset = 0;
7546
7547	switch (cdb->options & RSO_OPTIONS_MASK) {
7548	case RSO_OPTIONS_ALL:
7549		all = (struct scsi_report_supported_opcodes_all *)
7550		    ctsio->kern_data_ptr;
7551		num = 0;
7552		for (i = 0; i < 256; i++) {
7553			entry = &ctl_cmd_table[i];
7554			if (entry->flags & CTL_CMD_FLAG_SA5) {
7555				for (j = 0; j < 32; j++) {
7556					sentry = &((const struct ctl_cmd_entry *)
7557					    entry->execute)[j];
7558					if (!ctl_cmd_applicable(
7559					    lun->be_lun->lun_type, sentry))
7560						continue;
7561					descr = &all->descr[num++];
7562					descr->opcode = i;
7563					scsi_ulto2b(j, descr->service_action);
7564					descr->flags = RSO_SERVACTV;
7565					scsi_ulto2b(sentry->length,
7566					    descr->cdb_length);
7567				}
7568			} else {
7569				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7570				    entry))
7571					continue;
7572				descr = &all->descr[num++];
7573				descr->opcode = i;
7574				scsi_ulto2b(0, descr->service_action);
7575				descr->flags = 0;
7576				scsi_ulto2b(entry->length, descr->cdb_length);
7577			}
7578		}
7579		scsi_ulto4b(
7580		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7581		    all->length);
7582		break;
7583	case RSO_OPTIONS_OC:
7584		one = (struct scsi_report_supported_opcodes_one *)
7585		    ctsio->kern_data_ptr;
7586		entry = &ctl_cmd_table[opcode];
7587		goto fill_one;
7588	case RSO_OPTIONS_OC_SA:
7589		one = (struct scsi_report_supported_opcodes_one *)
7590		    ctsio->kern_data_ptr;
7591		entry = &ctl_cmd_table[opcode];
7592		entry = &((const struct ctl_cmd_entry *)
7593		    entry->execute)[service_action];
7594fill_one:
7595		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7596			one->support = 3;
7597			scsi_ulto2b(entry->length, one->cdb_length);
7598			one->cdb_usage[0] = opcode;
7599			memcpy(&one->cdb_usage[1], entry->usage,
7600			    entry->length - 1);
7601		} else
7602			one->support = 1;
7603		break;
7604	}
7605
7606	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7607	ctsio->be_move_done = ctl_config_move_done;
7608
7609	ctl_datamove((union ctl_io *)ctsio);
7610	return(retval);
7611}
7612
7613int
7614ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7615{
7616	struct ctl_lun *lun;
7617	struct scsi_report_supported_tmf *cdb;
7618	struct scsi_report_supported_tmf_data *data;
7619	int retval;
7620	int alloc_len, total_len;
7621
7622	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7623
7624	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7625	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7626
7627	retval = CTL_RETVAL_COMPLETE;
7628
7629	total_len = sizeof(struct scsi_report_supported_tmf_data);
7630	alloc_len = scsi_4btoul(cdb->length);
7631
7632	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7633
7634	ctsio->kern_sg_entries = 0;
7635
7636	if (total_len < alloc_len) {
7637		ctsio->residual = alloc_len - total_len;
7638		ctsio->kern_data_len = total_len;
7639		ctsio->kern_total_len = total_len;
7640	} else {
7641		ctsio->residual = 0;
7642		ctsio->kern_data_len = alloc_len;
7643		ctsio->kern_total_len = alloc_len;
7644	}
7645	ctsio->kern_data_resid = 0;
7646	ctsio->kern_rel_offset = 0;
7647
7648	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7649	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7650	data->byte2 |= RST_ITNRS;
7651
7652	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7653	ctsio->be_move_done = ctl_config_move_done;
7654
7655	ctl_datamove((union ctl_io *)ctsio);
7656	return (retval);
7657}
7658
7659int
7660ctl_report_timestamp(struct ctl_scsiio *ctsio)
7661{
7662	struct ctl_lun *lun;
7663	struct scsi_report_timestamp *cdb;
7664	struct scsi_report_timestamp_data *data;
7665	struct timeval tv;
7666	int64_t timestamp;
7667	int retval;
7668	int alloc_len, total_len;
7669
7670	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7671
7672	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7673	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7674
7675	retval = CTL_RETVAL_COMPLETE;
7676
7677	total_len = sizeof(struct scsi_report_timestamp_data);
7678	alloc_len = scsi_4btoul(cdb->length);
7679
7680	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7681
7682	ctsio->kern_sg_entries = 0;
7683
7684	if (total_len < alloc_len) {
7685		ctsio->residual = alloc_len - total_len;
7686		ctsio->kern_data_len = total_len;
7687		ctsio->kern_total_len = total_len;
7688	} else {
7689		ctsio->residual = 0;
7690		ctsio->kern_data_len = alloc_len;
7691		ctsio->kern_total_len = alloc_len;
7692	}
7693	ctsio->kern_data_resid = 0;
7694	ctsio->kern_rel_offset = 0;
7695
7696	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7697	scsi_ulto2b(sizeof(*data) - 2, data->length);
7698	data->origin = RTS_ORIG_OUTSIDE;
7699	getmicrotime(&tv);
7700	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7701	scsi_ulto4b(timestamp >> 16, data->timestamp);
7702	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7703
7704	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7705	ctsio->be_move_done = ctl_config_move_done;
7706
7707	ctl_datamove((union ctl_io *)ctsio);
7708	return (retval);
7709}
7710
7711int
7712ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7713{
7714	struct scsi_per_res_in *cdb;
7715	int alloc_len, total_len = 0;
7716	/* struct scsi_per_res_in_rsrv in_data; */
7717	struct ctl_lun *lun;
7718	struct ctl_softc *softc;
7719
7720	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7721
7722	softc = control_softc;
7723
7724	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7725
7726	alloc_len = scsi_2btoul(cdb->length);
7727
7728	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7729
7730retry:
7731	mtx_lock(&lun->lun_lock);
7732	switch (cdb->action) {
7733	case SPRI_RK: /* read keys */
7734		total_len = sizeof(struct scsi_per_res_in_keys) +
7735			lun->pr_key_count *
7736			sizeof(struct scsi_per_res_key);
7737		break;
7738	case SPRI_RR: /* read reservation */
7739		if (lun->flags & CTL_LUN_PR_RESERVED)
7740			total_len = sizeof(struct scsi_per_res_in_rsrv);
7741		else
7742			total_len = sizeof(struct scsi_per_res_in_header);
7743		break;
7744	case SPRI_RC: /* report capabilities */
7745		total_len = sizeof(struct scsi_per_res_cap);
7746		break;
7747	case SPRI_RS: /* read full status */
7748		total_len = sizeof(struct scsi_per_res_in_header) +
7749		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7750		    lun->pr_key_count;
7751		break;
7752	default:
7753		panic("Invalid PR type %x", cdb->action);
7754	}
7755	mtx_unlock(&lun->lun_lock);
7756
7757	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7758
7759	if (total_len < alloc_len) {
7760		ctsio->residual = alloc_len - total_len;
7761		ctsio->kern_data_len = total_len;
7762		ctsio->kern_total_len = total_len;
7763	} else {
7764		ctsio->residual = 0;
7765		ctsio->kern_data_len = alloc_len;
7766		ctsio->kern_total_len = alloc_len;
7767	}
7768
7769	ctsio->kern_data_resid = 0;
7770	ctsio->kern_rel_offset = 0;
7771	ctsio->kern_sg_entries = 0;
7772
7773	mtx_lock(&lun->lun_lock);
7774	switch (cdb->action) {
7775	case SPRI_RK: { // read keys
7776        struct scsi_per_res_in_keys *res_keys;
7777		int i, key_count;
7778
7779		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7780
7781		/*
7782		 * We had to drop the lock to allocate our buffer, which
7783		 * leaves time for someone to come in with another
7784		 * persistent reservation.  (That is unlikely, though,
7785		 * since this should be the only persistent reservation
7786		 * command active right now.)
7787		 */
7788		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7789		    (lun->pr_key_count *
7790		     sizeof(struct scsi_per_res_key)))){
7791			mtx_unlock(&lun->lun_lock);
7792			free(ctsio->kern_data_ptr, M_CTL);
7793			printf("%s: reservation length changed, retrying\n",
7794			       __func__);
7795			goto retry;
7796		}
7797
7798		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7799
7800		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7801			     lun->pr_key_count, res_keys->header.length);
7802
7803		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7804			if (!lun->per_res[i].registered)
7805				continue;
7806
7807			/*
7808			 * We used lun->pr_key_count to calculate the
7809			 * size to allocate.  If it turns out the number of
7810			 * initiators with the registered flag set is
7811			 * larger than that (i.e. they haven't been kept in
7812			 * sync), we've got a problem.
7813			 */
7814			if (key_count >= lun->pr_key_count) {
7815#ifdef NEEDTOPORT
7816				csevent_log(CSC_CTL | CSC_SHELF_SW |
7817					    CTL_PR_ERROR,
7818					    csevent_LogType_Fault,
7819					    csevent_AlertLevel_Yellow,
7820					    csevent_FRU_ShelfController,
7821					    csevent_FRU_Firmware,
7822				        csevent_FRU_Unknown,
7823					    "registered keys %d >= key "
7824					    "count %d", key_count,
7825					    lun->pr_key_count);
7826#endif
7827				key_count++;
7828				continue;
7829			}
7830			memcpy(res_keys->keys[key_count].key,
7831			       lun->per_res[i].res_key.key,
7832			       ctl_min(sizeof(res_keys->keys[key_count].key),
7833			       sizeof(lun->per_res[i].res_key)));
7834			key_count++;
7835		}
7836		break;
7837	}
7838	case SPRI_RR: { // read reservation
7839		struct scsi_per_res_in_rsrv *res;
7840		int tmp_len, header_only;
7841
7842		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7843
7844		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7845
7846		if (lun->flags & CTL_LUN_PR_RESERVED)
7847		{
7848			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7849			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7850				    res->header.length);
7851			header_only = 0;
7852		} else {
7853			tmp_len = sizeof(struct scsi_per_res_in_header);
7854			scsi_ulto4b(0, res->header.length);
7855			header_only = 1;
7856		}
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 (tmp_len != total_len) {
7866			mtx_unlock(&lun->lun_lock);
7867			free(ctsio->kern_data_ptr, M_CTL);
7868			printf("%s: reservation status changed, retrying\n",
7869			       __func__);
7870			goto retry;
7871		}
7872
7873		/*
7874		 * No reservation held, so we're done.
7875		 */
7876		if (header_only != 0)
7877			break;
7878
7879		/*
7880		 * If the registration is an All Registrants type, the key
7881		 * is 0, since it doesn't really matter.
7882		 */
7883		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7884			memcpy(res->data.reservation,
7885			       &lun->per_res[lun->pr_res_idx].res_key,
7886			       sizeof(struct scsi_per_res_key));
7887		}
7888		res->data.scopetype = lun->res_type;
7889		break;
7890	}
7891	case SPRI_RC:     //report capabilities
7892	{
7893		struct scsi_per_res_cap *res_cap;
7894		uint16_t type_mask;
7895
7896		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7897		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7898		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7899		type_mask = SPRI_TM_WR_EX_AR |
7900			    SPRI_TM_EX_AC_RO |
7901			    SPRI_TM_WR_EX_RO |
7902			    SPRI_TM_EX_AC |
7903			    SPRI_TM_WR_EX |
7904			    SPRI_TM_EX_AC_AR;
7905		scsi_ulto2b(type_mask, res_cap->type_mask);
7906		break;
7907	}
7908	case SPRI_RS: { // read full status
7909		struct scsi_per_res_in_full *res_status;
7910		struct scsi_per_res_in_full_desc *res_desc;
7911		struct ctl_port *port;
7912		int i, len;
7913
7914		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7915
7916		/*
7917		 * We had to drop the lock to allocate our buffer, which
7918		 * leaves time for someone to come in with another
7919		 * persistent reservation.  (That is unlikely, though,
7920		 * since this should be the only persistent reservation
7921		 * command active right now.)
7922		 */
7923		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7924		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7925		     lun->pr_key_count)){
7926			mtx_unlock(&lun->lun_lock);
7927			free(ctsio->kern_data_ptr, M_CTL);
7928			printf("%s: reservation length changed, retrying\n",
7929			       __func__);
7930			goto retry;
7931		}
7932
7933		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7934
7935		res_desc = &res_status->desc[0];
7936		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7937			if (!lun->per_res[i].registered)
7938				continue;
7939
7940			memcpy(&res_desc->res_key, &lun->per_res[i].res_key.key,
7941			    sizeof(res_desc->res_key));
7942			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7943			    (lun->pr_res_idx == i ||
7944			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7945				res_desc->flags = SPRI_FULL_R_HOLDER;
7946				res_desc->scopetype = lun->res_type;
7947			}
7948			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7949			    res_desc->rel_trgt_port_id);
7950			len = 0;
7951			port = softc->ctl_ports[
7952			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7953			if (port != NULL)
7954				len = ctl_create_iid(port,
7955				    i % CTL_MAX_INIT_PER_PORT,
7956				    res_desc->transport_id);
7957			scsi_ulto4b(len, res_desc->additional_length);
7958			res_desc = (struct scsi_per_res_in_full_desc *)
7959			    &res_desc->transport_id[len];
7960		}
7961		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7962		    res_status->header.length);
7963		break;
7964	}
7965	default:
7966		/*
7967		 * This is a bug, because we just checked for this above,
7968		 * and should have returned an error.
7969		 */
7970		panic("Invalid PR type %x", cdb->action);
7971		break; /* NOTREACHED */
7972	}
7973	mtx_unlock(&lun->lun_lock);
7974
7975	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7976	ctsio->be_move_done = ctl_config_move_done;
7977
7978	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7979			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7980			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7981			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7982			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7983
7984	ctl_datamove((union ctl_io *)ctsio);
7985
7986	return (CTL_RETVAL_COMPLETE);
7987}
7988
7989/*
7990 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7991 * it should return.
7992 */
7993static int
7994ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7995		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7996		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7997		struct scsi_per_res_out_parms* param)
7998{
7999	union ctl_ha_msg persis_io;
8000	int retval, i;
8001	int isc_retval;
8002
8003	retval = 0;
8004
8005	mtx_lock(&lun->lun_lock);
8006	if (sa_res_key == 0) {
8007		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8008			/* validate scope and type */
8009			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8010			     SPR_LU_SCOPE) {
8011				mtx_unlock(&lun->lun_lock);
8012				ctl_set_invalid_field(/*ctsio*/ ctsio,
8013						      /*sks_valid*/ 1,
8014						      /*command*/ 1,
8015						      /*field*/ 2,
8016						      /*bit_valid*/ 1,
8017						      /*bit*/ 4);
8018				ctl_done((union ctl_io *)ctsio);
8019				return (1);
8020			}
8021
8022		        if (type>8 || type==2 || type==4 || type==0) {
8023				mtx_unlock(&lun->lun_lock);
8024				ctl_set_invalid_field(/*ctsio*/ ctsio,
8025       	           				      /*sks_valid*/ 1,
8026						      /*command*/ 1,
8027						      /*field*/ 2,
8028						      /*bit_valid*/ 1,
8029						      /*bit*/ 0);
8030				ctl_done((union ctl_io *)ctsio);
8031				return (1);
8032		        }
8033
8034			/* temporarily unregister this nexus */
8035			lun->per_res[residx].registered = 0;
8036
8037			/*
8038			 * Unregister everybody else and build UA for
8039			 * them
8040			 */
8041			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8042				if (lun->per_res[i].registered == 0)
8043					continue;
8044
8045				if (!persis_offset
8046				 && i <CTL_MAX_INITIATORS)
8047					lun->pending_ua[i] |=
8048						CTL_UA_REG_PREEMPT;
8049				else if (persis_offset
8050				      && i >= persis_offset)
8051					lun->pending_ua[i-persis_offset] |=
8052						CTL_UA_REG_PREEMPT;
8053				lun->per_res[i].registered = 0;
8054				memset(&lun->per_res[i].res_key, 0,
8055				       sizeof(struct scsi_per_res_key));
8056			}
8057			lun->per_res[residx].registered = 1;
8058			lun->pr_key_count = 1;
8059			lun->res_type = type;
8060			if (lun->res_type != SPR_TYPE_WR_EX_AR
8061			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8062				lun->pr_res_idx = residx;
8063
8064			/* send msg to other side */
8065			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8066			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8067			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8068			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8069			persis_io.pr.pr_info.res_type = type;
8070			memcpy(persis_io.pr.pr_info.sa_res_key,
8071			       param->serv_act_res_key,
8072			       sizeof(param->serv_act_res_key));
8073			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8074			     &persis_io, sizeof(persis_io), 0)) >
8075			     CTL_HA_STATUS_SUCCESS) {
8076				printf("CTL:Persis Out error returned "
8077				       "from ctl_ha_msg_send %d\n",
8078				       isc_retval);
8079			}
8080		} else {
8081			/* not all registrants */
8082			mtx_unlock(&lun->lun_lock);
8083			free(ctsio->kern_data_ptr, M_CTL);
8084			ctl_set_invalid_field(ctsio,
8085					      /*sks_valid*/ 1,
8086					      /*command*/ 0,
8087					      /*field*/ 8,
8088					      /*bit_valid*/ 0,
8089					      /*bit*/ 0);
8090			ctl_done((union ctl_io *)ctsio);
8091			return (1);
8092		}
8093	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8094		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8095		int found = 0;
8096
8097		if (res_key == sa_res_key) {
8098			/* special case */
8099			/*
8100			 * The spec implies this is not good but doesn't
8101			 * say what to do. There are two choices either
8102			 * generate a res conflict or check condition
8103			 * with illegal field in parameter data. Since
8104			 * that is what is done when the sa_res_key is
8105			 * zero I'll take that approach since this has
8106			 * to do with the sa_res_key.
8107			 */
8108			mtx_unlock(&lun->lun_lock);
8109			free(ctsio->kern_data_ptr, M_CTL);
8110			ctl_set_invalid_field(ctsio,
8111					      /*sks_valid*/ 1,
8112					      /*command*/ 0,
8113					      /*field*/ 8,
8114					      /*bit_valid*/ 0,
8115					      /*bit*/ 0);
8116			ctl_done((union ctl_io *)ctsio);
8117			return (1);
8118		}
8119
8120		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8121			if (lun->per_res[i].registered
8122			 && memcmp(param->serv_act_res_key,
8123			    lun->per_res[i].res_key.key,
8124			    sizeof(struct scsi_per_res_key)) != 0)
8125				continue;
8126
8127			found = 1;
8128			lun->per_res[i].registered = 0;
8129			memset(&lun->per_res[i].res_key, 0,
8130			       sizeof(struct scsi_per_res_key));
8131			lun->pr_key_count--;
8132
8133			if (!persis_offset && i < CTL_MAX_INITIATORS)
8134				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8135			else if (persis_offset && i >= persis_offset)
8136				lun->pending_ua[i-persis_offset] |=
8137					CTL_UA_REG_PREEMPT;
8138		}
8139		if (!found) {
8140			mtx_unlock(&lun->lun_lock);
8141			free(ctsio->kern_data_ptr, M_CTL);
8142			ctl_set_reservation_conflict(ctsio);
8143			ctl_done((union ctl_io *)ctsio);
8144			return (CTL_RETVAL_COMPLETE);
8145		}
8146		/* send msg to other side */
8147		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8148		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8149		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8150		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8151		persis_io.pr.pr_info.res_type = type;
8152		memcpy(persis_io.pr.pr_info.sa_res_key,
8153		       param->serv_act_res_key,
8154		       sizeof(param->serv_act_res_key));
8155		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8156		     &persis_io, sizeof(persis_io), 0)) >
8157		     CTL_HA_STATUS_SUCCESS) {
8158			printf("CTL:Persis Out error returned from "
8159			       "ctl_ha_msg_send %d\n", isc_retval);
8160		}
8161	} else {
8162		/* Reserved but not all registrants */
8163		/* sa_res_key is res holder */
8164		if (memcmp(param->serv_act_res_key,
8165                   lun->per_res[lun->pr_res_idx].res_key.key,
8166                   sizeof(struct scsi_per_res_key)) == 0) {
8167			/* validate scope and type */
8168			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8169			     SPR_LU_SCOPE) {
8170				mtx_unlock(&lun->lun_lock);
8171				ctl_set_invalid_field(/*ctsio*/ ctsio,
8172						      /*sks_valid*/ 1,
8173						      /*command*/ 1,
8174						      /*field*/ 2,
8175						      /*bit_valid*/ 1,
8176						      /*bit*/ 4);
8177				ctl_done((union ctl_io *)ctsio);
8178				return (1);
8179			}
8180
8181			if (type>8 || type==2 || type==4 || type==0) {
8182				mtx_unlock(&lun->lun_lock);
8183				ctl_set_invalid_field(/*ctsio*/ ctsio,
8184						      /*sks_valid*/ 1,
8185						      /*command*/ 1,
8186						      /*field*/ 2,
8187						      /*bit_valid*/ 1,
8188						      /*bit*/ 0);
8189				ctl_done((union ctl_io *)ctsio);
8190				return (1);
8191			}
8192
8193			/*
8194			 * Do the following:
8195			 * if sa_res_key != res_key remove all
8196			 * registrants w/sa_res_key and generate UA
8197			 * for these registrants(Registrations
8198			 * Preempted) if it wasn't an exclusive
8199			 * reservation generate UA(Reservations
8200			 * Preempted) for all other registered nexuses
8201			 * if the type has changed. Establish the new
8202			 * reservation and holder. If res_key and
8203			 * sa_res_key are the same do the above
8204			 * except don't unregister the res holder.
8205			 */
8206
8207			/*
8208			 * Temporarily unregister so it won't get
8209			 * removed or UA generated
8210			 */
8211			lun->per_res[residx].registered = 0;
8212			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8213				if (lun->per_res[i].registered == 0)
8214					continue;
8215
8216				if (memcmp(param->serv_act_res_key,
8217				    lun->per_res[i].res_key.key,
8218				    sizeof(struct scsi_per_res_key)) == 0) {
8219					lun->per_res[i].registered = 0;
8220					memset(&lun->per_res[i].res_key,
8221					       0,
8222					       sizeof(struct scsi_per_res_key));
8223					lun->pr_key_count--;
8224
8225					if (!persis_offset
8226					 && i < CTL_MAX_INITIATORS)
8227						lun->pending_ua[i] |=
8228							CTL_UA_REG_PREEMPT;
8229					else if (persis_offset
8230					      && i >= persis_offset)
8231						lun->pending_ua[i-persis_offset] |=
8232						  CTL_UA_REG_PREEMPT;
8233				} else if (type != lun->res_type
8234					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8235					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8236						if (!persis_offset
8237						 && i < CTL_MAX_INITIATORS)
8238							lun->pending_ua[i] |=
8239							CTL_UA_RES_RELEASE;
8240						else if (persis_offset
8241						      && i >= persis_offset)
8242							lun->pending_ua[
8243							i-persis_offset] |=
8244							CTL_UA_RES_RELEASE;
8245				}
8246			}
8247			lun->per_res[residx].registered = 1;
8248			lun->res_type = type;
8249			if (lun->res_type != SPR_TYPE_WR_EX_AR
8250			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8251				lun->pr_res_idx = residx;
8252			else
8253				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8254
8255			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8256			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8257			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8258			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8259			persis_io.pr.pr_info.res_type = type;
8260			memcpy(persis_io.pr.pr_info.sa_res_key,
8261			       param->serv_act_res_key,
8262			       sizeof(param->serv_act_res_key));
8263			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8264			     &persis_io, sizeof(persis_io), 0)) >
8265			     CTL_HA_STATUS_SUCCESS) {
8266				printf("CTL:Persis Out error returned "
8267				       "from ctl_ha_msg_send %d\n",
8268				       isc_retval);
8269			}
8270		} else {
8271			/*
8272			 * sa_res_key is not the res holder just
8273			 * remove registrants
8274			 */
8275			int found=0;
8276
8277			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8278				if (memcmp(param->serv_act_res_key,
8279				    lun->per_res[i].res_key.key,
8280				    sizeof(struct scsi_per_res_key)) != 0)
8281					continue;
8282
8283				found = 1;
8284				lun->per_res[i].registered = 0;
8285				memset(&lun->per_res[i].res_key, 0,
8286				       sizeof(struct scsi_per_res_key));
8287				lun->pr_key_count--;
8288
8289				if (!persis_offset
8290				 && i < CTL_MAX_INITIATORS)
8291					lun->pending_ua[i] |=
8292						CTL_UA_REG_PREEMPT;
8293				else if (persis_offset
8294				      && i >= persis_offset)
8295					lun->pending_ua[i-persis_offset] |=
8296						CTL_UA_REG_PREEMPT;
8297			}
8298
8299			if (!found) {
8300				mtx_unlock(&lun->lun_lock);
8301				free(ctsio->kern_data_ptr, M_CTL);
8302				ctl_set_reservation_conflict(ctsio);
8303				ctl_done((union ctl_io *)ctsio);
8304		        	return (1);
8305			}
8306			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8307			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8308			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8309			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8310			persis_io.pr.pr_info.res_type = type;
8311			memcpy(persis_io.pr.pr_info.sa_res_key,
8312			       param->serv_act_res_key,
8313			       sizeof(param->serv_act_res_key));
8314			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8315			     &persis_io, sizeof(persis_io), 0)) >
8316			     CTL_HA_STATUS_SUCCESS) {
8317				printf("CTL:Persis Out error returned "
8318				       "from ctl_ha_msg_send %d\n",
8319				isc_retval);
8320			}
8321		}
8322	}
8323
8324	lun->PRGeneration++;
8325	mtx_unlock(&lun->lun_lock);
8326
8327	return (retval);
8328}
8329
8330static void
8331ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8332{
8333	int i;
8334
8335	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8336	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8337	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8338		   msg->pr.pr_info.sa_res_key,
8339		   sizeof(struct scsi_per_res_key)) != 0) {
8340		uint64_t sa_res_key;
8341		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8342
8343		if (sa_res_key == 0) {
8344			/* temporarily unregister this nexus */
8345			lun->per_res[msg->pr.pr_info.residx].registered = 0;
8346
8347			/*
8348			 * Unregister everybody else and build UA for
8349			 * them
8350			 */
8351			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8352				if (lun->per_res[i].registered == 0)
8353					continue;
8354
8355				if (!persis_offset
8356				 && i < CTL_MAX_INITIATORS)
8357					lun->pending_ua[i] |=
8358						CTL_UA_REG_PREEMPT;
8359				else if (persis_offset && i >= persis_offset)
8360					lun->pending_ua[i - persis_offset] |=
8361						CTL_UA_REG_PREEMPT;
8362				lun->per_res[i].registered = 0;
8363				memset(&lun->per_res[i].res_key, 0,
8364				       sizeof(struct scsi_per_res_key));
8365			}
8366
8367			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8368			lun->pr_key_count = 1;
8369			lun->res_type = msg->pr.pr_info.res_type;
8370			if (lun->res_type != SPR_TYPE_WR_EX_AR
8371			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8372				lun->pr_res_idx = msg->pr.pr_info.residx;
8373		} else {
8374		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8375				if (memcmp(msg->pr.pr_info.sa_res_key,
8376		                   lun->per_res[i].res_key.key,
8377		                   sizeof(struct scsi_per_res_key)) != 0)
8378					continue;
8379
8380				lun->per_res[i].registered = 0;
8381				memset(&lun->per_res[i].res_key, 0,
8382				       sizeof(struct scsi_per_res_key));
8383				lun->pr_key_count--;
8384
8385				if (!persis_offset
8386				 && i < persis_offset)
8387					lun->pending_ua[i] |=
8388						CTL_UA_REG_PREEMPT;
8389				else if (persis_offset
8390				      && i >= persis_offset)
8391					lun->pending_ua[i - persis_offset] |=
8392						CTL_UA_REG_PREEMPT;
8393			}
8394		}
8395	} else {
8396		/*
8397		 * Temporarily unregister so it won't get removed
8398		 * or UA generated
8399		 */
8400		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8401		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8402			if (lun->per_res[i].registered == 0)
8403				continue;
8404
8405			if (memcmp(msg->pr.pr_info.sa_res_key,
8406	                   lun->per_res[i].res_key.key,
8407	                   sizeof(struct scsi_per_res_key)) == 0) {
8408				lun->per_res[i].registered = 0;
8409				memset(&lun->per_res[i].res_key, 0,
8410				       sizeof(struct scsi_per_res_key));
8411				lun->pr_key_count--;
8412				if (!persis_offset
8413				 && i < CTL_MAX_INITIATORS)
8414					lun->pending_ua[i] |=
8415						CTL_UA_REG_PREEMPT;
8416				else if (persis_offset
8417				      && i >= persis_offset)
8418					lun->pending_ua[i - persis_offset] |=
8419						CTL_UA_REG_PREEMPT;
8420			} else if (msg->pr.pr_info.res_type != lun->res_type
8421				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8422				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8423					if (!persis_offset
8424					 && i < persis_offset)
8425						lun->pending_ua[i] |=
8426							CTL_UA_RES_RELEASE;
8427					else if (persis_offset
8428					      && i >= persis_offset)
8429					lun->pending_ua[i - persis_offset] |=
8430						CTL_UA_RES_RELEASE;
8431			}
8432		}
8433		lun->per_res[msg->pr.pr_info.residx].registered = 1;
8434		lun->res_type = msg->pr.pr_info.res_type;
8435		if (lun->res_type != SPR_TYPE_WR_EX_AR
8436		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8437			lun->pr_res_idx = msg->pr.pr_info.residx;
8438		else
8439			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8440	}
8441	lun->PRGeneration++;
8442
8443}
8444
8445
8446int
8447ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8448{
8449	int retval;
8450	int isc_retval;
8451	u_int32_t param_len;
8452	struct scsi_per_res_out *cdb;
8453	struct ctl_lun *lun;
8454	struct scsi_per_res_out_parms* param;
8455	struct ctl_softc *softc;
8456	uint32_t residx;
8457	uint64_t res_key, sa_res_key;
8458	uint8_t type;
8459	union ctl_ha_msg persis_io;
8460	int    i;
8461
8462	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8463
8464	retval = CTL_RETVAL_COMPLETE;
8465
8466	softc = control_softc;
8467
8468	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8469	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8470
8471	/*
8472	 * We only support whole-LUN scope.  The scope & type are ignored for
8473	 * register, register and ignore existing key and clear.
8474	 * We sometimes ignore scope and type on preempts too!!
8475	 * Verify reservation type here as well.
8476	 */
8477	type = cdb->scope_type & SPR_TYPE_MASK;
8478	if ((cdb->action == SPRO_RESERVE)
8479	 || (cdb->action == SPRO_RELEASE)) {
8480		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8481			ctl_set_invalid_field(/*ctsio*/ ctsio,
8482					      /*sks_valid*/ 1,
8483					      /*command*/ 1,
8484					      /*field*/ 2,
8485					      /*bit_valid*/ 1,
8486					      /*bit*/ 4);
8487			ctl_done((union ctl_io *)ctsio);
8488			return (CTL_RETVAL_COMPLETE);
8489		}
8490
8491		if (type>8 || type==2 || type==4 || type==0) {
8492			ctl_set_invalid_field(/*ctsio*/ ctsio,
8493					      /*sks_valid*/ 1,
8494					      /*command*/ 1,
8495					      /*field*/ 2,
8496					      /*bit_valid*/ 1,
8497					      /*bit*/ 0);
8498			ctl_done((union ctl_io *)ctsio);
8499			return (CTL_RETVAL_COMPLETE);
8500		}
8501	}
8502
8503	param_len = scsi_4btoul(cdb->length);
8504
8505	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8506		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8507		ctsio->kern_data_len = param_len;
8508		ctsio->kern_total_len = param_len;
8509		ctsio->kern_data_resid = 0;
8510		ctsio->kern_rel_offset = 0;
8511		ctsio->kern_sg_entries = 0;
8512		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8513		ctsio->be_move_done = ctl_config_move_done;
8514		ctl_datamove((union ctl_io *)ctsio);
8515
8516		return (CTL_RETVAL_COMPLETE);
8517	}
8518
8519	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8520
8521	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8522	res_key = scsi_8btou64(param->res_key.key);
8523	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8524
8525	/*
8526	 * Validate the reservation key here except for SPRO_REG_IGNO
8527	 * This must be done for all other service actions
8528	 */
8529	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8530		mtx_lock(&lun->lun_lock);
8531		if (lun->per_res[residx].registered) {
8532		    if (memcmp(param->res_key.key,
8533			       lun->per_res[residx].res_key.key,
8534			       ctl_min(sizeof(param->res_key),
8535			       sizeof(lun->per_res[residx].res_key))) != 0) {
8536				/*
8537				 * The current key passed in doesn't match
8538				 * the one the initiator previously
8539				 * registered.
8540				 */
8541				mtx_unlock(&lun->lun_lock);
8542				free(ctsio->kern_data_ptr, M_CTL);
8543				ctl_set_reservation_conflict(ctsio);
8544				ctl_done((union ctl_io *)ctsio);
8545				return (CTL_RETVAL_COMPLETE);
8546			}
8547		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8548			/*
8549			 * We are not registered
8550			 */
8551			mtx_unlock(&lun->lun_lock);
8552			free(ctsio->kern_data_ptr, M_CTL);
8553			ctl_set_reservation_conflict(ctsio);
8554			ctl_done((union ctl_io *)ctsio);
8555			return (CTL_RETVAL_COMPLETE);
8556		} else if (res_key != 0) {
8557			/*
8558			 * We are not registered and trying to register but
8559			 * the register key isn't zero.
8560			 */
8561			mtx_unlock(&lun->lun_lock);
8562			free(ctsio->kern_data_ptr, M_CTL);
8563			ctl_set_reservation_conflict(ctsio);
8564			ctl_done((union ctl_io *)ctsio);
8565			return (CTL_RETVAL_COMPLETE);
8566		}
8567		mtx_unlock(&lun->lun_lock);
8568	}
8569
8570	switch (cdb->action & SPRO_ACTION_MASK) {
8571	case SPRO_REGISTER:
8572	case SPRO_REG_IGNO: {
8573
8574#if 0
8575		printf("Registration received\n");
8576#endif
8577
8578		/*
8579		 * We don't support any of these options, as we report in
8580		 * the read capabilities request (see
8581		 * ctl_persistent_reserve_in(), above).
8582		 */
8583		if ((param->flags & SPR_SPEC_I_PT)
8584		 || (param->flags & SPR_ALL_TG_PT)
8585		 || (param->flags & SPR_APTPL)) {
8586			int bit_ptr;
8587
8588			if (param->flags & SPR_APTPL)
8589				bit_ptr = 0;
8590			else if (param->flags & SPR_ALL_TG_PT)
8591				bit_ptr = 2;
8592			else /* SPR_SPEC_I_PT */
8593				bit_ptr = 3;
8594
8595			free(ctsio->kern_data_ptr, M_CTL);
8596			ctl_set_invalid_field(ctsio,
8597					      /*sks_valid*/ 1,
8598					      /*command*/ 0,
8599					      /*field*/ 20,
8600					      /*bit_valid*/ 1,
8601					      /*bit*/ bit_ptr);
8602			ctl_done((union ctl_io *)ctsio);
8603			return (CTL_RETVAL_COMPLETE);
8604		}
8605
8606		mtx_lock(&lun->lun_lock);
8607
8608		/*
8609		 * The initiator wants to clear the
8610		 * key/unregister.
8611		 */
8612		if (sa_res_key == 0) {
8613			if ((res_key == 0
8614			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8615			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8616			  && !lun->per_res[residx].registered)) {
8617				mtx_unlock(&lun->lun_lock);
8618				goto done;
8619			}
8620
8621			lun->per_res[residx].registered = 0;
8622			memset(&lun->per_res[residx].res_key,
8623			       0, sizeof(lun->per_res[residx].res_key));
8624			lun->pr_key_count--;
8625
8626			if (residx == lun->pr_res_idx) {
8627				lun->flags &= ~CTL_LUN_PR_RESERVED;
8628				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8629
8630				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8631				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8632				 && lun->pr_key_count) {
8633					/*
8634					 * If the reservation is a registrants
8635					 * only type we need to generate a UA
8636					 * for other registered inits.  The
8637					 * sense code should be RESERVATIONS
8638					 * RELEASED
8639					 */
8640
8641					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8642						if (lun->per_res[
8643						    i+persis_offset].registered
8644						    == 0)
8645							continue;
8646						lun->pending_ua[i] |=
8647							CTL_UA_RES_RELEASE;
8648					}
8649				}
8650				lun->res_type = 0;
8651			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8652				if (lun->pr_key_count==0) {
8653					lun->flags &= ~CTL_LUN_PR_RESERVED;
8654					lun->res_type = 0;
8655					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8656				}
8657			}
8658			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8659			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8660			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8661			persis_io.pr.pr_info.residx = residx;
8662			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8663			     &persis_io, sizeof(persis_io), 0 )) >
8664			     CTL_HA_STATUS_SUCCESS) {
8665				printf("CTL:Persis Out error returned from "
8666				       "ctl_ha_msg_send %d\n", isc_retval);
8667			}
8668		} else /* sa_res_key != 0 */ {
8669
8670			/*
8671			 * If we aren't registered currently then increment
8672			 * the key count and set the registered flag.
8673			 */
8674			if (!lun->per_res[residx].registered) {
8675				lun->pr_key_count++;
8676				lun->per_res[residx].registered = 1;
8677			}
8678
8679			memcpy(&lun->per_res[residx].res_key,
8680			       param->serv_act_res_key,
8681			       ctl_min(sizeof(param->serv_act_res_key),
8682			       sizeof(lun->per_res[residx].res_key)));
8683
8684			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8685			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8686			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8687			persis_io.pr.pr_info.residx = residx;
8688			memcpy(persis_io.pr.pr_info.sa_res_key,
8689			       param->serv_act_res_key,
8690			       sizeof(param->serv_act_res_key));
8691			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8692			     &persis_io, sizeof(persis_io), 0)) >
8693			     CTL_HA_STATUS_SUCCESS) {
8694				printf("CTL:Persis Out error returned from "
8695				       "ctl_ha_msg_send %d\n", isc_retval);
8696			}
8697		}
8698		lun->PRGeneration++;
8699		mtx_unlock(&lun->lun_lock);
8700
8701		break;
8702	}
8703	case SPRO_RESERVE:
8704#if 0
8705                printf("Reserve executed type %d\n", type);
8706#endif
8707		mtx_lock(&lun->lun_lock);
8708		if (lun->flags & CTL_LUN_PR_RESERVED) {
8709			/*
8710			 * if this isn't the reservation holder and it's
8711			 * not a "all registrants" type or if the type is
8712			 * different then we have a conflict
8713			 */
8714			if ((lun->pr_res_idx != residx
8715			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8716			 || lun->res_type != type) {
8717				mtx_unlock(&lun->lun_lock);
8718				free(ctsio->kern_data_ptr, M_CTL);
8719				ctl_set_reservation_conflict(ctsio);
8720				ctl_done((union ctl_io *)ctsio);
8721				return (CTL_RETVAL_COMPLETE);
8722			}
8723			mtx_unlock(&lun->lun_lock);
8724		} else /* create a reservation */ {
8725			/*
8726			 * If it's not an "all registrants" type record
8727			 * reservation holder
8728			 */
8729			if (type != SPR_TYPE_WR_EX_AR
8730			 && type != SPR_TYPE_EX_AC_AR)
8731				lun->pr_res_idx = residx; /* Res holder */
8732			else
8733				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8734
8735			lun->flags |= CTL_LUN_PR_RESERVED;
8736			lun->res_type = type;
8737
8738			mtx_unlock(&lun->lun_lock);
8739
8740			/* send msg to other side */
8741			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8742			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8743			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8744			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8745			persis_io.pr.pr_info.res_type = type;
8746			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8747			     &persis_io, sizeof(persis_io), 0)) >
8748			     CTL_HA_STATUS_SUCCESS) {
8749				printf("CTL:Persis Out error returned from "
8750				       "ctl_ha_msg_send %d\n", isc_retval);
8751			}
8752		}
8753		break;
8754
8755	case SPRO_RELEASE:
8756		mtx_lock(&lun->lun_lock);
8757		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8758			/* No reservation exists return good status */
8759			mtx_unlock(&lun->lun_lock);
8760			goto done;
8761		}
8762		/*
8763		 * Is this nexus a reservation holder?
8764		 */
8765		if (lun->pr_res_idx != residx
8766		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8767			/*
8768			 * not a res holder return good status but
8769			 * do nothing
8770			 */
8771			mtx_unlock(&lun->lun_lock);
8772			goto done;
8773		}
8774
8775		if (lun->res_type != type) {
8776			mtx_unlock(&lun->lun_lock);
8777			free(ctsio->kern_data_ptr, M_CTL);
8778			ctl_set_illegal_pr_release(ctsio);
8779			ctl_done((union ctl_io *)ctsio);
8780			return (CTL_RETVAL_COMPLETE);
8781		}
8782
8783		/* okay to release */
8784		lun->flags &= ~CTL_LUN_PR_RESERVED;
8785		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8786		lun->res_type = 0;
8787
8788		/*
8789		 * if this isn't an exclusive access
8790		 * res generate UA for all other
8791		 * registrants.
8792		 */
8793		if (type != SPR_TYPE_EX_AC
8794		 && type != SPR_TYPE_WR_EX) {
8795			/*
8796			 * temporarily unregister so we don't generate UA
8797			 */
8798			lun->per_res[residx].registered = 0;
8799
8800			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8801				if (lun->per_res[i+persis_offset].registered
8802				    == 0)
8803					continue;
8804				lun->pending_ua[i] |=
8805					CTL_UA_RES_RELEASE;
8806			}
8807
8808			lun->per_res[residx].registered = 1;
8809		}
8810		mtx_unlock(&lun->lun_lock);
8811		/* Send msg to other side */
8812		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8813		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8814		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8815		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8816		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8817			printf("CTL:Persis Out error returned from "
8818			       "ctl_ha_msg_send %d\n", isc_retval);
8819		}
8820		break;
8821
8822	case SPRO_CLEAR:
8823		/* send msg to other side */
8824
8825		mtx_lock(&lun->lun_lock);
8826		lun->flags &= ~CTL_LUN_PR_RESERVED;
8827		lun->res_type = 0;
8828		lun->pr_key_count = 0;
8829		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8830
8831
8832		memset(&lun->per_res[residx].res_key,
8833		       0, sizeof(lun->per_res[residx].res_key));
8834		lun->per_res[residx].registered = 0;
8835
8836		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8837			if (lun->per_res[i].registered) {
8838				if (!persis_offset && i < CTL_MAX_INITIATORS)
8839					lun->pending_ua[i] |=
8840						CTL_UA_RES_PREEMPT;
8841				else if (persis_offset && i >= persis_offset)
8842					lun->pending_ua[i-persis_offset] |=
8843					    CTL_UA_RES_PREEMPT;
8844
8845				memset(&lun->per_res[i].res_key,
8846				       0, sizeof(struct scsi_per_res_key));
8847				lun->per_res[i].registered = 0;
8848			}
8849		lun->PRGeneration++;
8850		mtx_unlock(&lun->lun_lock);
8851		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8852		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8853		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8854		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8855		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8856			printf("CTL:Persis Out error returned from "
8857			       "ctl_ha_msg_send %d\n", isc_retval);
8858		}
8859		break;
8860
8861	case SPRO_PREEMPT: {
8862		int nretval;
8863
8864		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8865					  residx, ctsio, cdb, param);
8866		if (nretval != 0)
8867			return (CTL_RETVAL_COMPLETE);
8868		break;
8869	}
8870	default:
8871		panic("Invalid PR type %x", cdb->action);
8872	}
8873
8874done:
8875	free(ctsio->kern_data_ptr, M_CTL);
8876	ctl_set_success(ctsio);
8877	ctl_done((union ctl_io *)ctsio);
8878
8879	return (retval);
8880}
8881
8882/*
8883 * This routine is for handling a message from the other SC pertaining to
8884 * persistent reserve out. All the error checking will have been done
8885 * so only perorming the action need be done here to keep the two
8886 * in sync.
8887 */
8888static void
8889ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8890{
8891	struct ctl_lun *lun;
8892	struct ctl_softc *softc;
8893	int i;
8894	uint32_t targ_lun;
8895
8896	softc = control_softc;
8897
8898	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8899	lun = softc->ctl_luns[targ_lun];
8900	mtx_lock(&lun->lun_lock);
8901	switch(msg->pr.pr_info.action) {
8902	case CTL_PR_REG_KEY:
8903		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8904			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8905			lun->pr_key_count++;
8906		}
8907		lun->PRGeneration++;
8908		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8909		       msg->pr.pr_info.sa_res_key,
8910		       sizeof(struct scsi_per_res_key));
8911		break;
8912
8913	case CTL_PR_UNREG_KEY:
8914		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8915		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8916		       0, sizeof(struct scsi_per_res_key));
8917		lun->pr_key_count--;
8918
8919		/* XXX Need to see if the reservation has been released */
8920		/* if so do we need to generate UA? */
8921		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8922			lun->flags &= ~CTL_LUN_PR_RESERVED;
8923			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8924
8925			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8926			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8927			 && lun->pr_key_count) {
8928				/*
8929				 * If the reservation is a registrants
8930				 * only type we need to generate a UA
8931				 * for other registered inits.  The
8932				 * sense code should be RESERVATIONS
8933				 * RELEASED
8934				 */
8935
8936				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8937					if (lun->per_res[i+
8938					    persis_offset].registered == 0)
8939						continue;
8940
8941					lun->pending_ua[i] |=
8942						CTL_UA_RES_RELEASE;
8943				}
8944			}
8945			lun->res_type = 0;
8946		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8947			if (lun->pr_key_count==0) {
8948				lun->flags &= ~CTL_LUN_PR_RESERVED;
8949				lun->res_type = 0;
8950				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8951			}
8952		}
8953		lun->PRGeneration++;
8954		break;
8955
8956	case CTL_PR_RESERVE:
8957		lun->flags |= CTL_LUN_PR_RESERVED;
8958		lun->res_type = msg->pr.pr_info.res_type;
8959		lun->pr_res_idx = msg->pr.pr_info.residx;
8960
8961		break;
8962
8963	case CTL_PR_RELEASE:
8964		/*
8965		 * if this isn't an exclusive access res generate UA for all
8966		 * other registrants.
8967		 */
8968		if (lun->res_type != SPR_TYPE_EX_AC
8969		 && lun->res_type != SPR_TYPE_WR_EX) {
8970			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8971				if (lun->per_res[i+persis_offset].registered)
8972					lun->pending_ua[i] |=
8973						CTL_UA_RES_RELEASE;
8974		}
8975
8976		lun->flags &= ~CTL_LUN_PR_RESERVED;
8977		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8978		lun->res_type = 0;
8979		break;
8980
8981	case CTL_PR_PREEMPT:
8982		ctl_pro_preempt_other(lun, msg);
8983		break;
8984	case CTL_PR_CLEAR:
8985		lun->flags &= ~CTL_LUN_PR_RESERVED;
8986		lun->res_type = 0;
8987		lun->pr_key_count = 0;
8988		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8989
8990		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8991			if (lun->per_res[i].registered == 0)
8992				continue;
8993			if (!persis_offset
8994			 && i < CTL_MAX_INITIATORS)
8995				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
8996			else if (persis_offset
8997			      && i >= persis_offset)
8998				lun->pending_ua[i-persis_offset] |=
8999					CTL_UA_RES_PREEMPT;
9000			memset(&lun->per_res[i].res_key, 0,
9001			       sizeof(struct scsi_per_res_key));
9002			lun->per_res[i].registered = 0;
9003		}
9004		lun->PRGeneration++;
9005		break;
9006	}
9007
9008	mtx_unlock(&lun->lun_lock);
9009}
9010
9011int
9012ctl_read_write(struct ctl_scsiio *ctsio)
9013{
9014	struct ctl_lun *lun;
9015	struct ctl_lba_len_flags *lbalen;
9016	uint64_t lba;
9017	uint32_t num_blocks;
9018	int flags, retval;
9019	int isread;
9020
9021	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9022
9023	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
9024
9025	flags = 0;
9026	retval = CTL_RETVAL_COMPLETE;
9027
9028	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
9029	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
9030	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
9031		uint32_t residx;
9032
9033		/*
9034		 * XXX KDM need a lock here.
9035		 */
9036		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9037		if ((lun->res_type == SPR_TYPE_EX_AC
9038		  && residx != lun->pr_res_idx)
9039		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
9040		   || lun->res_type == SPR_TYPE_EX_AC_AR)
9041		  && !lun->per_res[residx].registered)) {
9042			ctl_set_reservation_conflict(ctsio);
9043			ctl_done((union ctl_io *)ctsio);
9044			return (CTL_RETVAL_COMPLETE);
9045	        }
9046	}
9047
9048	switch (ctsio->cdb[0]) {
9049	case READ_6:
9050	case WRITE_6: {
9051		struct scsi_rw_6 *cdb;
9052
9053		cdb = (struct scsi_rw_6 *)ctsio->cdb;
9054
9055		lba = scsi_3btoul(cdb->addr);
9056		/* only 5 bits are valid in the most significant address byte */
9057		lba &= 0x1fffff;
9058		num_blocks = cdb->length;
9059		/*
9060		 * This is correct according to SBC-2.
9061		 */
9062		if (num_blocks == 0)
9063			num_blocks = 256;
9064		break;
9065	}
9066	case READ_10:
9067	case WRITE_10: {
9068		struct scsi_rw_10 *cdb;
9069
9070		cdb = (struct scsi_rw_10 *)ctsio->cdb;
9071		if (cdb->byte2 & SRW10_FUA)
9072			flags |= CTL_LLF_FUA;
9073		if (cdb->byte2 & SRW10_DPO)
9074			flags |= CTL_LLF_DPO;
9075		lba = scsi_4btoul(cdb->addr);
9076		num_blocks = scsi_2btoul(cdb->length);
9077		break;
9078	}
9079	case WRITE_VERIFY_10: {
9080		struct scsi_write_verify_10 *cdb;
9081
9082		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9083		flags |= CTL_LLF_FUA;
9084		if (cdb->byte2 & SWV_DPO)
9085			flags |= CTL_LLF_DPO;
9086		lba = scsi_4btoul(cdb->addr);
9087		num_blocks = scsi_2btoul(cdb->length);
9088		break;
9089	}
9090	case READ_12:
9091	case WRITE_12: {
9092		struct scsi_rw_12 *cdb;
9093
9094		cdb = (struct scsi_rw_12 *)ctsio->cdb;
9095		if (cdb->byte2 & SRW12_FUA)
9096			flags |= CTL_LLF_FUA;
9097		if (cdb->byte2 & SRW12_DPO)
9098			flags |= CTL_LLF_DPO;
9099		lba = scsi_4btoul(cdb->addr);
9100		num_blocks = scsi_4btoul(cdb->length);
9101		break;
9102	}
9103	case WRITE_VERIFY_12: {
9104		struct scsi_write_verify_12 *cdb;
9105
9106		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9107		flags |= CTL_LLF_FUA;
9108		if (cdb->byte2 & SWV_DPO)
9109			flags |= CTL_LLF_DPO;
9110		lba = scsi_4btoul(cdb->addr);
9111		num_blocks = scsi_4btoul(cdb->length);
9112		break;
9113	}
9114	case READ_16:
9115	case WRITE_16: {
9116		struct scsi_rw_16 *cdb;
9117
9118		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9119		if (cdb->byte2 & SRW12_FUA)
9120			flags |= CTL_LLF_FUA;
9121		if (cdb->byte2 & SRW12_DPO)
9122			flags |= CTL_LLF_DPO;
9123		lba = scsi_8btou64(cdb->addr);
9124		num_blocks = scsi_4btoul(cdb->length);
9125		break;
9126	}
9127	case WRITE_VERIFY_16: {
9128		struct scsi_write_verify_16 *cdb;
9129
9130		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9131		flags |= CTL_LLF_FUA;
9132		if (cdb->byte2 & SWV_DPO)
9133			flags |= CTL_LLF_DPO;
9134		lba = scsi_8btou64(cdb->addr);
9135		num_blocks = scsi_4btoul(cdb->length);
9136		break;
9137	}
9138	default:
9139		/*
9140		 * We got a command we don't support.  This shouldn't
9141		 * happen, commands should be filtered out above us.
9142		 */
9143		ctl_set_invalid_opcode(ctsio);
9144		ctl_done((union ctl_io *)ctsio);
9145
9146		return (CTL_RETVAL_COMPLETE);
9147		break; /* NOTREACHED */
9148	}
9149
9150	/*
9151	 * The first check is to make sure we're in bounds, the second
9152	 * check is to catch wrap-around problems.  If the lba + num blocks
9153	 * is less than the lba, then we've wrapped around and the block
9154	 * range is invalid anyway.
9155	 */
9156	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9157	 || ((lba + num_blocks) < lba)) {
9158		ctl_set_lba_out_of_range(ctsio);
9159		ctl_done((union ctl_io *)ctsio);
9160		return (CTL_RETVAL_COMPLETE);
9161	}
9162
9163	/*
9164	 * According to SBC-3, a transfer length of 0 is not an error.
9165	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9166	 * translates to 256 blocks for those commands.
9167	 */
9168	if (num_blocks == 0) {
9169		ctl_set_success(ctsio);
9170		ctl_done((union ctl_io *)ctsio);
9171		return (CTL_RETVAL_COMPLETE);
9172	}
9173
9174	/* Set FUA and/or DPO if caches are disabled. */
9175	if (isread) {
9176		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9177		    SCP_RCD) != 0)
9178			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9179	} else {
9180		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9181		    SCP_WCE) == 0)
9182			flags |= CTL_LLF_FUA;
9183	}
9184
9185	lbalen = (struct ctl_lba_len_flags *)
9186	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9187	lbalen->lba = lba;
9188	lbalen->len = num_blocks;
9189	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9190
9191	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9192	ctsio->kern_rel_offset = 0;
9193
9194	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9195
9196	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9197
9198	return (retval);
9199}
9200
9201static int
9202ctl_cnw_cont(union ctl_io *io)
9203{
9204	struct ctl_scsiio *ctsio;
9205	struct ctl_lun *lun;
9206	struct ctl_lba_len_flags *lbalen;
9207	int retval;
9208
9209	ctsio = &io->scsiio;
9210	ctsio->io_hdr.status = CTL_STATUS_NONE;
9211	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9212	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9213	lbalen = (struct ctl_lba_len_flags *)
9214	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9215	lbalen->flags &= ~CTL_LLF_COMPARE;
9216	lbalen->flags |= CTL_LLF_WRITE;
9217
9218	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9219	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9220	return (retval);
9221}
9222
9223int
9224ctl_cnw(struct ctl_scsiio *ctsio)
9225{
9226	struct ctl_lun *lun;
9227	struct ctl_lba_len_flags *lbalen;
9228	uint64_t lba;
9229	uint32_t num_blocks;
9230	int flags, retval;
9231
9232	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9233
9234	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9235
9236	flags = 0;
9237	retval = CTL_RETVAL_COMPLETE;
9238
9239	switch (ctsio->cdb[0]) {
9240	case COMPARE_AND_WRITE: {
9241		struct scsi_compare_and_write *cdb;
9242
9243		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9244		if (cdb->byte2 & SRW10_FUA)
9245			flags |= CTL_LLF_FUA;
9246		if (cdb->byte2 & SRW10_DPO)
9247			flags |= CTL_LLF_DPO;
9248		lba = scsi_8btou64(cdb->addr);
9249		num_blocks = cdb->length;
9250		break;
9251	}
9252	default:
9253		/*
9254		 * We got a command we don't support.  This shouldn't
9255		 * happen, commands should be filtered out above us.
9256		 */
9257		ctl_set_invalid_opcode(ctsio);
9258		ctl_done((union ctl_io *)ctsio);
9259
9260		return (CTL_RETVAL_COMPLETE);
9261		break; /* NOTREACHED */
9262	}
9263
9264	/*
9265	 * The first check is to make sure we're in bounds, the second
9266	 * check is to catch wrap-around problems.  If the lba + num blocks
9267	 * is less than the lba, then we've wrapped around and the block
9268	 * range is invalid anyway.
9269	 */
9270	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9271	 || ((lba + num_blocks) < lba)) {
9272		ctl_set_lba_out_of_range(ctsio);
9273		ctl_done((union ctl_io *)ctsio);
9274		return (CTL_RETVAL_COMPLETE);
9275	}
9276
9277	/*
9278	 * According to SBC-3, a transfer length of 0 is not an error.
9279	 */
9280	if (num_blocks == 0) {
9281		ctl_set_success(ctsio);
9282		ctl_done((union ctl_io *)ctsio);
9283		return (CTL_RETVAL_COMPLETE);
9284	}
9285
9286	/* Set FUA if write cache is disabled. */
9287	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9288	    SCP_WCE) == 0)
9289		flags |= CTL_LLF_FUA;
9290
9291	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9292	ctsio->kern_rel_offset = 0;
9293
9294	/*
9295	 * Set the IO_CONT flag, so that if this I/O gets passed to
9296	 * ctl_data_submit_done(), it'll get passed back to
9297	 * ctl_ctl_cnw_cont() for further processing.
9298	 */
9299	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9300	ctsio->io_cont = ctl_cnw_cont;
9301
9302	lbalen = (struct ctl_lba_len_flags *)
9303	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9304	lbalen->lba = lba;
9305	lbalen->len = num_blocks;
9306	lbalen->flags = CTL_LLF_COMPARE | flags;
9307
9308	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9309	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9310	return (retval);
9311}
9312
9313int
9314ctl_verify(struct ctl_scsiio *ctsio)
9315{
9316	struct ctl_lun *lun;
9317	struct ctl_lba_len_flags *lbalen;
9318	uint64_t lba;
9319	uint32_t num_blocks;
9320	int bytchk, flags;
9321	int retval;
9322
9323	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9324
9325	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9326
9327	bytchk = 0;
9328	flags = CTL_LLF_FUA;
9329	retval = CTL_RETVAL_COMPLETE;
9330
9331	switch (ctsio->cdb[0]) {
9332	case VERIFY_10: {
9333		struct scsi_verify_10 *cdb;
9334
9335		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9336		if (cdb->byte2 & SVFY_BYTCHK)
9337			bytchk = 1;
9338		if (cdb->byte2 & SVFY_DPO)
9339			flags |= CTL_LLF_DPO;
9340		lba = scsi_4btoul(cdb->addr);
9341		num_blocks = scsi_2btoul(cdb->length);
9342		break;
9343	}
9344	case VERIFY_12: {
9345		struct scsi_verify_12 *cdb;
9346
9347		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9348		if (cdb->byte2 & SVFY_BYTCHK)
9349			bytchk = 1;
9350		if (cdb->byte2 & SVFY_DPO)
9351			flags |= CTL_LLF_DPO;
9352		lba = scsi_4btoul(cdb->addr);
9353		num_blocks = scsi_4btoul(cdb->length);
9354		break;
9355	}
9356	case VERIFY_16: {
9357		struct scsi_rw_16 *cdb;
9358
9359		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9360		if (cdb->byte2 & SVFY_BYTCHK)
9361			bytchk = 1;
9362		if (cdb->byte2 & SVFY_DPO)
9363			flags |= CTL_LLF_DPO;
9364		lba = scsi_8btou64(cdb->addr);
9365		num_blocks = scsi_4btoul(cdb->length);
9366		break;
9367	}
9368	default:
9369		/*
9370		 * We got a command we don't support.  This shouldn't
9371		 * happen, commands should be filtered out above us.
9372		 */
9373		ctl_set_invalid_opcode(ctsio);
9374		ctl_done((union ctl_io *)ctsio);
9375		return (CTL_RETVAL_COMPLETE);
9376	}
9377
9378	/*
9379	 * The first check is to make sure we're in bounds, the second
9380	 * check is to catch wrap-around problems.  If the lba + num blocks
9381	 * is less than the lba, then we've wrapped around and the block
9382	 * range is invalid anyway.
9383	 */
9384	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9385	 || ((lba + num_blocks) < lba)) {
9386		ctl_set_lba_out_of_range(ctsio);
9387		ctl_done((union ctl_io *)ctsio);
9388		return (CTL_RETVAL_COMPLETE);
9389	}
9390
9391	/*
9392	 * According to SBC-3, a transfer length of 0 is not an error.
9393	 */
9394	if (num_blocks == 0) {
9395		ctl_set_success(ctsio);
9396		ctl_done((union ctl_io *)ctsio);
9397		return (CTL_RETVAL_COMPLETE);
9398	}
9399
9400	lbalen = (struct ctl_lba_len_flags *)
9401	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9402	lbalen->lba = lba;
9403	lbalen->len = num_blocks;
9404	if (bytchk) {
9405		lbalen->flags = CTL_LLF_COMPARE | flags;
9406		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9407	} else {
9408		lbalen->flags = CTL_LLF_VERIFY | flags;
9409		ctsio->kern_total_len = 0;
9410	}
9411	ctsio->kern_rel_offset = 0;
9412
9413	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9414	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9415	return (retval);
9416}
9417
9418int
9419ctl_report_luns(struct ctl_scsiio *ctsio)
9420{
9421	struct scsi_report_luns *cdb;
9422	struct scsi_report_luns_data *lun_data;
9423	struct ctl_lun *lun, *request_lun;
9424	int num_luns, retval;
9425	uint32_t alloc_len, lun_datalen;
9426	int num_filled, well_known;
9427	uint32_t initidx, targ_lun_id, lun_id;
9428
9429	retval = CTL_RETVAL_COMPLETE;
9430	well_known = 0;
9431
9432	cdb = (struct scsi_report_luns *)ctsio->cdb;
9433
9434	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9435
9436	mtx_lock(&control_softc->ctl_lock);
9437	num_luns = control_softc->num_luns;
9438	mtx_unlock(&control_softc->ctl_lock);
9439
9440	switch (cdb->select_report) {
9441	case RPL_REPORT_DEFAULT:
9442	case RPL_REPORT_ALL:
9443		break;
9444	case RPL_REPORT_WELLKNOWN:
9445		well_known = 1;
9446		num_luns = 0;
9447		break;
9448	default:
9449		ctl_set_invalid_field(ctsio,
9450				      /*sks_valid*/ 1,
9451				      /*command*/ 1,
9452				      /*field*/ 2,
9453				      /*bit_valid*/ 0,
9454				      /*bit*/ 0);
9455		ctl_done((union ctl_io *)ctsio);
9456		return (retval);
9457		break; /* NOTREACHED */
9458	}
9459
9460	alloc_len = scsi_4btoul(cdb->length);
9461	/*
9462	 * The initiator has to allocate at least 16 bytes for this request,
9463	 * so he can at least get the header and the first LUN.  Otherwise
9464	 * we reject the request (per SPC-3 rev 14, section 6.21).
9465	 */
9466	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9467	    sizeof(struct scsi_report_luns_lundata))) {
9468		ctl_set_invalid_field(ctsio,
9469				      /*sks_valid*/ 1,
9470				      /*command*/ 1,
9471				      /*field*/ 6,
9472				      /*bit_valid*/ 0,
9473				      /*bit*/ 0);
9474		ctl_done((union ctl_io *)ctsio);
9475		return (retval);
9476	}
9477
9478	request_lun = (struct ctl_lun *)
9479		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9480
9481	lun_datalen = sizeof(*lun_data) +
9482		(num_luns * sizeof(struct scsi_report_luns_lundata));
9483
9484	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9485	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9486	ctsio->kern_sg_entries = 0;
9487
9488	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9489
9490	mtx_lock(&control_softc->ctl_lock);
9491	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9492		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9493		if (lun_id >= CTL_MAX_LUNS)
9494			continue;
9495		lun = control_softc->ctl_luns[lun_id];
9496		if (lun == NULL)
9497			continue;
9498
9499		if (targ_lun_id <= 0xff) {
9500			/*
9501			 * Peripheral addressing method, bus number 0.
9502			 */
9503			lun_data->luns[num_filled].lundata[0] =
9504				RPL_LUNDATA_ATYP_PERIPH;
9505			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9506			num_filled++;
9507		} else if (targ_lun_id <= 0x3fff) {
9508			/*
9509			 * Flat addressing method.
9510			 */
9511			lun_data->luns[num_filled].lundata[0] =
9512				RPL_LUNDATA_ATYP_FLAT |
9513				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9514#ifdef OLDCTLHEADERS
9515				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9516				(targ_lun_id & SRLD_BUS_LUN_MASK);
9517#endif
9518			lun_data->luns[num_filled].lundata[1] =
9519#ifdef OLDCTLHEADERS
9520				targ_lun_id >> SRLD_BUS_LUN_BITS;
9521#endif
9522				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9523			num_filled++;
9524		} else {
9525			printf("ctl_report_luns: bogus LUN number %jd, "
9526			       "skipping\n", (intmax_t)targ_lun_id);
9527		}
9528		/*
9529		 * According to SPC-3, rev 14 section 6.21:
9530		 *
9531		 * "The execution of a REPORT LUNS command to any valid and
9532		 * installed logical unit shall clear the REPORTED LUNS DATA
9533		 * HAS CHANGED unit attention condition for all logical
9534		 * units of that target with respect to the requesting
9535		 * initiator. A valid and installed logical unit is one
9536		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9537		 * INQUIRY data (see 6.4.2)."
9538		 *
9539		 * If request_lun is NULL, the LUN this report luns command
9540		 * was issued to is either disabled or doesn't exist. In that
9541		 * case, we shouldn't clear any pending lun change unit
9542		 * attention.
9543		 */
9544		if (request_lun != NULL) {
9545			mtx_lock(&lun->lun_lock);
9546			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9547			mtx_unlock(&lun->lun_lock);
9548		}
9549	}
9550	mtx_unlock(&control_softc->ctl_lock);
9551
9552	/*
9553	 * It's quite possible that we've returned fewer LUNs than we allocated
9554	 * space for.  Trim it.
9555	 */
9556	lun_datalen = sizeof(*lun_data) +
9557		(num_filled * sizeof(struct scsi_report_luns_lundata));
9558
9559	if (lun_datalen < alloc_len) {
9560		ctsio->residual = alloc_len - lun_datalen;
9561		ctsio->kern_data_len = lun_datalen;
9562		ctsio->kern_total_len = lun_datalen;
9563	} else {
9564		ctsio->residual = 0;
9565		ctsio->kern_data_len = alloc_len;
9566		ctsio->kern_total_len = alloc_len;
9567	}
9568	ctsio->kern_data_resid = 0;
9569	ctsio->kern_rel_offset = 0;
9570	ctsio->kern_sg_entries = 0;
9571
9572	/*
9573	 * We set this to the actual data length, regardless of how much
9574	 * space we actually have to return results.  If the user looks at
9575	 * this value, he'll know whether or not he allocated enough space
9576	 * and reissue the command if necessary.  We don't support well
9577	 * known logical units, so if the user asks for that, return none.
9578	 */
9579	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9580
9581	/*
9582	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9583	 * this request.
9584	 */
9585	ctsio->scsi_status = SCSI_STATUS_OK;
9586
9587	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9588	ctsio->be_move_done = ctl_config_move_done;
9589	ctl_datamove((union ctl_io *)ctsio);
9590
9591	return (retval);
9592}
9593
9594int
9595ctl_request_sense(struct ctl_scsiio *ctsio)
9596{
9597	struct scsi_request_sense *cdb;
9598	struct scsi_sense_data *sense_ptr;
9599	struct ctl_lun *lun;
9600	uint32_t initidx;
9601	int have_error;
9602	scsi_sense_data_type sense_format;
9603
9604	cdb = (struct scsi_request_sense *)ctsio->cdb;
9605
9606	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9607
9608	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9609
9610	/*
9611	 * Determine which sense format the user wants.
9612	 */
9613	if (cdb->byte2 & SRS_DESC)
9614		sense_format = SSD_TYPE_DESC;
9615	else
9616		sense_format = SSD_TYPE_FIXED;
9617
9618	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9619	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9620	ctsio->kern_sg_entries = 0;
9621
9622	/*
9623	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9624	 * larger than the largest allowed value for the length field in the
9625	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9626	 */
9627	ctsio->residual = 0;
9628	ctsio->kern_data_len = cdb->length;
9629	ctsio->kern_total_len = cdb->length;
9630
9631	ctsio->kern_data_resid = 0;
9632	ctsio->kern_rel_offset = 0;
9633	ctsio->kern_sg_entries = 0;
9634
9635	/*
9636	 * If we don't have a LUN, we don't have any pending sense.
9637	 */
9638	if (lun == NULL)
9639		goto no_sense;
9640
9641	have_error = 0;
9642	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9643	/*
9644	 * Check for pending sense, and then for pending unit attentions.
9645	 * Pending sense gets returned first, then pending unit attentions.
9646	 */
9647	mtx_lock(&lun->lun_lock);
9648#ifdef CTL_WITH_CA
9649	if (ctl_is_set(lun->have_ca, initidx)) {
9650		scsi_sense_data_type stored_format;
9651
9652		/*
9653		 * Check to see which sense format was used for the stored
9654		 * sense data.
9655		 */
9656		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9657
9658		/*
9659		 * If the user requested a different sense format than the
9660		 * one we stored, then we need to convert it to the other
9661		 * format.  If we're going from descriptor to fixed format
9662		 * sense data, we may lose things in translation, depending
9663		 * on what options were used.
9664		 *
9665		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9666		 * for some reason we'll just copy it out as-is.
9667		 */
9668		if ((stored_format == SSD_TYPE_FIXED)
9669		 && (sense_format == SSD_TYPE_DESC))
9670			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9671			    &lun->pending_sense[initidx],
9672			    (struct scsi_sense_data_desc *)sense_ptr);
9673		else if ((stored_format == SSD_TYPE_DESC)
9674		      && (sense_format == SSD_TYPE_FIXED))
9675			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9676			    &lun->pending_sense[initidx],
9677			    (struct scsi_sense_data_fixed *)sense_ptr);
9678		else
9679			memcpy(sense_ptr, &lun->pending_sense[initidx],
9680			       ctl_min(sizeof(*sense_ptr),
9681			       sizeof(lun->pending_sense[initidx])));
9682
9683		ctl_clear_mask(lun->have_ca, initidx);
9684		have_error = 1;
9685	} else
9686#endif
9687	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9688		ctl_ua_type ua_type;
9689
9690		ua_type = ctl_build_ua(lun->pending_ua[initidx],
9691				       sense_ptr, sense_format);
9692		if (ua_type != CTL_UA_NONE) {
9693			have_error = 1;
9694			/* We're reporting this UA, so clear it */
9695			lun->pending_ua[initidx] &= ~ua_type;
9696		}
9697	}
9698	mtx_unlock(&lun->lun_lock);
9699
9700	/*
9701	 * We already have a pending error, return it.
9702	 */
9703	if (have_error != 0) {
9704		/*
9705		 * We report the SCSI status as OK, since the status of the
9706		 * request sense command itself is OK.
9707		 */
9708		ctsio->scsi_status = SCSI_STATUS_OK;
9709
9710		/*
9711		 * We report 0 for the sense length, because we aren't doing
9712		 * autosense in this case.  We're reporting sense as
9713		 * parameter data.
9714		 */
9715		ctsio->sense_len = 0;
9716		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9717		ctsio->be_move_done = ctl_config_move_done;
9718		ctl_datamove((union ctl_io *)ctsio);
9719
9720		return (CTL_RETVAL_COMPLETE);
9721	}
9722
9723no_sense:
9724
9725	/*
9726	 * No sense information to report, so we report that everything is
9727	 * okay.
9728	 */
9729	ctl_set_sense_data(sense_ptr,
9730			   lun,
9731			   sense_format,
9732			   /*current_error*/ 1,
9733			   /*sense_key*/ SSD_KEY_NO_SENSE,
9734			   /*asc*/ 0x00,
9735			   /*ascq*/ 0x00,
9736			   SSD_ELEM_NONE);
9737
9738	ctsio->scsi_status = SCSI_STATUS_OK;
9739
9740	/*
9741	 * We report 0 for the sense length, because we aren't doing
9742	 * autosense in this case.  We're reporting sense as parameter data.
9743	 */
9744	ctsio->sense_len = 0;
9745	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9746	ctsio->be_move_done = ctl_config_move_done;
9747	ctl_datamove((union ctl_io *)ctsio);
9748
9749	return (CTL_RETVAL_COMPLETE);
9750}
9751
9752int
9753ctl_tur(struct ctl_scsiio *ctsio)
9754{
9755	struct ctl_lun *lun;
9756
9757	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9758
9759	CTL_DEBUG_PRINT(("ctl_tur\n"));
9760
9761	if (lun == NULL)
9762		return (EINVAL);
9763
9764	ctsio->scsi_status = SCSI_STATUS_OK;
9765	ctsio->io_hdr.status = CTL_SUCCESS;
9766
9767	ctl_done((union ctl_io *)ctsio);
9768
9769	return (CTL_RETVAL_COMPLETE);
9770}
9771
9772#ifdef notyet
9773static int
9774ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9775{
9776
9777}
9778#endif
9779
9780static int
9781ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9782{
9783	struct scsi_vpd_supported_pages *pages;
9784	int sup_page_size;
9785	struct ctl_lun *lun;
9786
9787	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9788
9789	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9790	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9791	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9792	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9793	ctsio->kern_sg_entries = 0;
9794
9795	if (sup_page_size < alloc_len) {
9796		ctsio->residual = alloc_len - sup_page_size;
9797		ctsio->kern_data_len = sup_page_size;
9798		ctsio->kern_total_len = sup_page_size;
9799	} else {
9800		ctsio->residual = 0;
9801		ctsio->kern_data_len = alloc_len;
9802		ctsio->kern_total_len = alloc_len;
9803	}
9804	ctsio->kern_data_resid = 0;
9805	ctsio->kern_rel_offset = 0;
9806	ctsio->kern_sg_entries = 0;
9807
9808	/*
9809	 * The control device is always connected.  The disk device, on the
9810	 * other hand, may not be online all the time.  Need to change this
9811	 * to figure out whether the disk device is actually online or not.
9812	 */
9813	if (lun != NULL)
9814		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9815				lun->be_lun->lun_type;
9816	else
9817		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9818
9819	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9820	/* Supported VPD pages */
9821	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9822	/* Serial Number */
9823	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9824	/* Device Identification */
9825	pages->page_list[2] = SVPD_DEVICE_ID;
9826	/* Extended INQUIRY Data */
9827	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9828	/* Mode Page Policy */
9829	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9830	/* SCSI Ports */
9831	pages->page_list[5] = SVPD_SCSI_PORTS;
9832	/* Third-party Copy */
9833	pages->page_list[6] = SVPD_SCSI_TPC;
9834	/* Block limits */
9835	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9836	/* Block Device Characteristics */
9837	pages->page_list[8] = SVPD_BDC;
9838	/* Logical Block Provisioning */
9839	pages->page_list[9] = SVPD_LBP;
9840
9841	ctsio->scsi_status = SCSI_STATUS_OK;
9842
9843	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9844	ctsio->be_move_done = ctl_config_move_done;
9845	ctl_datamove((union ctl_io *)ctsio);
9846
9847	return (CTL_RETVAL_COMPLETE);
9848}
9849
9850static int
9851ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9852{
9853	struct scsi_vpd_unit_serial_number *sn_ptr;
9854	struct ctl_lun *lun;
9855
9856	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9857
9858	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9859	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9860	ctsio->kern_sg_entries = 0;
9861
9862	if (sizeof(*sn_ptr) < alloc_len) {
9863		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9864		ctsio->kern_data_len = sizeof(*sn_ptr);
9865		ctsio->kern_total_len = sizeof(*sn_ptr);
9866	} else {
9867		ctsio->residual = 0;
9868		ctsio->kern_data_len = alloc_len;
9869		ctsio->kern_total_len = alloc_len;
9870	}
9871	ctsio->kern_data_resid = 0;
9872	ctsio->kern_rel_offset = 0;
9873	ctsio->kern_sg_entries = 0;
9874
9875	/*
9876	 * The control device is always connected.  The disk device, on the
9877	 * other hand, may not be online all the time.  Need to change this
9878	 * to figure out whether the disk device is actually online or not.
9879	 */
9880	if (lun != NULL)
9881		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9882				  lun->be_lun->lun_type;
9883	else
9884		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9885
9886	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9887	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9888	/*
9889	 * If we don't have a LUN, we just leave the serial number as
9890	 * all spaces.
9891	 */
9892	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9893	if (lun != NULL) {
9894		strncpy((char *)sn_ptr->serial_num,
9895			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9896	}
9897	ctsio->scsi_status = SCSI_STATUS_OK;
9898
9899	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9900	ctsio->be_move_done = ctl_config_move_done;
9901	ctl_datamove((union ctl_io *)ctsio);
9902
9903	return (CTL_RETVAL_COMPLETE);
9904}
9905
9906
9907static int
9908ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9909{
9910	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9911	struct ctl_lun *lun;
9912	int data_len;
9913
9914	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9915
9916	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9917	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9918
9919	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9920	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9921	ctsio->kern_sg_entries = 0;
9922
9923	if (data_len < alloc_len) {
9924		ctsio->residual = alloc_len - data_len;
9925		ctsio->kern_data_len = data_len;
9926		ctsio->kern_total_len = data_len;
9927	} else {
9928		ctsio->residual = 0;
9929		ctsio->kern_data_len = alloc_len;
9930		ctsio->kern_total_len = alloc_len;
9931	}
9932	ctsio->kern_data_resid = 0;
9933	ctsio->kern_rel_offset = 0;
9934	ctsio->kern_sg_entries = 0;
9935
9936	/*
9937	 * The control device is always connected.  The disk device, on the
9938	 * other hand, may not be online all the time.
9939	 */
9940	if (lun != NULL)
9941		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9942				     lun->be_lun->lun_type;
9943	else
9944		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9945	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9946	eid_ptr->page_length = data_len - 4;
9947	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9948	eid_ptr->flags3 = SVPD_EID_V_SUP;
9949
9950	ctsio->scsi_status = SCSI_STATUS_OK;
9951	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9952	ctsio->be_move_done = ctl_config_move_done;
9953	ctl_datamove((union ctl_io *)ctsio);
9954
9955	return (CTL_RETVAL_COMPLETE);
9956}
9957
9958static int
9959ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9960{
9961	struct scsi_vpd_mode_page_policy *mpp_ptr;
9962	struct ctl_lun *lun;
9963	int data_len;
9964
9965	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9966
9967	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9968	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9969
9970	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9971	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9972	ctsio->kern_sg_entries = 0;
9973
9974	if (data_len < alloc_len) {
9975		ctsio->residual = alloc_len - data_len;
9976		ctsio->kern_data_len = data_len;
9977		ctsio->kern_total_len = data_len;
9978	} else {
9979		ctsio->residual = 0;
9980		ctsio->kern_data_len = alloc_len;
9981		ctsio->kern_total_len = alloc_len;
9982	}
9983	ctsio->kern_data_resid = 0;
9984	ctsio->kern_rel_offset = 0;
9985	ctsio->kern_sg_entries = 0;
9986
9987	/*
9988	 * The control device is always connected.  The disk device, on the
9989	 * other hand, may not be online all the time.
9990	 */
9991	if (lun != NULL)
9992		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9993				     lun->be_lun->lun_type;
9994	else
9995		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9996	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9997	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9998	mpp_ptr->descr[0].page_code = 0x3f;
9999	mpp_ptr->descr[0].subpage_code = 0xff;
10000	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
10001
10002	ctsio->scsi_status = SCSI_STATUS_OK;
10003	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10004	ctsio->be_move_done = ctl_config_move_done;
10005	ctl_datamove((union ctl_io *)ctsio);
10006
10007	return (CTL_RETVAL_COMPLETE);
10008}
10009
10010static int
10011ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
10012{
10013	struct scsi_vpd_device_id *devid_ptr;
10014	struct scsi_vpd_id_descriptor *desc;
10015	struct ctl_softc *ctl_softc;
10016	struct ctl_lun *lun;
10017	struct ctl_port *port;
10018	int data_len;
10019	uint8_t proto;
10020
10021	ctl_softc = control_softc;
10022
10023	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
10024	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10025
10026	data_len = sizeof(struct scsi_vpd_device_id) +
10027	    sizeof(struct scsi_vpd_id_descriptor) +
10028		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
10029	    sizeof(struct scsi_vpd_id_descriptor) +
10030		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
10031	if (lun && lun->lun_devid)
10032		data_len += lun->lun_devid->len;
10033	if (port->port_devid)
10034		data_len += port->port_devid->len;
10035	if (port->target_devid)
10036		data_len += port->target_devid->len;
10037
10038	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10039	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
10040	ctsio->kern_sg_entries = 0;
10041
10042	if (data_len < alloc_len) {
10043		ctsio->residual = alloc_len - data_len;
10044		ctsio->kern_data_len = data_len;
10045		ctsio->kern_total_len = data_len;
10046	} else {
10047		ctsio->residual = 0;
10048		ctsio->kern_data_len = alloc_len;
10049		ctsio->kern_total_len = alloc_len;
10050	}
10051	ctsio->kern_data_resid = 0;
10052	ctsio->kern_rel_offset = 0;
10053	ctsio->kern_sg_entries = 0;
10054
10055	/*
10056	 * The control device is always connected.  The disk device, on the
10057	 * other hand, may not be online all the time.
10058	 */
10059	if (lun != NULL)
10060		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10061				     lun->be_lun->lun_type;
10062	else
10063		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10064	devid_ptr->page_code = SVPD_DEVICE_ID;
10065	scsi_ulto2b(data_len - 4, devid_ptr->length);
10066
10067	if (port->port_type == CTL_PORT_FC)
10068		proto = SCSI_PROTO_FC << 4;
10069	else if (port->port_type == CTL_PORT_ISCSI)
10070		proto = SCSI_PROTO_ISCSI << 4;
10071	else
10072		proto = SCSI_PROTO_SPI << 4;
10073	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
10074
10075	/*
10076	 * We're using a LUN association here.  i.e., this device ID is a
10077	 * per-LUN identifier.
10078	 */
10079	if (lun && lun->lun_devid) {
10080		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
10081		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10082		    lun->lun_devid->len);
10083	}
10084
10085	/*
10086	 * This is for the WWPN which is a port association.
10087	 */
10088	if (port->port_devid) {
10089		memcpy(desc, port->port_devid->data, port->port_devid->len);
10090		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10091		    port->port_devid->len);
10092	}
10093
10094	/*
10095	 * This is for the Relative Target Port(type 4h) identifier
10096	 */
10097	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10098	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10099	    SVPD_ID_TYPE_RELTARG;
10100	desc->length = 4;
10101	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
10102	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10103	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
10104
10105	/*
10106	 * This is for the Target Port Group(type 5h) identifier
10107	 */
10108	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10109	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10110	    SVPD_ID_TYPE_TPORTGRP;
10111	desc->length = 4;
10112	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
10113	    &desc->identifier[2]);
10114	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10115	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10116
10117	/*
10118	 * This is for the Target identifier
10119	 */
10120	if (port->target_devid) {
10121		memcpy(desc, port->target_devid->data, port->target_devid->len);
10122	}
10123
10124	ctsio->scsi_status = SCSI_STATUS_OK;
10125	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10126	ctsio->be_move_done = ctl_config_move_done;
10127	ctl_datamove((union ctl_io *)ctsio);
10128
10129	return (CTL_RETVAL_COMPLETE);
10130}
10131
10132static int
10133ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10134{
10135	struct ctl_softc *softc = control_softc;
10136	struct scsi_vpd_scsi_ports *sp;
10137	struct scsi_vpd_port_designation *pd;
10138	struct scsi_vpd_port_designation_cont *pdc;
10139	struct ctl_lun *lun;
10140	struct ctl_port *port;
10141	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10142	int num_target_port_groups, single;
10143
10144	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10145
10146	single = ctl_is_single;
10147	if (single)
10148		num_target_port_groups = 1;
10149	else
10150		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10151	num_target_ports = 0;
10152	iid_len = 0;
10153	id_len = 0;
10154	mtx_lock(&softc->ctl_lock);
10155	STAILQ_FOREACH(port, &softc->port_list, links) {
10156		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10157			continue;
10158		if (lun != NULL &&
10159		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10160		    CTL_MAX_LUNS)
10161			continue;
10162		num_target_ports++;
10163		if (port->init_devid)
10164			iid_len += port->init_devid->len;
10165		if (port->port_devid)
10166			id_len += port->port_devid->len;
10167	}
10168	mtx_unlock(&softc->ctl_lock);
10169
10170	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10171	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10172	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10173	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10174	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10175	ctsio->kern_sg_entries = 0;
10176
10177	if (data_len < alloc_len) {
10178		ctsio->residual = alloc_len - data_len;
10179		ctsio->kern_data_len = data_len;
10180		ctsio->kern_total_len = data_len;
10181	} else {
10182		ctsio->residual = 0;
10183		ctsio->kern_data_len = alloc_len;
10184		ctsio->kern_total_len = alloc_len;
10185	}
10186	ctsio->kern_data_resid = 0;
10187	ctsio->kern_rel_offset = 0;
10188	ctsio->kern_sg_entries = 0;
10189
10190	/*
10191	 * The control device is always connected.  The disk device, on the
10192	 * other hand, may not be online all the time.  Need to change this
10193	 * to figure out whether the disk device is actually online or not.
10194	 */
10195	if (lun != NULL)
10196		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10197				  lun->be_lun->lun_type;
10198	else
10199		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10200
10201	sp->page_code = SVPD_SCSI_PORTS;
10202	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10203	    sp->page_length);
10204	pd = &sp->design[0];
10205
10206	mtx_lock(&softc->ctl_lock);
10207	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10208		pg = 0;
10209	else
10210		pg = 1;
10211	for (g = 0; g < num_target_port_groups; g++) {
10212		STAILQ_FOREACH(port, &softc->port_list, links) {
10213			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10214				continue;
10215			if (lun != NULL &&
10216			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10217			    CTL_MAX_LUNS)
10218				continue;
10219			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10220			scsi_ulto2b(p, pd->relative_port_id);
10221			if (port->init_devid && g == pg) {
10222				iid_len = port->init_devid->len;
10223				memcpy(pd->initiator_transportid,
10224				    port->init_devid->data, port->init_devid->len);
10225			} else
10226				iid_len = 0;
10227			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10228			pdc = (struct scsi_vpd_port_designation_cont *)
10229			    (&pd->initiator_transportid[iid_len]);
10230			if (port->port_devid && g == pg) {
10231				id_len = port->port_devid->len;
10232				memcpy(pdc->target_port_descriptors,
10233				    port->port_devid->data, port->port_devid->len);
10234			} else
10235				id_len = 0;
10236			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10237			pd = (struct scsi_vpd_port_designation *)
10238			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10239		}
10240	}
10241	mtx_unlock(&softc->ctl_lock);
10242
10243	ctsio->scsi_status = SCSI_STATUS_OK;
10244	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10245	ctsio->be_move_done = ctl_config_move_done;
10246	ctl_datamove((union ctl_io *)ctsio);
10247
10248	return (CTL_RETVAL_COMPLETE);
10249}
10250
10251static int
10252ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10253{
10254	struct scsi_vpd_block_limits *bl_ptr;
10255	struct ctl_lun *lun;
10256	int bs;
10257
10258	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10259
10260	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10261	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10262	ctsio->kern_sg_entries = 0;
10263
10264	if (sizeof(*bl_ptr) < alloc_len) {
10265		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10266		ctsio->kern_data_len = sizeof(*bl_ptr);
10267		ctsio->kern_total_len = sizeof(*bl_ptr);
10268	} else {
10269		ctsio->residual = 0;
10270		ctsio->kern_data_len = alloc_len;
10271		ctsio->kern_total_len = alloc_len;
10272	}
10273	ctsio->kern_data_resid = 0;
10274	ctsio->kern_rel_offset = 0;
10275	ctsio->kern_sg_entries = 0;
10276
10277	/*
10278	 * The control device is always connected.  The disk device, on the
10279	 * other hand, may not be online all the time.  Need to change this
10280	 * to figure out whether the disk device is actually online or not.
10281	 */
10282	if (lun != NULL)
10283		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10284				  lun->be_lun->lun_type;
10285	else
10286		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10287
10288	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10289	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
10290	bl_ptr->max_cmp_write_len = 0xff;
10291	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10292	if (lun != NULL) {
10293		bs = lun->be_lun->blocksize;
10294		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10295		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10296			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10297			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10298			if (lun->be_lun->pblockexp != 0) {
10299				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10300				    bl_ptr->opt_unmap_grain);
10301				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10302				    bl_ptr->unmap_grain_align);
10303			}
10304		}
10305	}
10306	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10307
10308	ctsio->scsi_status = SCSI_STATUS_OK;
10309	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10310	ctsio->be_move_done = ctl_config_move_done;
10311	ctl_datamove((union ctl_io *)ctsio);
10312
10313	return (CTL_RETVAL_COMPLETE);
10314}
10315
10316static int
10317ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10318{
10319	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10320	struct ctl_lun *lun;
10321
10322	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10323
10324	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10325	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10326	ctsio->kern_sg_entries = 0;
10327
10328	if (sizeof(*bdc_ptr) < alloc_len) {
10329		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10330		ctsio->kern_data_len = sizeof(*bdc_ptr);
10331		ctsio->kern_total_len = sizeof(*bdc_ptr);
10332	} else {
10333		ctsio->residual = 0;
10334		ctsio->kern_data_len = alloc_len;
10335		ctsio->kern_total_len = alloc_len;
10336	}
10337	ctsio->kern_data_resid = 0;
10338	ctsio->kern_rel_offset = 0;
10339	ctsio->kern_sg_entries = 0;
10340
10341	/*
10342	 * The control device is always connected.  The disk device, on the
10343	 * other hand, may not be online all the time.  Need to change this
10344	 * to figure out whether the disk device is actually online or not.
10345	 */
10346	if (lun != NULL)
10347		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10348				  lun->be_lun->lun_type;
10349	else
10350		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10351	bdc_ptr->page_code = SVPD_BDC;
10352	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10353	scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10354	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10355
10356	ctsio->scsi_status = SCSI_STATUS_OK;
10357	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10358	ctsio->be_move_done = ctl_config_move_done;
10359	ctl_datamove((union ctl_io *)ctsio);
10360
10361	return (CTL_RETVAL_COMPLETE);
10362}
10363
10364static int
10365ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10366{
10367	struct scsi_vpd_logical_block_prov *lbp_ptr;
10368	struct ctl_lun *lun;
10369
10370	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10371
10372	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10373	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10374	ctsio->kern_sg_entries = 0;
10375
10376	if (sizeof(*lbp_ptr) < alloc_len) {
10377		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10378		ctsio->kern_data_len = sizeof(*lbp_ptr);
10379		ctsio->kern_total_len = sizeof(*lbp_ptr);
10380	} else {
10381		ctsio->residual = 0;
10382		ctsio->kern_data_len = alloc_len;
10383		ctsio->kern_total_len = alloc_len;
10384	}
10385	ctsio->kern_data_resid = 0;
10386	ctsio->kern_rel_offset = 0;
10387	ctsio->kern_sg_entries = 0;
10388
10389	/*
10390	 * The control device is always connected.  The disk device, on the
10391	 * other hand, may not be online all the time.  Need to change this
10392	 * to figure out whether the disk device is actually online or not.
10393	 */
10394	if (lun != NULL)
10395		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10396				  lun->be_lun->lun_type;
10397	else
10398		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10399
10400	lbp_ptr->page_code = SVPD_LBP;
10401	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10402	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10403		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10404		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10405		lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10406	}
10407
10408	ctsio->scsi_status = SCSI_STATUS_OK;
10409	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10410	ctsio->be_move_done = ctl_config_move_done;
10411	ctl_datamove((union ctl_io *)ctsio);
10412
10413	return (CTL_RETVAL_COMPLETE);
10414}
10415
10416static int
10417ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10418{
10419	struct scsi_inquiry *cdb;
10420	struct ctl_lun *lun;
10421	int alloc_len, retval;
10422
10423	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10424	cdb = (struct scsi_inquiry *)ctsio->cdb;
10425
10426	retval = CTL_RETVAL_COMPLETE;
10427
10428	alloc_len = scsi_2btoul(cdb->length);
10429
10430	switch (cdb->page_code) {
10431	case SVPD_SUPPORTED_PAGES:
10432		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10433		break;
10434	case SVPD_UNIT_SERIAL_NUMBER:
10435		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10436		break;
10437	case SVPD_DEVICE_ID:
10438		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10439		break;
10440	case SVPD_EXTENDED_INQUIRY_DATA:
10441		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10442		break;
10443	case SVPD_MODE_PAGE_POLICY:
10444		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10445		break;
10446	case SVPD_SCSI_PORTS:
10447		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10448		break;
10449	case SVPD_SCSI_TPC:
10450		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10451		break;
10452	case SVPD_BLOCK_LIMITS:
10453		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10454		break;
10455	case SVPD_BDC:
10456		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10457		break;
10458	case SVPD_LBP:
10459		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10460		break;
10461	default:
10462		ctl_set_invalid_field(ctsio,
10463				      /*sks_valid*/ 1,
10464				      /*command*/ 1,
10465				      /*field*/ 2,
10466				      /*bit_valid*/ 0,
10467				      /*bit*/ 0);
10468		ctl_done((union ctl_io *)ctsio);
10469		retval = CTL_RETVAL_COMPLETE;
10470		break;
10471	}
10472
10473	return (retval);
10474}
10475
10476static int
10477ctl_inquiry_std(struct ctl_scsiio *ctsio)
10478{
10479	struct scsi_inquiry_data *inq_ptr;
10480	struct scsi_inquiry *cdb;
10481	struct ctl_softc *ctl_softc;
10482	struct ctl_lun *lun;
10483	char *val;
10484	uint32_t alloc_len;
10485	ctl_port_type port_type;
10486
10487	ctl_softc = control_softc;
10488
10489	/*
10490	 * Figure out whether we're talking to a Fibre Channel port or not.
10491	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10492	 * SCSI front ends.
10493	 */
10494	port_type = ctl_softc->ctl_ports[
10495	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10496	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10497		port_type = CTL_PORT_SCSI;
10498
10499	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10500	cdb = (struct scsi_inquiry *)ctsio->cdb;
10501	alloc_len = scsi_2btoul(cdb->length);
10502
10503	/*
10504	 * We malloc the full inquiry data size here and fill it
10505	 * in.  If the user only asks for less, we'll give him
10506	 * that much.
10507	 */
10508	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
10509	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10510	ctsio->kern_sg_entries = 0;
10511	ctsio->kern_data_resid = 0;
10512	ctsio->kern_rel_offset = 0;
10513
10514	if (sizeof(*inq_ptr) < alloc_len) {
10515		ctsio->residual = alloc_len - sizeof(*inq_ptr);
10516		ctsio->kern_data_len = sizeof(*inq_ptr);
10517		ctsio->kern_total_len = sizeof(*inq_ptr);
10518	} else {
10519		ctsio->residual = 0;
10520		ctsio->kern_data_len = alloc_len;
10521		ctsio->kern_total_len = alloc_len;
10522	}
10523
10524	/*
10525	 * If we have a LUN configured, report it as connected.  Otherwise,
10526	 * report that it is offline or no device is supported, depending
10527	 * on the value of inquiry_pq_no_lun.
10528	 *
10529	 * According to the spec (SPC-4 r34), the peripheral qualifier
10530	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10531	 *
10532	 * "A peripheral device having the specified peripheral device type
10533	 * is not connected to this logical unit. However, the device
10534	 * server is capable of supporting the specified peripheral device
10535	 * type on this logical unit."
10536	 *
10537	 * According to the same spec, the peripheral qualifier
10538	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10539	 *
10540	 * "The device server is not capable of supporting a peripheral
10541	 * device on this logical unit. For this peripheral qualifier the
10542	 * peripheral device type shall be set to 1Fh. All other peripheral
10543	 * device type values are reserved for this peripheral qualifier."
10544	 *
10545	 * Given the text, it would seem that we probably want to report that
10546	 * the LUN is offline here.  There is no LUN connected, but we can
10547	 * support a LUN at the given LUN number.
10548	 *
10549	 * In the real world, though, it sounds like things are a little
10550	 * different:
10551	 *
10552	 * - Linux, when presented with a LUN with the offline peripheral
10553	 *   qualifier, will create an sg driver instance for it.  So when
10554	 *   you attach it to CTL, you wind up with a ton of sg driver
10555	 *   instances.  (One for every LUN that Linux bothered to probe.)
10556	 *   Linux does this despite the fact that it issues a REPORT LUNs
10557	 *   to LUN 0 to get the inventory of supported LUNs.
10558	 *
10559	 * - There is other anecdotal evidence (from Emulex folks) about
10560	 *   arrays that use the offline peripheral qualifier for LUNs that
10561	 *   are on the "passive" path in an active/passive array.
10562	 *
10563	 * So the solution is provide a hopefully reasonable default
10564	 * (return bad/no LUN) and allow the user to change the behavior
10565	 * with a tunable/sysctl variable.
10566	 */
10567	if (lun != NULL)
10568		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10569				  lun->be_lun->lun_type;
10570	else if (ctl_softc->inquiry_pq_no_lun == 0)
10571		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10572	else
10573		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10574
10575	/* RMB in byte 2 is 0 */
10576	inq_ptr->version = SCSI_REV_SPC4;
10577
10578	/*
10579	 * According to SAM-3, even if a device only supports a single
10580	 * level of LUN addressing, it should still set the HISUP bit:
10581	 *
10582	 * 4.9.1 Logical unit numbers overview
10583	 *
10584	 * All logical unit number formats described in this standard are
10585	 * hierarchical in structure even when only a single level in that
10586	 * hierarchy is used. The HISUP bit shall be set to one in the
10587	 * standard INQUIRY data (see SPC-2) when any logical unit number
10588	 * format described in this standard is used.  Non-hierarchical
10589	 * formats are outside the scope of this standard.
10590	 *
10591	 * Therefore we set the HiSup bit here.
10592	 *
10593	 * The reponse format is 2, per SPC-3.
10594	 */
10595	inq_ptr->response_format = SID_HiSup | 2;
10596
10597	inq_ptr->additional_length =
10598	    offsetof(struct scsi_inquiry_data, vendor_specific1) -
10599	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10600	CTL_DEBUG_PRINT(("additional_length = %d\n",
10601			 inq_ptr->additional_length));
10602
10603	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10604	/* 16 bit addressing */
10605	if (port_type == CTL_PORT_SCSI)
10606		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10607	/* XXX set the SID_MultiP bit here if we're actually going to
10608	   respond on multiple ports */
10609	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10610
10611	/* 16 bit data bus, synchronous transfers */
10612	if (port_type == CTL_PORT_SCSI)
10613		inq_ptr->flags = SID_WBus16 | SID_Sync;
10614	/*
10615	 * XXX KDM do we want to support tagged queueing on the control
10616	 * device at all?
10617	 */
10618	if ((lun == NULL)
10619	 || (lun->be_lun->lun_type != T_PROCESSOR))
10620		inq_ptr->flags |= SID_CmdQue;
10621	/*
10622	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10623	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10624	 * name and 4 bytes for the revision.
10625	 */
10626	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10627	    "vendor")) == NULL) {
10628		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10629	} else {
10630		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10631		strncpy(inq_ptr->vendor, val,
10632		    min(sizeof(inq_ptr->vendor), strlen(val)));
10633	}
10634	if (lun == NULL) {
10635		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10636		    sizeof(inq_ptr->product));
10637	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10638		switch (lun->be_lun->lun_type) {
10639		case T_DIRECT:
10640			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10641			    sizeof(inq_ptr->product));
10642			break;
10643		case T_PROCESSOR:
10644			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10645			    sizeof(inq_ptr->product));
10646			break;
10647		default:
10648			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10649			    sizeof(inq_ptr->product));
10650			break;
10651		}
10652	} else {
10653		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10654		strncpy(inq_ptr->product, val,
10655		    min(sizeof(inq_ptr->product), strlen(val)));
10656	}
10657
10658	/*
10659	 * XXX make this a macro somewhere so it automatically gets
10660	 * incremented when we make changes.
10661	 */
10662	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10663	    "revision")) == NULL) {
10664		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10665	} else {
10666		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10667		strncpy(inq_ptr->revision, val,
10668		    min(sizeof(inq_ptr->revision), strlen(val)));
10669	}
10670
10671	/*
10672	 * For parallel SCSI, we support double transition and single
10673	 * transition clocking.  We also support QAS (Quick Arbitration
10674	 * and Selection) and Information Unit transfers on both the
10675	 * control and array devices.
10676	 */
10677	if (port_type == CTL_PORT_SCSI)
10678		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10679				    SID_SPI_IUS;
10680
10681	/* SAM-5 (no version claimed) */
10682	scsi_ulto2b(0x00A0, inq_ptr->version1);
10683	/* SPC-4 (no version claimed) */
10684	scsi_ulto2b(0x0460, inq_ptr->version2);
10685	if (port_type == CTL_PORT_FC) {
10686		/* FCP-2 ANSI INCITS.350:2003 */
10687		scsi_ulto2b(0x0917, inq_ptr->version3);
10688	} else if (port_type == CTL_PORT_SCSI) {
10689		/* SPI-4 ANSI INCITS.362:200x */
10690		scsi_ulto2b(0x0B56, inq_ptr->version3);
10691	} else if (port_type == CTL_PORT_ISCSI) {
10692		/* iSCSI (no version claimed) */
10693		scsi_ulto2b(0x0960, inq_ptr->version3);
10694	} else if (port_type == CTL_PORT_SAS) {
10695		/* SAS (no version claimed) */
10696		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10697	}
10698
10699	if (lun == NULL) {
10700		/* SBC-3 (no version claimed) */
10701		scsi_ulto2b(0x04C0, inq_ptr->version4);
10702	} else {
10703		switch (lun->be_lun->lun_type) {
10704		case T_DIRECT:
10705			/* SBC-3 (no version claimed) */
10706			scsi_ulto2b(0x04C0, inq_ptr->version4);
10707			break;
10708		case T_PROCESSOR:
10709		default:
10710			break;
10711		}
10712	}
10713
10714	ctsio->scsi_status = SCSI_STATUS_OK;
10715	if (ctsio->kern_data_len > 0) {
10716		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10717		ctsio->be_move_done = ctl_config_move_done;
10718		ctl_datamove((union ctl_io *)ctsio);
10719	} else {
10720		ctsio->io_hdr.status = CTL_SUCCESS;
10721		ctl_done((union ctl_io *)ctsio);
10722	}
10723
10724	return (CTL_RETVAL_COMPLETE);
10725}
10726
10727int
10728ctl_inquiry(struct ctl_scsiio *ctsio)
10729{
10730	struct scsi_inquiry *cdb;
10731	int retval;
10732
10733	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10734
10735	cdb = (struct scsi_inquiry *)ctsio->cdb;
10736	if (cdb->byte2 & SI_EVPD)
10737		retval = ctl_inquiry_evpd(ctsio);
10738	else if (cdb->page_code == 0)
10739		retval = ctl_inquiry_std(ctsio);
10740	else {
10741		ctl_set_invalid_field(ctsio,
10742				      /*sks_valid*/ 1,
10743				      /*command*/ 1,
10744				      /*field*/ 2,
10745				      /*bit_valid*/ 0,
10746				      /*bit*/ 0);
10747		ctl_done((union ctl_io *)ctsio);
10748		return (CTL_RETVAL_COMPLETE);
10749	}
10750
10751	return (retval);
10752}
10753
10754/*
10755 * For known CDB types, parse the LBA and length.
10756 */
10757static int
10758ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10759{
10760	if (io->io_hdr.io_type != CTL_IO_SCSI)
10761		return (1);
10762
10763	switch (io->scsiio.cdb[0]) {
10764	case COMPARE_AND_WRITE: {
10765		struct scsi_compare_and_write *cdb;
10766
10767		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10768
10769		*lba = scsi_8btou64(cdb->addr);
10770		*len = cdb->length;
10771		break;
10772	}
10773	case READ_6:
10774	case WRITE_6: {
10775		struct scsi_rw_6 *cdb;
10776
10777		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10778
10779		*lba = scsi_3btoul(cdb->addr);
10780		/* only 5 bits are valid in the most significant address byte */
10781		*lba &= 0x1fffff;
10782		*len = cdb->length;
10783		break;
10784	}
10785	case READ_10:
10786	case WRITE_10: {
10787		struct scsi_rw_10 *cdb;
10788
10789		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10790
10791		*lba = scsi_4btoul(cdb->addr);
10792		*len = scsi_2btoul(cdb->length);
10793		break;
10794	}
10795	case WRITE_VERIFY_10: {
10796		struct scsi_write_verify_10 *cdb;
10797
10798		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10799
10800		*lba = scsi_4btoul(cdb->addr);
10801		*len = scsi_2btoul(cdb->length);
10802		break;
10803	}
10804	case READ_12:
10805	case WRITE_12: {
10806		struct scsi_rw_12 *cdb;
10807
10808		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10809
10810		*lba = scsi_4btoul(cdb->addr);
10811		*len = scsi_4btoul(cdb->length);
10812		break;
10813	}
10814	case WRITE_VERIFY_12: {
10815		struct scsi_write_verify_12 *cdb;
10816
10817		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10818
10819		*lba = scsi_4btoul(cdb->addr);
10820		*len = scsi_4btoul(cdb->length);
10821		break;
10822	}
10823	case READ_16:
10824	case WRITE_16: {
10825		struct scsi_rw_16 *cdb;
10826
10827		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10828
10829		*lba = scsi_8btou64(cdb->addr);
10830		*len = scsi_4btoul(cdb->length);
10831		break;
10832	}
10833	case WRITE_VERIFY_16: {
10834		struct scsi_write_verify_16 *cdb;
10835
10836		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10837
10838
10839		*lba = scsi_8btou64(cdb->addr);
10840		*len = scsi_4btoul(cdb->length);
10841		break;
10842	}
10843	case WRITE_SAME_10: {
10844		struct scsi_write_same_10 *cdb;
10845
10846		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10847
10848		*lba = scsi_4btoul(cdb->addr);
10849		*len = scsi_2btoul(cdb->length);
10850		break;
10851	}
10852	case WRITE_SAME_16: {
10853		struct scsi_write_same_16 *cdb;
10854
10855		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10856
10857		*lba = scsi_8btou64(cdb->addr);
10858		*len = scsi_4btoul(cdb->length);
10859		break;
10860	}
10861	case VERIFY_10: {
10862		struct scsi_verify_10 *cdb;
10863
10864		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10865
10866		*lba = scsi_4btoul(cdb->addr);
10867		*len = scsi_2btoul(cdb->length);
10868		break;
10869	}
10870	case VERIFY_12: {
10871		struct scsi_verify_12 *cdb;
10872
10873		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10874
10875		*lba = scsi_4btoul(cdb->addr);
10876		*len = scsi_4btoul(cdb->length);
10877		break;
10878	}
10879	case VERIFY_16: {
10880		struct scsi_verify_16 *cdb;
10881
10882		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10883
10884		*lba = scsi_8btou64(cdb->addr);
10885		*len = scsi_4btoul(cdb->length);
10886		break;
10887	}
10888	case UNMAP: {
10889		*lba = 0;
10890		*len = UINT64_MAX;
10891		break;
10892	}
10893	default:
10894		return (1);
10895		break; /* NOTREACHED */
10896	}
10897
10898	return (0);
10899}
10900
10901static ctl_action
10902ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10903{
10904	uint64_t endlba1, endlba2;
10905
10906	endlba1 = lba1 + len1 - 1;
10907	endlba2 = lba2 + len2 - 1;
10908
10909	if ((endlba1 < lba2)
10910	 || (endlba2 < lba1))
10911		return (CTL_ACTION_PASS);
10912	else
10913		return (CTL_ACTION_BLOCK);
10914}
10915
10916static int
10917ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10918{
10919	struct ctl_ptr_len_flags *ptrlen;
10920	struct scsi_unmap_desc *buf, *end, *range;
10921	uint64_t lba;
10922	uint32_t len;
10923
10924	/* If not UNMAP -- go other way. */
10925	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10926	    io->scsiio.cdb[0] != UNMAP)
10927		return (CTL_ACTION_ERROR);
10928
10929	/* If UNMAP without data -- block and wait for data. */
10930	ptrlen = (struct ctl_ptr_len_flags *)
10931	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10932	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10933	    ptrlen->ptr == NULL)
10934		return (CTL_ACTION_BLOCK);
10935
10936	/* UNMAP with data -- check for collision. */
10937	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10938	end = buf + ptrlen->len / sizeof(*buf);
10939	for (range = buf; range < end; range++) {
10940		lba = scsi_8btou64(range->lba);
10941		len = scsi_4btoul(range->length);
10942		if ((lba < lba2 + len2) && (lba + len > lba2))
10943			return (CTL_ACTION_BLOCK);
10944	}
10945	return (CTL_ACTION_PASS);
10946}
10947
10948static ctl_action
10949ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10950{
10951	uint64_t lba1, lba2;
10952	uint64_t len1, len2;
10953	int retval;
10954
10955	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10956		return (CTL_ACTION_ERROR);
10957
10958	retval = ctl_extent_check_unmap(io2, lba1, len1);
10959	if (retval != CTL_ACTION_ERROR)
10960		return (retval);
10961
10962	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10963		return (CTL_ACTION_ERROR);
10964
10965	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10966}
10967
10968static ctl_action
10969ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10970    union ctl_io *ooa_io)
10971{
10972	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10973	ctl_serialize_action *serialize_row;
10974
10975	/*
10976	 * The initiator attempted multiple untagged commands at the same
10977	 * time.  Can't do that.
10978	 */
10979	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10980	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10981	 && ((pending_io->io_hdr.nexus.targ_port ==
10982	      ooa_io->io_hdr.nexus.targ_port)
10983	  && (pending_io->io_hdr.nexus.initid.id ==
10984	      ooa_io->io_hdr.nexus.initid.id))
10985	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10986		return (CTL_ACTION_OVERLAP);
10987
10988	/*
10989	 * The initiator attempted to send multiple tagged commands with
10990	 * the same ID.  (It's fine if different initiators have the same
10991	 * tag ID.)
10992	 *
10993	 * Even if all of those conditions are true, we don't kill the I/O
10994	 * if the command ahead of us has been aborted.  We won't end up
10995	 * sending it to the FETD, and it's perfectly legal to resend a
10996	 * command with the same tag number as long as the previous
10997	 * instance of this tag number has been aborted somehow.
10998	 */
10999	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11000	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11001	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11002	 && ((pending_io->io_hdr.nexus.targ_port ==
11003	      ooa_io->io_hdr.nexus.targ_port)
11004	  && (pending_io->io_hdr.nexus.initid.id ==
11005	      ooa_io->io_hdr.nexus.initid.id))
11006	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
11007		return (CTL_ACTION_OVERLAP_TAG);
11008
11009	/*
11010	 * If we get a head of queue tag, SAM-3 says that we should
11011	 * immediately execute it.
11012	 *
11013	 * What happens if this command would normally block for some other
11014	 * reason?  e.g. a request sense with a head of queue tag
11015	 * immediately after a write.  Normally that would block, but this
11016	 * will result in its getting executed immediately...
11017	 *
11018	 * We currently return "pass" instead of "skip", so we'll end up
11019	 * going through the rest of the queue to check for overlapped tags.
11020	 *
11021	 * XXX KDM check for other types of blockage first??
11022	 */
11023	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11024		return (CTL_ACTION_PASS);
11025
11026	/*
11027	 * Ordered tags have to block until all items ahead of them
11028	 * have completed.  If we get called with an ordered tag, we always
11029	 * block, if something else is ahead of us in the queue.
11030	 */
11031	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11032		return (CTL_ACTION_BLOCK);
11033
11034	/*
11035	 * Simple tags get blocked until all head of queue and ordered tags
11036	 * ahead of them have completed.  I'm lumping untagged commands in
11037	 * with simple tags here.  XXX KDM is that the right thing to do?
11038	 */
11039	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11040	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11041	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11042	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11043		return (CTL_ACTION_BLOCK);
11044
11045	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11046	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11047
11048	serialize_row = ctl_serialize_table[ooa_entry->seridx];
11049
11050	switch (serialize_row[pending_entry->seridx]) {
11051	case CTL_SER_BLOCK:
11052		return (CTL_ACTION_BLOCK);
11053	case CTL_SER_EXTENT:
11054		return (ctl_extent_check(pending_io, ooa_io));
11055	case CTL_SER_EXTENTOPT:
11056		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11057		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11058			return (ctl_extent_check(pending_io, ooa_io));
11059		/* FALLTHROUGH */
11060	case CTL_SER_PASS:
11061		return (CTL_ACTION_PASS);
11062	case CTL_SER_BLOCKOPT:
11063		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11064		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11065			return (CTL_ACTION_BLOCK);
11066		return (CTL_ACTION_PASS);
11067	case CTL_SER_SKIP:
11068		return (CTL_ACTION_SKIP);
11069	default:
11070		panic("invalid serialization value %d",
11071		      serialize_row[pending_entry->seridx]);
11072	}
11073
11074	return (CTL_ACTION_ERROR);
11075}
11076
11077/*
11078 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11079 * Assumptions:
11080 * - pending_io is generally either incoming, or on the blocked queue
11081 * - starting I/O is the I/O we want to start the check with.
11082 */
11083static ctl_action
11084ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11085	      union ctl_io *starting_io)
11086{
11087	union ctl_io *ooa_io;
11088	ctl_action action;
11089
11090	mtx_assert(&lun->lun_lock, MA_OWNED);
11091
11092	/*
11093	 * Run back along the OOA queue, starting with the current
11094	 * blocked I/O and going through every I/O before it on the
11095	 * queue.  If starting_io is NULL, we'll just end up returning
11096	 * CTL_ACTION_PASS.
11097	 */
11098	for (ooa_io = starting_io; ooa_io != NULL;
11099	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11100	     ooa_links)){
11101
11102		/*
11103		 * This routine just checks to see whether
11104		 * cur_blocked is blocked by ooa_io, which is ahead
11105		 * of it in the queue.  It doesn't queue/dequeue
11106		 * cur_blocked.
11107		 */
11108		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11109		switch (action) {
11110		case CTL_ACTION_BLOCK:
11111		case CTL_ACTION_OVERLAP:
11112		case CTL_ACTION_OVERLAP_TAG:
11113		case CTL_ACTION_SKIP:
11114		case CTL_ACTION_ERROR:
11115			return (action);
11116			break; /* NOTREACHED */
11117		case CTL_ACTION_PASS:
11118			break;
11119		default:
11120			panic("invalid action %d", action);
11121			break;  /* NOTREACHED */
11122		}
11123	}
11124
11125	return (CTL_ACTION_PASS);
11126}
11127
11128/*
11129 * Assumptions:
11130 * - An I/O has just completed, and has been removed from the per-LUN OOA
11131 *   queue, so some items on the blocked queue may now be unblocked.
11132 */
11133static int
11134ctl_check_blocked(struct ctl_lun *lun)
11135{
11136	union ctl_io *cur_blocked, *next_blocked;
11137
11138	mtx_assert(&lun->lun_lock, MA_OWNED);
11139
11140	/*
11141	 * Run forward from the head of the blocked queue, checking each
11142	 * entry against the I/Os prior to it on the OOA queue to see if
11143	 * there is still any blockage.
11144	 *
11145	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11146	 * with our removing a variable on it while it is traversing the
11147	 * list.
11148	 */
11149	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11150	     cur_blocked != NULL; cur_blocked = next_blocked) {
11151		union ctl_io *prev_ooa;
11152		ctl_action action;
11153
11154		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11155							  blocked_links);
11156
11157		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11158						      ctl_ooaq, ooa_links);
11159
11160		/*
11161		 * If cur_blocked happens to be the first item in the OOA
11162		 * queue now, prev_ooa will be NULL, and the action
11163		 * returned will just be CTL_ACTION_PASS.
11164		 */
11165		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11166
11167		switch (action) {
11168		case CTL_ACTION_BLOCK:
11169			/* Nothing to do here, still blocked */
11170			break;
11171		case CTL_ACTION_OVERLAP:
11172		case CTL_ACTION_OVERLAP_TAG:
11173			/*
11174			 * This shouldn't happen!  In theory we've already
11175			 * checked this command for overlap...
11176			 */
11177			break;
11178		case CTL_ACTION_PASS:
11179		case CTL_ACTION_SKIP: {
11180			struct ctl_softc *softc;
11181			const struct ctl_cmd_entry *entry;
11182			uint32_t initidx;
11183			int isc_retval;
11184
11185			/*
11186			 * The skip case shouldn't happen, this transaction
11187			 * should have never made it onto the blocked queue.
11188			 */
11189			/*
11190			 * This I/O is no longer blocked, we can remove it
11191			 * from the blocked queue.  Since this is a TAILQ
11192			 * (doubly linked list), we can do O(1) removals
11193			 * from any place on the list.
11194			 */
11195			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11196				     blocked_links);
11197			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11198
11199			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11200				/*
11201				 * Need to send IO back to original side to
11202				 * run
11203				 */
11204				union ctl_ha_msg msg_info;
11205
11206				msg_info.hdr.original_sc =
11207					cur_blocked->io_hdr.original_sc;
11208				msg_info.hdr.serializing_sc = cur_blocked;
11209				msg_info.hdr.msg_type = CTL_MSG_R2R;
11210				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11211				     &msg_info, sizeof(msg_info), 0)) >
11212				     CTL_HA_STATUS_SUCCESS) {
11213					printf("CTL:Check Blocked error from "
11214					       "ctl_ha_msg_send %d\n",
11215					       isc_retval);
11216				}
11217				break;
11218			}
11219			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11220			softc = control_softc;
11221
11222			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11223
11224			/*
11225			 * Check this I/O for LUN state changes that may
11226			 * have happened while this command was blocked.
11227			 * The LUN state may have been changed by a command
11228			 * ahead of us in the queue, so we need to re-check
11229			 * for any states that can be caused by SCSI
11230			 * commands.
11231			 */
11232			if (ctl_scsiio_lun_check(softc, lun, entry,
11233						 &cur_blocked->scsiio) == 0) {
11234				cur_blocked->io_hdr.flags |=
11235				                      CTL_FLAG_IS_WAS_ON_RTR;
11236				ctl_enqueue_rtr(cur_blocked);
11237			} else
11238				ctl_done(cur_blocked);
11239			break;
11240		}
11241		default:
11242			/*
11243			 * This probably shouldn't happen -- we shouldn't
11244			 * get CTL_ACTION_ERROR, or anything else.
11245			 */
11246			break;
11247		}
11248	}
11249
11250	return (CTL_RETVAL_COMPLETE);
11251}
11252
11253/*
11254 * This routine (with one exception) checks LUN flags that can be set by
11255 * commands ahead of us in the OOA queue.  These flags have to be checked
11256 * when a command initially comes in, and when we pull a command off the
11257 * blocked queue and are preparing to execute it.  The reason we have to
11258 * check these flags for commands on the blocked queue is that the LUN
11259 * state may have been changed by a command ahead of us while we're on the
11260 * blocked queue.
11261 *
11262 * Ordering is somewhat important with these checks, so please pay
11263 * careful attention to the placement of any new checks.
11264 */
11265static int
11266ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11267    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11268{
11269	int retval;
11270	uint32_t residx;
11271
11272	retval = 0;
11273
11274	mtx_assert(&lun->lun_lock, MA_OWNED);
11275
11276	/*
11277	 * If this shelf is a secondary shelf controller, we have to reject
11278	 * any media access commands.
11279	 */
11280#if 0
11281	/* No longer needed for HA */
11282	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11283	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11284		ctl_set_lun_standby(ctsio);
11285		retval = 1;
11286		goto bailout;
11287	}
11288#endif
11289
11290	/*
11291	 * Check for a reservation conflict.  If this command isn't allowed
11292	 * even on reserved LUNs, and if this initiator isn't the one who
11293	 * reserved us, reject the command with a reservation conflict.
11294	 */
11295	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11296	if ((lun->flags & CTL_LUN_RESERVED)
11297	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11298		if (lun->res_idx != residx) {
11299			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11300			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11301			retval = 1;
11302			goto bailout;
11303		}
11304	}
11305
11306	if ((lun->flags & CTL_LUN_PR_RESERVED)
11307	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11308		/*
11309		 * if we aren't registered or it's a res holder type
11310		 * reservation and this isn't the res holder then set a
11311		 * conflict.
11312		 * NOTE: Commands which might be allowed on write exclusive
11313		 * type reservations are checked in the particular command
11314		 * for a conflict. Read and SSU are the only ones.
11315		 */
11316		if (!lun->per_res[residx].registered
11317		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11318			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11319			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11320			retval = 1;
11321			goto bailout;
11322		}
11323
11324	}
11325
11326	if ((lun->flags & CTL_LUN_OFFLINE)
11327	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11328		ctl_set_lun_not_ready(ctsio);
11329		retval = 1;
11330		goto bailout;
11331	}
11332
11333	/*
11334	 * If the LUN is stopped, see if this particular command is allowed
11335	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11336	 */
11337	if ((lun->flags & CTL_LUN_STOPPED)
11338	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11339		/* "Logical unit not ready, initializing cmd. required" */
11340		ctl_set_lun_stopped(ctsio);
11341		retval = 1;
11342		goto bailout;
11343	}
11344
11345	if ((lun->flags & CTL_LUN_INOPERABLE)
11346	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11347		/* "Medium format corrupted" */
11348		ctl_set_medium_format_corrupted(ctsio);
11349		retval = 1;
11350		goto bailout;
11351	}
11352
11353bailout:
11354	return (retval);
11355
11356}
11357
11358static void
11359ctl_failover_io(union ctl_io *io, int have_lock)
11360{
11361	ctl_set_busy(&io->scsiio);
11362	ctl_done(io);
11363}
11364
11365static void
11366ctl_failover(void)
11367{
11368	struct ctl_lun *lun;
11369	struct ctl_softc *ctl_softc;
11370	union ctl_io *next_io, *pending_io;
11371	union ctl_io *io;
11372	int lun_idx;
11373	int i;
11374
11375	ctl_softc = control_softc;
11376
11377	mtx_lock(&ctl_softc->ctl_lock);
11378	/*
11379	 * Remove any cmds from the other SC from the rtr queue.  These
11380	 * will obviously only be for LUNs for which we're the primary.
11381	 * We can't send status or get/send data for these commands.
11382	 * Since they haven't been executed yet, we can just remove them.
11383	 * We'll either abort them or delete them below, depending on
11384	 * which HA mode we're in.
11385	 */
11386#ifdef notyet
11387	mtx_lock(&ctl_softc->queue_lock);
11388	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11389	     io != NULL; io = next_io) {
11390		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11391		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11392			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11393				      ctl_io_hdr, links);
11394	}
11395	mtx_unlock(&ctl_softc->queue_lock);
11396#endif
11397
11398	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11399		lun = ctl_softc->ctl_luns[lun_idx];
11400		if (lun==NULL)
11401			continue;
11402
11403		/*
11404		 * Processor LUNs are primary on both sides.
11405		 * XXX will this always be true?
11406		 */
11407		if (lun->be_lun->lun_type == T_PROCESSOR)
11408			continue;
11409
11410		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11411		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11412			printf("FAILOVER: primary lun %d\n", lun_idx);
11413		        /*
11414			 * Remove all commands from the other SC. First from the
11415			 * blocked queue then from the ooa queue. Once we have
11416			 * removed them. Call ctl_check_blocked to see if there
11417			 * is anything that can run.
11418			 */
11419			for (io = (union ctl_io *)TAILQ_FIRST(
11420			     &lun->blocked_queue); io != NULL; io = next_io) {
11421
11422		        	next_io = (union ctl_io *)TAILQ_NEXT(
11423				    &io->io_hdr, blocked_links);
11424
11425				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11426					TAILQ_REMOVE(&lun->blocked_queue,
11427						     &io->io_hdr,blocked_links);
11428					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11429					TAILQ_REMOVE(&lun->ooa_queue,
11430						     &io->io_hdr, ooa_links);
11431
11432					ctl_free_io(io);
11433				}
11434			}
11435
11436			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11437	     		     io != NULL; io = next_io) {
11438
11439		        	next_io = (union ctl_io *)TAILQ_NEXT(
11440				    &io->io_hdr, ooa_links);
11441
11442				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11443
11444					TAILQ_REMOVE(&lun->ooa_queue,
11445						&io->io_hdr,
11446					     	ooa_links);
11447
11448					ctl_free_io(io);
11449				}
11450			}
11451			ctl_check_blocked(lun);
11452		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11453			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11454
11455			printf("FAILOVER: primary lun %d\n", lun_idx);
11456			/*
11457			 * Abort all commands from the other SC.  We can't
11458			 * send status back for them now.  These should get
11459			 * cleaned up when they are completed or come out
11460			 * for a datamove operation.
11461			 */
11462			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11463	     		     io != NULL; io = next_io) {
11464		        	next_io = (union ctl_io *)TAILQ_NEXT(
11465					&io->io_hdr, ooa_links);
11466
11467				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11468					io->io_hdr.flags |= CTL_FLAG_ABORT;
11469			}
11470		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11471			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11472
11473			printf("FAILOVER: secondary lun %d\n", lun_idx);
11474
11475			lun->flags |= CTL_LUN_PRIMARY_SC;
11476
11477			/*
11478			 * We send all I/O that was sent to this controller
11479			 * and redirected to the other side back with
11480			 * busy status, and have the initiator retry it.
11481			 * Figuring out how much data has been transferred,
11482			 * etc. and picking up where we left off would be
11483			 * very tricky.
11484			 *
11485			 * XXX KDM need to remove I/O from the blocked
11486			 * queue as well!
11487			 */
11488			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11489			     &lun->ooa_queue); pending_io != NULL;
11490			     pending_io = next_io) {
11491
11492				next_io =  (union ctl_io *)TAILQ_NEXT(
11493					&pending_io->io_hdr, ooa_links);
11494
11495				pending_io->io_hdr.flags &=
11496					~CTL_FLAG_SENT_2OTHER_SC;
11497
11498				if (pending_io->io_hdr.flags &
11499				    CTL_FLAG_IO_ACTIVE) {
11500					pending_io->io_hdr.flags |=
11501						CTL_FLAG_FAILOVER;
11502				} else {
11503					ctl_set_busy(&pending_io->scsiio);
11504					ctl_done(pending_io);
11505				}
11506			}
11507
11508			/*
11509			 * Build Unit Attention
11510			 */
11511			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11512				lun->pending_ua[i] |=
11513				                     CTL_UA_ASYM_ACC_CHANGE;
11514			}
11515		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11516			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11517			printf("FAILOVER: secondary lun %d\n", lun_idx);
11518			/*
11519			 * if the first io on the OOA is not on the RtR queue
11520			 * add it.
11521			 */
11522			lun->flags |= CTL_LUN_PRIMARY_SC;
11523
11524			pending_io = (union ctl_io *)TAILQ_FIRST(
11525			    &lun->ooa_queue);
11526			if (pending_io==NULL) {
11527				printf("Nothing on OOA queue\n");
11528				continue;
11529			}
11530
11531			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11532			if ((pending_io->io_hdr.flags &
11533			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11534				pending_io->io_hdr.flags |=
11535				    CTL_FLAG_IS_WAS_ON_RTR;
11536				ctl_enqueue_rtr(pending_io);
11537			}
11538#if 0
11539			else
11540			{
11541				printf("Tag 0x%04x is running\n",
11542				      pending_io->scsiio.tag_num);
11543			}
11544#endif
11545
11546			next_io = (union ctl_io *)TAILQ_NEXT(
11547			    &pending_io->io_hdr, ooa_links);
11548			for (pending_io=next_io; pending_io != NULL;
11549			     pending_io = next_io) {
11550				pending_io->io_hdr.flags &=
11551				    ~CTL_FLAG_SENT_2OTHER_SC;
11552				next_io = (union ctl_io *)TAILQ_NEXT(
11553					&pending_io->io_hdr, ooa_links);
11554				if (pending_io->io_hdr.flags &
11555				    CTL_FLAG_IS_WAS_ON_RTR) {
11556#if 0
11557				        printf("Tag 0x%04x is running\n",
11558				      		pending_io->scsiio.tag_num);
11559#endif
11560					continue;
11561				}
11562
11563				switch (ctl_check_ooa(lun, pending_io,
11564			            (union ctl_io *)TAILQ_PREV(
11565				    &pending_io->io_hdr, ctl_ooaq,
11566				    ooa_links))) {
11567
11568				case CTL_ACTION_BLOCK:
11569					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11570							  &pending_io->io_hdr,
11571							  blocked_links);
11572					pending_io->io_hdr.flags |=
11573					    CTL_FLAG_BLOCKED;
11574					break;
11575				case CTL_ACTION_PASS:
11576				case CTL_ACTION_SKIP:
11577					pending_io->io_hdr.flags |=
11578					    CTL_FLAG_IS_WAS_ON_RTR;
11579					ctl_enqueue_rtr(pending_io);
11580					break;
11581				case CTL_ACTION_OVERLAP:
11582					ctl_set_overlapped_cmd(
11583					    (struct ctl_scsiio *)pending_io);
11584					ctl_done(pending_io);
11585					break;
11586				case CTL_ACTION_OVERLAP_TAG:
11587					ctl_set_overlapped_tag(
11588					    (struct ctl_scsiio *)pending_io,
11589					    pending_io->scsiio.tag_num & 0xff);
11590					ctl_done(pending_io);
11591					break;
11592				case CTL_ACTION_ERROR:
11593				default:
11594					ctl_set_internal_failure(
11595						(struct ctl_scsiio *)pending_io,
11596						0,  // sks_valid
11597						0); //retry count
11598					ctl_done(pending_io);
11599					break;
11600				}
11601			}
11602
11603			/*
11604			 * Build Unit Attention
11605			 */
11606			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11607				lun->pending_ua[i] |=
11608				                     CTL_UA_ASYM_ACC_CHANGE;
11609			}
11610		} else {
11611			panic("Unhandled HA mode failover, LUN flags = %#x, "
11612			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11613		}
11614	}
11615	ctl_pause_rtr = 0;
11616	mtx_unlock(&ctl_softc->ctl_lock);
11617}
11618
11619static int
11620ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11621{
11622	struct ctl_lun *lun;
11623	const struct ctl_cmd_entry *entry;
11624	uint32_t initidx, targ_lun;
11625	int retval;
11626
11627	retval = 0;
11628
11629	lun = NULL;
11630
11631	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11632	if ((targ_lun < CTL_MAX_LUNS)
11633	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11634		lun = ctl_softc->ctl_luns[targ_lun];
11635		/*
11636		 * If the LUN is invalid, pretend that it doesn't exist.
11637		 * It will go away as soon as all pending I/O has been
11638		 * completed.
11639		 */
11640		if (lun->flags & CTL_LUN_DISABLED) {
11641			lun = NULL;
11642		} else {
11643			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11644			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11645				lun->be_lun;
11646			if (lun->be_lun->lun_type == T_PROCESSOR) {
11647				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11648			}
11649
11650			/*
11651			 * Every I/O goes into the OOA queue for a
11652			 * particular LUN, and stays there until completion.
11653			 */
11654			mtx_lock(&lun->lun_lock);
11655			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11656			    ooa_links);
11657		}
11658	} else {
11659		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11660		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11661	}
11662
11663	/* Get command entry and return error if it is unsuppotyed. */
11664	entry = ctl_validate_command(ctsio);
11665	if (entry == NULL) {
11666		if (lun)
11667			mtx_unlock(&lun->lun_lock);
11668		return (retval);
11669	}
11670
11671	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11672	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11673
11674	/*
11675	 * Check to see whether we can send this command to LUNs that don't
11676	 * exist.  This should pretty much only be the case for inquiry
11677	 * and request sense.  Further checks, below, really require having
11678	 * a LUN, so we can't really check the command anymore.  Just put
11679	 * it on the rtr queue.
11680	 */
11681	if (lun == NULL) {
11682		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11683			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11684			ctl_enqueue_rtr((union ctl_io *)ctsio);
11685			return (retval);
11686		}
11687
11688		ctl_set_unsupported_lun(ctsio);
11689		ctl_done((union ctl_io *)ctsio);
11690		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11691		return (retval);
11692	} else {
11693		/*
11694		 * Make sure we support this particular command on this LUN.
11695		 * e.g., we don't support writes to the control LUN.
11696		 */
11697		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11698			mtx_unlock(&lun->lun_lock);
11699			ctl_set_invalid_opcode(ctsio);
11700			ctl_done((union ctl_io *)ctsio);
11701			return (retval);
11702		}
11703	}
11704
11705	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11706
11707#ifdef CTL_WITH_CA
11708	/*
11709	 * If we've got a request sense, it'll clear the contingent
11710	 * allegiance condition.  Otherwise, if we have a CA condition for
11711	 * this initiator, clear it, because it sent down a command other
11712	 * than request sense.
11713	 */
11714	if ((ctsio->cdb[0] != REQUEST_SENSE)
11715	 && (ctl_is_set(lun->have_ca, initidx)))
11716		ctl_clear_mask(lun->have_ca, initidx);
11717#endif
11718
11719	/*
11720	 * If the command has this flag set, it handles its own unit
11721	 * attention reporting, we shouldn't do anything.  Otherwise we
11722	 * check for any pending unit attentions, and send them back to the
11723	 * initiator.  We only do this when a command initially comes in,
11724	 * not when we pull it off the blocked queue.
11725	 *
11726	 * According to SAM-3, section 5.3.2, the order that things get
11727	 * presented back to the host is basically unit attentions caused
11728	 * by some sort of reset event, busy status, reservation conflicts
11729	 * or task set full, and finally any other status.
11730	 *
11731	 * One issue here is that some of the unit attentions we report
11732	 * don't fall into the "reset" category (e.g. "reported luns data
11733	 * has changed").  So reporting it here, before the reservation
11734	 * check, may be technically wrong.  I guess the only thing to do
11735	 * would be to check for and report the reset events here, and then
11736	 * check for the other unit attention types after we check for a
11737	 * reservation conflict.
11738	 *
11739	 * XXX KDM need to fix this
11740	 */
11741	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11742		ctl_ua_type ua_type;
11743
11744		ua_type = lun->pending_ua[initidx];
11745		if (ua_type != CTL_UA_NONE) {
11746			scsi_sense_data_type sense_format;
11747
11748			if (lun != NULL)
11749				sense_format = (lun->flags &
11750				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11751				    SSD_TYPE_FIXED;
11752			else
11753				sense_format = SSD_TYPE_FIXED;
11754
11755			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11756					       sense_format);
11757			if (ua_type != CTL_UA_NONE) {
11758				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11759				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11760						       CTL_AUTOSENSE;
11761				ctsio->sense_len = SSD_FULL_SIZE;
11762				lun->pending_ua[initidx] &= ~ua_type;
11763				mtx_unlock(&lun->lun_lock);
11764				ctl_done((union ctl_io *)ctsio);
11765				return (retval);
11766			}
11767		}
11768	}
11769
11770
11771	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11772		mtx_unlock(&lun->lun_lock);
11773		ctl_done((union ctl_io *)ctsio);
11774		return (retval);
11775	}
11776
11777	/*
11778	 * XXX CHD this is where we want to send IO to other side if
11779	 * this LUN is secondary on this SC. We will need to make a copy
11780	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11781	 * the copy we send as FROM_OTHER.
11782	 * We also need to stuff the address of the original IO so we can
11783	 * find it easily. Something similar will need be done on the other
11784	 * side so when we are done we can find the copy.
11785	 */
11786	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11787		union ctl_ha_msg msg_info;
11788		int isc_retval;
11789
11790		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11791
11792		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11793		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11794#if 0
11795		printf("1. ctsio %p\n", ctsio);
11796#endif
11797		msg_info.hdr.serializing_sc = NULL;
11798		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11799		msg_info.scsi.tag_num = ctsio->tag_num;
11800		msg_info.scsi.tag_type = ctsio->tag_type;
11801		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11802
11803		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11804
11805		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11806		    (void *)&msg_info, sizeof(msg_info), 0)) >
11807		    CTL_HA_STATUS_SUCCESS) {
11808			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11809			       isc_retval);
11810			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11811		} else {
11812#if 0
11813			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11814#endif
11815		}
11816
11817		/*
11818		 * XXX KDM this I/O is off the incoming queue, but hasn't
11819		 * been inserted on any other queue.  We may need to come
11820		 * up with a holding queue while we wait for serialization
11821		 * so that we have an idea of what we're waiting for from
11822		 * the other side.
11823		 */
11824		mtx_unlock(&lun->lun_lock);
11825		return (retval);
11826	}
11827
11828	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11829			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11830			      ctl_ooaq, ooa_links))) {
11831	case CTL_ACTION_BLOCK:
11832		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11833		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11834				  blocked_links);
11835		mtx_unlock(&lun->lun_lock);
11836		return (retval);
11837	case CTL_ACTION_PASS:
11838	case CTL_ACTION_SKIP:
11839		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11840		mtx_unlock(&lun->lun_lock);
11841		ctl_enqueue_rtr((union ctl_io *)ctsio);
11842		break;
11843	case CTL_ACTION_OVERLAP:
11844		mtx_unlock(&lun->lun_lock);
11845		ctl_set_overlapped_cmd(ctsio);
11846		ctl_done((union ctl_io *)ctsio);
11847		break;
11848	case CTL_ACTION_OVERLAP_TAG:
11849		mtx_unlock(&lun->lun_lock);
11850		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11851		ctl_done((union ctl_io *)ctsio);
11852		break;
11853	case CTL_ACTION_ERROR:
11854	default:
11855		mtx_unlock(&lun->lun_lock);
11856		ctl_set_internal_failure(ctsio,
11857					 /*sks_valid*/ 0,
11858					 /*retry_count*/ 0);
11859		ctl_done((union ctl_io *)ctsio);
11860		break;
11861	}
11862	return (retval);
11863}
11864
11865const struct ctl_cmd_entry *
11866ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11867{
11868	const struct ctl_cmd_entry *entry;
11869	int service_action;
11870
11871	entry = &ctl_cmd_table[ctsio->cdb[0]];
11872	if (sa)
11873		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11874	if (entry->flags & CTL_CMD_FLAG_SA5) {
11875		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11876		entry = &((const struct ctl_cmd_entry *)
11877		    entry->execute)[service_action];
11878	}
11879	return (entry);
11880}
11881
11882const struct ctl_cmd_entry *
11883ctl_validate_command(struct ctl_scsiio *ctsio)
11884{
11885	const struct ctl_cmd_entry *entry;
11886	int i, sa;
11887	uint8_t diff;
11888
11889	entry = ctl_get_cmd_entry(ctsio, &sa);
11890	if (entry->execute == NULL) {
11891		if (sa)
11892			ctl_set_invalid_field(ctsio,
11893					      /*sks_valid*/ 1,
11894					      /*command*/ 1,
11895					      /*field*/ 1,
11896					      /*bit_valid*/ 1,
11897					      /*bit*/ 4);
11898		else
11899			ctl_set_invalid_opcode(ctsio);
11900		ctl_done((union ctl_io *)ctsio);
11901		return (NULL);
11902	}
11903	KASSERT(entry->length > 0,
11904	    ("Not defined length for command 0x%02x/0x%02x",
11905	     ctsio->cdb[0], ctsio->cdb[1]));
11906	for (i = 1; i < entry->length; i++) {
11907		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11908		if (diff == 0)
11909			continue;
11910		ctl_set_invalid_field(ctsio,
11911				      /*sks_valid*/ 1,
11912				      /*command*/ 1,
11913				      /*field*/ i,
11914				      /*bit_valid*/ 1,
11915				      /*bit*/ fls(diff) - 1);
11916		ctl_done((union ctl_io *)ctsio);
11917		return (NULL);
11918	}
11919	return (entry);
11920}
11921
11922static int
11923ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11924{
11925
11926	switch (lun_type) {
11927	case T_PROCESSOR:
11928		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11929		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11930			return (0);
11931		break;
11932	case T_DIRECT:
11933		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11934		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11935			return (0);
11936		break;
11937	default:
11938		return (0);
11939	}
11940	return (1);
11941}
11942
11943static int
11944ctl_scsiio(struct ctl_scsiio *ctsio)
11945{
11946	int retval;
11947	const struct ctl_cmd_entry *entry;
11948
11949	retval = CTL_RETVAL_COMPLETE;
11950
11951	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11952
11953	entry = ctl_get_cmd_entry(ctsio, NULL);
11954
11955	/*
11956	 * If this I/O has been aborted, just send it straight to
11957	 * ctl_done() without executing it.
11958	 */
11959	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11960		ctl_done((union ctl_io *)ctsio);
11961		goto bailout;
11962	}
11963
11964	/*
11965	 * All the checks should have been handled by ctl_scsiio_precheck().
11966	 * We should be clear now to just execute the I/O.
11967	 */
11968	retval = entry->execute(ctsio);
11969
11970bailout:
11971	return (retval);
11972}
11973
11974/*
11975 * Since we only implement one target right now, a bus reset simply resets
11976 * our single target.
11977 */
11978static int
11979ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11980{
11981	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11982}
11983
11984static int
11985ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11986		 ctl_ua_type ua_type)
11987{
11988	struct ctl_lun *lun;
11989	int retval;
11990
11991	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11992		union ctl_ha_msg msg_info;
11993
11994		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11995		msg_info.hdr.nexus = io->io_hdr.nexus;
11996		if (ua_type==CTL_UA_TARG_RESET)
11997			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11998		else
11999			msg_info.task.task_action = CTL_TASK_BUS_RESET;
12000		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12001		msg_info.hdr.original_sc = NULL;
12002		msg_info.hdr.serializing_sc = NULL;
12003		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12004		    (void *)&msg_info, sizeof(msg_info), 0)) {
12005		}
12006	}
12007	retval = 0;
12008
12009	mtx_lock(&ctl_softc->ctl_lock);
12010	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
12011		retval += ctl_lun_reset(lun, io, ua_type);
12012	mtx_unlock(&ctl_softc->ctl_lock);
12013
12014	return (retval);
12015}
12016
12017/*
12018 * The LUN should always be set.  The I/O is optional, and is used to
12019 * distinguish between I/Os sent by this initiator, and by other
12020 * initiators.  We set unit attention for initiators other than this one.
12021 * SAM-3 is vague on this point.  It does say that a unit attention should
12022 * be established for other initiators when a LUN is reset (see section
12023 * 5.7.3), but it doesn't specifically say that the unit attention should
12024 * be established for this particular initiator when a LUN is reset.  Here
12025 * is the relevant text, from SAM-3 rev 8:
12026 *
12027 * 5.7.2 When a SCSI initiator port aborts its own tasks
12028 *
12029 * When a SCSI initiator port causes its own task(s) to be aborted, no
12030 * notification that the task(s) have been aborted shall be returned to
12031 * the SCSI initiator port other than the completion response for the
12032 * command or task management function action that caused the task(s) to
12033 * be aborted and notification(s) associated with related effects of the
12034 * action (e.g., a reset unit attention condition).
12035 *
12036 * XXX KDM for now, we're setting unit attention for all initiators.
12037 */
12038static int
12039ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
12040{
12041	union ctl_io *xio;
12042#if 0
12043	uint32_t initindex;
12044#endif
12045	int i;
12046
12047	mtx_lock(&lun->lun_lock);
12048	/*
12049	 * Run through the OOA queue and abort each I/O.
12050	 */
12051#if 0
12052	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12053#endif
12054	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12055	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12056		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
12057	}
12058
12059	/*
12060	 * This version sets unit attention for every
12061	 */
12062#if 0
12063	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12064	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12065		if (initindex == i)
12066			continue;
12067		lun->pending_ua[i] |= ua_type;
12068	}
12069#endif
12070
12071	/*
12072	 * A reset (any kind, really) clears reservations established with
12073	 * RESERVE/RELEASE.  It does not clear reservations established
12074	 * with PERSISTENT RESERVE OUT, but we don't support that at the
12075	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
12076	 * reservations made with the RESERVE/RELEASE commands, because
12077	 * those commands are obsolete in SPC-3.
12078	 */
12079	lun->flags &= ~CTL_LUN_RESERVED;
12080
12081	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12082#ifdef CTL_WITH_CA
12083		ctl_clear_mask(lun->have_ca, i);
12084#endif
12085		lun->pending_ua[i] |= ua_type;
12086	}
12087	mtx_unlock(&lun->lun_lock);
12088
12089	return (0);
12090}
12091
12092static void
12093ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
12094    int other_sc)
12095{
12096	union ctl_io *xio;
12097
12098	mtx_assert(&lun->lun_lock, MA_OWNED);
12099
12100	/*
12101	 * Run through the OOA queue and attempt to find the given I/O.
12102	 * The target port, initiator ID, tag type and tag number have to
12103	 * match the values that we got from the initiator.  If we have an
12104	 * untagged command to abort, simply abort the first untagged command
12105	 * we come to.  We only allow one untagged command at a time of course.
12106	 */
12107	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12108	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12109
12110		if ((targ_port == UINT32_MAX ||
12111		     targ_port == xio->io_hdr.nexus.targ_port) &&
12112		    (init_id == UINT32_MAX ||
12113		     init_id == xio->io_hdr.nexus.initid.id)) {
12114			if (targ_port != xio->io_hdr.nexus.targ_port ||
12115			    init_id != xio->io_hdr.nexus.initid.id)
12116				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12117			xio->io_hdr.flags |= CTL_FLAG_ABORT;
12118			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12119				union ctl_ha_msg msg_info;
12120
12121				msg_info.hdr.nexus = xio->io_hdr.nexus;
12122				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12123				msg_info.task.tag_num = xio->scsiio.tag_num;
12124				msg_info.task.tag_type = xio->scsiio.tag_type;
12125				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12126				msg_info.hdr.original_sc = NULL;
12127				msg_info.hdr.serializing_sc = NULL;
12128				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12129				    (void *)&msg_info, sizeof(msg_info), 0);
12130			}
12131		}
12132	}
12133}
12134
12135static int
12136ctl_abort_task_set(union ctl_io *io)
12137{
12138	struct ctl_softc *softc = control_softc;
12139	struct ctl_lun *lun;
12140	uint32_t targ_lun;
12141
12142	/*
12143	 * Look up the LUN.
12144	 */
12145	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12146	mtx_lock(&softc->ctl_lock);
12147	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12148		lun = softc->ctl_luns[targ_lun];
12149	else {
12150		mtx_unlock(&softc->ctl_lock);
12151		return (1);
12152	}
12153
12154	mtx_lock(&lun->lun_lock);
12155	mtx_unlock(&softc->ctl_lock);
12156	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12157		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12158		    io->io_hdr.nexus.initid.id,
12159		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12160	} else { /* CTL_TASK_CLEAR_TASK_SET */
12161		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12162		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12163	}
12164	mtx_unlock(&lun->lun_lock);
12165	return (0);
12166}
12167
12168static int
12169ctl_i_t_nexus_reset(union ctl_io *io)
12170{
12171	struct ctl_softc *softc = control_softc;
12172	struct ctl_lun *lun;
12173	uint32_t initindex, residx;
12174
12175	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12176	residx = ctl_get_resindex(&io->io_hdr.nexus);
12177	mtx_lock(&softc->ctl_lock);
12178	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12179		mtx_lock(&lun->lun_lock);
12180		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12181		    io->io_hdr.nexus.initid.id,
12182		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12183#ifdef CTL_WITH_CA
12184		ctl_clear_mask(lun->have_ca, initindex);
12185#endif
12186		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
12187			lun->flags &= ~CTL_LUN_RESERVED;
12188		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12189		mtx_unlock(&lun->lun_lock);
12190	}
12191	mtx_unlock(&softc->ctl_lock);
12192	return (0);
12193}
12194
12195static int
12196ctl_abort_task(union ctl_io *io)
12197{
12198	union ctl_io *xio;
12199	struct ctl_lun *lun;
12200	struct ctl_softc *ctl_softc;
12201#if 0
12202	struct sbuf sb;
12203	char printbuf[128];
12204#endif
12205	int found;
12206	uint32_t targ_lun;
12207
12208	ctl_softc = control_softc;
12209	found = 0;
12210
12211	/*
12212	 * Look up the LUN.
12213	 */
12214	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12215	mtx_lock(&ctl_softc->ctl_lock);
12216	if ((targ_lun < CTL_MAX_LUNS)
12217	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12218		lun = ctl_softc->ctl_luns[targ_lun];
12219	else {
12220		mtx_unlock(&ctl_softc->ctl_lock);
12221		return (1);
12222	}
12223
12224#if 0
12225	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12226	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12227#endif
12228
12229	mtx_lock(&lun->lun_lock);
12230	mtx_unlock(&ctl_softc->ctl_lock);
12231	/*
12232	 * Run through the OOA queue and attempt to find the given I/O.
12233	 * The target port, initiator ID, tag type and tag number have to
12234	 * match the values that we got from the initiator.  If we have an
12235	 * untagged command to abort, simply abort the first untagged command
12236	 * we come to.  We only allow one untagged command at a time of course.
12237	 */
12238#if 0
12239	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12240#endif
12241	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12242	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12243#if 0
12244		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12245
12246		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12247			    lun->lun, xio->scsiio.tag_num,
12248			    xio->scsiio.tag_type,
12249			    (xio->io_hdr.blocked_links.tqe_prev
12250			    == NULL) ? "" : " BLOCKED",
12251			    (xio->io_hdr.flags &
12252			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12253			    (xio->io_hdr.flags &
12254			    CTL_FLAG_ABORT) ? " ABORT" : "",
12255			    (xio->io_hdr.flags &
12256			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12257		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12258		sbuf_finish(&sb);
12259		printf("%s\n", sbuf_data(&sb));
12260#endif
12261
12262		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12263		 && (xio->io_hdr.nexus.initid.id ==
12264		     io->io_hdr.nexus.initid.id)) {
12265			/*
12266			 * If the abort says that the task is untagged, the
12267			 * task in the queue must be untagged.  Otherwise,
12268			 * we just check to see whether the tag numbers
12269			 * match.  This is because the QLogic firmware
12270			 * doesn't pass back the tag type in an abort
12271			 * request.
12272			 */
12273#if 0
12274			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12275			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12276			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12277#endif
12278			/*
12279			 * XXX KDM we've got problems with FC, because it
12280			 * doesn't send down a tag type with aborts.  So we
12281			 * can only really go by the tag number...
12282			 * This may cause problems with parallel SCSI.
12283			 * Need to figure that out!!
12284			 */
12285			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12286				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12287				found = 1;
12288				if ((io->io_hdr.flags &
12289				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12290				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12291					union ctl_ha_msg msg_info;
12292
12293					io->io_hdr.flags |=
12294					                CTL_FLAG_SENT_2OTHER_SC;
12295					msg_info.hdr.nexus = io->io_hdr.nexus;
12296					msg_info.task.task_action =
12297						CTL_TASK_ABORT_TASK;
12298					msg_info.task.tag_num =
12299						io->taskio.tag_num;
12300					msg_info.task.tag_type =
12301						io->taskio.tag_type;
12302					msg_info.hdr.msg_type =
12303						CTL_MSG_MANAGE_TASKS;
12304					msg_info.hdr.original_sc = NULL;
12305					msg_info.hdr.serializing_sc = NULL;
12306#if 0
12307					printf("Sent Abort to other side\n");
12308#endif
12309					if (CTL_HA_STATUS_SUCCESS !=
12310					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12311		    				(void *)&msg_info,
12312						sizeof(msg_info), 0)) {
12313					}
12314				}
12315#if 0
12316				printf("ctl_abort_task: found I/O to abort\n");
12317#endif
12318				break;
12319			}
12320		}
12321	}
12322	mtx_unlock(&lun->lun_lock);
12323
12324	if (found == 0) {
12325		/*
12326		 * This isn't really an error.  It's entirely possible for
12327		 * the abort and command completion to cross on the wire.
12328		 * This is more of an informative/diagnostic error.
12329		 */
12330#if 0
12331		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12332		       "%d:%d:%d:%d tag %d type %d\n",
12333		       io->io_hdr.nexus.initid.id,
12334		       io->io_hdr.nexus.targ_port,
12335		       io->io_hdr.nexus.targ_target.id,
12336		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12337		       io->taskio.tag_type);
12338#endif
12339	}
12340	return (0);
12341}
12342
12343static void
12344ctl_run_task(union ctl_io *io)
12345{
12346	struct ctl_softc *ctl_softc = control_softc;
12347	int retval = 1;
12348	const char *task_desc;
12349
12350	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12351
12352	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12353	    ("ctl_run_task: Unextected io_type %d\n",
12354	     io->io_hdr.io_type));
12355
12356	task_desc = ctl_scsi_task_string(&io->taskio);
12357	if (task_desc != NULL) {
12358#ifdef NEEDTOPORT
12359		csevent_log(CSC_CTL | CSC_SHELF_SW |
12360			    CTL_TASK_REPORT,
12361			    csevent_LogType_Trace,
12362			    csevent_Severity_Information,
12363			    csevent_AlertLevel_Green,
12364			    csevent_FRU_Firmware,
12365			    csevent_FRU_Unknown,
12366			    "CTL: received task: %s",task_desc);
12367#endif
12368	} else {
12369#ifdef NEEDTOPORT
12370		csevent_log(CSC_CTL | CSC_SHELF_SW |
12371			    CTL_TASK_REPORT,
12372			    csevent_LogType_Trace,
12373			    csevent_Severity_Information,
12374			    csevent_AlertLevel_Green,
12375			    csevent_FRU_Firmware,
12376			    csevent_FRU_Unknown,
12377			    "CTL: received unknown task "
12378			    "type: %d (%#x)",
12379			    io->taskio.task_action,
12380			    io->taskio.task_action);
12381#endif
12382	}
12383	switch (io->taskio.task_action) {
12384	case CTL_TASK_ABORT_TASK:
12385		retval = ctl_abort_task(io);
12386		break;
12387	case CTL_TASK_ABORT_TASK_SET:
12388	case CTL_TASK_CLEAR_TASK_SET:
12389		retval = ctl_abort_task_set(io);
12390		break;
12391	case CTL_TASK_CLEAR_ACA:
12392		break;
12393	case CTL_TASK_I_T_NEXUS_RESET:
12394		retval = ctl_i_t_nexus_reset(io);
12395		break;
12396	case CTL_TASK_LUN_RESET: {
12397		struct ctl_lun *lun;
12398		uint32_t targ_lun;
12399
12400		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12401		mtx_lock(&ctl_softc->ctl_lock);
12402		if ((targ_lun < CTL_MAX_LUNS)
12403		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12404			lun = ctl_softc->ctl_luns[targ_lun];
12405		else {
12406			mtx_unlock(&ctl_softc->ctl_lock);
12407			retval = 1;
12408			break;
12409		}
12410
12411		if (!(io->io_hdr.flags &
12412		    CTL_FLAG_FROM_OTHER_SC)) {
12413			union ctl_ha_msg msg_info;
12414
12415			io->io_hdr.flags |=
12416				CTL_FLAG_SENT_2OTHER_SC;
12417			msg_info.hdr.msg_type =
12418				CTL_MSG_MANAGE_TASKS;
12419			msg_info.hdr.nexus = io->io_hdr.nexus;
12420			msg_info.task.task_action =
12421				CTL_TASK_LUN_RESET;
12422			msg_info.hdr.original_sc = NULL;
12423			msg_info.hdr.serializing_sc = NULL;
12424			if (CTL_HA_STATUS_SUCCESS !=
12425			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12426			    (void *)&msg_info,
12427			    sizeof(msg_info), 0)) {
12428			}
12429		}
12430
12431		retval = ctl_lun_reset(lun, io,
12432				       CTL_UA_LUN_RESET);
12433		mtx_unlock(&ctl_softc->ctl_lock);
12434		break;
12435	}
12436	case CTL_TASK_TARGET_RESET:
12437		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12438		break;
12439	case CTL_TASK_BUS_RESET:
12440		retval = ctl_bus_reset(ctl_softc, io);
12441		break;
12442	case CTL_TASK_PORT_LOGIN:
12443		break;
12444	case CTL_TASK_PORT_LOGOUT:
12445		break;
12446	default:
12447		printf("ctl_run_task: got unknown task management event %d\n",
12448		       io->taskio.task_action);
12449		break;
12450	}
12451	if (retval == 0)
12452		io->io_hdr.status = CTL_SUCCESS;
12453	else
12454		io->io_hdr.status = CTL_ERROR;
12455	ctl_done(io);
12456}
12457
12458/*
12459 * For HA operation.  Handle commands that come in from the other
12460 * controller.
12461 */
12462static void
12463ctl_handle_isc(union ctl_io *io)
12464{
12465	int free_io;
12466	struct ctl_lun *lun;
12467	struct ctl_softc *ctl_softc;
12468	uint32_t targ_lun;
12469
12470	ctl_softc = control_softc;
12471
12472	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12473	lun = ctl_softc->ctl_luns[targ_lun];
12474
12475	switch (io->io_hdr.msg_type) {
12476	case CTL_MSG_SERIALIZE:
12477		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12478		break;
12479	case CTL_MSG_R2R: {
12480		const struct ctl_cmd_entry *entry;
12481
12482		/*
12483		 * This is only used in SER_ONLY mode.
12484		 */
12485		free_io = 0;
12486		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12487		mtx_lock(&lun->lun_lock);
12488		if (ctl_scsiio_lun_check(ctl_softc, lun,
12489		    entry, (struct ctl_scsiio *)io) != 0) {
12490			mtx_unlock(&lun->lun_lock);
12491			ctl_done(io);
12492			break;
12493		}
12494		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12495		mtx_unlock(&lun->lun_lock);
12496		ctl_enqueue_rtr(io);
12497		break;
12498	}
12499	case CTL_MSG_FINISH_IO:
12500		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12501			free_io = 0;
12502			ctl_done(io);
12503		} else {
12504			free_io = 1;
12505			mtx_lock(&lun->lun_lock);
12506			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12507				     ooa_links);
12508			ctl_check_blocked(lun);
12509			mtx_unlock(&lun->lun_lock);
12510		}
12511		break;
12512	case CTL_MSG_PERS_ACTION:
12513		ctl_hndl_per_res_out_on_other_sc(
12514			(union ctl_ha_msg *)&io->presio.pr_msg);
12515		free_io = 1;
12516		break;
12517	case CTL_MSG_BAD_JUJU:
12518		free_io = 0;
12519		ctl_done(io);
12520		break;
12521	case CTL_MSG_DATAMOVE:
12522		/* Only used in XFER mode */
12523		free_io = 0;
12524		ctl_datamove_remote(io);
12525		break;
12526	case CTL_MSG_DATAMOVE_DONE:
12527		/* Only used in XFER mode */
12528		free_io = 0;
12529		io->scsiio.be_move_done(io);
12530		break;
12531	default:
12532		free_io = 1;
12533		printf("%s: Invalid message type %d\n",
12534		       __func__, io->io_hdr.msg_type);
12535		break;
12536	}
12537	if (free_io)
12538		ctl_free_io(io);
12539
12540}
12541
12542
12543/*
12544 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12545 * there is no match.
12546 */
12547static ctl_lun_error_pattern
12548ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12549{
12550	const struct ctl_cmd_entry *entry;
12551	ctl_lun_error_pattern filtered_pattern, pattern;
12552
12553	pattern = desc->error_pattern;
12554
12555	/*
12556	 * XXX KDM we need more data passed into this function to match a
12557	 * custom pattern, and we actually need to implement custom pattern
12558	 * matching.
12559	 */
12560	if (pattern & CTL_LUN_PAT_CMD)
12561		return (CTL_LUN_PAT_CMD);
12562
12563	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12564		return (CTL_LUN_PAT_ANY);
12565
12566	entry = ctl_get_cmd_entry(ctsio, NULL);
12567
12568	filtered_pattern = entry->pattern & pattern;
12569
12570	/*
12571	 * If the user requested specific flags in the pattern (e.g.
12572	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12573	 * flags.
12574	 *
12575	 * If the user did not specify any flags, it doesn't matter whether
12576	 * or not the command supports the flags.
12577	 */
12578	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12579	     (pattern & ~CTL_LUN_PAT_MASK))
12580		return (CTL_LUN_PAT_NONE);
12581
12582	/*
12583	 * If the user asked for a range check, see if the requested LBA
12584	 * range overlaps with this command's LBA range.
12585	 */
12586	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12587		uint64_t lba1;
12588		uint64_t len1;
12589		ctl_action action;
12590		int retval;
12591
12592		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12593		if (retval != 0)
12594			return (CTL_LUN_PAT_NONE);
12595
12596		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12597					      desc->lba_range.len);
12598		/*
12599		 * A "pass" means that the LBA ranges don't overlap, so
12600		 * this doesn't match the user's range criteria.
12601		 */
12602		if (action == CTL_ACTION_PASS)
12603			return (CTL_LUN_PAT_NONE);
12604	}
12605
12606	return (filtered_pattern);
12607}
12608
12609static void
12610ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12611{
12612	struct ctl_error_desc *desc, *desc2;
12613
12614	mtx_assert(&lun->lun_lock, MA_OWNED);
12615
12616	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12617		ctl_lun_error_pattern pattern;
12618		/*
12619		 * Check to see whether this particular command matches
12620		 * the pattern in the descriptor.
12621		 */
12622		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12623		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12624			continue;
12625
12626		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12627		case CTL_LUN_INJ_ABORTED:
12628			ctl_set_aborted(&io->scsiio);
12629			break;
12630		case CTL_LUN_INJ_MEDIUM_ERR:
12631			ctl_set_medium_error(&io->scsiio);
12632			break;
12633		case CTL_LUN_INJ_UA:
12634			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12635			 * OCCURRED */
12636			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12637			break;
12638		case CTL_LUN_INJ_CUSTOM:
12639			/*
12640			 * We're assuming the user knows what he is doing.
12641			 * Just copy the sense information without doing
12642			 * checks.
12643			 */
12644			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12645			      ctl_min(sizeof(desc->custom_sense),
12646				      sizeof(io->scsiio.sense_data)));
12647			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12648			io->scsiio.sense_len = SSD_FULL_SIZE;
12649			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12650			break;
12651		case CTL_LUN_INJ_NONE:
12652		default:
12653			/*
12654			 * If this is an error injection type we don't know
12655			 * about, clear the continuous flag (if it is set)
12656			 * so it will get deleted below.
12657			 */
12658			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12659			break;
12660		}
12661		/*
12662		 * By default, each error injection action is a one-shot
12663		 */
12664		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12665			continue;
12666
12667		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12668
12669		free(desc, M_CTL);
12670	}
12671}
12672
12673#ifdef CTL_IO_DELAY
12674static void
12675ctl_datamove_timer_wakeup(void *arg)
12676{
12677	union ctl_io *io;
12678
12679	io = (union ctl_io *)arg;
12680
12681	ctl_datamove(io);
12682}
12683#endif /* CTL_IO_DELAY */
12684
12685void
12686ctl_datamove(union ctl_io *io)
12687{
12688	void (*fe_datamove)(union ctl_io *io);
12689
12690	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12691
12692	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12693
12694#ifdef CTL_TIME_IO
12695	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12696		char str[256];
12697		char path_str[64];
12698		struct sbuf sb;
12699
12700		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12701		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12702
12703		sbuf_cat(&sb, path_str);
12704		switch (io->io_hdr.io_type) {
12705		case CTL_IO_SCSI:
12706			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12707			sbuf_printf(&sb, "\n");
12708			sbuf_cat(&sb, path_str);
12709			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12710				    io->scsiio.tag_num, io->scsiio.tag_type);
12711			break;
12712		case CTL_IO_TASK:
12713			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12714				    "Tag Type: %d\n", io->taskio.task_action,
12715				    io->taskio.tag_num, io->taskio.tag_type);
12716			break;
12717		default:
12718			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12719			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12720			break;
12721		}
12722		sbuf_cat(&sb, path_str);
12723		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12724			    (intmax_t)time_uptime - io->io_hdr.start_time);
12725		sbuf_finish(&sb);
12726		printf("%s", sbuf_data(&sb));
12727	}
12728#endif /* CTL_TIME_IO */
12729
12730#ifdef CTL_IO_DELAY
12731	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12732		struct ctl_lun *lun;
12733
12734		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12735
12736		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12737	} else {
12738		struct ctl_lun *lun;
12739
12740		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12741		if ((lun != NULL)
12742		 && (lun->delay_info.datamove_delay > 0)) {
12743			struct callout *callout;
12744
12745			callout = (struct callout *)&io->io_hdr.timer_bytes;
12746			callout_init(callout, /*mpsafe*/ 1);
12747			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12748			callout_reset(callout,
12749				      lun->delay_info.datamove_delay * hz,
12750				      ctl_datamove_timer_wakeup, io);
12751			if (lun->delay_info.datamove_type ==
12752			    CTL_DELAY_TYPE_ONESHOT)
12753				lun->delay_info.datamove_delay = 0;
12754			return;
12755		}
12756	}
12757#endif
12758
12759	/*
12760	 * This command has been aborted.  Set the port status, so we fail
12761	 * the data move.
12762	 */
12763	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12764		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12765		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12766		       io->io_hdr.nexus.targ_port,
12767		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12768		       io->io_hdr.nexus.targ_lun);
12769		io->io_hdr.port_status = 31337;
12770		/*
12771		 * Note that the backend, in this case, will get the
12772		 * callback in its context.  In other cases it may get
12773		 * called in the frontend's interrupt thread context.
12774		 */
12775		io->scsiio.be_move_done(io);
12776		return;
12777	}
12778
12779	/*
12780	 * If we're in XFER mode and this I/O is from the other shelf
12781	 * controller, we need to send the DMA to the other side to
12782	 * actually transfer the data to/from the host.  In serialize only
12783	 * mode the transfer happens below CTL and ctl_datamove() is only
12784	 * called on the machine that originally received the I/O.
12785	 */
12786	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12787	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12788		union ctl_ha_msg msg;
12789		uint32_t sg_entries_sent;
12790		int do_sg_copy;
12791		int i;
12792
12793		memset(&msg, 0, sizeof(msg));
12794		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12795		msg.hdr.original_sc = io->io_hdr.original_sc;
12796		msg.hdr.serializing_sc = io;
12797		msg.hdr.nexus = io->io_hdr.nexus;
12798		msg.dt.flags = io->io_hdr.flags;
12799		/*
12800		 * We convert everything into a S/G list here.  We can't
12801		 * pass by reference, only by value between controllers.
12802		 * So we can't pass a pointer to the S/G list, only as many
12803		 * S/G entries as we can fit in here.  If it's possible for
12804		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12805		 * then we need to break this up into multiple transfers.
12806		 */
12807		if (io->scsiio.kern_sg_entries == 0) {
12808			msg.dt.kern_sg_entries = 1;
12809			/*
12810			 * If this is in cached memory, flush the cache
12811			 * before we send the DMA request to the other
12812			 * controller.  We want to do this in either the
12813			 * read or the write case.  The read case is
12814			 * straightforward.  In the write case, we want to
12815			 * make sure nothing is in the local cache that
12816			 * could overwrite the DMAed data.
12817			 */
12818			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12819				/*
12820				 * XXX KDM use bus_dmamap_sync() here.
12821				 */
12822			}
12823
12824			/*
12825			 * Convert to a physical address if this is a
12826			 * virtual address.
12827			 */
12828			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12829				msg.dt.sg_list[0].addr =
12830					io->scsiio.kern_data_ptr;
12831			} else {
12832				/*
12833				 * XXX KDM use busdma here!
12834				 */
12835#if 0
12836				msg.dt.sg_list[0].addr = (void *)
12837					vtophys(io->scsiio.kern_data_ptr);
12838#endif
12839			}
12840
12841			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12842			do_sg_copy = 0;
12843		} else {
12844			struct ctl_sg_entry *sgl;
12845
12846			do_sg_copy = 1;
12847			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12848			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12849			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12850				/*
12851				 * XXX KDM use bus_dmamap_sync() here.
12852				 */
12853			}
12854		}
12855
12856		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12857		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12858		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12859		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12860		msg.dt.sg_sequence = 0;
12861
12862		/*
12863		 * Loop until we've sent all of the S/G entries.  On the
12864		 * other end, we'll recompose these S/G entries into one
12865		 * contiguous list before passing it to the
12866		 */
12867		for (sg_entries_sent = 0; sg_entries_sent <
12868		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12869			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12870				sizeof(msg.dt.sg_list[0])),
12871				msg.dt.kern_sg_entries - sg_entries_sent);
12872
12873			if (do_sg_copy != 0) {
12874				struct ctl_sg_entry *sgl;
12875				int j;
12876
12877				sgl = (struct ctl_sg_entry *)
12878					io->scsiio.kern_data_ptr;
12879				/*
12880				 * If this is in cached memory, flush the cache
12881				 * before we send the DMA request to the other
12882				 * controller.  We want to do this in either
12883				 * the * read or the write case.  The read
12884				 * case is straightforward.  In the write
12885				 * case, we want to make sure nothing is
12886				 * in the local cache that could overwrite
12887				 * the DMAed data.
12888				 */
12889
12890				for (i = sg_entries_sent, j = 0;
12891				     i < msg.dt.cur_sg_entries; i++, j++) {
12892					if ((io->io_hdr.flags &
12893					     CTL_FLAG_NO_DATASYNC) == 0) {
12894						/*
12895						 * XXX KDM use bus_dmamap_sync()
12896						 */
12897					}
12898					if ((io->io_hdr.flags &
12899					     CTL_FLAG_BUS_ADDR) == 0) {
12900						/*
12901						 * XXX KDM use busdma.
12902						 */
12903#if 0
12904						msg.dt.sg_list[j].addr =(void *)
12905						       vtophys(sgl[i].addr);
12906#endif
12907					} else {
12908						msg.dt.sg_list[j].addr =
12909							sgl[i].addr;
12910					}
12911					msg.dt.sg_list[j].len = sgl[i].len;
12912				}
12913			}
12914
12915			sg_entries_sent += msg.dt.cur_sg_entries;
12916			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12917				msg.dt.sg_last = 1;
12918			else
12919				msg.dt.sg_last = 0;
12920
12921			/*
12922			 * XXX KDM drop and reacquire the lock here?
12923			 */
12924			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12925			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12926				/*
12927				 * XXX do something here.
12928				 */
12929			}
12930
12931			msg.dt.sent_sg_entries = sg_entries_sent;
12932		}
12933		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12934		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12935			ctl_failover_io(io, /*have_lock*/ 0);
12936
12937	} else {
12938
12939		/*
12940		 * Lookup the fe_datamove() function for this particular
12941		 * front end.
12942		 */
12943		fe_datamove =
12944		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12945
12946		fe_datamove(io);
12947	}
12948}
12949
12950static void
12951ctl_send_datamove_done(union ctl_io *io, int have_lock)
12952{
12953	union ctl_ha_msg msg;
12954	int isc_status;
12955
12956	memset(&msg, 0, sizeof(msg));
12957
12958	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12959	msg.hdr.original_sc = io;
12960	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12961	msg.hdr.nexus = io->io_hdr.nexus;
12962	msg.hdr.status = io->io_hdr.status;
12963	msg.scsi.tag_num = io->scsiio.tag_num;
12964	msg.scsi.tag_type = io->scsiio.tag_type;
12965	msg.scsi.scsi_status = io->scsiio.scsi_status;
12966	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12967	       sizeof(io->scsiio.sense_data));
12968	msg.scsi.sense_len = io->scsiio.sense_len;
12969	msg.scsi.sense_residual = io->scsiio.sense_residual;
12970	msg.scsi.fetd_status = io->io_hdr.port_status;
12971	msg.scsi.residual = io->scsiio.residual;
12972	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12973
12974	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12975		ctl_failover_io(io, /*have_lock*/ have_lock);
12976		return;
12977	}
12978
12979	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12980	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12981		/* XXX do something if this fails */
12982	}
12983
12984}
12985
12986/*
12987 * The DMA to the remote side is done, now we need to tell the other side
12988 * we're done so it can continue with its data movement.
12989 */
12990static void
12991ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12992{
12993	union ctl_io *io;
12994
12995	io = rq->context;
12996
12997	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12998		printf("%s: ISC DMA write failed with error %d", __func__,
12999		       rq->ret);
13000		ctl_set_internal_failure(&io->scsiio,
13001					 /*sks_valid*/ 1,
13002					 /*retry_count*/ rq->ret);
13003	}
13004
13005	ctl_dt_req_free(rq);
13006
13007	/*
13008	 * In this case, we had to malloc the memory locally.  Free it.
13009	 */
13010	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13011		int i;
13012		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13013			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13014	}
13015	/*
13016	 * The data is in local and remote memory, so now we need to send
13017	 * status (good or back) back to the other side.
13018	 */
13019	ctl_send_datamove_done(io, /*have_lock*/ 0);
13020}
13021
13022/*
13023 * We've moved the data from the host/controller into local memory.  Now we
13024 * need to push it over to the remote controller's memory.
13025 */
13026static int
13027ctl_datamove_remote_dm_write_cb(union ctl_io *io)
13028{
13029	int retval;
13030
13031	retval = 0;
13032
13033	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
13034					  ctl_datamove_remote_write_cb);
13035
13036	return (retval);
13037}
13038
13039static void
13040ctl_datamove_remote_write(union ctl_io *io)
13041{
13042	int retval;
13043	void (*fe_datamove)(union ctl_io *io);
13044
13045	/*
13046	 * - Get the data from the host/HBA into local memory.
13047	 * - DMA memory from the local controller to the remote controller.
13048	 * - Send status back to the remote controller.
13049	 */
13050
13051	retval = ctl_datamove_remote_sgl_setup(io);
13052	if (retval != 0)
13053		return;
13054
13055	/* Switch the pointer over so the FETD knows what to do */
13056	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13057
13058	/*
13059	 * Use a custom move done callback, since we need to send completion
13060	 * back to the other controller, not to the backend on this side.
13061	 */
13062	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
13063
13064	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13065
13066	fe_datamove(io);
13067
13068	return;
13069
13070}
13071
13072static int
13073ctl_datamove_remote_dm_read_cb(union ctl_io *io)
13074{
13075#if 0
13076	char str[256];
13077	char path_str[64];
13078	struct sbuf sb;
13079#endif
13080
13081	/*
13082	 * In this case, we had to malloc the memory locally.  Free it.
13083	 */
13084	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13085		int i;
13086		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13087			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13088	}
13089
13090#if 0
13091	scsi_path_string(io, path_str, sizeof(path_str));
13092	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13093	sbuf_cat(&sb, path_str);
13094	scsi_command_string(&io->scsiio, NULL, &sb);
13095	sbuf_printf(&sb, "\n");
13096	sbuf_cat(&sb, path_str);
13097	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13098		    io->scsiio.tag_num, io->scsiio.tag_type);
13099	sbuf_cat(&sb, path_str);
13100	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
13101		    io->io_hdr.flags, io->io_hdr.status);
13102	sbuf_finish(&sb);
13103	printk("%s", sbuf_data(&sb));
13104#endif
13105
13106
13107	/*
13108	 * The read is done, now we need to send status (good or bad) back
13109	 * to the other side.
13110	 */
13111	ctl_send_datamove_done(io, /*have_lock*/ 0);
13112
13113	return (0);
13114}
13115
13116static void
13117ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
13118{
13119	union ctl_io *io;
13120	void (*fe_datamove)(union ctl_io *io);
13121
13122	io = rq->context;
13123
13124	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13125		printf("%s: ISC DMA read failed with error %d", __func__,
13126		       rq->ret);
13127		ctl_set_internal_failure(&io->scsiio,
13128					 /*sks_valid*/ 1,
13129					 /*retry_count*/ rq->ret);
13130	}
13131
13132	ctl_dt_req_free(rq);
13133
13134	/* Switch the pointer over so the FETD knows what to do */
13135	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13136
13137	/*
13138	 * Use a custom move done callback, since we need to send completion
13139	 * back to the other controller, not to the backend on this side.
13140	 */
13141	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13142
13143	/* XXX KDM add checks like the ones in ctl_datamove? */
13144
13145	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13146
13147	fe_datamove(io);
13148}
13149
13150static int
13151ctl_datamove_remote_sgl_setup(union ctl_io *io)
13152{
13153	struct ctl_sg_entry *local_sglist, *remote_sglist;
13154	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13155	struct ctl_softc *softc;
13156	int retval;
13157	int i;
13158
13159	retval = 0;
13160	softc = control_softc;
13161
13162	local_sglist = io->io_hdr.local_sglist;
13163	local_dma_sglist = io->io_hdr.local_dma_sglist;
13164	remote_sglist = io->io_hdr.remote_sglist;
13165	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13166
13167	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13168		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13169			local_sglist[i].len = remote_sglist[i].len;
13170
13171			/*
13172			 * XXX Detect the situation where the RS-level I/O
13173			 * redirector on the other side has already read the
13174			 * data off of the AOR RS on this side, and
13175			 * transferred it to remote (mirror) memory on the
13176			 * other side.  Since we already have the data in
13177			 * memory here, we just need to use it.
13178			 *
13179			 * XXX KDM this can probably be removed once we
13180			 * get the cache device code in and take the
13181			 * current AOR implementation out.
13182			 */
13183#ifdef NEEDTOPORT
13184			if ((remote_sglist[i].addr >=
13185			     (void *)vtophys(softc->mirr->addr))
13186			 && (remote_sglist[i].addr <
13187			     ((void *)vtophys(softc->mirr->addr) +
13188			     CacheMirrorOffset))) {
13189				local_sglist[i].addr = remote_sglist[i].addr -
13190					CacheMirrorOffset;
13191				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13192				     CTL_FLAG_DATA_IN)
13193					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13194			} else {
13195				local_sglist[i].addr = remote_sglist[i].addr +
13196					CacheMirrorOffset;
13197			}
13198#endif
13199#if 0
13200			printf("%s: local %p, remote %p, len %d\n",
13201			       __func__, local_sglist[i].addr,
13202			       remote_sglist[i].addr, local_sglist[i].len);
13203#endif
13204		}
13205	} else {
13206		uint32_t len_to_go;
13207
13208		/*
13209		 * In this case, we don't have automatically allocated
13210		 * memory for this I/O on this controller.  This typically
13211		 * happens with internal CTL I/O -- e.g. inquiry, mode
13212		 * sense, etc.  Anything coming from RAIDCore will have
13213		 * a mirror area available.
13214		 */
13215		len_to_go = io->scsiio.kern_data_len;
13216
13217		/*
13218		 * Clear the no datasync flag, we have to use malloced
13219		 * buffers.
13220		 */
13221		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13222
13223		/*
13224		 * The difficult thing here is that the size of the various
13225		 * S/G segments may be different than the size from the
13226		 * remote controller.  That'll make it harder when DMAing
13227		 * the data back to the other side.
13228		 */
13229		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13230		     sizeof(io->io_hdr.remote_sglist[0])) &&
13231		     (len_to_go > 0); i++) {
13232			local_sglist[i].len = ctl_min(len_to_go, 131072);
13233			CTL_SIZE_8B(local_dma_sglist[i].len,
13234				    local_sglist[i].len);
13235			local_sglist[i].addr =
13236				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13237
13238			local_dma_sglist[i].addr = local_sglist[i].addr;
13239
13240			if (local_sglist[i].addr == NULL) {
13241				int j;
13242
13243				printf("malloc failed for %zd bytes!",
13244				       local_dma_sglist[i].len);
13245				for (j = 0; j < i; j++) {
13246					free(local_sglist[j].addr, M_CTL);
13247				}
13248				ctl_set_internal_failure(&io->scsiio,
13249							 /*sks_valid*/ 1,
13250							 /*retry_count*/ 4857);
13251				retval = 1;
13252				goto bailout_error;
13253
13254			}
13255			/* XXX KDM do we need a sync here? */
13256
13257			len_to_go -= local_sglist[i].len;
13258		}
13259		/*
13260		 * Reset the number of S/G entries accordingly.  The
13261		 * original number of S/G entries is available in
13262		 * rem_sg_entries.
13263		 */
13264		io->scsiio.kern_sg_entries = i;
13265
13266#if 0
13267		printf("%s: kern_sg_entries = %d\n", __func__,
13268		       io->scsiio.kern_sg_entries);
13269		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13270			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13271			       local_sglist[i].addr, local_sglist[i].len,
13272			       local_dma_sglist[i].len);
13273#endif
13274	}
13275
13276
13277	return (retval);
13278
13279bailout_error:
13280
13281	ctl_send_datamove_done(io, /*have_lock*/ 0);
13282
13283	return (retval);
13284}
13285
13286static int
13287ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13288			 ctl_ha_dt_cb callback)
13289{
13290	struct ctl_ha_dt_req *rq;
13291	struct ctl_sg_entry *remote_sglist, *local_sglist;
13292	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13293	uint32_t local_used, remote_used, total_used;
13294	int retval;
13295	int i, j;
13296
13297	retval = 0;
13298
13299	rq = ctl_dt_req_alloc();
13300
13301	/*
13302	 * If we failed to allocate the request, and if the DMA didn't fail
13303	 * anyway, set busy status.  This is just a resource allocation
13304	 * failure.
13305	 */
13306	if ((rq == NULL)
13307	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13308		ctl_set_busy(&io->scsiio);
13309
13310	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13311
13312		if (rq != NULL)
13313			ctl_dt_req_free(rq);
13314
13315		/*
13316		 * The data move failed.  We need to return status back
13317		 * to the other controller.  No point in trying to DMA
13318		 * data to the remote controller.
13319		 */
13320
13321		ctl_send_datamove_done(io, /*have_lock*/ 0);
13322
13323		retval = 1;
13324
13325		goto bailout;
13326	}
13327
13328	local_sglist = io->io_hdr.local_sglist;
13329	local_dma_sglist = io->io_hdr.local_dma_sglist;
13330	remote_sglist = io->io_hdr.remote_sglist;
13331	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13332	local_used = 0;
13333	remote_used = 0;
13334	total_used = 0;
13335
13336	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13337		rq->ret = CTL_HA_STATUS_SUCCESS;
13338		rq->context = io;
13339		callback(rq);
13340		goto bailout;
13341	}
13342
13343	/*
13344	 * Pull/push the data over the wire from/to the other controller.
13345	 * This takes into account the possibility that the local and
13346	 * remote sglists may not be identical in terms of the size of
13347	 * the elements and the number of elements.
13348	 *
13349	 * One fundamental assumption here is that the length allocated for
13350	 * both the local and remote sglists is identical.  Otherwise, we've
13351	 * essentially got a coding error of some sort.
13352	 */
13353	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13354		int isc_ret;
13355		uint32_t cur_len, dma_length;
13356		uint8_t *tmp_ptr;
13357
13358		rq->id = CTL_HA_DATA_CTL;
13359		rq->command = command;
13360		rq->context = io;
13361
13362		/*
13363		 * Both pointers should be aligned.  But it is possible
13364		 * that the allocation length is not.  They should both
13365		 * also have enough slack left over at the end, though,
13366		 * to round up to the next 8 byte boundary.
13367		 */
13368		cur_len = ctl_min(local_sglist[i].len - local_used,
13369				  remote_sglist[j].len - remote_used);
13370
13371		/*
13372		 * In this case, we have a size issue and need to decrease
13373		 * the size, except in the case where we actually have less
13374		 * than 8 bytes left.  In that case, we need to increase
13375		 * the DMA length to get the last bit.
13376		 */
13377		if ((cur_len & 0x7) != 0) {
13378			if (cur_len > 0x7) {
13379				cur_len = cur_len - (cur_len & 0x7);
13380				dma_length = cur_len;
13381			} else {
13382				CTL_SIZE_8B(dma_length, cur_len);
13383			}
13384
13385		} else
13386			dma_length = cur_len;
13387
13388		/*
13389		 * If we had to allocate memory for this I/O, instead of using
13390		 * the non-cached mirror memory, we'll need to flush the cache
13391		 * before trying to DMA to the other controller.
13392		 *
13393		 * We could end up doing this multiple times for the same
13394		 * segment if we have a larger local segment than remote
13395		 * segment.  That shouldn't be an issue.
13396		 */
13397		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13398			/*
13399			 * XXX KDM use bus_dmamap_sync() here.
13400			 */
13401		}
13402
13403		rq->size = dma_length;
13404
13405		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13406		tmp_ptr += local_used;
13407
13408		/* Use physical addresses when talking to ISC hardware */
13409		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13410			/* XXX KDM use busdma */
13411#if 0
13412			rq->local = vtophys(tmp_ptr);
13413#endif
13414		} else
13415			rq->local = tmp_ptr;
13416
13417		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13418		tmp_ptr += remote_used;
13419		rq->remote = tmp_ptr;
13420
13421		rq->callback = NULL;
13422
13423		local_used += cur_len;
13424		if (local_used >= local_sglist[i].len) {
13425			i++;
13426			local_used = 0;
13427		}
13428
13429		remote_used += cur_len;
13430		if (remote_used >= remote_sglist[j].len) {
13431			j++;
13432			remote_used = 0;
13433		}
13434		total_used += cur_len;
13435
13436		if (total_used >= io->scsiio.kern_data_len)
13437			rq->callback = callback;
13438
13439		if ((rq->size & 0x7) != 0) {
13440			printf("%s: warning: size %d is not on 8b boundary\n",
13441			       __func__, rq->size);
13442		}
13443		if (((uintptr_t)rq->local & 0x7) != 0) {
13444			printf("%s: warning: local %p not on 8b boundary\n",
13445			       __func__, rq->local);
13446		}
13447		if (((uintptr_t)rq->remote & 0x7) != 0) {
13448			printf("%s: warning: remote %p not on 8b boundary\n",
13449			       __func__, rq->local);
13450		}
13451#if 0
13452		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13453		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13454		       rq->local, rq->remote, rq->size);
13455#endif
13456
13457		isc_ret = ctl_dt_single(rq);
13458		if (isc_ret == CTL_HA_STATUS_WAIT)
13459			continue;
13460
13461		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13462			rq->ret = CTL_HA_STATUS_SUCCESS;
13463		} else {
13464			rq->ret = isc_ret;
13465		}
13466		callback(rq);
13467		goto bailout;
13468	}
13469
13470bailout:
13471	return (retval);
13472
13473}
13474
13475static void
13476ctl_datamove_remote_read(union ctl_io *io)
13477{
13478	int retval;
13479	int i;
13480
13481	/*
13482	 * This will send an error to the other controller in the case of a
13483	 * failure.
13484	 */
13485	retval = ctl_datamove_remote_sgl_setup(io);
13486	if (retval != 0)
13487		return;
13488
13489	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13490					  ctl_datamove_remote_read_cb);
13491	if ((retval != 0)
13492	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13493		/*
13494		 * Make sure we free memory if there was an error..  The
13495		 * ctl_datamove_remote_xfer() function will send the
13496		 * datamove done message, or call the callback with an
13497		 * error if there is a problem.
13498		 */
13499		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13500			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13501	}
13502
13503	return;
13504}
13505
13506/*
13507 * Process a datamove request from the other controller.  This is used for
13508 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13509 * first.  Once that is complete, the data gets DMAed into the remote
13510 * controller's memory.  For reads, we DMA from the remote controller's
13511 * memory into our memory first, and then move it out to the FETD.
13512 */
13513static void
13514ctl_datamove_remote(union ctl_io *io)
13515{
13516	struct ctl_softc *softc;
13517
13518	softc = control_softc;
13519
13520	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13521
13522	/*
13523	 * Note that we look for an aborted I/O here, but don't do some of
13524	 * the other checks that ctl_datamove() normally does.
13525	 * We don't need to run the datamove delay code, since that should
13526	 * have been done if need be on the other controller.
13527	 */
13528	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13529		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13530		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13531		       io->io_hdr.nexus.targ_port,
13532		       io->io_hdr.nexus.targ_target.id,
13533		       io->io_hdr.nexus.targ_lun);
13534		io->io_hdr.port_status = 31338;
13535		ctl_send_datamove_done(io, /*have_lock*/ 0);
13536		return;
13537	}
13538
13539	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13540		ctl_datamove_remote_write(io);
13541	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13542		ctl_datamove_remote_read(io);
13543	} else {
13544		union ctl_ha_msg msg;
13545		struct scsi_sense_data *sense;
13546		uint8_t sks[3];
13547		int retry_count;
13548
13549		memset(&msg, 0, sizeof(msg));
13550
13551		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13552		msg.hdr.status = CTL_SCSI_ERROR;
13553		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13554
13555		retry_count = 4243;
13556
13557		sense = &msg.scsi.sense_data;
13558		sks[0] = SSD_SCS_VALID;
13559		sks[1] = (retry_count >> 8) & 0xff;
13560		sks[2] = retry_count & 0xff;
13561
13562		/* "Internal target failure" */
13563		scsi_set_sense_data(sense,
13564				    /*sense_format*/ SSD_TYPE_NONE,
13565				    /*current_error*/ 1,
13566				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13567				    /*asc*/ 0x44,
13568				    /*ascq*/ 0x00,
13569				    /*type*/ SSD_ELEM_SKS,
13570				    /*size*/ sizeof(sks),
13571				    /*data*/ sks,
13572				    SSD_ELEM_NONE);
13573
13574		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13575		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13576			ctl_failover_io(io, /*have_lock*/ 1);
13577			return;
13578		}
13579
13580		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13581		    CTL_HA_STATUS_SUCCESS) {
13582			/* XXX KDM what to do if this fails? */
13583		}
13584		return;
13585	}
13586
13587}
13588
13589static int
13590ctl_process_done(union ctl_io *io)
13591{
13592	struct ctl_lun *lun;
13593	struct ctl_softc *ctl_softc;
13594	void (*fe_done)(union ctl_io *io);
13595	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13596
13597	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13598
13599	fe_done =
13600	    control_softc->ctl_ports[targ_port]->fe_done;
13601
13602#ifdef CTL_TIME_IO
13603	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13604		char str[256];
13605		char path_str[64];
13606		struct sbuf sb;
13607
13608		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13609		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13610
13611		sbuf_cat(&sb, path_str);
13612		switch (io->io_hdr.io_type) {
13613		case CTL_IO_SCSI:
13614			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13615			sbuf_printf(&sb, "\n");
13616			sbuf_cat(&sb, path_str);
13617			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13618				    io->scsiio.tag_num, io->scsiio.tag_type);
13619			break;
13620		case CTL_IO_TASK:
13621			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13622				    "Tag Type: %d\n", io->taskio.task_action,
13623				    io->taskio.tag_num, io->taskio.tag_type);
13624			break;
13625		default:
13626			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13627			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13628			break;
13629		}
13630		sbuf_cat(&sb, path_str);
13631		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13632			    (intmax_t)time_uptime - io->io_hdr.start_time);
13633		sbuf_finish(&sb);
13634		printf("%s", sbuf_data(&sb));
13635	}
13636#endif /* CTL_TIME_IO */
13637
13638	switch (io->io_hdr.io_type) {
13639	case CTL_IO_SCSI:
13640		break;
13641	case CTL_IO_TASK:
13642		if (bootverbose || verbose > 0)
13643			ctl_io_error_print(io, NULL);
13644		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13645			ctl_free_io(io);
13646		else
13647			fe_done(io);
13648		return (CTL_RETVAL_COMPLETE);
13649		break;
13650	default:
13651		printf("ctl_process_done: invalid io type %d\n",
13652		       io->io_hdr.io_type);
13653		panic("ctl_process_done: invalid io type %d\n",
13654		      io->io_hdr.io_type);
13655		break; /* NOTREACHED */
13656	}
13657
13658	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13659	if (lun == NULL) {
13660		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13661				 io->io_hdr.nexus.targ_mapped_lun));
13662		fe_done(io);
13663		goto bailout;
13664	}
13665	ctl_softc = lun->ctl_softc;
13666
13667	mtx_lock(&lun->lun_lock);
13668
13669	/*
13670	 * Check to see if we have any errors to inject here.  We only
13671	 * inject errors for commands that don't already have errors set.
13672	 */
13673	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13674	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13675		ctl_inject_error(lun, io);
13676
13677	/*
13678	 * XXX KDM how do we treat commands that aren't completed
13679	 * successfully?
13680	 *
13681	 * XXX KDM should we also track I/O latency?
13682	 */
13683	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13684	    io->io_hdr.io_type == CTL_IO_SCSI) {
13685#ifdef CTL_TIME_IO
13686		struct bintime cur_bt;
13687#endif
13688		int type;
13689
13690		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13691		    CTL_FLAG_DATA_IN)
13692			type = CTL_STATS_READ;
13693		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13694		    CTL_FLAG_DATA_OUT)
13695			type = CTL_STATS_WRITE;
13696		else
13697			type = CTL_STATS_NO_IO;
13698
13699		lun->stats.ports[targ_port].bytes[type] +=
13700		    io->scsiio.kern_total_len;
13701		lun->stats.ports[targ_port].operations[type]++;
13702#ifdef CTL_TIME_IO
13703		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13704		   &io->io_hdr.dma_bt);
13705		lun->stats.ports[targ_port].num_dmas[type] +=
13706		    io->io_hdr.num_dmas;
13707		getbintime(&cur_bt);
13708		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13709		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13710#endif
13711	}
13712
13713	/*
13714	 * Remove this from the OOA queue.
13715	 */
13716	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13717
13718	/*
13719	 * Run through the blocked queue on this LUN and see if anything
13720	 * has become unblocked, now that this transaction is done.
13721	 */
13722	ctl_check_blocked(lun);
13723
13724	/*
13725	 * If the LUN has been invalidated, free it if there is nothing
13726	 * left on its OOA queue.
13727	 */
13728	if ((lun->flags & CTL_LUN_INVALID)
13729	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13730		mtx_unlock(&lun->lun_lock);
13731		mtx_lock(&ctl_softc->ctl_lock);
13732		ctl_free_lun(lun);
13733		mtx_unlock(&ctl_softc->ctl_lock);
13734	} else
13735		mtx_unlock(&lun->lun_lock);
13736
13737	/*
13738	 * If this command has been aborted, make sure we set the status
13739	 * properly.  The FETD is responsible for freeing the I/O and doing
13740	 * whatever it needs to do to clean up its state.
13741	 */
13742	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13743		ctl_set_task_aborted(&io->scsiio);
13744
13745	/*
13746	 * We print out status for every task management command.  For SCSI
13747	 * commands, we filter out any unit attention errors; they happen
13748	 * on every boot, and would clutter up the log.  Note:  task
13749	 * management commands aren't printed here, they are printed above,
13750	 * since they should never even make it down here.
13751	 */
13752	switch (io->io_hdr.io_type) {
13753	case CTL_IO_SCSI: {
13754		int error_code, sense_key, asc, ascq;
13755
13756		sense_key = 0;
13757
13758		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13759		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13760			/*
13761			 * Since this is just for printing, no need to
13762			 * show errors here.
13763			 */
13764			scsi_extract_sense_len(&io->scsiio.sense_data,
13765					       io->scsiio.sense_len,
13766					       &error_code,
13767					       &sense_key,
13768					       &asc,
13769					       &ascq,
13770					       /*show_errors*/ 0);
13771		}
13772
13773		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13774		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13775		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13776		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13777
13778			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13779				ctl_softc->skipped_prints++;
13780			} else {
13781				uint32_t skipped_prints;
13782
13783				skipped_prints = ctl_softc->skipped_prints;
13784
13785				ctl_softc->skipped_prints = 0;
13786				ctl_softc->last_print_jiffies = time_uptime;
13787
13788				if (skipped_prints > 0) {
13789#ifdef NEEDTOPORT
13790					csevent_log(CSC_CTL | CSC_SHELF_SW |
13791					    CTL_ERROR_REPORT,
13792					    csevent_LogType_Trace,
13793					    csevent_Severity_Information,
13794					    csevent_AlertLevel_Green,
13795					    csevent_FRU_Firmware,
13796					    csevent_FRU_Unknown,
13797					    "High CTL error volume, %d prints "
13798					    "skipped", skipped_prints);
13799#endif
13800				}
13801				if (bootverbose || verbose > 0)
13802					ctl_io_error_print(io, NULL);
13803			}
13804		}
13805		break;
13806	}
13807	case CTL_IO_TASK:
13808		if (bootverbose || verbose > 0)
13809			ctl_io_error_print(io, NULL);
13810		break;
13811	default:
13812		break;
13813	}
13814
13815	/*
13816	 * Tell the FETD or the other shelf controller we're done with this
13817	 * command.  Note that only SCSI commands get to this point.  Task
13818	 * management commands are completed above.
13819	 *
13820	 * We only send status to the other controller if we're in XFER
13821	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13822	 * received the I/O (from CTL's perspective), and so the status is
13823	 * generated there.
13824	 *
13825	 * XXX KDM if we hold the lock here, we could cause a deadlock
13826	 * if the frontend comes back in in this context to queue
13827	 * something.
13828	 */
13829	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13830	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13831		union ctl_ha_msg msg;
13832
13833		memset(&msg, 0, sizeof(msg));
13834		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13835		msg.hdr.original_sc = io->io_hdr.original_sc;
13836		msg.hdr.nexus = io->io_hdr.nexus;
13837		msg.hdr.status = io->io_hdr.status;
13838		msg.scsi.scsi_status = io->scsiio.scsi_status;
13839		msg.scsi.tag_num = io->scsiio.tag_num;
13840		msg.scsi.tag_type = io->scsiio.tag_type;
13841		msg.scsi.sense_len = io->scsiio.sense_len;
13842		msg.scsi.sense_residual = io->scsiio.sense_residual;
13843		msg.scsi.residual = io->scsiio.residual;
13844		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13845		       sizeof(io->scsiio.sense_data));
13846		/*
13847		 * We copy this whether or not this is an I/O-related
13848		 * command.  Otherwise, we'd have to go and check to see
13849		 * whether it's a read/write command, and it really isn't
13850		 * worth it.
13851		 */
13852		memcpy(&msg.scsi.lbalen,
13853		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13854		       sizeof(msg.scsi.lbalen));
13855
13856		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13857				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13858			/* XXX do something here */
13859		}
13860
13861		ctl_free_io(io);
13862	} else
13863		fe_done(io);
13864
13865bailout:
13866
13867	return (CTL_RETVAL_COMPLETE);
13868}
13869
13870#ifdef CTL_WITH_CA
13871/*
13872 * Front end should call this if it doesn't do autosense.  When the request
13873 * sense comes back in from the initiator, we'll dequeue this and send it.
13874 */
13875int
13876ctl_queue_sense(union ctl_io *io)
13877{
13878	struct ctl_lun *lun;
13879	struct ctl_softc *ctl_softc;
13880	uint32_t initidx, targ_lun;
13881
13882	ctl_softc = control_softc;
13883
13884	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13885
13886	/*
13887	 * LUN lookup will likely move to the ctl_work_thread() once we
13888	 * have our new queueing infrastructure (that doesn't put things on
13889	 * a per-LUN queue initially).  That is so that we can handle
13890	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13891	 * can't deal with that right now.
13892	 */
13893	mtx_lock(&ctl_softc->ctl_lock);
13894
13895	/*
13896	 * If we don't have a LUN for this, just toss the sense
13897	 * information.
13898	 */
13899	targ_lun = io->io_hdr.nexus.targ_lun;
13900	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13901	if ((targ_lun < CTL_MAX_LUNS)
13902	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13903		lun = ctl_softc->ctl_luns[targ_lun];
13904	else
13905		goto bailout;
13906
13907	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13908
13909	mtx_lock(&lun->lun_lock);
13910	/*
13911	 * Already have CA set for this LUN...toss the sense information.
13912	 */
13913	if (ctl_is_set(lun->have_ca, initidx)) {
13914		mtx_unlock(&lun->lun_lock);
13915		goto bailout;
13916	}
13917
13918	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13919	       ctl_min(sizeof(lun->pending_sense[initidx]),
13920	       sizeof(io->scsiio.sense_data)));
13921	ctl_set_mask(lun->have_ca, initidx);
13922	mtx_unlock(&lun->lun_lock);
13923
13924bailout:
13925	mtx_unlock(&ctl_softc->ctl_lock);
13926
13927	ctl_free_io(io);
13928
13929	return (CTL_RETVAL_COMPLETE);
13930}
13931#endif
13932
13933/*
13934 * Primary command inlet from frontend ports.  All SCSI and task I/O
13935 * requests must go through this function.
13936 */
13937int
13938ctl_queue(union ctl_io *io)
13939{
13940	struct ctl_softc *ctl_softc;
13941
13942	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13943
13944	ctl_softc = control_softc;
13945
13946#ifdef CTL_TIME_IO
13947	io->io_hdr.start_time = time_uptime;
13948	getbintime(&io->io_hdr.start_bt);
13949#endif /* CTL_TIME_IO */
13950
13951	/* Map FE-specific LUN ID into global one. */
13952	io->io_hdr.nexus.targ_mapped_lun =
13953	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13954
13955	switch (io->io_hdr.io_type) {
13956	case CTL_IO_SCSI:
13957	case CTL_IO_TASK:
13958		ctl_enqueue_incoming(io);
13959		break;
13960	default:
13961		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13962		return (EINVAL);
13963	}
13964
13965	return (CTL_RETVAL_COMPLETE);
13966}
13967
13968#ifdef CTL_IO_DELAY
13969static void
13970ctl_done_timer_wakeup(void *arg)
13971{
13972	union ctl_io *io;
13973
13974	io = (union ctl_io *)arg;
13975	ctl_done(io);
13976}
13977#endif /* CTL_IO_DELAY */
13978
13979void
13980ctl_done(union ctl_io *io)
13981{
13982	struct ctl_softc *ctl_softc;
13983
13984	ctl_softc = control_softc;
13985
13986	/*
13987	 * Enable this to catch duplicate completion issues.
13988	 */
13989#if 0
13990	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13991		printf("%s: type %d msg %d cdb %x iptl: "
13992		       "%d:%d:%d:%d tag 0x%04x "
13993		       "flag %#x status %x\n",
13994			__func__,
13995			io->io_hdr.io_type,
13996			io->io_hdr.msg_type,
13997			io->scsiio.cdb[0],
13998			io->io_hdr.nexus.initid.id,
13999			io->io_hdr.nexus.targ_port,
14000			io->io_hdr.nexus.targ_target.id,
14001			io->io_hdr.nexus.targ_lun,
14002			(io->io_hdr.io_type ==
14003			CTL_IO_TASK) ?
14004			io->taskio.tag_num :
14005			io->scsiio.tag_num,
14006		        io->io_hdr.flags,
14007			io->io_hdr.status);
14008	} else
14009		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
14010#endif
14011
14012	/*
14013	 * This is an internal copy of an I/O, and should not go through
14014	 * the normal done processing logic.
14015	 */
14016	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
14017		return;
14018
14019	/*
14020	 * We need to send a msg to the serializing shelf to finish the IO
14021	 * as well.  We don't send a finish message to the other shelf if
14022	 * this is a task management command.  Task management commands
14023	 * aren't serialized in the OOA queue, but rather just executed on
14024	 * both shelf controllers for commands that originated on that
14025	 * controller.
14026	 */
14027	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
14028	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
14029		union ctl_ha_msg msg_io;
14030
14031		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
14032		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
14033		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
14034		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
14035		}
14036		/* continue on to finish IO */
14037	}
14038#ifdef CTL_IO_DELAY
14039	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
14040		struct ctl_lun *lun;
14041
14042		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14043
14044		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
14045	} else {
14046		struct ctl_lun *lun;
14047
14048		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14049
14050		if ((lun != NULL)
14051		 && (lun->delay_info.done_delay > 0)) {
14052			struct callout *callout;
14053
14054			callout = (struct callout *)&io->io_hdr.timer_bytes;
14055			callout_init(callout, /*mpsafe*/ 1);
14056			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
14057			callout_reset(callout,
14058				      lun->delay_info.done_delay * hz,
14059				      ctl_done_timer_wakeup, io);
14060			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
14061				lun->delay_info.done_delay = 0;
14062			return;
14063		}
14064	}
14065#endif /* CTL_IO_DELAY */
14066
14067	ctl_enqueue_done(io);
14068}
14069
14070int
14071ctl_isc(struct ctl_scsiio *ctsio)
14072{
14073	struct ctl_lun *lun;
14074	int retval;
14075
14076	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14077
14078	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
14079
14080	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
14081
14082	retval = lun->backend->data_submit((union ctl_io *)ctsio);
14083
14084	return (retval);
14085}
14086
14087
14088static void
14089ctl_work_thread(void *arg)
14090{
14091	struct ctl_thread *thr = (struct ctl_thread *)arg;
14092	struct ctl_softc *softc = thr->ctl_softc;
14093	union ctl_io *io;
14094	int retval;
14095
14096	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
14097
14098	for (;;) {
14099		retval = 0;
14100
14101		/*
14102		 * We handle the queues in this order:
14103		 * - ISC
14104		 * - done queue (to free up resources, unblock other commands)
14105		 * - RtR queue
14106		 * - incoming queue
14107		 *
14108		 * If those queues are empty, we break out of the loop and
14109		 * go to sleep.
14110		 */
14111		mtx_lock(&thr->queue_lock);
14112		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
14113		if (io != NULL) {
14114			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
14115			mtx_unlock(&thr->queue_lock);
14116			ctl_handle_isc(io);
14117			continue;
14118		}
14119		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
14120		if (io != NULL) {
14121			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
14122			/* clear any blocked commands, call fe_done */
14123			mtx_unlock(&thr->queue_lock);
14124			retval = ctl_process_done(io);
14125			continue;
14126		}
14127		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
14128		if (io != NULL) {
14129			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
14130			mtx_unlock(&thr->queue_lock);
14131			if (io->io_hdr.io_type == CTL_IO_TASK)
14132				ctl_run_task(io);
14133			else
14134				ctl_scsiio_precheck(softc, &io->scsiio);
14135			continue;
14136		}
14137		if (!ctl_pause_rtr) {
14138			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
14139			if (io != NULL) {
14140				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
14141				mtx_unlock(&thr->queue_lock);
14142				retval = ctl_scsiio(&io->scsiio);
14143				if (retval != CTL_RETVAL_COMPLETE)
14144					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
14145				continue;
14146			}
14147		}
14148
14149		/* Sleep until we have something to do. */
14150		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
14151	}
14152}
14153
14154static void
14155ctl_lun_thread(void *arg)
14156{
14157	struct ctl_softc *softc = (struct ctl_softc *)arg;
14158	struct ctl_be_lun *be_lun;
14159	int retval;
14160
14161	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14162
14163	for (;;) {
14164		retval = 0;
14165		mtx_lock(&softc->ctl_lock);
14166		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14167		if (be_lun != NULL) {
14168			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14169			mtx_unlock(&softc->ctl_lock);
14170			ctl_create_lun(be_lun);
14171			continue;
14172		}
14173
14174		/* Sleep until we have something to do. */
14175		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14176		    PDROP | PRIBIO, "-", 0);
14177	}
14178}
14179
14180static void
14181ctl_enqueue_incoming(union ctl_io *io)
14182{
14183	struct ctl_softc *softc = control_softc;
14184	struct ctl_thread *thr;
14185	u_int idx;
14186
14187	idx = (io->io_hdr.nexus.targ_port * 127 +
14188	       io->io_hdr.nexus.initid.id) % worker_threads;
14189	thr = &softc->threads[idx];
14190	mtx_lock(&thr->queue_lock);
14191	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14192	mtx_unlock(&thr->queue_lock);
14193	wakeup(thr);
14194}
14195
14196static void
14197ctl_enqueue_rtr(union ctl_io *io)
14198{
14199	struct ctl_softc *softc = control_softc;
14200	struct ctl_thread *thr;
14201
14202	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14203	mtx_lock(&thr->queue_lock);
14204	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14205	mtx_unlock(&thr->queue_lock);
14206	wakeup(thr);
14207}
14208
14209static void
14210ctl_enqueue_done(union ctl_io *io)
14211{
14212	struct ctl_softc *softc = control_softc;
14213	struct ctl_thread *thr;
14214
14215	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14216	mtx_lock(&thr->queue_lock);
14217	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14218	mtx_unlock(&thr->queue_lock);
14219	wakeup(thr);
14220}
14221
14222static void
14223ctl_enqueue_isc(union ctl_io *io)
14224{
14225	struct ctl_softc *softc = control_softc;
14226	struct ctl_thread *thr;
14227
14228	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14229	mtx_lock(&thr->queue_lock);
14230	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14231	mtx_unlock(&thr->queue_lock);
14232	wakeup(thr);
14233}
14234
14235/* Initialization and failover */
14236
14237void
14238ctl_init_isc_msg(void)
14239{
14240	printf("CTL: Still calling this thing\n");
14241}
14242
14243/*
14244 * Init component
14245 * 	Initializes component into configuration defined by bootMode
14246 *	(see hasc-sv.c)
14247 *  	returns hasc_Status:
14248 * 		OK
14249 *		ERROR - fatal error
14250 */
14251static ctl_ha_comp_status
14252ctl_isc_init(struct ctl_ha_component *c)
14253{
14254	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14255
14256	c->status = ret;
14257	return ret;
14258}
14259
14260/* Start component
14261 * 	Starts component in state requested. If component starts successfully,
14262 *	it must set its own state to the requestrd state
14263 *	When requested state is HASC_STATE_HA, the component may refine it
14264 * 	by adding _SLAVE or _MASTER flags.
14265 *	Currently allowed state transitions are:
14266 *	UNKNOWN->HA		- initial startup
14267 *	UNKNOWN->SINGLE - initial startup when no parter detected
14268 *	HA->SINGLE		- failover
14269 * returns ctl_ha_comp_status:
14270 * 		OK	- component successfully started in requested state
14271 *		FAILED  - could not start the requested state, failover may
14272 * 			  be possible
14273 *		ERROR	- fatal error detected, no future startup possible
14274 */
14275static ctl_ha_comp_status
14276ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14277{
14278	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14279
14280	printf("%s: go\n", __func__);
14281
14282	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14283	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14284		ctl_is_single = 0;
14285		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14286		    != CTL_HA_STATUS_SUCCESS) {
14287			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14288			ret = CTL_HA_COMP_STATUS_ERROR;
14289		}
14290	} else if (CTL_HA_STATE_IS_HA(c->state)
14291		&& CTL_HA_STATE_IS_SINGLE(state)){
14292		// HA->SINGLE transition
14293	        ctl_failover();
14294		ctl_is_single = 1;
14295	} else {
14296		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14297		       c->state, state);
14298		ret = CTL_HA_COMP_STATUS_ERROR;
14299	}
14300	if (CTL_HA_STATE_IS_SINGLE(state))
14301		ctl_is_single = 1;
14302
14303	c->state = state;
14304	c->status = ret;
14305	return ret;
14306}
14307
14308/*
14309 * Quiesce component
14310 * The component must clear any error conditions (set status to OK) and
14311 * prepare itself to another Start call
14312 * returns ctl_ha_comp_status:
14313 * 	OK
14314 *	ERROR
14315 */
14316static ctl_ha_comp_status
14317ctl_isc_quiesce(struct ctl_ha_component *c)
14318{
14319	int ret = CTL_HA_COMP_STATUS_OK;
14320
14321	ctl_pause_rtr = 1;
14322	c->status = ret;
14323	return ret;
14324}
14325
14326struct ctl_ha_component ctl_ha_component_ctlisc =
14327{
14328	.name = "CTL ISC",
14329	.state = CTL_HA_STATE_UNKNOWN,
14330	.init = ctl_isc_init,
14331	.start = ctl_isc_start,
14332	.quiesce = ctl_isc_quiesce
14333};
14334
14335/*
14336 *  vim: ts=8
14337 */
14338