ctl.c revision 271951
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 271951 2014-09-22 10:05:36Z 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.flags & CTL_FLAG_ABORT) == 0 &&
5284	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5285	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5286		io->scsiio.io_cont(io);
5287		return;
5288	}
5289	/*
5290	 * Since a configuration write can be done for commands that actually
5291	 * have data allocated, like write buffer, and commands that have
5292	 * no data, like start/stop unit, we need to check here.
5293	 */
5294	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5295		buf = io->scsiio.kern_data_ptr;
5296	else
5297		buf = NULL;
5298	ctl_done(io);
5299	if (buf)
5300		free(buf, M_CTL);
5301}
5302
5303/*
5304 * SCSI release command.
5305 */
5306int
5307ctl_scsi_release(struct ctl_scsiio *ctsio)
5308{
5309	int length, longid, thirdparty_id, resv_id;
5310	struct ctl_softc *ctl_softc;
5311	struct ctl_lun *lun;
5312	uint32_t residx;
5313
5314	length = 0;
5315	resv_id = 0;
5316
5317	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5318
5319	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5320	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5321	ctl_softc = control_softc;
5322
5323	switch (ctsio->cdb[0]) {
5324	case RELEASE_10: {
5325		struct scsi_release_10 *cdb;
5326
5327		cdb = (struct scsi_release_10 *)ctsio->cdb;
5328
5329		if (cdb->byte2 & SR10_LONGID)
5330			longid = 1;
5331		else
5332			thirdparty_id = cdb->thirdparty_id;
5333
5334		resv_id = cdb->resv_id;
5335		length = scsi_2btoul(cdb->length);
5336		break;
5337	}
5338	}
5339
5340
5341	/*
5342	 * XXX KDM right now, we only support LUN reservation.  We don't
5343	 * support 3rd party reservations, or extent reservations, which
5344	 * might actually need the parameter list.  If we've gotten this
5345	 * far, we've got a LUN reservation.  Anything else got kicked out
5346	 * above.  So, according to SPC, ignore the length.
5347	 */
5348	length = 0;
5349
5350	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5351	 && (length > 0)) {
5352		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5353		ctsio->kern_data_len = length;
5354		ctsio->kern_total_len = length;
5355		ctsio->kern_data_resid = 0;
5356		ctsio->kern_rel_offset = 0;
5357		ctsio->kern_sg_entries = 0;
5358		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5359		ctsio->be_move_done = ctl_config_move_done;
5360		ctl_datamove((union ctl_io *)ctsio);
5361
5362		return (CTL_RETVAL_COMPLETE);
5363	}
5364
5365	if (length > 0)
5366		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5367
5368	mtx_lock(&lun->lun_lock);
5369
5370	/*
5371	 * According to SPC, it is not an error for an intiator to attempt
5372	 * to release a reservation on a LUN that isn't reserved, or that
5373	 * is reserved by another initiator.  The reservation can only be
5374	 * released, though, by the initiator who made it or by one of
5375	 * several reset type events.
5376	 */
5377	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5378			lun->flags &= ~CTL_LUN_RESERVED;
5379
5380	mtx_unlock(&lun->lun_lock);
5381
5382	ctsio->scsi_status = SCSI_STATUS_OK;
5383	ctsio->io_hdr.status = CTL_SUCCESS;
5384
5385	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5386		free(ctsio->kern_data_ptr, M_CTL);
5387		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5388	}
5389
5390	ctl_done((union ctl_io *)ctsio);
5391	return (CTL_RETVAL_COMPLETE);
5392}
5393
5394int
5395ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5396{
5397	int extent, thirdparty, longid;
5398	int resv_id, length;
5399	uint64_t thirdparty_id;
5400	struct ctl_softc *ctl_softc;
5401	struct ctl_lun *lun;
5402	uint32_t residx;
5403
5404	extent = 0;
5405	thirdparty = 0;
5406	longid = 0;
5407	resv_id = 0;
5408	length = 0;
5409	thirdparty_id = 0;
5410
5411	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5412
5413	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5414	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5415	ctl_softc = control_softc;
5416
5417	switch (ctsio->cdb[0]) {
5418	case RESERVE_10: {
5419		struct scsi_reserve_10 *cdb;
5420
5421		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5422
5423		if (cdb->byte2 & SR10_LONGID)
5424			longid = 1;
5425		else
5426			thirdparty_id = cdb->thirdparty_id;
5427
5428		resv_id = cdb->resv_id;
5429		length = scsi_2btoul(cdb->length);
5430		break;
5431	}
5432	}
5433
5434	/*
5435	 * XXX KDM right now, we only support LUN reservation.  We don't
5436	 * support 3rd party reservations, or extent reservations, which
5437	 * might actually need the parameter list.  If we've gotten this
5438	 * far, we've got a LUN reservation.  Anything else got kicked out
5439	 * above.  So, according to SPC, ignore the length.
5440	 */
5441	length = 0;
5442
5443	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5444	 && (length > 0)) {
5445		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5446		ctsio->kern_data_len = length;
5447		ctsio->kern_total_len = length;
5448		ctsio->kern_data_resid = 0;
5449		ctsio->kern_rel_offset = 0;
5450		ctsio->kern_sg_entries = 0;
5451		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5452		ctsio->be_move_done = ctl_config_move_done;
5453		ctl_datamove((union ctl_io *)ctsio);
5454
5455		return (CTL_RETVAL_COMPLETE);
5456	}
5457
5458	if (length > 0)
5459		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5460
5461	mtx_lock(&lun->lun_lock);
5462	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5463		ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5464		ctsio->io_hdr.status = CTL_SCSI_ERROR;
5465		goto bailout;
5466	}
5467
5468	lun->flags |= CTL_LUN_RESERVED;
5469	lun->res_idx = residx;
5470
5471	ctsio->scsi_status = SCSI_STATUS_OK;
5472	ctsio->io_hdr.status = CTL_SUCCESS;
5473
5474bailout:
5475	mtx_unlock(&lun->lun_lock);
5476
5477	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5478		free(ctsio->kern_data_ptr, M_CTL);
5479		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5480	}
5481
5482	ctl_done((union ctl_io *)ctsio);
5483	return (CTL_RETVAL_COMPLETE);
5484}
5485
5486int
5487ctl_start_stop(struct ctl_scsiio *ctsio)
5488{
5489	struct scsi_start_stop_unit *cdb;
5490	struct ctl_lun *lun;
5491	struct ctl_softc *ctl_softc;
5492	int retval;
5493
5494	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5495
5496	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5497	ctl_softc = control_softc;
5498	retval = 0;
5499
5500	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5501
5502	/*
5503	 * XXX KDM
5504	 * We don't support the immediate bit on a stop unit.  In order to
5505	 * do that, we would need to code up a way to know that a stop is
5506	 * pending, and hold off any new commands until it completes, one
5507	 * way or another.  Then we could accept or reject those commands
5508	 * depending on its status.  We would almost need to do the reverse
5509	 * of what we do below for an immediate start -- return the copy of
5510	 * the ctl_io to the FETD with status to send to the host (and to
5511	 * free the copy!) and then free the original I/O once the stop
5512	 * actually completes.  That way, the OOA queue mechanism can work
5513	 * to block commands that shouldn't proceed.  Another alternative
5514	 * would be to put the copy in the queue in place of the original,
5515	 * and return the original back to the caller.  That could be
5516	 * slightly safer..
5517	 */
5518	if ((cdb->byte2 & SSS_IMMED)
5519	 && ((cdb->how & SSS_START) == 0)) {
5520		ctl_set_invalid_field(ctsio,
5521				      /*sks_valid*/ 1,
5522				      /*command*/ 1,
5523				      /*field*/ 1,
5524				      /*bit_valid*/ 1,
5525				      /*bit*/ 0);
5526		ctl_done((union ctl_io *)ctsio);
5527		return (CTL_RETVAL_COMPLETE);
5528	}
5529
5530	if ((lun->flags & CTL_LUN_PR_RESERVED)
5531	 && ((cdb->how & SSS_START)==0)) {
5532		uint32_t residx;
5533
5534		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5535		if (!lun->per_res[residx].registered
5536		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5537
5538			ctl_set_reservation_conflict(ctsio);
5539			ctl_done((union ctl_io *)ctsio);
5540			return (CTL_RETVAL_COMPLETE);
5541		}
5542	}
5543
5544	/*
5545	 * If there is no backend on this device, we can't start or stop
5546	 * it.  In theory we shouldn't get any start/stop commands in the
5547	 * first place at this level if the LUN doesn't have a backend.
5548	 * That should get stopped by the command decode code.
5549	 */
5550	if (lun->backend == NULL) {
5551		ctl_set_invalid_opcode(ctsio);
5552		ctl_done((union ctl_io *)ctsio);
5553		return (CTL_RETVAL_COMPLETE);
5554	}
5555
5556	/*
5557	 * XXX KDM Copan-specific offline behavior.
5558	 * Figure out a reasonable way to port this?
5559	 */
5560#ifdef NEEDTOPORT
5561	mtx_lock(&lun->lun_lock);
5562
5563	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5564	 && (lun->flags & CTL_LUN_OFFLINE)) {
5565		/*
5566		 * If the LUN is offline, and the on/offline bit isn't set,
5567		 * reject the start or stop.  Otherwise, let it through.
5568		 */
5569		mtx_unlock(&lun->lun_lock);
5570		ctl_set_lun_not_ready(ctsio);
5571		ctl_done((union ctl_io *)ctsio);
5572	} else {
5573		mtx_unlock(&lun->lun_lock);
5574#endif /* NEEDTOPORT */
5575		/*
5576		 * This could be a start or a stop when we're online,
5577		 * or a stop/offline or start/online.  A start or stop when
5578		 * we're offline is covered in the case above.
5579		 */
5580		/*
5581		 * In the non-immediate case, we send the request to
5582		 * the backend and return status to the user when
5583		 * it is done.
5584		 *
5585		 * In the immediate case, we allocate a new ctl_io
5586		 * to hold a copy of the request, and send that to
5587		 * the backend.  We then set good status on the
5588		 * user's request and return it immediately.
5589		 */
5590		if (cdb->byte2 & SSS_IMMED) {
5591			union ctl_io *new_io;
5592
5593			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5594			if (new_io == NULL) {
5595				ctl_set_busy(ctsio);
5596				ctl_done((union ctl_io *)ctsio);
5597			} else {
5598				ctl_copy_io((union ctl_io *)ctsio,
5599					    new_io);
5600				retval = lun->backend->config_write(new_io);
5601				ctl_set_success(ctsio);
5602				ctl_done((union ctl_io *)ctsio);
5603			}
5604		} else {
5605			retval = lun->backend->config_write(
5606				(union ctl_io *)ctsio);
5607		}
5608#ifdef NEEDTOPORT
5609	}
5610#endif
5611	return (retval);
5612}
5613
5614/*
5615 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5616 * we don't really do anything with the LBA and length fields if the user
5617 * passes them in.  Instead we'll just flush out the cache for the entire
5618 * LUN.
5619 */
5620int
5621ctl_sync_cache(struct ctl_scsiio *ctsio)
5622{
5623	struct ctl_lun *lun;
5624	struct ctl_softc *ctl_softc;
5625	uint64_t starting_lba;
5626	uint32_t block_count;
5627	int retval;
5628
5629	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5630
5631	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5632	ctl_softc = control_softc;
5633	retval = 0;
5634
5635	switch (ctsio->cdb[0]) {
5636	case SYNCHRONIZE_CACHE: {
5637		struct scsi_sync_cache *cdb;
5638		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5639
5640		starting_lba = scsi_4btoul(cdb->begin_lba);
5641		block_count = scsi_2btoul(cdb->lb_count);
5642		break;
5643	}
5644	case SYNCHRONIZE_CACHE_16: {
5645		struct scsi_sync_cache_16 *cdb;
5646		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5647
5648		starting_lba = scsi_8btou64(cdb->begin_lba);
5649		block_count = scsi_4btoul(cdb->lb_count);
5650		break;
5651	}
5652	default:
5653		ctl_set_invalid_opcode(ctsio);
5654		ctl_done((union ctl_io *)ctsio);
5655		goto bailout;
5656		break; /* NOTREACHED */
5657	}
5658
5659	/*
5660	 * We check the LBA and length, but don't do anything with them.
5661	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5662	 * get flushed.  This check will just help satisfy anyone who wants
5663	 * to see an error for an out of range LBA.
5664	 */
5665	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5666		ctl_set_lba_out_of_range(ctsio);
5667		ctl_done((union ctl_io *)ctsio);
5668		goto bailout;
5669	}
5670
5671	/*
5672	 * If this LUN has no backend, we can't flush the cache anyway.
5673	 */
5674	if (lun->backend == NULL) {
5675		ctl_set_invalid_opcode(ctsio);
5676		ctl_done((union ctl_io *)ctsio);
5677		goto bailout;
5678	}
5679
5680	/*
5681	 * Check to see whether we're configured to send the SYNCHRONIZE
5682	 * CACHE command directly to the back end.
5683	 */
5684	mtx_lock(&lun->lun_lock);
5685	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5686	 && (++(lun->sync_count) >= lun->sync_interval)) {
5687		lun->sync_count = 0;
5688		mtx_unlock(&lun->lun_lock);
5689		retval = lun->backend->config_write((union ctl_io *)ctsio);
5690	} else {
5691		mtx_unlock(&lun->lun_lock);
5692		ctl_set_success(ctsio);
5693		ctl_done((union ctl_io *)ctsio);
5694	}
5695
5696bailout:
5697
5698	return (retval);
5699}
5700
5701int
5702ctl_format(struct ctl_scsiio *ctsio)
5703{
5704	struct scsi_format *cdb;
5705	struct ctl_lun *lun;
5706	struct ctl_softc *ctl_softc;
5707	int length, defect_list_len;
5708
5709	CTL_DEBUG_PRINT(("ctl_format\n"));
5710
5711	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5712	ctl_softc = control_softc;
5713
5714	cdb = (struct scsi_format *)ctsio->cdb;
5715
5716	length = 0;
5717	if (cdb->byte2 & SF_FMTDATA) {
5718		if (cdb->byte2 & SF_LONGLIST)
5719			length = sizeof(struct scsi_format_header_long);
5720		else
5721			length = sizeof(struct scsi_format_header_short);
5722	}
5723
5724	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5725	 && (length > 0)) {
5726		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5727		ctsio->kern_data_len = length;
5728		ctsio->kern_total_len = length;
5729		ctsio->kern_data_resid = 0;
5730		ctsio->kern_rel_offset = 0;
5731		ctsio->kern_sg_entries = 0;
5732		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5733		ctsio->be_move_done = ctl_config_move_done;
5734		ctl_datamove((union ctl_io *)ctsio);
5735
5736		return (CTL_RETVAL_COMPLETE);
5737	}
5738
5739	defect_list_len = 0;
5740
5741	if (cdb->byte2 & SF_FMTDATA) {
5742		if (cdb->byte2 & SF_LONGLIST) {
5743			struct scsi_format_header_long *header;
5744
5745			header = (struct scsi_format_header_long *)
5746				ctsio->kern_data_ptr;
5747
5748			defect_list_len = scsi_4btoul(header->defect_list_len);
5749			if (defect_list_len != 0) {
5750				ctl_set_invalid_field(ctsio,
5751						      /*sks_valid*/ 1,
5752						      /*command*/ 0,
5753						      /*field*/ 2,
5754						      /*bit_valid*/ 0,
5755						      /*bit*/ 0);
5756				goto bailout;
5757			}
5758		} else {
5759			struct scsi_format_header_short *header;
5760
5761			header = (struct scsi_format_header_short *)
5762				ctsio->kern_data_ptr;
5763
5764			defect_list_len = scsi_2btoul(header->defect_list_len);
5765			if (defect_list_len != 0) {
5766				ctl_set_invalid_field(ctsio,
5767						      /*sks_valid*/ 1,
5768						      /*command*/ 0,
5769						      /*field*/ 2,
5770						      /*bit_valid*/ 0,
5771						      /*bit*/ 0);
5772				goto bailout;
5773			}
5774		}
5775	}
5776
5777	/*
5778	 * The format command will clear out the "Medium format corrupted"
5779	 * status if set by the configuration code.  That status is really
5780	 * just a way to notify the host that we have lost the media, and
5781	 * get them to issue a command that will basically make them think
5782	 * they're blowing away the media.
5783	 */
5784	mtx_lock(&lun->lun_lock);
5785	lun->flags &= ~CTL_LUN_INOPERABLE;
5786	mtx_unlock(&lun->lun_lock);
5787
5788	ctsio->scsi_status = SCSI_STATUS_OK;
5789	ctsio->io_hdr.status = CTL_SUCCESS;
5790bailout:
5791
5792	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5793		free(ctsio->kern_data_ptr, M_CTL);
5794		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5795	}
5796
5797	ctl_done((union ctl_io *)ctsio);
5798	return (CTL_RETVAL_COMPLETE);
5799}
5800
5801int
5802ctl_read_buffer(struct ctl_scsiio *ctsio)
5803{
5804	struct scsi_read_buffer *cdb;
5805	struct ctl_lun *lun;
5806	int buffer_offset, len;
5807	static uint8_t descr[4];
5808	static uint8_t echo_descr[4] = { 0 };
5809
5810	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5811
5812	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5813	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5814
5815	if (lun->flags & CTL_LUN_PR_RESERVED) {
5816		uint32_t residx;
5817
5818		/*
5819		 * XXX KDM need a lock here.
5820		 */
5821		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5822		if ((lun->res_type == SPR_TYPE_EX_AC
5823		  && residx != lun->pr_res_idx)
5824		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5825		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5826		  && !lun->per_res[residx].registered)) {
5827			ctl_set_reservation_conflict(ctsio);
5828			ctl_done((union ctl_io *)ctsio);
5829			return (CTL_RETVAL_COMPLETE);
5830	        }
5831	}
5832
5833	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5834	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5835	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5836		ctl_set_invalid_field(ctsio,
5837				      /*sks_valid*/ 1,
5838				      /*command*/ 1,
5839				      /*field*/ 1,
5840				      /*bit_valid*/ 1,
5841				      /*bit*/ 4);
5842		ctl_done((union ctl_io *)ctsio);
5843		return (CTL_RETVAL_COMPLETE);
5844	}
5845
5846	len = scsi_3btoul(cdb->length);
5847	buffer_offset = scsi_3btoul(cdb->offset);
5848
5849	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5850		ctl_set_invalid_field(ctsio,
5851				      /*sks_valid*/ 1,
5852				      /*command*/ 1,
5853				      /*field*/ 6,
5854				      /*bit_valid*/ 0,
5855				      /*bit*/ 0);
5856		ctl_done((union ctl_io *)ctsio);
5857		return (CTL_RETVAL_COMPLETE);
5858	}
5859
5860	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5861		descr[0] = 0;
5862		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5863		ctsio->kern_data_ptr = descr;
5864		len = min(len, sizeof(descr));
5865	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5866		ctsio->kern_data_ptr = echo_descr;
5867		len = min(len, sizeof(echo_descr));
5868	} else
5869		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5870	ctsio->kern_data_len = len;
5871	ctsio->kern_total_len = len;
5872	ctsio->kern_data_resid = 0;
5873	ctsio->kern_rel_offset = 0;
5874	ctsio->kern_sg_entries = 0;
5875	ctsio->be_move_done = ctl_config_move_done;
5876	ctl_datamove((union ctl_io *)ctsio);
5877
5878	return (CTL_RETVAL_COMPLETE);
5879}
5880
5881int
5882ctl_write_buffer(struct ctl_scsiio *ctsio)
5883{
5884	struct scsi_write_buffer *cdb;
5885	struct ctl_lun *lun;
5886	int buffer_offset, len;
5887
5888	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5889
5890	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5891	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5892
5893	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5894		ctl_set_invalid_field(ctsio,
5895				      /*sks_valid*/ 1,
5896				      /*command*/ 1,
5897				      /*field*/ 1,
5898				      /*bit_valid*/ 1,
5899				      /*bit*/ 4);
5900		ctl_done((union ctl_io *)ctsio);
5901		return (CTL_RETVAL_COMPLETE);
5902	}
5903
5904	len = scsi_3btoul(cdb->length);
5905	buffer_offset = scsi_3btoul(cdb->offset);
5906
5907	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5908		ctl_set_invalid_field(ctsio,
5909				      /*sks_valid*/ 1,
5910				      /*command*/ 1,
5911				      /*field*/ 6,
5912				      /*bit_valid*/ 0,
5913				      /*bit*/ 0);
5914		ctl_done((union ctl_io *)ctsio);
5915		return (CTL_RETVAL_COMPLETE);
5916	}
5917
5918	/*
5919	 * If we've got a kernel request that hasn't been malloced yet,
5920	 * malloc it and tell the caller the data buffer is here.
5921	 */
5922	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5923		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5924		ctsio->kern_data_len = len;
5925		ctsio->kern_total_len = len;
5926		ctsio->kern_data_resid = 0;
5927		ctsio->kern_rel_offset = 0;
5928		ctsio->kern_sg_entries = 0;
5929		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5930		ctsio->be_move_done = ctl_config_move_done;
5931		ctl_datamove((union ctl_io *)ctsio);
5932
5933		return (CTL_RETVAL_COMPLETE);
5934	}
5935
5936	ctl_done((union ctl_io *)ctsio);
5937
5938	return (CTL_RETVAL_COMPLETE);
5939}
5940
5941int
5942ctl_write_same(struct ctl_scsiio *ctsio)
5943{
5944	struct ctl_lun *lun;
5945	struct ctl_lba_len_flags *lbalen;
5946	uint64_t lba;
5947	uint32_t num_blocks;
5948	int len, retval;
5949	uint8_t byte2;
5950
5951	retval = CTL_RETVAL_COMPLETE;
5952
5953	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5954
5955	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5956
5957	switch (ctsio->cdb[0]) {
5958	case WRITE_SAME_10: {
5959		struct scsi_write_same_10 *cdb;
5960
5961		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5962
5963		lba = scsi_4btoul(cdb->addr);
5964		num_blocks = scsi_2btoul(cdb->length);
5965		byte2 = cdb->byte2;
5966		break;
5967	}
5968	case WRITE_SAME_16: {
5969		struct scsi_write_same_16 *cdb;
5970
5971		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5972
5973		lba = scsi_8btou64(cdb->addr);
5974		num_blocks = scsi_4btoul(cdb->length);
5975		byte2 = cdb->byte2;
5976		break;
5977	}
5978	default:
5979		/*
5980		 * We got a command we don't support.  This shouldn't
5981		 * happen, commands should be filtered out above us.
5982		 */
5983		ctl_set_invalid_opcode(ctsio);
5984		ctl_done((union ctl_io *)ctsio);
5985
5986		return (CTL_RETVAL_COMPLETE);
5987		break; /* NOTREACHED */
5988	}
5989
5990	/* NDOB flag can be used only together with UNMAP */
5991	if ((byte2 & (SWS_NDOB | SWS_UNMAP)) == SWS_NDOB) {
5992		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5993		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5994		ctl_done((union ctl_io *)ctsio);
5995		return (CTL_RETVAL_COMPLETE);
5996	}
5997
5998	/*
5999	 * The first check is to make sure we're in bounds, the second
6000	 * check is to catch wrap-around problems.  If the lba + num blocks
6001	 * is less than the lba, then we've wrapped around and the block
6002	 * range is invalid anyway.
6003	 */
6004	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6005	 || ((lba + num_blocks) < lba)) {
6006		ctl_set_lba_out_of_range(ctsio);
6007		ctl_done((union ctl_io *)ctsio);
6008		return (CTL_RETVAL_COMPLETE);
6009	}
6010
6011	/* Zero number of blocks means "to the last logical block" */
6012	if (num_blocks == 0) {
6013		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
6014			ctl_set_invalid_field(ctsio,
6015					      /*sks_valid*/ 0,
6016					      /*command*/ 1,
6017					      /*field*/ 0,
6018					      /*bit_valid*/ 0,
6019					      /*bit*/ 0);
6020			ctl_done((union ctl_io *)ctsio);
6021			return (CTL_RETVAL_COMPLETE);
6022		}
6023		num_blocks = (lun->be_lun->maxlba + 1) - lba;
6024	}
6025
6026	len = lun->be_lun->blocksize;
6027
6028	/*
6029	 * If we've got a kernel request that hasn't been malloced yet,
6030	 * malloc it and tell the caller the data buffer is here.
6031	 */
6032	if ((byte2 & SWS_NDOB) == 0 &&
6033	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6034		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6035		ctsio->kern_data_len = len;
6036		ctsio->kern_total_len = len;
6037		ctsio->kern_data_resid = 0;
6038		ctsio->kern_rel_offset = 0;
6039		ctsio->kern_sg_entries = 0;
6040		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6041		ctsio->be_move_done = ctl_config_move_done;
6042		ctl_datamove((union ctl_io *)ctsio);
6043
6044		return (CTL_RETVAL_COMPLETE);
6045	}
6046
6047	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6048	lbalen->lba = lba;
6049	lbalen->len = num_blocks;
6050	lbalen->flags = byte2;
6051	retval = lun->backend->config_write((union ctl_io *)ctsio);
6052
6053	return (retval);
6054}
6055
6056int
6057ctl_unmap(struct ctl_scsiio *ctsio)
6058{
6059	struct ctl_lun *lun;
6060	struct scsi_unmap *cdb;
6061	struct ctl_ptr_len_flags *ptrlen;
6062	struct scsi_unmap_header *hdr;
6063	struct scsi_unmap_desc *buf, *end, *endnz, *range;
6064	uint64_t lba;
6065	uint32_t num_blocks;
6066	int len, retval;
6067	uint8_t byte2;
6068
6069	retval = CTL_RETVAL_COMPLETE;
6070
6071	CTL_DEBUG_PRINT(("ctl_unmap\n"));
6072
6073	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6074	cdb = (struct scsi_unmap *)ctsio->cdb;
6075
6076	len = scsi_2btoul(cdb->length);
6077	byte2 = cdb->byte2;
6078
6079	/*
6080	 * If we've got a kernel request that hasn't been malloced yet,
6081	 * malloc it and tell the caller the data buffer is here.
6082	 */
6083	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6084		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6085		ctsio->kern_data_len = len;
6086		ctsio->kern_total_len = len;
6087		ctsio->kern_data_resid = 0;
6088		ctsio->kern_rel_offset = 0;
6089		ctsio->kern_sg_entries = 0;
6090		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6091		ctsio->be_move_done = ctl_config_move_done;
6092		ctl_datamove((union ctl_io *)ctsio);
6093
6094		return (CTL_RETVAL_COMPLETE);
6095	}
6096
6097	len = ctsio->kern_total_len - ctsio->kern_data_resid;
6098	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6099	if (len < sizeof (*hdr) ||
6100	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6101	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6102	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6103		ctl_set_invalid_field(ctsio,
6104				      /*sks_valid*/ 0,
6105				      /*command*/ 0,
6106				      /*field*/ 0,
6107				      /*bit_valid*/ 0,
6108				      /*bit*/ 0);
6109		ctl_done((union ctl_io *)ctsio);
6110		return (CTL_RETVAL_COMPLETE);
6111	}
6112	len = scsi_2btoul(hdr->desc_length);
6113	buf = (struct scsi_unmap_desc *)(hdr + 1);
6114	end = buf + len / sizeof(*buf);
6115
6116	endnz = buf;
6117	for (range = buf; range < end; range++) {
6118		lba = scsi_8btou64(range->lba);
6119		num_blocks = scsi_4btoul(range->length);
6120		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6121		 || ((lba + num_blocks) < lba)) {
6122			ctl_set_lba_out_of_range(ctsio);
6123			ctl_done((union ctl_io *)ctsio);
6124			return (CTL_RETVAL_COMPLETE);
6125		}
6126		if (num_blocks != 0)
6127			endnz = range + 1;
6128	}
6129
6130	/*
6131	 * Block backend can not handle zero last range.
6132	 * Filter it out and return if there is nothing left.
6133	 */
6134	len = (uint8_t *)endnz - (uint8_t *)buf;
6135	if (len == 0) {
6136		ctl_set_success(ctsio);
6137		ctl_done((union ctl_io *)ctsio);
6138		return (CTL_RETVAL_COMPLETE);
6139	}
6140
6141	mtx_lock(&lun->lun_lock);
6142	ptrlen = (struct ctl_ptr_len_flags *)
6143	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6144	ptrlen->ptr = (void *)buf;
6145	ptrlen->len = len;
6146	ptrlen->flags = byte2;
6147	ctl_check_blocked(lun);
6148	mtx_unlock(&lun->lun_lock);
6149
6150	retval = lun->backend->config_write((union ctl_io *)ctsio);
6151	return (retval);
6152}
6153
6154/*
6155 * Note that this function currently doesn't actually do anything inside
6156 * CTL to enforce things if the DQue bit is turned on.
6157 *
6158 * Also note that this function can't be used in the default case, because
6159 * the DQue bit isn't set in the changeable mask for the control mode page
6160 * anyway.  This is just here as an example for how to implement a page
6161 * handler, and a placeholder in case we want to allow the user to turn
6162 * tagged queueing on and off.
6163 *
6164 * The D_SENSE bit handling is functional, however, and will turn
6165 * descriptor sense on and off for a given LUN.
6166 */
6167int
6168ctl_control_page_handler(struct ctl_scsiio *ctsio,
6169			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6170{
6171	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6172	struct ctl_lun *lun;
6173	struct ctl_softc *softc;
6174	int set_ua;
6175	uint32_t initidx;
6176
6177	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6178	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6179	set_ua = 0;
6180
6181	user_cp = (struct scsi_control_page *)page_ptr;
6182	current_cp = (struct scsi_control_page *)
6183		(page_index->page_data + (page_index->page_len *
6184		CTL_PAGE_CURRENT));
6185	saved_cp = (struct scsi_control_page *)
6186		(page_index->page_data + (page_index->page_len *
6187		CTL_PAGE_SAVED));
6188
6189	softc = control_softc;
6190
6191	mtx_lock(&lun->lun_lock);
6192	if (((current_cp->rlec & SCP_DSENSE) == 0)
6193	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6194		/*
6195		 * Descriptor sense is currently turned off and the user
6196		 * wants to turn it on.
6197		 */
6198		current_cp->rlec |= SCP_DSENSE;
6199		saved_cp->rlec |= SCP_DSENSE;
6200		lun->flags |= CTL_LUN_SENSE_DESC;
6201		set_ua = 1;
6202	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6203		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6204		/*
6205		 * Descriptor sense is currently turned on, and the user
6206		 * wants to turn it off.
6207		 */
6208		current_cp->rlec &= ~SCP_DSENSE;
6209		saved_cp->rlec &= ~SCP_DSENSE;
6210		lun->flags &= ~CTL_LUN_SENSE_DESC;
6211		set_ua = 1;
6212	}
6213	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6214	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6215		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6216		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6217		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6218		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6219		set_ua = 1;
6220	}
6221	if (set_ua != 0) {
6222		int i;
6223		/*
6224		 * Let other initiators know that the mode
6225		 * parameters for this LUN have changed.
6226		 */
6227		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6228			if (i == initidx)
6229				continue;
6230
6231			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6232		}
6233	}
6234	mtx_unlock(&lun->lun_lock);
6235
6236	return (0);
6237}
6238
6239int
6240ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6241		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6242{
6243	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6244	struct ctl_lun *lun;
6245	int set_ua;
6246	uint32_t initidx;
6247
6248	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6249	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6250	set_ua = 0;
6251
6252	user_cp = (struct scsi_caching_page *)page_ptr;
6253	current_cp = (struct scsi_caching_page *)
6254		(page_index->page_data + (page_index->page_len *
6255		CTL_PAGE_CURRENT));
6256	saved_cp = (struct scsi_caching_page *)
6257		(page_index->page_data + (page_index->page_len *
6258		CTL_PAGE_SAVED));
6259
6260	mtx_lock(&lun->lun_lock);
6261	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6262	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6263		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6264		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6265		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6266		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6267		set_ua = 1;
6268	}
6269	if (set_ua != 0) {
6270		int i;
6271		/*
6272		 * Let other initiators know that the mode
6273		 * parameters for this LUN have changed.
6274		 */
6275		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6276			if (i == initidx)
6277				continue;
6278
6279			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6280		}
6281	}
6282	mtx_unlock(&lun->lun_lock);
6283
6284	return (0);
6285}
6286
6287int
6288ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6289		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6290{
6291	return (0);
6292}
6293
6294int
6295ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6296			   struct ctl_page_index *page_index, int pc)
6297{
6298	struct copan_power_subpage *page;
6299
6300	page = (struct copan_power_subpage *)page_index->page_data +
6301		(page_index->page_len * pc);
6302
6303	switch (pc) {
6304	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6305		/*
6306		 * We don't update the changable bits for this page.
6307		 */
6308		break;
6309	case SMS_PAGE_CTRL_CURRENT >> 6:
6310	case SMS_PAGE_CTRL_DEFAULT >> 6:
6311	case SMS_PAGE_CTRL_SAVED >> 6:
6312#ifdef NEEDTOPORT
6313		ctl_update_power_subpage(page);
6314#endif
6315		break;
6316	default:
6317#ifdef NEEDTOPORT
6318		EPRINT(0, "Invalid PC %d!!", pc);
6319#endif
6320		break;
6321	}
6322	return (0);
6323}
6324
6325
6326int
6327ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6328		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6329{
6330	struct copan_aps_subpage *user_sp;
6331	struct copan_aps_subpage *current_sp;
6332	union ctl_modepage_info *modepage_info;
6333	struct ctl_softc *softc;
6334	struct ctl_lun *lun;
6335	int retval;
6336
6337	retval = CTL_RETVAL_COMPLETE;
6338	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6339		     (page_index->page_len * CTL_PAGE_CURRENT));
6340	softc = control_softc;
6341	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6342
6343	user_sp = (struct copan_aps_subpage *)page_ptr;
6344
6345	modepage_info = (union ctl_modepage_info *)
6346		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6347
6348	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6349	modepage_info->header.subpage = page_index->subpage;
6350	modepage_info->aps.lock_active = user_sp->lock_active;
6351
6352	mtx_lock(&softc->ctl_lock);
6353
6354	/*
6355	 * If there is a request to lock the LUN and another LUN is locked
6356	 * this is an error. If the requested LUN is already locked ignore
6357	 * the request. If no LUN is locked attempt to lock it.
6358	 * if there is a request to unlock the LUN and the LUN is currently
6359	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6360	 * if another LUN is locked or no LUN is locked.
6361	 */
6362	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6363		if (softc->aps_locked_lun == lun->lun) {
6364			/*
6365			 * This LUN is already locked, so we're done.
6366			 */
6367			retval = CTL_RETVAL_COMPLETE;
6368		} else if (softc->aps_locked_lun == 0) {
6369			/*
6370			 * No one has the lock, pass the request to the
6371			 * backend.
6372			 */
6373			retval = lun->backend->config_write(
6374				(union ctl_io *)ctsio);
6375		} else {
6376			/*
6377			 * Someone else has the lock, throw out the request.
6378			 */
6379			ctl_set_already_locked(ctsio);
6380			free(ctsio->kern_data_ptr, M_CTL);
6381			ctl_done((union ctl_io *)ctsio);
6382
6383			/*
6384			 * Set the return value so that ctl_do_mode_select()
6385			 * won't try to complete the command.  We already
6386			 * completed it here.
6387			 */
6388			retval = CTL_RETVAL_ERROR;
6389		}
6390	} else if (softc->aps_locked_lun == lun->lun) {
6391		/*
6392		 * This LUN is locked, so pass the unlock request to the
6393		 * backend.
6394		 */
6395		retval = lun->backend->config_write((union ctl_io *)ctsio);
6396	}
6397	mtx_unlock(&softc->ctl_lock);
6398
6399	return (retval);
6400}
6401
6402int
6403ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6404				struct ctl_page_index *page_index,
6405				uint8_t *page_ptr)
6406{
6407	uint8_t *c;
6408	int i;
6409
6410	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6411	ctl_time_io_secs =
6412		(c[0] << 8) |
6413		(c[1] << 0) |
6414		0;
6415	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6416	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6417	printf("page data:");
6418	for (i=0; i<8; i++)
6419		printf(" %.2x",page_ptr[i]);
6420	printf("\n");
6421	return (0);
6422}
6423
6424int
6425ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6426			       struct ctl_page_index *page_index,
6427			       int pc)
6428{
6429	struct copan_debugconf_subpage *page;
6430
6431	page = (struct copan_debugconf_subpage *)page_index->page_data +
6432		(page_index->page_len * pc);
6433
6434	switch (pc) {
6435	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6436	case SMS_PAGE_CTRL_DEFAULT >> 6:
6437	case SMS_PAGE_CTRL_SAVED >> 6:
6438		/*
6439		 * We don't update the changable or default bits for this page.
6440		 */
6441		break;
6442	case SMS_PAGE_CTRL_CURRENT >> 6:
6443		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6444		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6445		break;
6446	default:
6447#ifdef NEEDTOPORT
6448		EPRINT(0, "Invalid PC %d!!", pc);
6449#endif /* NEEDTOPORT */
6450		break;
6451	}
6452	return (0);
6453}
6454
6455
6456static int
6457ctl_do_mode_select(union ctl_io *io)
6458{
6459	struct scsi_mode_page_header *page_header;
6460	struct ctl_page_index *page_index;
6461	struct ctl_scsiio *ctsio;
6462	int control_dev, page_len;
6463	int page_len_offset, page_len_size;
6464	union ctl_modepage_info *modepage_info;
6465	struct ctl_lun *lun;
6466	int *len_left, *len_used;
6467	int retval, i;
6468
6469	ctsio = &io->scsiio;
6470	page_index = NULL;
6471	page_len = 0;
6472	retval = CTL_RETVAL_COMPLETE;
6473
6474	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6475
6476	if (lun->be_lun->lun_type != T_DIRECT)
6477		control_dev = 1;
6478	else
6479		control_dev = 0;
6480
6481	modepage_info = (union ctl_modepage_info *)
6482		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6483	len_left = &modepage_info->header.len_left;
6484	len_used = &modepage_info->header.len_used;
6485
6486do_next_page:
6487
6488	page_header = (struct scsi_mode_page_header *)
6489		(ctsio->kern_data_ptr + *len_used);
6490
6491	if (*len_left == 0) {
6492		free(ctsio->kern_data_ptr, M_CTL);
6493		ctl_set_success(ctsio);
6494		ctl_done((union ctl_io *)ctsio);
6495		return (CTL_RETVAL_COMPLETE);
6496	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6497
6498		free(ctsio->kern_data_ptr, M_CTL);
6499		ctl_set_param_len_error(ctsio);
6500		ctl_done((union ctl_io *)ctsio);
6501		return (CTL_RETVAL_COMPLETE);
6502
6503	} else if ((page_header->page_code & SMPH_SPF)
6504		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6505
6506		free(ctsio->kern_data_ptr, M_CTL);
6507		ctl_set_param_len_error(ctsio);
6508		ctl_done((union ctl_io *)ctsio);
6509		return (CTL_RETVAL_COMPLETE);
6510	}
6511
6512
6513	/*
6514	 * XXX KDM should we do something with the block descriptor?
6515	 */
6516	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6517
6518		if ((control_dev != 0)
6519		 && (lun->mode_pages.index[i].page_flags &
6520		     CTL_PAGE_FLAG_DISK_ONLY))
6521			continue;
6522
6523		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6524		    (page_header->page_code & SMPH_PC_MASK))
6525			continue;
6526
6527		/*
6528		 * If neither page has a subpage code, then we've got a
6529		 * match.
6530		 */
6531		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6532		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6533			page_index = &lun->mode_pages.index[i];
6534			page_len = page_header->page_length;
6535			break;
6536		}
6537
6538		/*
6539		 * If both pages have subpages, then the subpage numbers
6540		 * have to match.
6541		 */
6542		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6543		  && (page_header->page_code & SMPH_SPF)) {
6544			struct scsi_mode_page_header_sp *sph;
6545
6546			sph = (struct scsi_mode_page_header_sp *)page_header;
6547
6548			if (lun->mode_pages.index[i].subpage ==
6549			    sph->subpage) {
6550				page_index = &lun->mode_pages.index[i];
6551				page_len = scsi_2btoul(sph->page_length);
6552				break;
6553			}
6554		}
6555	}
6556
6557	/*
6558	 * If we couldn't find the page, or if we don't have a mode select
6559	 * handler for it, send back an error to the user.
6560	 */
6561	if ((page_index == NULL)
6562	 || (page_index->select_handler == NULL)) {
6563		ctl_set_invalid_field(ctsio,
6564				      /*sks_valid*/ 1,
6565				      /*command*/ 0,
6566				      /*field*/ *len_used,
6567				      /*bit_valid*/ 0,
6568				      /*bit*/ 0);
6569		free(ctsio->kern_data_ptr, M_CTL);
6570		ctl_done((union ctl_io *)ctsio);
6571		return (CTL_RETVAL_COMPLETE);
6572	}
6573
6574	if (page_index->page_code & SMPH_SPF) {
6575		page_len_offset = 2;
6576		page_len_size = 2;
6577	} else {
6578		page_len_size = 1;
6579		page_len_offset = 1;
6580	}
6581
6582	/*
6583	 * If the length the initiator gives us isn't the one we specify in
6584	 * the mode page header, or if they didn't specify enough data in
6585	 * the CDB to avoid truncating this page, kick out the request.
6586	 */
6587	if ((page_len != (page_index->page_len - page_len_offset -
6588			  page_len_size))
6589	 || (*len_left < page_index->page_len)) {
6590
6591
6592		ctl_set_invalid_field(ctsio,
6593				      /*sks_valid*/ 1,
6594				      /*command*/ 0,
6595				      /*field*/ *len_used + page_len_offset,
6596				      /*bit_valid*/ 0,
6597				      /*bit*/ 0);
6598		free(ctsio->kern_data_ptr, M_CTL);
6599		ctl_done((union ctl_io *)ctsio);
6600		return (CTL_RETVAL_COMPLETE);
6601	}
6602
6603	/*
6604	 * Run through the mode page, checking to make sure that the bits
6605	 * the user changed are actually legal for him to change.
6606	 */
6607	for (i = 0; i < page_index->page_len; i++) {
6608		uint8_t *user_byte, *change_mask, *current_byte;
6609		int bad_bit;
6610		int j;
6611
6612		user_byte = (uint8_t *)page_header + i;
6613		change_mask = page_index->page_data +
6614			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6615		current_byte = page_index->page_data +
6616			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6617
6618		/*
6619		 * Check to see whether the user set any bits in this byte
6620		 * that he is not allowed to set.
6621		 */
6622		if ((*user_byte & ~(*change_mask)) ==
6623		    (*current_byte & ~(*change_mask)))
6624			continue;
6625
6626		/*
6627		 * Go through bit by bit to determine which one is illegal.
6628		 */
6629		bad_bit = 0;
6630		for (j = 7; j >= 0; j--) {
6631			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6632			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6633				bad_bit = i;
6634				break;
6635			}
6636		}
6637		ctl_set_invalid_field(ctsio,
6638				      /*sks_valid*/ 1,
6639				      /*command*/ 0,
6640				      /*field*/ *len_used + i,
6641				      /*bit_valid*/ 1,
6642				      /*bit*/ bad_bit);
6643		free(ctsio->kern_data_ptr, M_CTL);
6644		ctl_done((union ctl_io *)ctsio);
6645		return (CTL_RETVAL_COMPLETE);
6646	}
6647
6648	/*
6649	 * Decrement these before we call the page handler, since we may
6650	 * end up getting called back one way or another before the handler
6651	 * returns to this context.
6652	 */
6653	*len_left -= page_index->page_len;
6654	*len_used += page_index->page_len;
6655
6656	retval = page_index->select_handler(ctsio, page_index,
6657					    (uint8_t *)page_header);
6658
6659	/*
6660	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6661	 * wait until this queued command completes to finish processing
6662	 * the mode page.  If it returns anything other than
6663	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6664	 * already set the sense information, freed the data pointer, and
6665	 * completed the io for us.
6666	 */
6667	if (retval != CTL_RETVAL_COMPLETE)
6668		goto bailout_no_done;
6669
6670	/*
6671	 * If the initiator sent us more than one page, parse the next one.
6672	 */
6673	if (*len_left > 0)
6674		goto do_next_page;
6675
6676	ctl_set_success(ctsio);
6677	free(ctsio->kern_data_ptr, M_CTL);
6678	ctl_done((union ctl_io *)ctsio);
6679
6680bailout_no_done:
6681
6682	return (CTL_RETVAL_COMPLETE);
6683
6684}
6685
6686int
6687ctl_mode_select(struct ctl_scsiio *ctsio)
6688{
6689	int param_len, pf, sp;
6690	int header_size, bd_len;
6691	int len_left, len_used;
6692	struct ctl_page_index *page_index;
6693	struct ctl_lun *lun;
6694	int control_dev, page_len;
6695	union ctl_modepage_info *modepage_info;
6696	int retval;
6697
6698	pf = 0;
6699	sp = 0;
6700	page_len = 0;
6701	len_used = 0;
6702	len_left = 0;
6703	retval = 0;
6704	bd_len = 0;
6705	page_index = NULL;
6706
6707	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6708
6709	if (lun->be_lun->lun_type != T_DIRECT)
6710		control_dev = 1;
6711	else
6712		control_dev = 0;
6713
6714	switch (ctsio->cdb[0]) {
6715	case MODE_SELECT_6: {
6716		struct scsi_mode_select_6 *cdb;
6717
6718		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6719
6720		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6721		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6722
6723		param_len = cdb->length;
6724		header_size = sizeof(struct scsi_mode_header_6);
6725		break;
6726	}
6727	case MODE_SELECT_10: {
6728		struct scsi_mode_select_10 *cdb;
6729
6730		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6731
6732		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6733		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6734
6735		param_len = scsi_2btoul(cdb->length);
6736		header_size = sizeof(struct scsi_mode_header_10);
6737		break;
6738	}
6739	default:
6740		ctl_set_invalid_opcode(ctsio);
6741		ctl_done((union ctl_io *)ctsio);
6742		return (CTL_RETVAL_COMPLETE);
6743		break; /* NOTREACHED */
6744	}
6745
6746	/*
6747	 * From SPC-3:
6748	 * "A parameter list length of zero indicates that the Data-Out Buffer
6749	 * shall be empty. This condition shall not be considered as an error."
6750	 */
6751	if (param_len == 0) {
6752		ctl_set_success(ctsio);
6753		ctl_done((union ctl_io *)ctsio);
6754		return (CTL_RETVAL_COMPLETE);
6755	}
6756
6757	/*
6758	 * Since we'll hit this the first time through, prior to
6759	 * allocation, we don't need to free a data buffer here.
6760	 */
6761	if (param_len < header_size) {
6762		ctl_set_param_len_error(ctsio);
6763		ctl_done((union ctl_io *)ctsio);
6764		return (CTL_RETVAL_COMPLETE);
6765	}
6766
6767	/*
6768	 * Allocate the data buffer and grab the user's data.  In theory,
6769	 * we shouldn't have to sanity check the parameter list length here
6770	 * because the maximum size is 64K.  We should be able to malloc
6771	 * that much without too many problems.
6772	 */
6773	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6774		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6775		ctsio->kern_data_len = param_len;
6776		ctsio->kern_total_len = param_len;
6777		ctsio->kern_data_resid = 0;
6778		ctsio->kern_rel_offset = 0;
6779		ctsio->kern_sg_entries = 0;
6780		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6781		ctsio->be_move_done = ctl_config_move_done;
6782		ctl_datamove((union ctl_io *)ctsio);
6783
6784		return (CTL_RETVAL_COMPLETE);
6785	}
6786
6787	switch (ctsio->cdb[0]) {
6788	case MODE_SELECT_6: {
6789		struct scsi_mode_header_6 *mh6;
6790
6791		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6792		bd_len = mh6->blk_desc_len;
6793		break;
6794	}
6795	case MODE_SELECT_10: {
6796		struct scsi_mode_header_10 *mh10;
6797
6798		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6799		bd_len = scsi_2btoul(mh10->blk_desc_len);
6800		break;
6801	}
6802	default:
6803		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6804		break;
6805	}
6806
6807	if (param_len < (header_size + bd_len)) {
6808		free(ctsio->kern_data_ptr, M_CTL);
6809		ctl_set_param_len_error(ctsio);
6810		ctl_done((union ctl_io *)ctsio);
6811		return (CTL_RETVAL_COMPLETE);
6812	}
6813
6814	/*
6815	 * Set the IO_CONT flag, so that if this I/O gets passed to
6816	 * ctl_config_write_done(), it'll get passed back to
6817	 * ctl_do_mode_select() for further processing, or completion if
6818	 * we're all done.
6819	 */
6820	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6821	ctsio->io_cont = ctl_do_mode_select;
6822
6823	modepage_info = (union ctl_modepage_info *)
6824		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6825
6826	memset(modepage_info, 0, sizeof(*modepage_info));
6827
6828	len_left = param_len - header_size - bd_len;
6829	len_used = header_size + bd_len;
6830
6831	modepage_info->header.len_left = len_left;
6832	modepage_info->header.len_used = len_used;
6833
6834	return (ctl_do_mode_select((union ctl_io *)ctsio));
6835}
6836
6837int
6838ctl_mode_sense(struct ctl_scsiio *ctsio)
6839{
6840	struct ctl_lun *lun;
6841	int pc, page_code, dbd, llba, subpage;
6842	int alloc_len, page_len, header_len, total_len;
6843	struct scsi_mode_block_descr *block_desc;
6844	struct ctl_page_index *page_index;
6845	int control_dev;
6846
6847	dbd = 0;
6848	llba = 0;
6849	block_desc = NULL;
6850	page_index = NULL;
6851
6852	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6853
6854	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6855
6856	if (lun->be_lun->lun_type != T_DIRECT)
6857		control_dev = 1;
6858	else
6859		control_dev = 0;
6860
6861	if (lun->flags & CTL_LUN_PR_RESERVED) {
6862		uint32_t residx;
6863
6864		/*
6865		 * XXX KDM need a lock here.
6866		 */
6867		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6868		if ((lun->res_type == SPR_TYPE_EX_AC
6869		  && residx != lun->pr_res_idx)
6870		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6871		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6872		  && !lun->per_res[residx].registered)) {
6873			ctl_set_reservation_conflict(ctsio);
6874			ctl_done((union ctl_io *)ctsio);
6875			return (CTL_RETVAL_COMPLETE);
6876		}
6877	}
6878
6879	switch (ctsio->cdb[0]) {
6880	case MODE_SENSE_6: {
6881		struct scsi_mode_sense_6 *cdb;
6882
6883		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6884
6885		header_len = sizeof(struct scsi_mode_hdr_6);
6886		if (cdb->byte2 & SMS_DBD)
6887			dbd = 1;
6888		else
6889			header_len += sizeof(struct scsi_mode_block_descr);
6890
6891		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6892		page_code = cdb->page & SMS_PAGE_CODE;
6893		subpage = cdb->subpage;
6894		alloc_len = cdb->length;
6895		break;
6896	}
6897	case MODE_SENSE_10: {
6898		struct scsi_mode_sense_10 *cdb;
6899
6900		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6901
6902		header_len = sizeof(struct scsi_mode_hdr_10);
6903
6904		if (cdb->byte2 & SMS_DBD)
6905			dbd = 1;
6906		else
6907			header_len += sizeof(struct scsi_mode_block_descr);
6908		if (cdb->byte2 & SMS10_LLBAA)
6909			llba = 1;
6910		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6911		page_code = cdb->page & SMS_PAGE_CODE;
6912		subpage = cdb->subpage;
6913		alloc_len = scsi_2btoul(cdb->length);
6914		break;
6915	}
6916	default:
6917		ctl_set_invalid_opcode(ctsio);
6918		ctl_done((union ctl_io *)ctsio);
6919		return (CTL_RETVAL_COMPLETE);
6920		break; /* NOTREACHED */
6921	}
6922
6923	/*
6924	 * We have to make a first pass through to calculate the size of
6925	 * the pages that match the user's query.  Then we allocate enough
6926	 * memory to hold it, and actually copy the data into the buffer.
6927	 */
6928	switch (page_code) {
6929	case SMS_ALL_PAGES_PAGE: {
6930		int i;
6931
6932		page_len = 0;
6933
6934		/*
6935		 * At the moment, values other than 0 and 0xff here are
6936		 * reserved according to SPC-3.
6937		 */
6938		if ((subpage != SMS_SUBPAGE_PAGE_0)
6939		 && (subpage != SMS_SUBPAGE_ALL)) {
6940			ctl_set_invalid_field(ctsio,
6941					      /*sks_valid*/ 1,
6942					      /*command*/ 1,
6943					      /*field*/ 3,
6944					      /*bit_valid*/ 0,
6945					      /*bit*/ 0);
6946			ctl_done((union ctl_io *)ctsio);
6947			return (CTL_RETVAL_COMPLETE);
6948		}
6949
6950		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6951			if ((control_dev != 0)
6952			 && (lun->mode_pages.index[i].page_flags &
6953			     CTL_PAGE_FLAG_DISK_ONLY))
6954				continue;
6955
6956			/*
6957			 * We don't use this subpage if the user didn't
6958			 * request all subpages.
6959			 */
6960			if ((lun->mode_pages.index[i].subpage != 0)
6961			 && (subpage == SMS_SUBPAGE_PAGE_0))
6962				continue;
6963
6964#if 0
6965			printf("found page %#x len %d\n",
6966			       lun->mode_pages.index[i].page_code &
6967			       SMPH_PC_MASK,
6968			       lun->mode_pages.index[i].page_len);
6969#endif
6970			page_len += lun->mode_pages.index[i].page_len;
6971		}
6972		break;
6973	}
6974	default: {
6975		int i;
6976
6977		page_len = 0;
6978
6979		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6980			/* Look for the right page code */
6981			if ((lun->mode_pages.index[i].page_code &
6982			     SMPH_PC_MASK) != page_code)
6983				continue;
6984
6985			/* Look for the right subpage or the subpage wildcard*/
6986			if ((lun->mode_pages.index[i].subpage != subpage)
6987			 && (subpage != SMS_SUBPAGE_ALL))
6988				continue;
6989
6990			/* Make sure the page is supported for this dev type */
6991			if ((control_dev != 0)
6992			 && (lun->mode_pages.index[i].page_flags &
6993			     CTL_PAGE_FLAG_DISK_ONLY))
6994				continue;
6995
6996#if 0
6997			printf("found page %#x len %d\n",
6998			       lun->mode_pages.index[i].page_code &
6999			       SMPH_PC_MASK,
7000			       lun->mode_pages.index[i].page_len);
7001#endif
7002
7003			page_len += lun->mode_pages.index[i].page_len;
7004		}
7005
7006		if (page_len == 0) {
7007			ctl_set_invalid_field(ctsio,
7008					      /*sks_valid*/ 1,
7009					      /*command*/ 1,
7010					      /*field*/ 2,
7011					      /*bit_valid*/ 1,
7012					      /*bit*/ 5);
7013			ctl_done((union ctl_io *)ctsio);
7014			return (CTL_RETVAL_COMPLETE);
7015		}
7016		break;
7017	}
7018	}
7019
7020	total_len = header_len + page_len;
7021#if 0
7022	printf("header_len = %d, page_len = %d, total_len = %d\n",
7023	       header_len, page_len, total_len);
7024#endif
7025
7026	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7027	ctsio->kern_sg_entries = 0;
7028	ctsio->kern_data_resid = 0;
7029	ctsio->kern_rel_offset = 0;
7030	if (total_len < alloc_len) {
7031		ctsio->residual = alloc_len - total_len;
7032		ctsio->kern_data_len = total_len;
7033		ctsio->kern_total_len = total_len;
7034	} else {
7035		ctsio->residual = 0;
7036		ctsio->kern_data_len = alloc_len;
7037		ctsio->kern_total_len = alloc_len;
7038	}
7039
7040	switch (ctsio->cdb[0]) {
7041	case MODE_SENSE_6: {
7042		struct scsi_mode_hdr_6 *header;
7043
7044		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7045
7046		header->datalen = ctl_min(total_len - 1, 254);
7047		if (control_dev == 0)
7048			header->dev_specific = 0x10; /* DPOFUA */
7049		if (dbd)
7050			header->block_descr_len = 0;
7051		else
7052			header->block_descr_len =
7053				sizeof(struct scsi_mode_block_descr);
7054		block_desc = (struct scsi_mode_block_descr *)&header[1];
7055		break;
7056	}
7057	case MODE_SENSE_10: {
7058		struct scsi_mode_hdr_10 *header;
7059		int datalen;
7060
7061		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7062
7063		datalen = ctl_min(total_len - 2, 65533);
7064		scsi_ulto2b(datalen, header->datalen);
7065		if (control_dev == 0)
7066			header->dev_specific = 0x10; /* DPOFUA */
7067		if (dbd)
7068			scsi_ulto2b(0, header->block_descr_len);
7069		else
7070			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7071				    header->block_descr_len);
7072		block_desc = (struct scsi_mode_block_descr *)&header[1];
7073		break;
7074	}
7075	default:
7076		panic("invalid CDB type %#x", ctsio->cdb[0]);
7077		break; /* NOTREACHED */
7078	}
7079
7080	/*
7081	 * If we've got a disk, use its blocksize in the block
7082	 * descriptor.  Otherwise, just set it to 0.
7083	 */
7084	if (dbd == 0) {
7085		if (control_dev == 0)
7086			scsi_ulto3b(lun->be_lun->blocksize,
7087				    block_desc->block_len);
7088		else
7089			scsi_ulto3b(0, block_desc->block_len);
7090	}
7091
7092	switch (page_code) {
7093	case SMS_ALL_PAGES_PAGE: {
7094		int i, data_used;
7095
7096		data_used = header_len;
7097		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7098			struct ctl_page_index *page_index;
7099
7100			page_index = &lun->mode_pages.index[i];
7101
7102			if ((control_dev != 0)
7103			 && (page_index->page_flags &
7104			    CTL_PAGE_FLAG_DISK_ONLY))
7105				continue;
7106
7107			/*
7108			 * We don't use this subpage if the user didn't
7109			 * request all subpages.  We already checked (above)
7110			 * to make sure the user only specified a subpage
7111			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7112			 */
7113			if ((page_index->subpage != 0)
7114			 && (subpage == SMS_SUBPAGE_PAGE_0))
7115				continue;
7116
7117			/*
7118			 * Call the handler, if it exists, to update the
7119			 * page to the latest values.
7120			 */
7121			if (page_index->sense_handler != NULL)
7122				page_index->sense_handler(ctsio, page_index,pc);
7123
7124			memcpy(ctsio->kern_data_ptr + data_used,
7125			       page_index->page_data +
7126			       (page_index->page_len * pc),
7127			       page_index->page_len);
7128			data_used += page_index->page_len;
7129		}
7130		break;
7131	}
7132	default: {
7133		int i, data_used;
7134
7135		data_used = header_len;
7136
7137		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7138			struct ctl_page_index *page_index;
7139
7140			page_index = &lun->mode_pages.index[i];
7141
7142			/* Look for the right page code */
7143			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7144				continue;
7145
7146			/* Look for the right subpage or the subpage wildcard*/
7147			if ((page_index->subpage != subpage)
7148			 && (subpage != SMS_SUBPAGE_ALL))
7149				continue;
7150
7151			/* Make sure the page is supported for this dev type */
7152			if ((control_dev != 0)
7153			 && (page_index->page_flags &
7154			     CTL_PAGE_FLAG_DISK_ONLY))
7155				continue;
7156
7157			/*
7158			 * Call the handler, if it exists, to update the
7159			 * page to the latest values.
7160			 */
7161			if (page_index->sense_handler != NULL)
7162				page_index->sense_handler(ctsio, page_index,pc);
7163
7164			memcpy(ctsio->kern_data_ptr + data_used,
7165			       page_index->page_data +
7166			       (page_index->page_len * pc),
7167			       page_index->page_len);
7168			data_used += page_index->page_len;
7169		}
7170		break;
7171	}
7172	}
7173
7174	ctsio->scsi_status = SCSI_STATUS_OK;
7175
7176	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7177	ctsio->be_move_done = ctl_config_move_done;
7178	ctl_datamove((union ctl_io *)ctsio);
7179
7180	return (CTL_RETVAL_COMPLETE);
7181}
7182
7183int
7184ctl_read_capacity(struct ctl_scsiio *ctsio)
7185{
7186	struct scsi_read_capacity *cdb;
7187	struct scsi_read_capacity_data *data;
7188	struct ctl_lun *lun;
7189	uint32_t lba;
7190
7191	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7192
7193	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7194
7195	lba = scsi_4btoul(cdb->addr);
7196	if (((cdb->pmi & SRC_PMI) == 0)
7197	 && (lba != 0)) {
7198		ctl_set_invalid_field(/*ctsio*/ ctsio,
7199				      /*sks_valid*/ 1,
7200				      /*command*/ 1,
7201				      /*field*/ 2,
7202				      /*bit_valid*/ 0,
7203				      /*bit*/ 0);
7204		ctl_done((union ctl_io *)ctsio);
7205		return (CTL_RETVAL_COMPLETE);
7206	}
7207
7208	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7209
7210	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7211	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7212	ctsio->residual = 0;
7213	ctsio->kern_data_len = sizeof(*data);
7214	ctsio->kern_total_len = sizeof(*data);
7215	ctsio->kern_data_resid = 0;
7216	ctsio->kern_rel_offset = 0;
7217	ctsio->kern_sg_entries = 0;
7218
7219	/*
7220	 * If the maximum LBA is greater than 0xfffffffe, the user must
7221	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7222	 * serivce action set.
7223	 */
7224	if (lun->be_lun->maxlba > 0xfffffffe)
7225		scsi_ulto4b(0xffffffff, data->addr);
7226	else
7227		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7228
7229	/*
7230	 * XXX KDM this may not be 512 bytes...
7231	 */
7232	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7233
7234	ctsio->scsi_status = SCSI_STATUS_OK;
7235
7236	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7237	ctsio->be_move_done = ctl_config_move_done;
7238	ctl_datamove((union ctl_io *)ctsio);
7239
7240	return (CTL_RETVAL_COMPLETE);
7241}
7242
7243int
7244ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7245{
7246	struct scsi_read_capacity_16 *cdb;
7247	struct scsi_read_capacity_data_long *data;
7248	struct ctl_lun *lun;
7249	uint64_t lba;
7250	uint32_t alloc_len;
7251
7252	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7253
7254	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7255
7256	alloc_len = scsi_4btoul(cdb->alloc_len);
7257	lba = scsi_8btou64(cdb->addr);
7258
7259	if ((cdb->reladr & SRC16_PMI)
7260	 && (lba != 0)) {
7261		ctl_set_invalid_field(/*ctsio*/ ctsio,
7262				      /*sks_valid*/ 1,
7263				      /*command*/ 1,
7264				      /*field*/ 2,
7265				      /*bit_valid*/ 0,
7266				      /*bit*/ 0);
7267		ctl_done((union ctl_io *)ctsio);
7268		return (CTL_RETVAL_COMPLETE);
7269	}
7270
7271	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7272
7273	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7274	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7275
7276	if (sizeof(*data) < alloc_len) {
7277		ctsio->residual = alloc_len - sizeof(*data);
7278		ctsio->kern_data_len = sizeof(*data);
7279		ctsio->kern_total_len = sizeof(*data);
7280	} else {
7281		ctsio->residual = 0;
7282		ctsio->kern_data_len = alloc_len;
7283		ctsio->kern_total_len = alloc_len;
7284	}
7285	ctsio->kern_data_resid = 0;
7286	ctsio->kern_rel_offset = 0;
7287	ctsio->kern_sg_entries = 0;
7288
7289	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7290	/* XXX KDM this may not be 512 bytes... */
7291	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7292	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7293	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7294	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7295		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7296
7297	ctsio->scsi_status = SCSI_STATUS_OK;
7298
7299	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7300	ctsio->be_move_done = ctl_config_move_done;
7301	ctl_datamove((union ctl_io *)ctsio);
7302
7303	return (CTL_RETVAL_COMPLETE);
7304}
7305
7306int
7307ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7308{
7309	struct scsi_maintenance_in *cdb;
7310	int retval;
7311	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7312	int num_target_port_groups, num_target_ports, single;
7313	struct ctl_lun *lun;
7314	struct ctl_softc *softc;
7315	struct ctl_port *port;
7316	struct scsi_target_group_data *rtg_ptr;
7317	struct scsi_target_group_data_extended *rtg_ext_ptr;
7318	struct scsi_target_port_group_descriptor *tpg_desc;
7319
7320	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7321
7322	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7323	softc = control_softc;
7324	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7325
7326	retval = CTL_RETVAL_COMPLETE;
7327
7328	switch (cdb->byte2 & STG_PDF_MASK) {
7329	case STG_PDF_LENGTH:
7330		ext = 0;
7331		break;
7332	case STG_PDF_EXTENDED:
7333		ext = 1;
7334		break;
7335	default:
7336		ctl_set_invalid_field(/*ctsio*/ ctsio,
7337				      /*sks_valid*/ 1,
7338				      /*command*/ 1,
7339				      /*field*/ 2,
7340				      /*bit_valid*/ 1,
7341				      /*bit*/ 5);
7342		ctl_done((union ctl_io *)ctsio);
7343		return(retval);
7344	}
7345
7346	single = ctl_is_single;
7347	if (single)
7348		num_target_port_groups = 1;
7349	else
7350		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7351	num_target_ports = 0;
7352	mtx_lock(&softc->ctl_lock);
7353	STAILQ_FOREACH(port, &softc->port_list, links) {
7354		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7355			continue;
7356		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7357			continue;
7358		num_target_ports++;
7359	}
7360	mtx_unlock(&softc->ctl_lock);
7361
7362	if (ext)
7363		total_len = sizeof(struct scsi_target_group_data_extended);
7364	else
7365		total_len = sizeof(struct scsi_target_group_data);
7366	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7367		num_target_port_groups +
7368	    sizeof(struct scsi_target_port_descriptor) *
7369		num_target_ports * num_target_port_groups;
7370
7371	alloc_len = scsi_4btoul(cdb->length);
7372
7373	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7374
7375	ctsio->kern_sg_entries = 0;
7376
7377	if (total_len < alloc_len) {
7378		ctsio->residual = alloc_len - total_len;
7379		ctsio->kern_data_len = total_len;
7380		ctsio->kern_total_len = total_len;
7381	} else {
7382		ctsio->residual = 0;
7383		ctsio->kern_data_len = alloc_len;
7384		ctsio->kern_total_len = alloc_len;
7385	}
7386	ctsio->kern_data_resid = 0;
7387	ctsio->kern_rel_offset = 0;
7388
7389	if (ext) {
7390		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7391		    ctsio->kern_data_ptr;
7392		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7393		rtg_ext_ptr->format_type = 0x10;
7394		rtg_ext_ptr->implicit_transition_time = 0;
7395		tpg_desc = &rtg_ext_ptr->groups[0];
7396	} else {
7397		rtg_ptr = (struct scsi_target_group_data *)
7398		    ctsio->kern_data_ptr;
7399		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7400		tpg_desc = &rtg_ptr->groups[0];
7401	}
7402
7403	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7404	mtx_lock(&softc->ctl_lock);
7405	for (g = 0; g < num_target_port_groups; g++) {
7406		if (g == pg)
7407			tpg_desc->pref_state = TPG_PRIMARY |
7408			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7409		else
7410			tpg_desc->pref_state =
7411			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7412		tpg_desc->support = TPG_AO_SUP;
7413		if (!single)
7414			tpg_desc->support |= TPG_AN_SUP;
7415		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7416		tpg_desc->status = TPG_IMPLICIT;
7417		pc = 0;
7418		STAILQ_FOREACH(port, &softc->port_list, links) {
7419			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7420				continue;
7421			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7422			    CTL_MAX_LUNS)
7423				continue;
7424			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7425			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7426			    relative_target_port_identifier);
7427			pc++;
7428		}
7429		tpg_desc->target_port_count = pc;
7430		tpg_desc = (struct scsi_target_port_group_descriptor *)
7431		    &tpg_desc->descriptors[pc];
7432	}
7433	mtx_unlock(&softc->ctl_lock);
7434
7435	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7436	ctsio->be_move_done = ctl_config_move_done;
7437
7438	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7439			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7440			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7441			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7442			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7443
7444	ctl_datamove((union ctl_io *)ctsio);
7445	return(retval);
7446}
7447
7448int
7449ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7450{
7451	struct ctl_lun *lun;
7452	struct scsi_report_supported_opcodes *cdb;
7453	const struct ctl_cmd_entry *entry, *sentry;
7454	struct scsi_report_supported_opcodes_all *all;
7455	struct scsi_report_supported_opcodes_descr *descr;
7456	struct scsi_report_supported_opcodes_one *one;
7457	int retval;
7458	int alloc_len, total_len;
7459	int opcode, service_action, i, j, num;
7460
7461	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7462
7463	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7464	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7465
7466	retval = CTL_RETVAL_COMPLETE;
7467
7468	opcode = cdb->requested_opcode;
7469	service_action = scsi_2btoul(cdb->requested_service_action);
7470	switch (cdb->options & RSO_OPTIONS_MASK) {
7471	case RSO_OPTIONS_ALL:
7472		num = 0;
7473		for (i = 0; i < 256; i++) {
7474			entry = &ctl_cmd_table[i];
7475			if (entry->flags & CTL_CMD_FLAG_SA5) {
7476				for (j = 0; j < 32; j++) {
7477					sentry = &((const struct ctl_cmd_entry *)
7478					    entry->execute)[j];
7479					if (ctl_cmd_applicable(
7480					    lun->be_lun->lun_type, sentry))
7481						num++;
7482				}
7483			} else {
7484				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7485				    entry))
7486					num++;
7487			}
7488		}
7489		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7490		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7491		break;
7492	case RSO_OPTIONS_OC:
7493		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7494			ctl_set_invalid_field(/*ctsio*/ ctsio,
7495					      /*sks_valid*/ 1,
7496					      /*command*/ 1,
7497					      /*field*/ 2,
7498					      /*bit_valid*/ 1,
7499					      /*bit*/ 2);
7500			ctl_done((union ctl_io *)ctsio);
7501			return (CTL_RETVAL_COMPLETE);
7502		}
7503		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7504		break;
7505	case RSO_OPTIONS_OC_SA:
7506		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7507		    service_action >= 32) {
7508			ctl_set_invalid_field(/*ctsio*/ ctsio,
7509					      /*sks_valid*/ 1,
7510					      /*command*/ 1,
7511					      /*field*/ 2,
7512					      /*bit_valid*/ 1,
7513					      /*bit*/ 2);
7514			ctl_done((union ctl_io *)ctsio);
7515			return (CTL_RETVAL_COMPLETE);
7516		}
7517		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7518		break;
7519	default:
7520		ctl_set_invalid_field(/*ctsio*/ ctsio,
7521				      /*sks_valid*/ 1,
7522				      /*command*/ 1,
7523				      /*field*/ 2,
7524				      /*bit_valid*/ 1,
7525				      /*bit*/ 2);
7526		ctl_done((union ctl_io *)ctsio);
7527		return (CTL_RETVAL_COMPLETE);
7528	}
7529
7530	alloc_len = scsi_4btoul(cdb->length);
7531
7532	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7533
7534	ctsio->kern_sg_entries = 0;
7535
7536	if (total_len < alloc_len) {
7537		ctsio->residual = alloc_len - total_len;
7538		ctsio->kern_data_len = total_len;
7539		ctsio->kern_total_len = total_len;
7540	} else {
7541		ctsio->residual = 0;
7542		ctsio->kern_data_len = alloc_len;
7543		ctsio->kern_total_len = alloc_len;
7544	}
7545	ctsio->kern_data_resid = 0;
7546	ctsio->kern_rel_offset = 0;
7547
7548	switch (cdb->options & RSO_OPTIONS_MASK) {
7549	case RSO_OPTIONS_ALL:
7550		all = (struct scsi_report_supported_opcodes_all *)
7551		    ctsio->kern_data_ptr;
7552		num = 0;
7553		for (i = 0; i < 256; i++) {
7554			entry = &ctl_cmd_table[i];
7555			if (entry->flags & CTL_CMD_FLAG_SA5) {
7556				for (j = 0; j < 32; j++) {
7557					sentry = &((const struct ctl_cmd_entry *)
7558					    entry->execute)[j];
7559					if (!ctl_cmd_applicable(
7560					    lun->be_lun->lun_type, sentry))
7561						continue;
7562					descr = &all->descr[num++];
7563					descr->opcode = i;
7564					scsi_ulto2b(j, descr->service_action);
7565					descr->flags = RSO_SERVACTV;
7566					scsi_ulto2b(sentry->length,
7567					    descr->cdb_length);
7568				}
7569			} else {
7570				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7571				    entry))
7572					continue;
7573				descr = &all->descr[num++];
7574				descr->opcode = i;
7575				scsi_ulto2b(0, descr->service_action);
7576				descr->flags = 0;
7577				scsi_ulto2b(entry->length, descr->cdb_length);
7578			}
7579		}
7580		scsi_ulto4b(
7581		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7582		    all->length);
7583		break;
7584	case RSO_OPTIONS_OC:
7585		one = (struct scsi_report_supported_opcodes_one *)
7586		    ctsio->kern_data_ptr;
7587		entry = &ctl_cmd_table[opcode];
7588		goto fill_one;
7589	case RSO_OPTIONS_OC_SA:
7590		one = (struct scsi_report_supported_opcodes_one *)
7591		    ctsio->kern_data_ptr;
7592		entry = &ctl_cmd_table[opcode];
7593		entry = &((const struct ctl_cmd_entry *)
7594		    entry->execute)[service_action];
7595fill_one:
7596		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7597			one->support = 3;
7598			scsi_ulto2b(entry->length, one->cdb_length);
7599			one->cdb_usage[0] = opcode;
7600			memcpy(&one->cdb_usage[1], entry->usage,
7601			    entry->length - 1);
7602		} else
7603			one->support = 1;
7604		break;
7605	}
7606
7607	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7608	ctsio->be_move_done = ctl_config_move_done;
7609
7610	ctl_datamove((union ctl_io *)ctsio);
7611	return(retval);
7612}
7613
7614int
7615ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7616{
7617	struct ctl_lun *lun;
7618	struct scsi_report_supported_tmf *cdb;
7619	struct scsi_report_supported_tmf_data *data;
7620	int retval;
7621	int alloc_len, total_len;
7622
7623	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7624
7625	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7626	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7627
7628	retval = CTL_RETVAL_COMPLETE;
7629
7630	total_len = sizeof(struct scsi_report_supported_tmf_data);
7631	alloc_len = scsi_4btoul(cdb->length);
7632
7633	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7634
7635	ctsio->kern_sg_entries = 0;
7636
7637	if (total_len < alloc_len) {
7638		ctsio->residual = alloc_len - total_len;
7639		ctsio->kern_data_len = total_len;
7640		ctsio->kern_total_len = total_len;
7641	} else {
7642		ctsio->residual = 0;
7643		ctsio->kern_data_len = alloc_len;
7644		ctsio->kern_total_len = alloc_len;
7645	}
7646	ctsio->kern_data_resid = 0;
7647	ctsio->kern_rel_offset = 0;
7648
7649	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7650	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7651	data->byte2 |= RST_ITNRS;
7652
7653	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7654	ctsio->be_move_done = ctl_config_move_done;
7655
7656	ctl_datamove((union ctl_io *)ctsio);
7657	return (retval);
7658}
7659
7660int
7661ctl_report_timestamp(struct ctl_scsiio *ctsio)
7662{
7663	struct ctl_lun *lun;
7664	struct scsi_report_timestamp *cdb;
7665	struct scsi_report_timestamp_data *data;
7666	struct timeval tv;
7667	int64_t timestamp;
7668	int retval;
7669	int alloc_len, total_len;
7670
7671	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7672
7673	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7674	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7675
7676	retval = CTL_RETVAL_COMPLETE;
7677
7678	total_len = sizeof(struct scsi_report_timestamp_data);
7679	alloc_len = scsi_4btoul(cdb->length);
7680
7681	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7682
7683	ctsio->kern_sg_entries = 0;
7684
7685	if (total_len < alloc_len) {
7686		ctsio->residual = alloc_len - total_len;
7687		ctsio->kern_data_len = total_len;
7688		ctsio->kern_total_len = total_len;
7689	} else {
7690		ctsio->residual = 0;
7691		ctsio->kern_data_len = alloc_len;
7692		ctsio->kern_total_len = alloc_len;
7693	}
7694	ctsio->kern_data_resid = 0;
7695	ctsio->kern_rel_offset = 0;
7696
7697	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7698	scsi_ulto2b(sizeof(*data) - 2, data->length);
7699	data->origin = RTS_ORIG_OUTSIDE;
7700	getmicrotime(&tv);
7701	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7702	scsi_ulto4b(timestamp >> 16, data->timestamp);
7703	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7704
7705	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7706	ctsio->be_move_done = ctl_config_move_done;
7707
7708	ctl_datamove((union ctl_io *)ctsio);
7709	return (retval);
7710}
7711
7712int
7713ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7714{
7715	struct scsi_per_res_in *cdb;
7716	int alloc_len, total_len = 0;
7717	/* struct scsi_per_res_in_rsrv in_data; */
7718	struct ctl_lun *lun;
7719	struct ctl_softc *softc;
7720
7721	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7722
7723	softc = control_softc;
7724
7725	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7726
7727	alloc_len = scsi_2btoul(cdb->length);
7728
7729	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7730
7731retry:
7732	mtx_lock(&lun->lun_lock);
7733	switch (cdb->action) {
7734	case SPRI_RK: /* read keys */
7735		total_len = sizeof(struct scsi_per_res_in_keys) +
7736			lun->pr_key_count *
7737			sizeof(struct scsi_per_res_key);
7738		break;
7739	case SPRI_RR: /* read reservation */
7740		if (lun->flags & CTL_LUN_PR_RESERVED)
7741			total_len = sizeof(struct scsi_per_res_in_rsrv);
7742		else
7743			total_len = sizeof(struct scsi_per_res_in_header);
7744		break;
7745	case SPRI_RC: /* report capabilities */
7746		total_len = sizeof(struct scsi_per_res_cap);
7747		break;
7748	case SPRI_RS: /* read full status */
7749		total_len = sizeof(struct scsi_per_res_in_header) +
7750		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7751		    lun->pr_key_count;
7752		break;
7753	default:
7754		panic("Invalid PR type %x", cdb->action);
7755	}
7756	mtx_unlock(&lun->lun_lock);
7757
7758	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7759
7760	if (total_len < alloc_len) {
7761		ctsio->residual = alloc_len - total_len;
7762		ctsio->kern_data_len = total_len;
7763		ctsio->kern_total_len = total_len;
7764	} else {
7765		ctsio->residual = 0;
7766		ctsio->kern_data_len = alloc_len;
7767		ctsio->kern_total_len = alloc_len;
7768	}
7769
7770	ctsio->kern_data_resid = 0;
7771	ctsio->kern_rel_offset = 0;
7772	ctsio->kern_sg_entries = 0;
7773
7774	mtx_lock(&lun->lun_lock);
7775	switch (cdb->action) {
7776	case SPRI_RK: { // read keys
7777        struct scsi_per_res_in_keys *res_keys;
7778		int i, key_count;
7779
7780		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7781
7782		/*
7783		 * We had to drop the lock to allocate our buffer, which
7784		 * leaves time for someone to come in with another
7785		 * persistent reservation.  (That is unlikely, though,
7786		 * since this should be the only persistent reservation
7787		 * command active right now.)
7788		 */
7789		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7790		    (lun->pr_key_count *
7791		     sizeof(struct scsi_per_res_key)))){
7792			mtx_unlock(&lun->lun_lock);
7793			free(ctsio->kern_data_ptr, M_CTL);
7794			printf("%s: reservation length changed, retrying\n",
7795			       __func__);
7796			goto retry;
7797		}
7798
7799		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7800
7801		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7802			     lun->pr_key_count, res_keys->header.length);
7803
7804		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7805			if (!lun->per_res[i].registered)
7806				continue;
7807
7808			/*
7809			 * We used lun->pr_key_count to calculate the
7810			 * size to allocate.  If it turns out the number of
7811			 * initiators with the registered flag set is
7812			 * larger than that (i.e. they haven't been kept in
7813			 * sync), we've got a problem.
7814			 */
7815			if (key_count >= lun->pr_key_count) {
7816#ifdef NEEDTOPORT
7817				csevent_log(CSC_CTL | CSC_SHELF_SW |
7818					    CTL_PR_ERROR,
7819					    csevent_LogType_Fault,
7820					    csevent_AlertLevel_Yellow,
7821					    csevent_FRU_ShelfController,
7822					    csevent_FRU_Firmware,
7823				        csevent_FRU_Unknown,
7824					    "registered keys %d >= key "
7825					    "count %d", key_count,
7826					    lun->pr_key_count);
7827#endif
7828				key_count++;
7829				continue;
7830			}
7831			memcpy(res_keys->keys[key_count].key,
7832			       lun->per_res[i].res_key.key,
7833			       ctl_min(sizeof(res_keys->keys[key_count].key),
7834			       sizeof(lun->per_res[i].res_key)));
7835			key_count++;
7836		}
7837		break;
7838	}
7839	case SPRI_RR: { // read reservation
7840		struct scsi_per_res_in_rsrv *res;
7841		int tmp_len, header_only;
7842
7843		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7844
7845		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7846
7847		if (lun->flags & CTL_LUN_PR_RESERVED)
7848		{
7849			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7850			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7851				    res->header.length);
7852			header_only = 0;
7853		} else {
7854			tmp_len = sizeof(struct scsi_per_res_in_header);
7855			scsi_ulto4b(0, res->header.length);
7856			header_only = 1;
7857		}
7858
7859		/*
7860		 * We had to drop the lock to allocate our buffer, which
7861		 * leaves time for someone to come in with another
7862		 * persistent reservation.  (That is unlikely, though,
7863		 * since this should be the only persistent reservation
7864		 * command active right now.)
7865		 */
7866		if (tmp_len != total_len) {
7867			mtx_unlock(&lun->lun_lock);
7868			free(ctsio->kern_data_ptr, M_CTL);
7869			printf("%s: reservation status changed, retrying\n",
7870			       __func__);
7871			goto retry;
7872		}
7873
7874		/*
7875		 * No reservation held, so we're done.
7876		 */
7877		if (header_only != 0)
7878			break;
7879
7880		/*
7881		 * If the registration is an All Registrants type, the key
7882		 * is 0, since it doesn't really matter.
7883		 */
7884		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7885			memcpy(res->data.reservation,
7886			       &lun->per_res[lun->pr_res_idx].res_key,
7887			       sizeof(struct scsi_per_res_key));
7888		}
7889		res->data.scopetype = lun->res_type;
7890		break;
7891	}
7892	case SPRI_RC:     //report capabilities
7893	{
7894		struct scsi_per_res_cap *res_cap;
7895		uint16_t type_mask;
7896
7897		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7898		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7899		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7900		type_mask = SPRI_TM_WR_EX_AR |
7901			    SPRI_TM_EX_AC_RO |
7902			    SPRI_TM_WR_EX_RO |
7903			    SPRI_TM_EX_AC |
7904			    SPRI_TM_WR_EX |
7905			    SPRI_TM_EX_AC_AR;
7906		scsi_ulto2b(type_mask, res_cap->type_mask);
7907		break;
7908	}
7909	case SPRI_RS: { // read full status
7910		struct scsi_per_res_in_full *res_status;
7911		struct scsi_per_res_in_full_desc *res_desc;
7912		struct ctl_port *port;
7913		int i, len;
7914
7915		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7916
7917		/*
7918		 * We had to drop the lock to allocate our buffer, which
7919		 * leaves time for someone to come in with another
7920		 * persistent reservation.  (That is unlikely, though,
7921		 * since this should be the only persistent reservation
7922		 * command active right now.)
7923		 */
7924		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7925		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7926		     lun->pr_key_count)){
7927			mtx_unlock(&lun->lun_lock);
7928			free(ctsio->kern_data_ptr, M_CTL);
7929			printf("%s: reservation length changed, retrying\n",
7930			       __func__);
7931			goto retry;
7932		}
7933
7934		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7935
7936		res_desc = &res_status->desc[0];
7937		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7938			if (!lun->per_res[i].registered)
7939				continue;
7940
7941			memcpy(&res_desc->res_key, &lun->per_res[i].res_key.key,
7942			    sizeof(res_desc->res_key));
7943			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7944			    (lun->pr_res_idx == i ||
7945			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7946				res_desc->flags = SPRI_FULL_R_HOLDER;
7947				res_desc->scopetype = lun->res_type;
7948			}
7949			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7950			    res_desc->rel_trgt_port_id);
7951			len = 0;
7952			port = softc->ctl_ports[
7953			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7954			if (port != NULL)
7955				len = ctl_create_iid(port,
7956				    i % CTL_MAX_INIT_PER_PORT,
7957				    res_desc->transport_id);
7958			scsi_ulto4b(len, res_desc->additional_length);
7959			res_desc = (struct scsi_per_res_in_full_desc *)
7960			    &res_desc->transport_id[len];
7961		}
7962		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7963		    res_status->header.length);
7964		break;
7965	}
7966	default:
7967		/*
7968		 * This is a bug, because we just checked for this above,
7969		 * and should have returned an error.
7970		 */
7971		panic("Invalid PR type %x", cdb->action);
7972		break; /* NOTREACHED */
7973	}
7974	mtx_unlock(&lun->lun_lock);
7975
7976	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7977	ctsio->be_move_done = ctl_config_move_done;
7978
7979	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7980			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7981			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7982			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7983			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7984
7985	ctl_datamove((union ctl_io *)ctsio);
7986
7987	return (CTL_RETVAL_COMPLETE);
7988}
7989
7990/*
7991 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7992 * it should return.
7993 */
7994static int
7995ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7996		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7997		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7998		struct scsi_per_res_out_parms* param)
7999{
8000	union ctl_ha_msg persis_io;
8001	int retval, i;
8002	int isc_retval;
8003
8004	retval = 0;
8005
8006	mtx_lock(&lun->lun_lock);
8007	if (sa_res_key == 0) {
8008		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8009			/* validate scope and type */
8010			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8011			     SPR_LU_SCOPE) {
8012				mtx_unlock(&lun->lun_lock);
8013				ctl_set_invalid_field(/*ctsio*/ ctsio,
8014						      /*sks_valid*/ 1,
8015						      /*command*/ 1,
8016						      /*field*/ 2,
8017						      /*bit_valid*/ 1,
8018						      /*bit*/ 4);
8019				ctl_done((union ctl_io *)ctsio);
8020				return (1);
8021			}
8022
8023		        if (type>8 || type==2 || type==4 || type==0) {
8024				mtx_unlock(&lun->lun_lock);
8025				ctl_set_invalid_field(/*ctsio*/ ctsio,
8026       	           				      /*sks_valid*/ 1,
8027						      /*command*/ 1,
8028						      /*field*/ 2,
8029						      /*bit_valid*/ 1,
8030						      /*bit*/ 0);
8031				ctl_done((union ctl_io *)ctsio);
8032				return (1);
8033		        }
8034
8035			/* temporarily unregister this nexus */
8036			lun->per_res[residx].registered = 0;
8037
8038			/*
8039			 * Unregister everybody else and build UA for
8040			 * them
8041			 */
8042			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8043				if (lun->per_res[i].registered == 0)
8044					continue;
8045
8046				if (!persis_offset
8047				 && i <CTL_MAX_INITIATORS)
8048					lun->pending_ua[i] |=
8049						CTL_UA_REG_PREEMPT;
8050				else if (persis_offset
8051				      && i >= persis_offset)
8052					lun->pending_ua[i-persis_offset] |=
8053						CTL_UA_REG_PREEMPT;
8054				lun->per_res[i].registered = 0;
8055				memset(&lun->per_res[i].res_key, 0,
8056				       sizeof(struct scsi_per_res_key));
8057			}
8058			lun->per_res[residx].registered = 1;
8059			lun->pr_key_count = 1;
8060			lun->res_type = type;
8061			if (lun->res_type != SPR_TYPE_WR_EX_AR
8062			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8063				lun->pr_res_idx = residx;
8064
8065			/* send msg to other side */
8066			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8067			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8068			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8069			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8070			persis_io.pr.pr_info.res_type = type;
8071			memcpy(persis_io.pr.pr_info.sa_res_key,
8072			       param->serv_act_res_key,
8073			       sizeof(param->serv_act_res_key));
8074			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8075			     &persis_io, sizeof(persis_io), 0)) >
8076			     CTL_HA_STATUS_SUCCESS) {
8077				printf("CTL:Persis Out error returned "
8078				       "from ctl_ha_msg_send %d\n",
8079				       isc_retval);
8080			}
8081		} else {
8082			/* not all registrants */
8083			mtx_unlock(&lun->lun_lock);
8084			free(ctsio->kern_data_ptr, M_CTL);
8085			ctl_set_invalid_field(ctsio,
8086					      /*sks_valid*/ 1,
8087					      /*command*/ 0,
8088					      /*field*/ 8,
8089					      /*bit_valid*/ 0,
8090					      /*bit*/ 0);
8091			ctl_done((union ctl_io *)ctsio);
8092			return (1);
8093		}
8094	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8095		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8096		int found = 0;
8097
8098		if (res_key == sa_res_key) {
8099			/* special case */
8100			/*
8101			 * The spec implies this is not good but doesn't
8102			 * say what to do. There are two choices either
8103			 * generate a res conflict or check condition
8104			 * with illegal field in parameter data. Since
8105			 * that is what is done when the sa_res_key is
8106			 * zero I'll take that approach since this has
8107			 * to do with the sa_res_key.
8108			 */
8109			mtx_unlock(&lun->lun_lock);
8110			free(ctsio->kern_data_ptr, M_CTL);
8111			ctl_set_invalid_field(ctsio,
8112					      /*sks_valid*/ 1,
8113					      /*command*/ 0,
8114					      /*field*/ 8,
8115					      /*bit_valid*/ 0,
8116					      /*bit*/ 0);
8117			ctl_done((union ctl_io *)ctsio);
8118			return (1);
8119		}
8120
8121		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8122			if (lun->per_res[i].registered
8123			 && memcmp(param->serv_act_res_key,
8124			    lun->per_res[i].res_key.key,
8125			    sizeof(struct scsi_per_res_key)) != 0)
8126				continue;
8127
8128			found = 1;
8129			lun->per_res[i].registered = 0;
8130			memset(&lun->per_res[i].res_key, 0,
8131			       sizeof(struct scsi_per_res_key));
8132			lun->pr_key_count--;
8133
8134			if (!persis_offset && i < CTL_MAX_INITIATORS)
8135				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8136			else if (persis_offset && i >= persis_offset)
8137				lun->pending_ua[i-persis_offset] |=
8138					CTL_UA_REG_PREEMPT;
8139		}
8140		if (!found) {
8141			mtx_unlock(&lun->lun_lock);
8142			free(ctsio->kern_data_ptr, M_CTL);
8143			ctl_set_reservation_conflict(ctsio);
8144			ctl_done((union ctl_io *)ctsio);
8145			return (CTL_RETVAL_COMPLETE);
8146		}
8147		/* send msg to other side */
8148		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8149		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8150		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8151		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8152		persis_io.pr.pr_info.res_type = type;
8153		memcpy(persis_io.pr.pr_info.sa_res_key,
8154		       param->serv_act_res_key,
8155		       sizeof(param->serv_act_res_key));
8156		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8157		     &persis_io, sizeof(persis_io), 0)) >
8158		     CTL_HA_STATUS_SUCCESS) {
8159			printf("CTL:Persis Out error returned from "
8160			       "ctl_ha_msg_send %d\n", isc_retval);
8161		}
8162	} else {
8163		/* Reserved but not all registrants */
8164		/* sa_res_key is res holder */
8165		if (memcmp(param->serv_act_res_key,
8166                   lun->per_res[lun->pr_res_idx].res_key.key,
8167                   sizeof(struct scsi_per_res_key)) == 0) {
8168			/* validate scope and type */
8169			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8170			     SPR_LU_SCOPE) {
8171				mtx_unlock(&lun->lun_lock);
8172				ctl_set_invalid_field(/*ctsio*/ ctsio,
8173						      /*sks_valid*/ 1,
8174						      /*command*/ 1,
8175						      /*field*/ 2,
8176						      /*bit_valid*/ 1,
8177						      /*bit*/ 4);
8178				ctl_done((union ctl_io *)ctsio);
8179				return (1);
8180			}
8181
8182			if (type>8 || type==2 || type==4 || type==0) {
8183				mtx_unlock(&lun->lun_lock);
8184				ctl_set_invalid_field(/*ctsio*/ ctsio,
8185						      /*sks_valid*/ 1,
8186						      /*command*/ 1,
8187						      /*field*/ 2,
8188						      /*bit_valid*/ 1,
8189						      /*bit*/ 0);
8190				ctl_done((union ctl_io *)ctsio);
8191				return (1);
8192			}
8193
8194			/*
8195			 * Do the following:
8196			 * if sa_res_key != res_key remove all
8197			 * registrants w/sa_res_key and generate UA
8198			 * for these registrants(Registrations
8199			 * Preempted) if it wasn't an exclusive
8200			 * reservation generate UA(Reservations
8201			 * Preempted) for all other registered nexuses
8202			 * if the type has changed. Establish the new
8203			 * reservation and holder. If res_key and
8204			 * sa_res_key are the same do the above
8205			 * except don't unregister the res holder.
8206			 */
8207
8208			/*
8209			 * Temporarily unregister so it won't get
8210			 * removed or UA generated
8211			 */
8212			lun->per_res[residx].registered = 0;
8213			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8214				if (lun->per_res[i].registered == 0)
8215					continue;
8216
8217				if (memcmp(param->serv_act_res_key,
8218				    lun->per_res[i].res_key.key,
8219				    sizeof(struct scsi_per_res_key)) == 0) {
8220					lun->per_res[i].registered = 0;
8221					memset(&lun->per_res[i].res_key,
8222					       0,
8223					       sizeof(struct scsi_per_res_key));
8224					lun->pr_key_count--;
8225
8226					if (!persis_offset
8227					 && i < CTL_MAX_INITIATORS)
8228						lun->pending_ua[i] |=
8229							CTL_UA_REG_PREEMPT;
8230					else if (persis_offset
8231					      && i >= persis_offset)
8232						lun->pending_ua[i-persis_offset] |=
8233						  CTL_UA_REG_PREEMPT;
8234				} else if (type != lun->res_type
8235					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8236					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8237						if (!persis_offset
8238						 && i < CTL_MAX_INITIATORS)
8239							lun->pending_ua[i] |=
8240							CTL_UA_RES_RELEASE;
8241						else if (persis_offset
8242						      && i >= persis_offset)
8243							lun->pending_ua[
8244							i-persis_offset] |=
8245							CTL_UA_RES_RELEASE;
8246				}
8247			}
8248			lun->per_res[residx].registered = 1;
8249			lun->res_type = type;
8250			if (lun->res_type != SPR_TYPE_WR_EX_AR
8251			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8252				lun->pr_res_idx = residx;
8253			else
8254				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8255
8256			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8257			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8258			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8259			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8260			persis_io.pr.pr_info.res_type = type;
8261			memcpy(persis_io.pr.pr_info.sa_res_key,
8262			       param->serv_act_res_key,
8263			       sizeof(param->serv_act_res_key));
8264			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8265			     &persis_io, sizeof(persis_io), 0)) >
8266			     CTL_HA_STATUS_SUCCESS) {
8267				printf("CTL:Persis Out error returned "
8268				       "from ctl_ha_msg_send %d\n",
8269				       isc_retval);
8270			}
8271		} else {
8272			/*
8273			 * sa_res_key is not the res holder just
8274			 * remove registrants
8275			 */
8276			int found=0;
8277
8278			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8279				if (memcmp(param->serv_act_res_key,
8280				    lun->per_res[i].res_key.key,
8281				    sizeof(struct scsi_per_res_key)) != 0)
8282					continue;
8283
8284				found = 1;
8285				lun->per_res[i].registered = 0;
8286				memset(&lun->per_res[i].res_key, 0,
8287				       sizeof(struct scsi_per_res_key));
8288				lun->pr_key_count--;
8289
8290				if (!persis_offset
8291				 && i < CTL_MAX_INITIATORS)
8292					lun->pending_ua[i] |=
8293						CTL_UA_REG_PREEMPT;
8294				else if (persis_offset
8295				      && i >= persis_offset)
8296					lun->pending_ua[i-persis_offset] |=
8297						CTL_UA_REG_PREEMPT;
8298			}
8299
8300			if (!found) {
8301				mtx_unlock(&lun->lun_lock);
8302				free(ctsio->kern_data_ptr, M_CTL);
8303				ctl_set_reservation_conflict(ctsio);
8304				ctl_done((union ctl_io *)ctsio);
8305		        	return (1);
8306			}
8307			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8308			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8309			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8310			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8311			persis_io.pr.pr_info.res_type = type;
8312			memcpy(persis_io.pr.pr_info.sa_res_key,
8313			       param->serv_act_res_key,
8314			       sizeof(param->serv_act_res_key));
8315			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8316			     &persis_io, sizeof(persis_io), 0)) >
8317			     CTL_HA_STATUS_SUCCESS) {
8318				printf("CTL:Persis Out error returned "
8319				       "from ctl_ha_msg_send %d\n",
8320				isc_retval);
8321			}
8322		}
8323	}
8324
8325	lun->PRGeneration++;
8326	mtx_unlock(&lun->lun_lock);
8327
8328	return (retval);
8329}
8330
8331static void
8332ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8333{
8334	int i;
8335
8336	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8337	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8338	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8339		   msg->pr.pr_info.sa_res_key,
8340		   sizeof(struct scsi_per_res_key)) != 0) {
8341		uint64_t sa_res_key;
8342		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8343
8344		if (sa_res_key == 0) {
8345			/* temporarily unregister this nexus */
8346			lun->per_res[msg->pr.pr_info.residx].registered = 0;
8347
8348			/*
8349			 * Unregister everybody else and build UA for
8350			 * them
8351			 */
8352			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8353				if (lun->per_res[i].registered == 0)
8354					continue;
8355
8356				if (!persis_offset
8357				 && i < CTL_MAX_INITIATORS)
8358					lun->pending_ua[i] |=
8359						CTL_UA_REG_PREEMPT;
8360				else if (persis_offset && i >= persis_offset)
8361					lun->pending_ua[i - persis_offset] |=
8362						CTL_UA_REG_PREEMPT;
8363				lun->per_res[i].registered = 0;
8364				memset(&lun->per_res[i].res_key, 0,
8365				       sizeof(struct scsi_per_res_key));
8366			}
8367
8368			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8369			lun->pr_key_count = 1;
8370			lun->res_type = msg->pr.pr_info.res_type;
8371			if (lun->res_type != SPR_TYPE_WR_EX_AR
8372			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8373				lun->pr_res_idx = msg->pr.pr_info.residx;
8374		} else {
8375		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8376				if (memcmp(msg->pr.pr_info.sa_res_key,
8377		                   lun->per_res[i].res_key.key,
8378		                   sizeof(struct scsi_per_res_key)) != 0)
8379					continue;
8380
8381				lun->per_res[i].registered = 0;
8382				memset(&lun->per_res[i].res_key, 0,
8383				       sizeof(struct scsi_per_res_key));
8384				lun->pr_key_count--;
8385
8386				if (!persis_offset
8387				 && i < persis_offset)
8388					lun->pending_ua[i] |=
8389						CTL_UA_REG_PREEMPT;
8390				else if (persis_offset
8391				      && i >= persis_offset)
8392					lun->pending_ua[i - persis_offset] |=
8393						CTL_UA_REG_PREEMPT;
8394			}
8395		}
8396	} else {
8397		/*
8398		 * Temporarily unregister so it won't get removed
8399		 * or UA generated
8400		 */
8401		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8402		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8403			if (lun->per_res[i].registered == 0)
8404				continue;
8405
8406			if (memcmp(msg->pr.pr_info.sa_res_key,
8407	                   lun->per_res[i].res_key.key,
8408	                   sizeof(struct scsi_per_res_key)) == 0) {
8409				lun->per_res[i].registered = 0;
8410				memset(&lun->per_res[i].res_key, 0,
8411				       sizeof(struct scsi_per_res_key));
8412				lun->pr_key_count--;
8413				if (!persis_offset
8414				 && i < CTL_MAX_INITIATORS)
8415					lun->pending_ua[i] |=
8416						CTL_UA_REG_PREEMPT;
8417				else if (persis_offset
8418				      && i >= persis_offset)
8419					lun->pending_ua[i - persis_offset] |=
8420						CTL_UA_REG_PREEMPT;
8421			} else if (msg->pr.pr_info.res_type != lun->res_type
8422				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8423				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8424					if (!persis_offset
8425					 && i < persis_offset)
8426						lun->pending_ua[i] |=
8427							CTL_UA_RES_RELEASE;
8428					else if (persis_offset
8429					      && i >= persis_offset)
8430					lun->pending_ua[i - persis_offset] |=
8431						CTL_UA_RES_RELEASE;
8432			}
8433		}
8434		lun->per_res[msg->pr.pr_info.residx].registered = 1;
8435		lun->res_type = msg->pr.pr_info.res_type;
8436		if (lun->res_type != SPR_TYPE_WR_EX_AR
8437		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8438			lun->pr_res_idx = msg->pr.pr_info.residx;
8439		else
8440			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8441	}
8442	lun->PRGeneration++;
8443
8444}
8445
8446
8447int
8448ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8449{
8450	int retval;
8451	int isc_retval;
8452	u_int32_t param_len;
8453	struct scsi_per_res_out *cdb;
8454	struct ctl_lun *lun;
8455	struct scsi_per_res_out_parms* param;
8456	struct ctl_softc *softc;
8457	uint32_t residx;
8458	uint64_t res_key, sa_res_key;
8459	uint8_t type;
8460	union ctl_ha_msg persis_io;
8461	int    i;
8462
8463	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8464
8465	retval = CTL_RETVAL_COMPLETE;
8466
8467	softc = control_softc;
8468
8469	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8470	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8471
8472	/*
8473	 * We only support whole-LUN scope.  The scope & type are ignored for
8474	 * register, register and ignore existing key and clear.
8475	 * We sometimes ignore scope and type on preempts too!!
8476	 * Verify reservation type here as well.
8477	 */
8478	type = cdb->scope_type & SPR_TYPE_MASK;
8479	if ((cdb->action == SPRO_RESERVE)
8480	 || (cdb->action == SPRO_RELEASE)) {
8481		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8482			ctl_set_invalid_field(/*ctsio*/ ctsio,
8483					      /*sks_valid*/ 1,
8484					      /*command*/ 1,
8485					      /*field*/ 2,
8486					      /*bit_valid*/ 1,
8487					      /*bit*/ 4);
8488			ctl_done((union ctl_io *)ctsio);
8489			return (CTL_RETVAL_COMPLETE);
8490		}
8491
8492		if (type>8 || type==2 || type==4 || type==0) {
8493			ctl_set_invalid_field(/*ctsio*/ ctsio,
8494					      /*sks_valid*/ 1,
8495					      /*command*/ 1,
8496					      /*field*/ 2,
8497					      /*bit_valid*/ 1,
8498					      /*bit*/ 0);
8499			ctl_done((union ctl_io *)ctsio);
8500			return (CTL_RETVAL_COMPLETE);
8501		}
8502	}
8503
8504	param_len = scsi_4btoul(cdb->length);
8505
8506	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8507		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8508		ctsio->kern_data_len = param_len;
8509		ctsio->kern_total_len = param_len;
8510		ctsio->kern_data_resid = 0;
8511		ctsio->kern_rel_offset = 0;
8512		ctsio->kern_sg_entries = 0;
8513		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8514		ctsio->be_move_done = ctl_config_move_done;
8515		ctl_datamove((union ctl_io *)ctsio);
8516
8517		return (CTL_RETVAL_COMPLETE);
8518	}
8519
8520	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8521
8522	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8523	res_key = scsi_8btou64(param->res_key.key);
8524	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8525
8526	/*
8527	 * Validate the reservation key here except for SPRO_REG_IGNO
8528	 * This must be done for all other service actions
8529	 */
8530	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8531		mtx_lock(&lun->lun_lock);
8532		if (lun->per_res[residx].registered) {
8533		    if (memcmp(param->res_key.key,
8534			       lun->per_res[residx].res_key.key,
8535			       ctl_min(sizeof(param->res_key),
8536			       sizeof(lun->per_res[residx].res_key))) != 0) {
8537				/*
8538				 * The current key passed in doesn't match
8539				 * the one the initiator previously
8540				 * registered.
8541				 */
8542				mtx_unlock(&lun->lun_lock);
8543				free(ctsio->kern_data_ptr, M_CTL);
8544				ctl_set_reservation_conflict(ctsio);
8545				ctl_done((union ctl_io *)ctsio);
8546				return (CTL_RETVAL_COMPLETE);
8547			}
8548		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8549			/*
8550			 * We are not registered
8551			 */
8552			mtx_unlock(&lun->lun_lock);
8553			free(ctsio->kern_data_ptr, M_CTL);
8554			ctl_set_reservation_conflict(ctsio);
8555			ctl_done((union ctl_io *)ctsio);
8556			return (CTL_RETVAL_COMPLETE);
8557		} else if (res_key != 0) {
8558			/*
8559			 * We are not registered and trying to register but
8560			 * the register key isn't zero.
8561			 */
8562			mtx_unlock(&lun->lun_lock);
8563			free(ctsio->kern_data_ptr, M_CTL);
8564			ctl_set_reservation_conflict(ctsio);
8565			ctl_done((union ctl_io *)ctsio);
8566			return (CTL_RETVAL_COMPLETE);
8567		}
8568		mtx_unlock(&lun->lun_lock);
8569	}
8570
8571	switch (cdb->action & SPRO_ACTION_MASK) {
8572	case SPRO_REGISTER:
8573	case SPRO_REG_IGNO: {
8574
8575#if 0
8576		printf("Registration received\n");
8577#endif
8578
8579		/*
8580		 * We don't support any of these options, as we report in
8581		 * the read capabilities request (see
8582		 * ctl_persistent_reserve_in(), above).
8583		 */
8584		if ((param->flags & SPR_SPEC_I_PT)
8585		 || (param->flags & SPR_ALL_TG_PT)
8586		 || (param->flags & SPR_APTPL)) {
8587			int bit_ptr;
8588
8589			if (param->flags & SPR_APTPL)
8590				bit_ptr = 0;
8591			else if (param->flags & SPR_ALL_TG_PT)
8592				bit_ptr = 2;
8593			else /* SPR_SPEC_I_PT */
8594				bit_ptr = 3;
8595
8596			free(ctsio->kern_data_ptr, M_CTL);
8597			ctl_set_invalid_field(ctsio,
8598					      /*sks_valid*/ 1,
8599					      /*command*/ 0,
8600					      /*field*/ 20,
8601					      /*bit_valid*/ 1,
8602					      /*bit*/ bit_ptr);
8603			ctl_done((union ctl_io *)ctsio);
8604			return (CTL_RETVAL_COMPLETE);
8605		}
8606
8607		mtx_lock(&lun->lun_lock);
8608
8609		/*
8610		 * The initiator wants to clear the
8611		 * key/unregister.
8612		 */
8613		if (sa_res_key == 0) {
8614			if ((res_key == 0
8615			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8616			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8617			  && !lun->per_res[residx].registered)) {
8618				mtx_unlock(&lun->lun_lock);
8619				goto done;
8620			}
8621
8622			lun->per_res[residx].registered = 0;
8623			memset(&lun->per_res[residx].res_key,
8624			       0, sizeof(lun->per_res[residx].res_key));
8625			lun->pr_key_count--;
8626
8627			if (residx == lun->pr_res_idx) {
8628				lun->flags &= ~CTL_LUN_PR_RESERVED;
8629				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8630
8631				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8632				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8633				 && lun->pr_key_count) {
8634					/*
8635					 * If the reservation is a registrants
8636					 * only type we need to generate a UA
8637					 * for other registered inits.  The
8638					 * sense code should be RESERVATIONS
8639					 * RELEASED
8640					 */
8641
8642					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8643						if (lun->per_res[
8644						    i+persis_offset].registered
8645						    == 0)
8646							continue;
8647						lun->pending_ua[i] |=
8648							CTL_UA_RES_RELEASE;
8649					}
8650				}
8651				lun->res_type = 0;
8652			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8653				if (lun->pr_key_count==0) {
8654					lun->flags &= ~CTL_LUN_PR_RESERVED;
8655					lun->res_type = 0;
8656					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8657				}
8658			}
8659			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8660			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8661			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8662			persis_io.pr.pr_info.residx = residx;
8663			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8664			     &persis_io, sizeof(persis_io), 0 )) >
8665			     CTL_HA_STATUS_SUCCESS) {
8666				printf("CTL:Persis Out error returned from "
8667				       "ctl_ha_msg_send %d\n", isc_retval);
8668			}
8669		} else /* sa_res_key != 0 */ {
8670
8671			/*
8672			 * If we aren't registered currently then increment
8673			 * the key count and set the registered flag.
8674			 */
8675			if (!lun->per_res[residx].registered) {
8676				lun->pr_key_count++;
8677				lun->per_res[residx].registered = 1;
8678			}
8679
8680			memcpy(&lun->per_res[residx].res_key,
8681			       param->serv_act_res_key,
8682			       ctl_min(sizeof(param->serv_act_res_key),
8683			       sizeof(lun->per_res[residx].res_key)));
8684
8685			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8686			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8687			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8688			persis_io.pr.pr_info.residx = residx;
8689			memcpy(persis_io.pr.pr_info.sa_res_key,
8690			       param->serv_act_res_key,
8691			       sizeof(param->serv_act_res_key));
8692			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8693			     &persis_io, sizeof(persis_io), 0)) >
8694			     CTL_HA_STATUS_SUCCESS) {
8695				printf("CTL:Persis Out error returned from "
8696				       "ctl_ha_msg_send %d\n", isc_retval);
8697			}
8698		}
8699		lun->PRGeneration++;
8700		mtx_unlock(&lun->lun_lock);
8701
8702		break;
8703	}
8704	case SPRO_RESERVE:
8705#if 0
8706                printf("Reserve executed type %d\n", type);
8707#endif
8708		mtx_lock(&lun->lun_lock);
8709		if (lun->flags & CTL_LUN_PR_RESERVED) {
8710			/*
8711			 * if this isn't the reservation holder and it's
8712			 * not a "all registrants" type or if the type is
8713			 * different then we have a conflict
8714			 */
8715			if ((lun->pr_res_idx != residx
8716			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8717			 || lun->res_type != type) {
8718				mtx_unlock(&lun->lun_lock);
8719				free(ctsio->kern_data_ptr, M_CTL);
8720				ctl_set_reservation_conflict(ctsio);
8721				ctl_done((union ctl_io *)ctsio);
8722				return (CTL_RETVAL_COMPLETE);
8723			}
8724			mtx_unlock(&lun->lun_lock);
8725		} else /* create a reservation */ {
8726			/*
8727			 * If it's not an "all registrants" type record
8728			 * reservation holder
8729			 */
8730			if (type != SPR_TYPE_WR_EX_AR
8731			 && type != SPR_TYPE_EX_AC_AR)
8732				lun->pr_res_idx = residx; /* Res holder */
8733			else
8734				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8735
8736			lun->flags |= CTL_LUN_PR_RESERVED;
8737			lun->res_type = type;
8738
8739			mtx_unlock(&lun->lun_lock);
8740
8741			/* send msg to other side */
8742			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8743			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8744			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8745			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8746			persis_io.pr.pr_info.res_type = type;
8747			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8748			     &persis_io, sizeof(persis_io), 0)) >
8749			     CTL_HA_STATUS_SUCCESS) {
8750				printf("CTL:Persis Out error returned from "
8751				       "ctl_ha_msg_send %d\n", isc_retval);
8752			}
8753		}
8754		break;
8755
8756	case SPRO_RELEASE:
8757		mtx_lock(&lun->lun_lock);
8758		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8759			/* No reservation exists return good status */
8760			mtx_unlock(&lun->lun_lock);
8761			goto done;
8762		}
8763		/*
8764		 * Is this nexus a reservation holder?
8765		 */
8766		if (lun->pr_res_idx != residx
8767		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8768			/*
8769			 * not a res holder return good status but
8770			 * do nothing
8771			 */
8772			mtx_unlock(&lun->lun_lock);
8773			goto done;
8774		}
8775
8776		if (lun->res_type != type) {
8777			mtx_unlock(&lun->lun_lock);
8778			free(ctsio->kern_data_ptr, M_CTL);
8779			ctl_set_illegal_pr_release(ctsio);
8780			ctl_done((union ctl_io *)ctsio);
8781			return (CTL_RETVAL_COMPLETE);
8782		}
8783
8784		/* okay to release */
8785		lun->flags &= ~CTL_LUN_PR_RESERVED;
8786		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8787		lun->res_type = 0;
8788
8789		/*
8790		 * if this isn't an exclusive access
8791		 * res generate UA for all other
8792		 * registrants.
8793		 */
8794		if (type != SPR_TYPE_EX_AC
8795		 && type != SPR_TYPE_WR_EX) {
8796			/*
8797			 * temporarily unregister so we don't generate UA
8798			 */
8799			lun->per_res[residx].registered = 0;
8800
8801			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8802				if (lun->per_res[i+persis_offset].registered
8803				    == 0)
8804					continue;
8805				lun->pending_ua[i] |=
8806					CTL_UA_RES_RELEASE;
8807			}
8808
8809			lun->per_res[residx].registered = 1;
8810		}
8811		mtx_unlock(&lun->lun_lock);
8812		/* Send msg to other side */
8813		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8814		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8815		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8816		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8817		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8818			printf("CTL:Persis Out error returned from "
8819			       "ctl_ha_msg_send %d\n", isc_retval);
8820		}
8821		break;
8822
8823	case SPRO_CLEAR:
8824		/* send msg to other side */
8825
8826		mtx_lock(&lun->lun_lock);
8827		lun->flags &= ~CTL_LUN_PR_RESERVED;
8828		lun->res_type = 0;
8829		lun->pr_key_count = 0;
8830		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8831
8832
8833		memset(&lun->per_res[residx].res_key,
8834		       0, sizeof(lun->per_res[residx].res_key));
8835		lun->per_res[residx].registered = 0;
8836
8837		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8838			if (lun->per_res[i].registered) {
8839				if (!persis_offset && i < CTL_MAX_INITIATORS)
8840					lun->pending_ua[i] |=
8841						CTL_UA_RES_PREEMPT;
8842				else if (persis_offset && i >= persis_offset)
8843					lun->pending_ua[i-persis_offset] |=
8844					    CTL_UA_RES_PREEMPT;
8845
8846				memset(&lun->per_res[i].res_key,
8847				       0, sizeof(struct scsi_per_res_key));
8848				lun->per_res[i].registered = 0;
8849			}
8850		lun->PRGeneration++;
8851		mtx_unlock(&lun->lun_lock);
8852		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8853		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8854		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8855		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8856		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8857			printf("CTL:Persis Out error returned from "
8858			       "ctl_ha_msg_send %d\n", isc_retval);
8859		}
8860		break;
8861
8862	case SPRO_PREEMPT: {
8863		int nretval;
8864
8865		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8866					  residx, ctsio, cdb, param);
8867		if (nretval != 0)
8868			return (CTL_RETVAL_COMPLETE);
8869		break;
8870	}
8871	default:
8872		panic("Invalid PR type %x", cdb->action);
8873	}
8874
8875done:
8876	free(ctsio->kern_data_ptr, M_CTL);
8877	ctl_set_success(ctsio);
8878	ctl_done((union ctl_io *)ctsio);
8879
8880	return (retval);
8881}
8882
8883/*
8884 * This routine is for handling a message from the other SC pertaining to
8885 * persistent reserve out. All the error checking will have been done
8886 * so only perorming the action need be done here to keep the two
8887 * in sync.
8888 */
8889static void
8890ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8891{
8892	struct ctl_lun *lun;
8893	struct ctl_softc *softc;
8894	int i;
8895	uint32_t targ_lun;
8896
8897	softc = control_softc;
8898
8899	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8900	lun = softc->ctl_luns[targ_lun];
8901	mtx_lock(&lun->lun_lock);
8902	switch(msg->pr.pr_info.action) {
8903	case CTL_PR_REG_KEY:
8904		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8905			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8906			lun->pr_key_count++;
8907		}
8908		lun->PRGeneration++;
8909		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8910		       msg->pr.pr_info.sa_res_key,
8911		       sizeof(struct scsi_per_res_key));
8912		break;
8913
8914	case CTL_PR_UNREG_KEY:
8915		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8916		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8917		       0, sizeof(struct scsi_per_res_key));
8918		lun->pr_key_count--;
8919
8920		/* XXX Need to see if the reservation has been released */
8921		/* if so do we need to generate UA? */
8922		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8923			lun->flags &= ~CTL_LUN_PR_RESERVED;
8924			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8925
8926			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8927			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8928			 && lun->pr_key_count) {
8929				/*
8930				 * If the reservation is a registrants
8931				 * only type we need to generate a UA
8932				 * for other registered inits.  The
8933				 * sense code should be RESERVATIONS
8934				 * RELEASED
8935				 */
8936
8937				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8938					if (lun->per_res[i+
8939					    persis_offset].registered == 0)
8940						continue;
8941
8942					lun->pending_ua[i] |=
8943						CTL_UA_RES_RELEASE;
8944				}
8945			}
8946			lun->res_type = 0;
8947		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8948			if (lun->pr_key_count==0) {
8949				lun->flags &= ~CTL_LUN_PR_RESERVED;
8950				lun->res_type = 0;
8951				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8952			}
8953		}
8954		lun->PRGeneration++;
8955		break;
8956
8957	case CTL_PR_RESERVE:
8958		lun->flags |= CTL_LUN_PR_RESERVED;
8959		lun->res_type = msg->pr.pr_info.res_type;
8960		lun->pr_res_idx = msg->pr.pr_info.residx;
8961
8962		break;
8963
8964	case CTL_PR_RELEASE:
8965		/*
8966		 * if this isn't an exclusive access res generate UA for all
8967		 * other registrants.
8968		 */
8969		if (lun->res_type != SPR_TYPE_EX_AC
8970		 && lun->res_type != SPR_TYPE_WR_EX) {
8971			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8972				if (lun->per_res[i+persis_offset].registered)
8973					lun->pending_ua[i] |=
8974						CTL_UA_RES_RELEASE;
8975		}
8976
8977		lun->flags &= ~CTL_LUN_PR_RESERVED;
8978		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8979		lun->res_type = 0;
8980		break;
8981
8982	case CTL_PR_PREEMPT:
8983		ctl_pro_preempt_other(lun, msg);
8984		break;
8985	case CTL_PR_CLEAR:
8986		lun->flags &= ~CTL_LUN_PR_RESERVED;
8987		lun->res_type = 0;
8988		lun->pr_key_count = 0;
8989		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8990
8991		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8992			if (lun->per_res[i].registered == 0)
8993				continue;
8994			if (!persis_offset
8995			 && i < CTL_MAX_INITIATORS)
8996				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
8997			else if (persis_offset
8998			      && i >= persis_offset)
8999				lun->pending_ua[i-persis_offset] |=
9000					CTL_UA_RES_PREEMPT;
9001			memset(&lun->per_res[i].res_key, 0,
9002			       sizeof(struct scsi_per_res_key));
9003			lun->per_res[i].registered = 0;
9004		}
9005		lun->PRGeneration++;
9006		break;
9007	}
9008
9009	mtx_unlock(&lun->lun_lock);
9010}
9011
9012int
9013ctl_read_write(struct ctl_scsiio *ctsio)
9014{
9015	struct ctl_lun *lun;
9016	struct ctl_lba_len_flags *lbalen;
9017	uint64_t lba;
9018	uint32_t num_blocks;
9019	int flags, retval;
9020	int isread;
9021
9022	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9023
9024	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
9025
9026	flags = 0;
9027	retval = CTL_RETVAL_COMPLETE;
9028
9029	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
9030	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
9031	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
9032		uint32_t residx;
9033
9034		/*
9035		 * XXX KDM need a lock here.
9036		 */
9037		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9038		if ((lun->res_type == SPR_TYPE_EX_AC
9039		  && residx != lun->pr_res_idx)
9040		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
9041		   || lun->res_type == SPR_TYPE_EX_AC_AR)
9042		  && !lun->per_res[residx].registered)) {
9043			ctl_set_reservation_conflict(ctsio);
9044			ctl_done((union ctl_io *)ctsio);
9045			return (CTL_RETVAL_COMPLETE);
9046	        }
9047	}
9048
9049	switch (ctsio->cdb[0]) {
9050	case READ_6:
9051	case WRITE_6: {
9052		struct scsi_rw_6 *cdb;
9053
9054		cdb = (struct scsi_rw_6 *)ctsio->cdb;
9055
9056		lba = scsi_3btoul(cdb->addr);
9057		/* only 5 bits are valid in the most significant address byte */
9058		lba &= 0x1fffff;
9059		num_blocks = cdb->length;
9060		/*
9061		 * This is correct according to SBC-2.
9062		 */
9063		if (num_blocks == 0)
9064			num_blocks = 256;
9065		break;
9066	}
9067	case READ_10:
9068	case WRITE_10: {
9069		struct scsi_rw_10 *cdb;
9070
9071		cdb = (struct scsi_rw_10 *)ctsio->cdb;
9072		if (cdb->byte2 & SRW10_FUA)
9073			flags |= CTL_LLF_FUA;
9074		if (cdb->byte2 & SRW10_DPO)
9075			flags |= CTL_LLF_DPO;
9076		lba = scsi_4btoul(cdb->addr);
9077		num_blocks = scsi_2btoul(cdb->length);
9078		break;
9079	}
9080	case WRITE_VERIFY_10: {
9081		struct scsi_write_verify_10 *cdb;
9082
9083		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9084		flags |= CTL_LLF_FUA;
9085		if (cdb->byte2 & SWV_DPO)
9086			flags |= CTL_LLF_DPO;
9087		lba = scsi_4btoul(cdb->addr);
9088		num_blocks = scsi_2btoul(cdb->length);
9089		break;
9090	}
9091	case READ_12:
9092	case WRITE_12: {
9093		struct scsi_rw_12 *cdb;
9094
9095		cdb = (struct scsi_rw_12 *)ctsio->cdb;
9096		if (cdb->byte2 & SRW12_FUA)
9097			flags |= CTL_LLF_FUA;
9098		if (cdb->byte2 & SRW12_DPO)
9099			flags |= CTL_LLF_DPO;
9100		lba = scsi_4btoul(cdb->addr);
9101		num_blocks = scsi_4btoul(cdb->length);
9102		break;
9103	}
9104	case WRITE_VERIFY_12: {
9105		struct scsi_write_verify_12 *cdb;
9106
9107		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9108		flags |= CTL_LLF_FUA;
9109		if (cdb->byte2 & SWV_DPO)
9110			flags |= CTL_LLF_DPO;
9111		lba = scsi_4btoul(cdb->addr);
9112		num_blocks = scsi_4btoul(cdb->length);
9113		break;
9114	}
9115	case READ_16:
9116	case WRITE_16: {
9117		struct scsi_rw_16 *cdb;
9118
9119		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9120		if (cdb->byte2 & SRW12_FUA)
9121			flags |= CTL_LLF_FUA;
9122		if (cdb->byte2 & SRW12_DPO)
9123			flags |= CTL_LLF_DPO;
9124		lba = scsi_8btou64(cdb->addr);
9125		num_blocks = scsi_4btoul(cdb->length);
9126		break;
9127	}
9128	case WRITE_VERIFY_16: {
9129		struct scsi_write_verify_16 *cdb;
9130
9131		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9132		flags |= CTL_LLF_FUA;
9133		if (cdb->byte2 & SWV_DPO)
9134			flags |= CTL_LLF_DPO;
9135		lba = scsi_8btou64(cdb->addr);
9136		num_blocks = scsi_4btoul(cdb->length);
9137		break;
9138	}
9139	default:
9140		/*
9141		 * We got a command we don't support.  This shouldn't
9142		 * happen, commands should be filtered out above us.
9143		 */
9144		ctl_set_invalid_opcode(ctsio);
9145		ctl_done((union ctl_io *)ctsio);
9146
9147		return (CTL_RETVAL_COMPLETE);
9148		break; /* NOTREACHED */
9149	}
9150
9151	/*
9152	 * The first check is to make sure we're in bounds, the second
9153	 * check is to catch wrap-around problems.  If the lba + num blocks
9154	 * is less than the lba, then we've wrapped around and the block
9155	 * range is invalid anyway.
9156	 */
9157	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9158	 || ((lba + num_blocks) < lba)) {
9159		ctl_set_lba_out_of_range(ctsio);
9160		ctl_done((union ctl_io *)ctsio);
9161		return (CTL_RETVAL_COMPLETE);
9162	}
9163
9164	/*
9165	 * According to SBC-3, a transfer length of 0 is not an error.
9166	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9167	 * translates to 256 blocks for those commands.
9168	 */
9169	if (num_blocks == 0) {
9170		ctl_set_success(ctsio);
9171		ctl_done((union ctl_io *)ctsio);
9172		return (CTL_RETVAL_COMPLETE);
9173	}
9174
9175	/* Set FUA and/or DPO if caches are disabled. */
9176	if (isread) {
9177		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9178		    SCP_RCD) != 0)
9179			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9180	} else {
9181		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9182		    SCP_WCE) == 0)
9183			flags |= CTL_LLF_FUA;
9184	}
9185
9186	lbalen = (struct ctl_lba_len_flags *)
9187	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9188	lbalen->lba = lba;
9189	lbalen->len = num_blocks;
9190	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9191
9192	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9193	ctsio->kern_rel_offset = 0;
9194
9195	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9196
9197	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9198
9199	return (retval);
9200}
9201
9202static int
9203ctl_cnw_cont(union ctl_io *io)
9204{
9205	struct ctl_scsiio *ctsio;
9206	struct ctl_lun *lun;
9207	struct ctl_lba_len_flags *lbalen;
9208	int retval;
9209
9210	ctsio = &io->scsiio;
9211	ctsio->io_hdr.status = CTL_STATUS_NONE;
9212	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9213	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9214	lbalen = (struct ctl_lba_len_flags *)
9215	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9216	lbalen->flags &= ~CTL_LLF_COMPARE;
9217	lbalen->flags |= CTL_LLF_WRITE;
9218
9219	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9220	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9221	return (retval);
9222}
9223
9224int
9225ctl_cnw(struct ctl_scsiio *ctsio)
9226{
9227	struct ctl_lun *lun;
9228	struct ctl_lba_len_flags *lbalen;
9229	uint64_t lba;
9230	uint32_t num_blocks;
9231	int flags, retval;
9232
9233	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9234
9235	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9236
9237	flags = 0;
9238	retval = CTL_RETVAL_COMPLETE;
9239
9240	switch (ctsio->cdb[0]) {
9241	case COMPARE_AND_WRITE: {
9242		struct scsi_compare_and_write *cdb;
9243
9244		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9245		if (cdb->byte2 & SRW10_FUA)
9246			flags |= CTL_LLF_FUA;
9247		if (cdb->byte2 & SRW10_DPO)
9248			flags |= CTL_LLF_DPO;
9249		lba = scsi_8btou64(cdb->addr);
9250		num_blocks = cdb->length;
9251		break;
9252	}
9253	default:
9254		/*
9255		 * We got a command we don't support.  This shouldn't
9256		 * happen, commands should be filtered out above us.
9257		 */
9258		ctl_set_invalid_opcode(ctsio);
9259		ctl_done((union ctl_io *)ctsio);
9260
9261		return (CTL_RETVAL_COMPLETE);
9262		break; /* NOTREACHED */
9263	}
9264
9265	/*
9266	 * The first check is to make sure we're in bounds, the second
9267	 * check is to catch wrap-around problems.  If the lba + num blocks
9268	 * is less than the lba, then we've wrapped around and the block
9269	 * range is invalid anyway.
9270	 */
9271	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9272	 || ((lba + num_blocks) < lba)) {
9273		ctl_set_lba_out_of_range(ctsio);
9274		ctl_done((union ctl_io *)ctsio);
9275		return (CTL_RETVAL_COMPLETE);
9276	}
9277
9278	/*
9279	 * According to SBC-3, a transfer length of 0 is not an error.
9280	 */
9281	if (num_blocks == 0) {
9282		ctl_set_success(ctsio);
9283		ctl_done((union ctl_io *)ctsio);
9284		return (CTL_RETVAL_COMPLETE);
9285	}
9286
9287	/* Set FUA if write cache is disabled. */
9288	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9289	    SCP_WCE) == 0)
9290		flags |= CTL_LLF_FUA;
9291
9292	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9293	ctsio->kern_rel_offset = 0;
9294
9295	/*
9296	 * Set the IO_CONT flag, so that if this I/O gets passed to
9297	 * ctl_data_submit_done(), it'll get passed back to
9298	 * ctl_ctl_cnw_cont() for further processing.
9299	 */
9300	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9301	ctsio->io_cont = ctl_cnw_cont;
9302
9303	lbalen = (struct ctl_lba_len_flags *)
9304	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9305	lbalen->lba = lba;
9306	lbalen->len = num_blocks;
9307	lbalen->flags = CTL_LLF_COMPARE | flags;
9308
9309	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9310	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9311	return (retval);
9312}
9313
9314int
9315ctl_verify(struct ctl_scsiio *ctsio)
9316{
9317	struct ctl_lun *lun;
9318	struct ctl_lba_len_flags *lbalen;
9319	uint64_t lba;
9320	uint32_t num_blocks;
9321	int bytchk, flags;
9322	int retval;
9323
9324	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9325
9326	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9327
9328	bytchk = 0;
9329	flags = CTL_LLF_FUA;
9330	retval = CTL_RETVAL_COMPLETE;
9331
9332	switch (ctsio->cdb[0]) {
9333	case VERIFY_10: {
9334		struct scsi_verify_10 *cdb;
9335
9336		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9337		if (cdb->byte2 & SVFY_BYTCHK)
9338			bytchk = 1;
9339		if (cdb->byte2 & SVFY_DPO)
9340			flags |= CTL_LLF_DPO;
9341		lba = scsi_4btoul(cdb->addr);
9342		num_blocks = scsi_2btoul(cdb->length);
9343		break;
9344	}
9345	case VERIFY_12: {
9346		struct scsi_verify_12 *cdb;
9347
9348		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9349		if (cdb->byte2 & SVFY_BYTCHK)
9350			bytchk = 1;
9351		if (cdb->byte2 & SVFY_DPO)
9352			flags |= CTL_LLF_DPO;
9353		lba = scsi_4btoul(cdb->addr);
9354		num_blocks = scsi_4btoul(cdb->length);
9355		break;
9356	}
9357	case VERIFY_16: {
9358		struct scsi_rw_16 *cdb;
9359
9360		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9361		if (cdb->byte2 & SVFY_BYTCHK)
9362			bytchk = 1;
9363		if (cdb->byte2 & SVFY_DPO)
9364			flags |= CTL_LLF_DPO;
9365		lba = scsi_8btou64(cdb->addr);
9366		num_blocks = scsi_4btoul(cdb->length);
9367		break;
9368	}
9369	default:
9370		/*
9371		 * We got a command we don't support.  This shouldn't
9372		 * happen, commands should be filtered out above us.
9373		 */
9374		ctl_set_invalid_opcode(ctsio);
9375		ctl_done((union ctl_io *)ctsio);
9376		return (CTL_RETVAL_COMPLETE);
9377	}
9378
9379	/*
9380	 * The first check is to make sure we're in bounds, the second
9381	 * check is to catch wrap-around problems.  If the lba + num blocks
9382	 * is less than the lba, then we've wrapped around and the block
9383	 * range is invalid anyway.
9384	 */
9385	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9386	 || ((lba + num_blocks) < lba)) {
9387		ctl_set_lba_out_of_range(ctsio);
9388		ctl_done((union ctl_io *)ctsio);
9389		return (CTL_RETVAL_COMPLETE);
9390	}
9391
9392	/*
9393	 * According to SBC-3, a transfer length of 0 is not an error.
9394	 */
9395	if (num_blocks == 0) {
9396		ctl_set_success(ctsio);
9397		ctl_done((union ctl_io *)ctsio);
9398		return (CTL_RETVAL_COMPLETE);
9399	}
9400
9401	lbalen = (struct ctl_lba_len_flags *)
9402	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9403	lbalen->lba = lba;
9404	lbalen->len = num_blocks;
9405	if (bytchk) {
9406		lbalen->flags = CTL_LLF_COMPARE | flags;
9407		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9408	} else {
9409		lbalen->flags = CTL_LLF_VERIFY | flags;
9410		ctsio->kern_total_len = 0;
9411	}
9412	ctsio->kern_rel_offset = 0;
9413
9414	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9415	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9416	return (retval);
9417}
9418
9419int
9420ctl_report_luns(struct ctl_scsiio *ctsio)
9421{
9422	struct scsi_report_luns *cdb;
9423	struct scsi_report_luns_data *lun_data;
9424	struct ctl_lun *lun, *request_lun;
9425	int num_luns, retval;
9426	uint32_t alloc_len, lun_datalen;
9427	int num_filled, well_known;
9428	uint32_t initidx, targ_lun_id, lun_id;
9429
9430	retval = CTL_RETVAL_COMPLETE;
9431	well_known = 0;
9432
9433	cdb = (struct scsi_report_luns *)ctsio->cdb;
9434
9435	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9436
9437	mtx_lock(&control_softc->ctl_lock);
9438	num_luns = control_softc->num_luns;
9439	mtx_unlock(&control_softc->ctl_lock);
9440
9441	switch (cdb->select_report) {
9442	case RPL_REPORT_DEFAULT:
9443	case RPL_REPORT_ALL:
9444		break;
9445	case RPL_REPORT_WELLKNOWN:
9446		well_known = 1;
9447		num_luns = 0;
9448		break;
9449	default:
9450		ctl_set_invalid_field(ctsio,
9451				      /*sks_valid*/ 1,
9452				      /*command*/ 1,
9453				      /*field*/ 2,
9454				      /*bit_valid*/ 0,
9455				      /*bit*/ 0);
9456		ctl_done((union ctl_io *)ctsio);
9457		return (retval);
9458		break; /* NOTREACHED */
9459	}
9460
9461	alloc_len = scsi_4btoul(cdb->length);
9462	/*
9463	 * The initiator has to allocate at least 16 bytes for this request,
9464	 * so he can at least get the header and the first LUN.  Otherwise
9465	 * we reject the request (per SPC-3 rev 14, section 6.21).
9466	 */
9467	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9468	    sizeof(struct scsi_report_luns_lundata))) {
9469		ctl_set_invalid_field(ctsio,
9470				      /*sks_valid*/ 1,
9471				      /*command*/ 1,
9472				      /*field*/ 6,
9473				      /*bit_valid*/ 0,
9474				      /*bit*/ 0);
9475		ctl_done((union ctl_io *)ctsio);
9476		return (retval);
9477	}
9478
9479	request_lun = (struct ctl_lun *)
9480		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9481
9482	lun_datalen = sizeof(*lun_data) +
9483		(num_luns * sizeof(struct scsi_report_luns_lundata));
9484
9485	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9486	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9487	ctsio->kern_sg_entries = 0;
9488
9489	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9490
9491	mtx_lock(&control_softc->ctl_lock);
9492	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9493		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9494		if (lun_id >= CTL_MAX_LUNS)
9495			continue;
9496		lun = control_softc->ctl_luns[lun_id];
9497		if (lun == NULL)
9498			continue;
9499
9500		if (targ_lun_id <= 0xff) {
9501			/*
9502			 * Peripheral addressing method, bus number 0.
9503			 */
9504			lun_data->luns[num_filled].lundata[0] =
9505				RPL_LUNDATA_ATYP_PERIPH;
9506			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9507			num_filled++;
9508		} else if (targ_lun_id <= 0x3fff) {
9509			/*
9510			 * Flat addressing method.
9511			 */
9512			lun_data->luns[num_filled].lundata[0] =
9513				RPL_LUNDATA_ATYP_FLAT |
9514				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9515#ifdef OLDCTLHEADERS
9516				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9517				(targ_lun_id & SRLD_BUS_LUN_MASK);
9518#endif
9519			lun_data->luns[num_filled].lundata[1] =
9520#ifdef OLDCTLHEADERS
9521				targ_lun_id >> SRLD_BUS_LUN_BITS;
9522#endif
9523				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9524			num_filled++;
9525		} else {
9526			printf("ctl_report_luns: bogus LUN number %jd, "
9527			       "skipping\n", (intmax_t)targ_lun_id);
9528		}
9529		/*
9530		 * According to SPC-3, rev 14 section 6.21:
9531		 *
9532		 * "The execution of a REPORT LUNS command to any valid and
9533		 * installed logical unit shall clear the REPORTED LUNS DATA
9534		 * HAS CHANGED unit attention condition for all logical
9535		 * units of that target with respect to the requesting
9536		 * initiator. A valid and installed logical unit is one
9537		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9538		 * INQUIRY data (see 6.4.2)."
9539		 *
9540		 * If request_lun is NULL, the LUN this report luns command
9541		 * was issued to is either disabled or doesn't exist. In that
9542		 * case, we shouldn't clear any pending lun change unit
9543		 * attention.
9544		 */
9545		if (request_lun != NULL) {
9546			mtx_lock(&lun->lun_lock);
9547			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9548			mtx_unlock(&lun->lun_lock);
9549		}
9550	}
9551	mtx_unlock(&control_softc->ctl_lock);
9552
9553	/*
9554	 * It's quite possible that we've returned fewer LUNs than we allocated
9555	 * space for.  Trim it.
9556	 */
9557	lun_datalen = sizeof(*lun_data) +
9558		(num_filled * sizeof(struct scsi_report_luns_lundata));
9559
9560	if (lun_datalen < alloc_len) {
9561		ctsio->residual = alloc_len - lun_datalen;
9562		ctsio->kern_data_len = lun_datalen;
9563		ctsio->kern_total_len = lun_datalen;
9564	} else {
9565		ctsio->residual = 0;
9566		ctsio->kern_data_len = alloc_len;
9567		ctsio->kern_total_len = alloc_len;
9568	}
9569	ctsio->kern_data_resid = 0;
9570	ctsio->kern_rel_offset = 0;
9571	ctsio->kern_sg_entries = 0;
9572
9573	/*
9574	 * We set this to the actual data length, regardless of how much
9575	 * space we actually have to return results.  If the user looks at
9576	 * this value, he'll know whether or not he allocated enough space
9577	 * and reissue the command if necessary.  We don't support well
9578	 * known logical units, so if the user asks for that, return none.
9579	 */
9580	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9581
9582	/*
9583	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9584	 * this request.
9585	 */
9586	ctsio->scsi_status = SCSI_STATUS_OK;
9587
9588	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9589	ctsio->be_move_done = ctl_config_move_done;
9590	ctl_datamove((union ctl_io *)ctsio);
9591
9592	return (retval);
9593}
9594
9595int
9596ctl_request_sense(struct ctl_scsiio *ctsio)
9597{
9598	struct scsi_request_sense *cdb;
9599	struct scsi_sense_data *sense_ptr;
9600	struct ctl_lun *lun;
9601	uint32_t initidx;
9602	int have_error;
9603	scsi_sense_data_type sense_format;
9604
9605	cdb = (struct scsi_request_sense *)ctsio->cdb;
9606
9607	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9608
9609	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9610
9611	/*
9612	 * Determine which sense format the user wants.
9613	 */
9614	if (cdb->byte2 & SRS_DESC)
9615		sense_format = SSD_TYPE_DESC;
9616	else
9617		sense_format = SSD_TYPE_FIXED;
9618
9619	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9620	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9621	ctsio->kern_sg_entries = 0;
9622
9623	/*
9624	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9625	 * larger than the largest allowed value for the length field in the
9626	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9627	 */
9628	ctsio->residual = 0;
9629	ctsio->kern_data_len = cdb->length;
9630	ctsio->kern_total_len = cdb->length;
9631
9632	ctsio->kern_data_resid = 0;
9633	ctsio->kern_rel_offset = 0;
9634	ctsio->kern_sg_entries = 0;
9635
9636	/*
9637	 * If we don't have a LUN, we don't have any pending sense.
9638	 */
9639	if (lun == NULL)
9640		goto no_sense;
9641
9642	have_error = 0;
9643	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9644	/*
9645	 * Check for pending sense, and then for pending unit attentions.
9646	 * Pending sense gets returned first, then pending unit attentions.
9647	 */
9648	mtx_lock(&lun->lun_lock);
9649#ifdef CTL_WITH_CA
9650	if (ctl_is_set(lun->have_ca, initidx)) {
9651		scsi_sense_data_type stored_format;
9652
9653		/*
9654		 * Check to see which sense format was used for the stored
9655		 * sense data.
9656		 */
9657		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9658
9659		/*
9660		 * If the user requested a different sense format than the
9661		 * one we stored, then we need to convert it to the other
9662		 * format.  If we're going from descriptor to fixed format
9663		 * sense data, we may lose things in translation, depending
9664		 * on what options were used.
9665		 *
9666		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9667		 * for some reason we'll just copy it out as-is.
9668		 */
9669		if ((stored_format == SSD_TYPE_FIXED)
9670		 && (sense_format == SSD_TYPE_DESC))
9671			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9672			    &lun->pending_sense[initidx],
9673			    (struct scsi_sense_data_desc *)sense_ptr);
9674		else if ((stored_format == SSD_TYPE_DESC)
9675		      && (sense_format == SSD_TYPE_FIXED))
9676			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9677			    &lun->pending_sense[initidx],
9678			    (struct scsi_sense_data_fixed *)sense_ptr);
9679		else
9680			memcpy(sense_ptr, &lun->pending_sense[initidx],
9681			       ctl_min(sizeof(*sense_ptr),
9682			       sizeof(lun->pending_sense[initidx])));
9683
9684		ctl_clear_mask(lun->have_ca, initidx);
9685		have_error = 1;
9686	} else
9687#endif
9688	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9689		ctl_ua_type ua_type;
9690
9691		ua_type = ctl_build_ua(lun->pending_ua[initidx],
9692				       sense_ptr, sense_format);
9693		if (ua_type != CTL_UA_NONE) {
9694			have_error = 1;
9695			/* We're reporting this UA, so clear it */
9696			lun->pending_ua[initidx] &= ~ua_type;
9697		}
9698	}
9699	mtx_unlock(&lun->lun_lock);
9700
9701	/*
9702	 * We already have a pending error, return it.
9703	 */
9704	if (have_error != 0) {
9705		/*
9706		 * We report the SCSI status as OK, since the status of the
9707		 * request sense command itself is OK.
9708		 */
9709		ctsio->scsi_status = SCSI_STATUS_OK;
9710
9711		/*
9712		 * We report 0 for the sense length, because we aren't doing
9713		 * autosense in this case.  We're reporting sense as
9714		 * parameter data.
9715		 */
9716		ctsio->sense_len = 0;
9717		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9718		ctsio->be_move_done = ctl_config_move_done;
9719		ctl_datamove((union ctl_io *)ctsio);
9720
9721		return (CTL_RETVAL_COMPLETE);
9722	}
9723
9724no_sense:
9725
9726	/*
9727	 * No sense information to report, so we report that everything is
9728	 * okay.
9729	 */
9730	ctl_set_sense_data(sense_ptr,
9731			   lun,
9732			   sense_format,
9733			   /*current_error*/ 1,
9734			   /*sense_key*/ SSD_KEY_NO_SENSE,
9735			   /*asc*/ 0x00,
9736			   /*ascq*/ 0x00,
9737			   SSD_ELEM_NONE);
9738
9739	ctsio->scsi_status = SCSI_STATUS_OK;
9740
9741	/*
9742	 * We report 0 for the sense length, because we aren't doing
9743	 * autosense in this case.  We're reporting sense as parameter data.
9744	 */
9745	ctsio->sense_len = 0;
9746	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9747	ctsio->be_move_done = ctl_config_move_done;
9748	ctl_datamove((union ctl_io *)ctsio);
9749
9750	return (CTL_RETVAL_COMPLETE);
9751}
9752
9753int
9754ctl_tur(struct ctl_scsiio *ctsio)
9755{
9756	struct ctl_lun *lun;
9757
9758	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9759
9760	CTL_DEBUG_PRINT(("ctl_tur\n"));
9761
9762	if (lun == NULL)
9763		return (EINVAL);
9764
9765	ctsio->scsi_status = SCSI_STATUS_OK;
9766	ctsio->io_hdr.status = CTL_SUCCESS;
9767
9768	ctl_done((union ctl_io *)ctsio);
9769
9770	return (CTL_RETVAL_COMPLETE);
9771}
9772
9773#ifdef notyet
9774static int
9775ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9776{
9777
9778}
9779#endif
9780
9781static int
9782ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9783{
9784	struct scsi_vpd_supported_pages *pages;
9785	int sup_page_size;
9786	struct ctl_lun *lun;
9787
9788	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9789
9790	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9791	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9792	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9793	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9794	ctsio->kern_sg_entries = 0;
9795
9796	if (sup_page_size < alloc_len) {
9797		ctsio->residual = alloc_len - sup_page_size;
9798		ctsio->kern_data_len = sup_page_size;
9799		ctsio->kern_total_len = sup_page_size;
9800	} else {
9801		ctsio->residual = 0;
9802		ctsio->kern_data_len = alloc_len;
9803		ctsio->kern_total_len = alloc_len;
9804	}
9805	ctsio->kern_data_resid = 0;
9806	ctsio->kern_rel_offset = 0;
9807	ctsio->kern_sg_entries = 0;
9808
9809	/*
9810	 * The control device is always connected.  The disk device, on the
9811	 * other hand, may not be online all the time.  Need to change this
9812	 * to figure out whether the disk device is actually online or not.
9813	 */
9814	if (lun != NULL)
9815		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9816				lun->be_lun->lun_type;
9817	else
9818		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9819
9820	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9821	/* Supported VPD pages */
9822	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9823	/* Serial Number */
9824	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9825	/* Device Identification */
9826	pages->page_list[2] = SVPD_DEVICE_ID;
9827	/* Extended INQUIRY Data */
9828	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9829	/* Mode Page Policy */
9830	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9831	/* SCSI Ports */
9832	pages->page_list[5] = SVPD_SCSI_PORTS;
9833	/* Third-party Copy */
9834	pages->page_list[6] = SVPD_SCSI_TPC;
9835	/* Block limits */
9836	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9837	/* Block Device Characteristics */
9838	pages->page_list[8] = SVPD_BDC;
9839	/* Logical Block Provisioning */
9840	pages->page_list[9] = SVPD_LBP;
9841
9842	ctsio->scsi_status = SCSI_STATUS_OK;
9843
9844	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9845	ctsio->be_move_done = ctl_config_move_done;
9846	ctl_datamove((union ctl_io *)ctsio);
9847
9848	return (CTL_RETVAL_COMPLETE);
9849}
9850
9851static int
9852ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9853{
9854	struct scsi_vpd_unit_serial_number *sn_ptr;
9855	struct ctl_lun *lun;
9856
9857	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9858
9859	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9860	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9861	ctsio->kern_sg_entries = 0;
9862
9863	if (sizeof(*sn_ptr) < alloc_len) {
9864		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9865		ctsio->kern_data_len = sizeof(*sn_ptr);
9866		ctsio->kern_total_len = sizeof(*sn_ptr);
9867	} else {
9868		ctsio->residual = 0;
9869		ctsio->kern_data_len = alloc_len;
9870		ctsio->kern_total_len = alloc_len;
9871	}
9872	ctsio->kern_data_resid = 0;
9873	ctsio->kern_rel_offset = 0;
9874	ctsio->kern_sg_entries = 0;
9875
9876	/*
9877	 * The control device is always connected.  The disk device, on the
9878	 * other hand, may not be online all the time.  Need to change this
9879	 * to figure out whether the disk device is actually online or not.
9880	 */
9881	if (lun != NULL)
9882		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9883				  lun->be_lun->lun_type;
9884	else
9885		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9886
9887	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9888	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9889	/*
9890	 * If we don't have a LUN, we just leave the serial number as
9891	 * all spaces.
9892	 */
9893	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9894	if (lun != NULL) {
9895		strncpy((char *)sn_ptr->serial_num,
9896			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9897	}
9898	ctsio->scsi_status = SCSI_STATUS_OK;
9899
9900	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9901	ctsio->be_move_done = ctl_config_move_done;
9902	ctl_datamove((union ctl_io *)ctsio);
9903
9904	return (CTL_RETVAL_COMPLETE);
9905}
9906
9907
9908static int
9909ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9910{
9911	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9912	struct ctl_lun *lun;
9913	int data_len;
9914
9915	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9916
9917	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9918	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9919
9920	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9921	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9922	ctsio->kern_sg_entries = 0;
9923
9924	if (data_len < alloc_len) {
9925		ctsio->residual = alloc_len - data_len;
9926		ctsio->kern_data_len = data_len;
9927		ctsio->kern_total_len = data_len;
9928	} else {
9929		ctsio->residual = 0;
9930		ctsio->kern_data_len = alloc_len;
9931		ctsio->kern_total_len = alloc_len;
9932	}
9933	ctsio->kern_data_resid = 0;
9934	ctsio->kern_rel_offset = 0;
9935	ctsio->kern_sg_entries = 0;
9936
9937	/*
9938	 * The control device is always connected.  The disk device, on the
9939	 * other hand, may not be online all the time.
9940	 */
9941	if (lun != NULL)
9942		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9943				     lun->be_lun->lun_type;
9944	else
9945		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9946	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9947	eid_ptr->page_length = data_len - 4;
9948	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9949	eid_ptr->flags3 = SVPD_EID_V_SUP;
9950
9951	ctsio->scsi_status = SCSI_STATUS_OK;
9952	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9953	ctsio->be_move_done = ctl_config_move_done;
9954	ctl_datamove((union ctl_io *)ctsio);
9955
9956	return (CTL_RETVAL_COMPLETE);
9957}
9958
9959static int
9960ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9961{
9962	struct scsi_vpd_mode_page_policy *mpp_ptr;
9963	struct ctl_lun *lun;
9964	int data_len;
9965
9966	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9967
9968	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9969	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9970
9971	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9972	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9973	ctsio->kern_sg_entries = 0;
9974
9975	if (data_len < alloc_len) {
9976		ctsio->residual = alloc_len - data_len;
9977		ctsio->kern_data_len = data_len;
9978		ctsio->kern_total_len = data_len;
9979	} else {
9980		ctsio->residual = 0;
9981		ctsio->kern_data_len = alloc_len;
9982		ctsio->kern_total_len = alloc_len;
9983	}
9984	ctsio->kern_data_resid = 0;
9985	ctsio->kern_rel_offset = 0;
9986	ctsio->kern_sg_entries = 0;
9987
9988	/*
9989	 * The control device is always connected.  The disk device, on the
9990	 * other hand, may not be online all the time.
9991	 */
9992	if (lun != NULL)
9993		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9994				     lun->be_lun->lun_type;
9995	else
9996		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9997	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9998	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9999	mpp_ptr->descr[0].page_code = 0x3f;
10000	mpp_ptr->descr[0].subpage_code = 0xff;
10001	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
10002
10003	ctsio->scsi_status = SCSI_STATUS_OK;
10004	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10005	ctsio->be_move_done = ctl_config_move_done;
10006	ctl_datamove((union ctl_io *)ctsio);
10007
10008	return (CTL_RETVAL_COMPLETE);
10009}
10010
10011static int
10012ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
10013{
10014	struct scsi_vpd_device_id *devid_ptr;
10015	struct scsi_vpd_id_descriptor *desc;
10016	struct ctl_softc *ctl_softc;
10017	struct ctl_lun *lun;
10018	struct ctl_port *port;
10019	int data_len;
10020	uint8_t proto;
10021
10022	ctl_softc = control_softc;
10023
10024	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
10025	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10026
10027	data_len = sizeof(struct scsi_vpd_device_id) +
10028	    sizeof(struct scsi_vpd_id_descriptor) +
10029		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
10030	    sizeof(struct scsi_vpd_id_descriptor) +
10031		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
10032	if (lun && lun->lun_devid)
10033		data_len += lun->lun_devid->len;
10034	if (port->port_devid)
10035		data_len += port->port_devid->len;
10036	if (port->target_devid)
10037		data_len += port->target_devid->len;
10038
10039	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10040	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
10041	ctsio->kern_sg_entries = 0;
10042
10043	if (data_len < alloc_len) {
10044		ctsio->residual = alloc_len - data_len;
10045		ctsio->kern_data_len = data_len;
10046		ctsio->kern_total_len = data_len;
10047	} else {
10048		ctsio->residual = 0;
10049		ctsio->kern_data_len = alloc_len;
10050		ctsio->kern_total_len = alloc_len;
10051	}
10052	ctsio->kern_data_resid = 0;
10053	ctsio->kern_rel_offset = 0;
10054	ctsio->kern_sg_entries = 0;
10055
10056	/*
10057	 * The control device is always connected.  The disk device, on the
10058	 * other hand, may not be online all the time.
10059	 */
10060	if (lun != NULL)
10061		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10062				     lun->be_lun->lun_type;
10063	else
10064		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10065	devid_ptr->page_code = SVPD_DEVICE_ID;
10066	scsi_ulto2b(data_len - 4, devid_ptr->length);
10067
10068	if (port->port_type == CTL_PORT_FC)
10069		proto = SCSI_PROTO_FC << 4;
10070	else if (port->port_type == CTL_PORT_ISCSI)
10071		proto = SCSI_PROTO_ISCSI << 4;
10072	else
10073		proto = SCSI_PROTO_SPI << 4;
10074	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
10075
10076	/*
10077	 * We're using a LUN association here.  i.e., this device ID is a
10078	 * per-LUN identifier.
10079	 */
10080	if (lun && lun->lun_devid) {
10081		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
10082		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10083		    lun->lun_devid->len);
10084	}
10085
10086	/*
10087	 * This is for the WWPN which is a port association.
10088	 */
10089	if (port->port_devid) {
10090		memcpy(desc, port->port_devid->data, port->port_devid->len);
10091		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10092		    port->port_devid->len);
10093	}
10094
10095	/*
10096	 * This is for the Relative Target Port(type 4h) identifier
10097	 */
10098	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10099	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10100	    SVPD_ID_TYPE_RELTARG;
10101	desc->length = 4;
10102	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
10103	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10104	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
10105
10106	/*
10107	 * This is for the Target Port Group(type 5h) identifier
10108	 */
10109	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10110	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10111	    SVPD_ID_TYPE_TPORTGRP;
10112	desc->length = 4;
10113	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
10114	    &desc->identifier[2]);
10115	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10116	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10117
10118	/*
10119	 * This is for the Target identifier
10120	 */
10121	if (port->target_devid) {
10122		memcpy(desc, port->target_devid->data, port->target_devid->len);
10123	}
10124
10125	ctsio->scsi_status = SCSI_STATUS_OK;
10126	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10127	ctsio->be_move_done = ctl_config_move_done;
10128	ctl_datamove((union ctl_io *)ctsio);
10129
10130	return (CTL_RETVAL_COMPLETE);
10131}
10132
10133static int
10134ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10135{
10136	struct ctl_softc *softc = control_softc;
10137	struct scsi_vpd_scsi_ports *sp;
10138	struct scsi_vpd_port_designation *pd;
10139	struct scsi_vpd_port_designation_cont *pdc;
10140	struct ctl_lun *lun;
10141	struct ctl_port *port;
10142	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10143	int num_target_port_groups, single;
10144
10145	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10146
10147	single = ctl_is_single;
10148	if (single)
10149		num_target_port_groups = 1;
10150	else
10151		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10152	num_target_ports = 0;
10153	iid_len = 0;
10154	id_len = 0;
10155	mtx_lock(&softc->ctl_lock);
10156	STAILQ_FOREACH(port, &softc->port_list, links) {
10157		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10158			continue;
10159		if (lun != NULL &&
10160		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10161		    CTL_MAX_LUNS)
10162			continue;
10163		num_target_ports++;
10164		if (port->init_devid)
10165			iid_len += port->init_devid->len;
10166		if (port->port_devid)
10167			id_len += port->port_devid->len;
10168	}
10169	mtx_unlock(&softc->ctl_lock);
10170
10171	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10172	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10173	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10174	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10175	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10176	ctsio->kern_sg_entries = 0;
10177
10178	if (data_len < alloc_len) {
10179		ctsio->residual = alloc_len - data_len;
10180		ctsio->kern_data_len = data_len;
10181		ctsio->kern_total_len = data_len;
10182	} else {
10183		ctsio->residual = 0;
10184		ctsio->kern_data_len = alloc_len;
10185		ctsio->kern_total_len = alloc_len;
10186	}
10187	ctsio->kern_data_resid = 0;
10188	ctsio->kern_rel_offset = 0;
10189	ctsio->kern_sg_entries = 0;
10190
10191	/*
10192	 * The control device is always connected.  The disk device, on the
10193	 * other hand, may not be online all the time.  Need to change this
10194	 * to figure out whether the disk device is actually online or not.
10195	 */
10196	if (lun != NULL)
10197		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10198				  lun->be_lun->lun_type;
10199	else
10200		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10201
10202	sp->page_code = SVPD_SCSI_PORTS;
10203	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10204	    sp->page_length);
10205	pd = &sp->design[0];
10206
10207	mtx_lock(&softc->ctl_lock);
10208	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10209		pg = 0;
10210	else
10211		pg = 1;
10212	for (g = 0; g < num_target_port_groups; g++) {
10213		STAILQ_FOREACH(port, &softc->port_list, links) {
10214			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10215				continue;
10216			if (lun != NULL &&
10217			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10218			    CTL_MAX_LUNS)
10219				continue;
10220			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10221			scsi_ulto2b(p, pd->relative_port_id);
10222			if (port->init_devid && g == pg) {
10223				iid_len = port->init_devid->len;
10224				memcpy(pd->initiator_transportid,
10225				    port->init_devid->data, port->init_devid->len);
10226			} else
10227				iid_len = 0;
10228			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10229			pdc = (struct scsi_vpd_port_designation_cont *)
10230			    (&pd->initiator_transportid[iid_len]);
10231			if (port->port_devid && g == pg) {
10232				id_len = port->port_devid->len;
10233				memcpy(pdc->target_port_descriptors,
10234				    port->port_devid->data, port->port_devid->len);
10235			} else
10236				id_len = 0;
10237			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10238			pd = (struct scsi_vpd_port_designation *)
10239			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10240		}
10241	}
10242	mtx_unlock(&softc->ctl_lock);
10243
10244	ctsio->scsi_status = SCSI_STATUS_OK;
10245	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10246	ctsio->be_move_done = ctl_config_move_done;
10247	ctl_datamove((union ctl_io *)ctsio);
10248
10249	return (CTL_RETVAL_COMPLETE);
10250}
10251
10252static int
10253ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10254{
10255	struct scsi_vpd_block_limits *bl_ptr;
10256	struct ctl_lun *lun;
10257	int bs;
10258
10259	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10260
10261	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10262	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10263	ctsio->kern_sg_entries = 0;
10264
10265	if (sizeof(*bl_ptr) < alloc_len) {
10266		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10267		ctsio->kern_data_len = sizeof(*bl_ptr);
10268		ctsio->kern_total_len = sizeof(*bl_ptr);
10269	} else {
10270		ctsio->residual = 0;
10271		ctsio->kern_data_len = alloc_len;
10272		ctsio->kern_total_len = alloc_len;
10273	}
10274	ctsio->kern_data_resid = 0;
10275	ctsio->kern_rel_offset = 0;
10276	ctsio->kern_sg_entries = 0;
10277
10278	/*
10279	 * The control device is always connected.  The disk device, on the
10280	 * other hand, may not be online all the time.  Need to change this
10281	 * to figure out whether the disk device is actually online or not.
10282	 */
10283	if (lun != NULL)
10284		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10285				  lun->be_lun->lun_type;
10286	else
10287		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10288
10289	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10290	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
10291	bl_ptr->max_cmp_write_len = 0xff;
10292	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10293	if (lun != NULL) {
10294		bs = lun->be_lun->blocksize;
10295		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10296		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10297			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10298			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10299			if (lun->be_lun->pblockexp != 0) {
10300				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10301				    bl_ptr->opt_unmap_grain);
10302				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10303				    bl_ptr->unmap_grain_align);
10304			}
10305		}
10306	}
10307	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10308
10309	ctsio->scsi_status = SCSI_STATUS_OK;
10310	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10311	ctsio->be_move_done = ctl_config_move_done;
10312	ctl_datamove((union ctl_io *)ctsio);
10313
10314	return (CTL_RETVAL_COMPLETE);
10315}
10316
10317static int
10318ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10319{
10320	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10321	struct ctl_lun *lun;
10322
10323	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10324
10325	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10326	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10327	ctsio->kern_sg_entries = 0;
10328
10329	if (sizeof(*bdc_ptr) < alloc_len) {
10330		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10331		ctsio->kern_data_len = sizeof(*bdc_ptr);
10332		ctsio->kern_total_len = sizeof(*bdc_ptr);
10333	} else {
10334		ctsio->residual = 0;
10335		ctsio->kern_data_len = alloc_len;
10336		ctsio->kern_total_len = alloc_len;
10337	}
10338	ctsio->kern_data_resid = 0;
10339	ctsio->kern_rel_offset = 0;
10340	ctsio->kern_sg_entries = 0;
10341
10342	/*
10343	 * The control device is always connected.  The disk device, on the
10344	 * other hand, may not be online all the time.  Need to change this
10345	 * to figure out whether the disk device is actually online or not.
10346	 */
10347	if (lun != NULL)
10348		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10349				  lun->be_lun->lun_type;
10350	else
10351		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10352	bdc_ptr->page_code = SVPD_BDC;
10353	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10354	scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10355	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10356
10357	ctsio->scsi_status = SCSI_STATUS_OK;
10358	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10359	ctsio->be_move_done = ctl_config_move_done;
10360	ctl_datamove((union ctl_io *)ctsio);
10361
10362	return (CTL_RETVAL_COMPLETE);
10363}
10364
10365static int
10366ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10367{
10368	struct scsi_vpd_logical_block_prov *lbp_ptr;
10369	struct ctl_lun *lun;
10370
10371	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10372
10373	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10374	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10375	ctsio->kern_sg_entries = 0;
10376
10377	if (sizeof(*lbp_ptr) < alloc_len) {
10378		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10379		ctsio->kern_data_len = sizeof(*lbp_ptr);
10380		ctsio->kern_total_len = sizeof(*lbp_ptr);
10381	} else {
10382		ctsio->residual = 0;
10383		ctsio->kern_data_len = alloc_len;
10384		ctsio->kern_total_len = alloc_len;
10385	}
10386	ctsio->kern_data_resid = 0;
10387	ctsio->kern_rel_offset = 0;
10388	ctsio->kern_sg_entries = 0;
10389
10390	/*
10391	 * The control device is always connected.  The disk device, on the
10392	 * other hand, may not be online all the time.  Need to change this
10393	 * to figure out whether the disk device is actually online or not.
10394	 */
10395	if (lun != NULL)
10396		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10397				  lun->be_lun->lun_type;
10398	else
10399		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10400
10401	lbp_ptr->page_code = SVPD_LBP;
10402	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10403	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10404		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10405		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10406		lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10407	}
10408
10409	ctsio->scsi_status = SCSI_STATUS_OK;
10410	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10411	ctsio->be_move_done = ctl_config_move_done;
10412	ctl_datamove((union ctl_io *)ctsio);
10413
10414	return (CTL_RETVAL_COMPLETE);
10415}
10416
10417static int
10418ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10419{
10420	struct scsi_inquiry *cdb;
10421	struct ctl_lun *lun;
10422	int alloc_len, retval;
10423
10424	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10425	cdb = (struct scsi_inquiry *)ctsio->cdb;
10426
10427	retval = CTL_RETVAL_COMPLETE;
10428
10429	alloc_len = scsi_2btoul(cdb->length);
10430
10431	switch (cdb->page_code) {
10432	case SVPD_SUPPORTED_PAGES:
10433		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10434		break;
10435	case SVPD_UNIT_SERIAL_NUMBER:
10436		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10437		break;
10438	case SVPD_DEVICE_ID:
10439		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10440		break;
10441	case SVPD_EXTENDED_INQUIRY_DATA:
10442		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10443		break;
10444	case SVPD_MODE_PAGE_POLICY:
10445		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10446		break;
10447	case SVPD_SCSI_PORTS:
10448		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10449		break;
10450	case SVPD_SCSI_TPC:
10451		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10452		break;
10453	case SVPD_BLOCK_LIMITS:
10454		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10455		break;
10456	case SVPD_BDC:
10457		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10458		break;
10459	case SVPD_LBP:
10460		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10461		break;
10462	default:
10463		ctl_set_invalid_field(ctsio,
10464				      /*sks_valid*/ 1,
10465				      /*command*/ 1,
10466				      /*field*/ 2,
10467				      /*bit_valid*/ 0,
10468				      /*bit*/ 0);
10469		ctl_done((union ctl_io *)ctsio);
10470		retval = CTL_RETVAL_COMPLETE;
10471		break;
10472	}
10473
10474	return (retval);
10475}
10476
10477static int
10478ctl_inquiry_std(struct ctl_scsiio *ctsio)
10479{
10480	struct scsi_inquiry_data *inq_ptr;
10481	struct scsi_inquiry *cdb;
10482	struct ctl_softc *ctl_softc;
10483	struct ctl_lun *lun;
10484	char *val;
10485	uint32_t alloc_len;
10486	ctl_port_type port_type;
10487
10488	ctl_softc = control_softc;
10489
10490	/*
10491	 * Figure out whether we're talking to a Fibre Channel port or not.
10492	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10493	 * SCSI front ends.
10494	 */
10495	port_type = ctl_softc->ctl_ports[
10496	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10497	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10498		port_type = CTL_PORT_SCSI;
10499
10500	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10501	cdb = (struct scsi_inquiry *)ctsio->cdb;
10502	alloc_len = scsi_2btoul(cdb->length);
10503
10504	/*
10505	 * We malloc the full inquiry data size here and fill it
10506	 * in.  If the user only asks for less, we'll give him
10507	 * that much.
10508	 */
10509	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
10510	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10511	ctsio->kern_sg_entries = 0;
10512	ctsio->kern_data_resid = 0;
10513	ctsio->kern_rel_offset = 0;
10514
10515	if (sizeof(*inq_ptr) < alloc_len) {
10516		ctsio->residual = alloc_len - sizeof(*inq_ptr);
10517		ctsio->kern_data_len = sizeof(*inq_ptr);
10518		ctsio->kern_total_len = sizeof(*inq_ptr);
10519	} else {
10520		ctsio->residual = 0;
10521		ctsio->kern_data_len = alloc_len;
10522		ctsio->kern_total_len = alloc_len;
10523	}
10524
10525	/*
10526	 * If we have a LUN configured, report it as connected.  Otherwise,
10527	 * report that it is offline or no device is supported, depending
10528	 * on the value of inquiry_pq_no_lun.
10529	 *
10530	 * According to the spec (SPC-4 r34), the peripheral qualifier
10531	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10532	 *
10533	 * "A peripheral device having the specified peripheral device type
10534	 * is not connected to this logical unit. However, the device
10535	 * server is capable of supporting the specified peripheral device
10536	 * type on this logical unit."
10537	 *
10538	 * According to the same spec, the peripheral qualifier
10539	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10540	 *
10541	 * "The device server is not capable of supporting a peripheral
10542	 * device on this logical unit. For this peripheral qualifier the
10543	 * peripheral device type shall be set to 1Fh. All other peripheral
10544	 * device type values are reserved for this peripheral qualifier."
10545	 *
10546	 * Given the text, it would seem that we probably want to report that
10547	 * the LUN is offline here.  There is no LUN connected, but we can
10548	 * support a LUN at the given LUN number.
10549	 *
10550	 * In the real world, though, it sounds like things are a little
10551	 * different:
10552	 *
10553	 * - Linux, when presented with a LUN with the offline peripheral
10554	 *   qualifier, will create an sg driver instance for it.  So when
10555	 *   you attach it to CTL, you wind up with a ton of sg driver
10556	 *   instances.  (One for every LUN that Linux bothered to probe.)
10557	 *   Linux does this despite the fact that it issues a REPORT LUNs
10558	 *   to LUN 0 to get the inventory of supported LUNs.
10559	 *
10560	 * - There is other anecdotal evidence (from Emulex folks) about
10561	 *   arrays that use the offline peripheral qualifier for LUNs that
10562	 *   are on the "passive" path in an active/passive array.
10563	 *
10564	 * So the solution is provide a hopefully reasonable default
10565	 * (return bad/no LUN) and allow the user to change the behavior
10566	 * with a tunable/sysctl variable.
10567	 */
10568	if (lun != NULL)
10569		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10570				  lun->be_lun->lun_type;
10571	else if (ctl_softc->inquiry_pq_no_lun == 0)
10572		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10573	else
10574		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10575
10576	/* RMB in byte 2 is 0 */
10577	inq_ptr->version = SCSI_REV_SPC4;
10578
10579	/*
10580	 * According to SAM-3, even if a device only supports a single
10581	 * level of LUN addressing, it should still set the HISUP bit:
10582	 *
10583	 * 4.9.1 Logical unit numbers overview
10584	 *
10585	 * All logical unit number formats described in this standard are
10586	 * hierarchical in structure even when only a single level in that
10587	 * hierarchy is used. The HISUP bit shall be set to one in the
10588	 * standard INQUIRY data (see SPC-2) when any logical unit number
10589	 * format described in this standard is used.  Non-hierarchical
10590	 * formats are outside the scope of this standard.
10591	 *
10592	 * Therefore we set the HiSup bit here.
10593	 *
10594	 * The reponse format is 2, per SPC-3.
10595	 */
10596	inq_ptr->response_format = SID_HiSup | 2;
10597
10598	inq_ptr->additional_length =
10599	    offsetof(struct scsi_inquiry_data, vendor_specific1) -
10600	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10601	CTL_DEBUG_PRINT(("additional_length = %d\n",
10602			 inq_ptr->additional_length));
10603
10604	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10605	/* 16 bit addressing */
10606	if (port_type == CTL_PORT_SCSI)
10607		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10608	/* XXX set the SID_MultiP bit here if we're actually going to
10609	   respond on multiple ports */
10610	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10611
10612	/* 16 bit data bus, synchronous transfers */
10613	if (port_type == CTL_PORT_SCSI)
10614		inq_ptr->flags = SID_WBus16 | SID_Sync;
10615	/*
10616	 * XXX KDM do we want to support tagged queueing on the control
10617	 * device at all?
10618	 */
10619	if ((lun == NULL)
10620	 || (lun->be_lun->lun_type != T_PROCESSOR))
10621		inq_ptr->flags |= SID_CmdQue;
10622	/*
10623	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10624	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10625	 * name and 4 bytes for the revision.
10626	 */
10627	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10628	    "vendor")) == NULL) {
10629		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10630	} else {
10631		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10632		strncpy(inq_ptr->vendor, val,
10633		    min(sizeof(inq_ptr->vendor), strlen(val)));
10634	}
10635	if (lun == NULL) {
10636		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10637		    sizeof(inq_ptr->product));
10638	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10639		switch (lun->be_lun->lun_type) {
10640		case T_DIRECT:
10641			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10642			    sizeof(inq_ptr->product));
10643			break;
10644		case T_PROCESSOR:
10645			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10646			    sizeof(inq_ptr->product));
10647			break;
10648		default:
10649			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10650			    sizeof(inq_ptr->product));
10651			break;
10652		}
10653	} else {
10654		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10655		strncpy(inq_ptr->product, val,
10656		    min(sizeof(inq_ptr->product), strlen(val)));
10657	}
10658
10659	/*
10660	 * XXX make this a macro somewhere so it automatically gets
10661	 * incremented when we make changes.
10662	 */
10663	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10664	    "revision")) == NULL) {
10665		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10666	} else {
10667		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10668		strncpy(inq_ptr->revision, val,
10669		    min(sizeof(inq_ptr->revision), strlen(val)));
10670	}
10671
10672	/*
10673	 * For parallel SCSI, we support double transition and single
10674	 * transition clocking.  We also support QAS (Quick Arbitration
10675	 * and Selection) and Information Unit transfers on both the
10676	 * control and array devices.
10677	 */
10678	if (port_type == CTL_PORT_SCSI)
10679		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10680				    SID_SPI_IUS;
10681
10682	/* SAM-5 (no version claimed) */
10683	scsi_ulto2b(0x00A0, inq_ptr->version1);
10684	/* SPC-4 (no version claimed) */
10685	scsi_ulto2b(0x0460, inq_ptr->version2);
10686	if (port_type == CTL_PORT_FC) {
10687		/* FCP-2 ANSI INCITS.350:2003 */
10688		scsi_ulto2b(0x0917, inq_ptr->version3);
10689	} else if (port_type == CTL_PORT_SCSI) {
10690		/* SPI-4 ANSI INCITS.362:200x */
10691		scsi_ulto2b(0x0B56, inq_ptr->version3);
10692	} else if (port_type == CTL_PORT_ISCSI) {
10693		/* iSCSI (no version claimed) */
10694		scsi_ulto2b(0x0960, inq_ptr->version3);
10695	} else if (port_type == CTL_PORT_SAS) {
10696		/* SAS (no version claimed) */
10697		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10698	}
10699
10700	if (lun == NULL) {
10701		/* SBC-3 (no version claimed) */
10702		scsi_ulto2b(0x04C0, inq_ptr->version4);
10703	} else {
10704		switch (lun->be_lun->lun_type) {
10705		case T_DIRECT:
10706			/* SBC-3 (no version claimed) */
10707			scsi_ulto2b(0x04C0, inq_ptr->version4);
10708			break;
10709		case T_PROCESSOR:
10710		default:
10711			break;
10712		}
10713	}
10714
10715	ctsio->scsi_status = SCSI_STATUS_OK;
10716	if (ctsio->kern_data_len > 0) {
10717		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10718		ctsio->be_move_done = ctl_config_move_done;
10719		ctl_datamove((union ctl_io *)ctsio);
10720	} else {
10721		ctsio->io_hdr.status = CTL_SUCCESS;
10722		ctl_done((union ctl_io *)ctsio);
10723	}
10724
10725	return (CTL_RETVAL_COMPLETE);
10726}
10727
10728int
10729ctl_inquiry(struct ctl_scsiio *ctsio)
10730{
10731	struct scsi_inquiry *cdb;
10732	int retval;
10733
10734	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10735
10736	cdb = (struct scsi_inquiry *)ctsio->cdb;
10737	if (cdb->byte2 & SI_EVPD)
10738		retval = ctl_inquiry_evpd(ctsio);
10739	else if (cdb->page_code == 0)
10740		retval = ctl_inquiry_std(ctsio);
10741	else {
10742		ctl_set_invalid_field(ctsio,
10743				      /*sks_valid*/ 1,
10744				      /*command*/ 1,
10745				      /*field*/ 2,
10746				      /*bit_valid*/ 0,
10747				      /*bit*/ 0);
10748		ctl_done((union ctl_io *)ctsio);
10749		return (CTL_RETVAL_COMPLETE);
10750	}
10751
10752	return (retval);
10753}
10754
10755/*
10756 * For known CDB types, parse the LBA and length.
10757 */
10758static int
10759ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10760{
10761	if (io->io_hdr.io_type != CTL_IO_SCSI)
10762		return (1);
10763
10764	switch (io->scsiio.cdb[0]) {
10765	case COMPARE_AND_WRITE: {
10766		struct scsi_compare_and_write *cdb;
10767
10768		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10769
10770		*lba = scsi_8btou64(cdb->addr);
10771		*len = cdb->length;
10772		break;
10773	}
10774	case READ_6:
10775	case WRITE_6: {
10776		struct scsi_rw_6 *cdb;
10777
10778		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10779
10780		*lba = scsi_3btoul(cdb->addr);
10781		/* only 5 bits are valid in the most significant address byte */
10782		*lba &= 0x1fffff;
10783		*len = cdb->length;
10784		break;
10785	}
10786	case READ_10:
10787	case WRITE_10: {
10788		struct scsi_rw_10 *cdb;
10789
10790		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10791
10792		*lba = scsi_4btoul(cdb->addr);
10793		*len = scsi_2btoul(cdb->length);
10794		break;
10795	}
10796	case WRITE_VERIFY_10: {
10797		struct scsi_write_verify_10 *cdb;
10798
10799		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10800
10801		*lba = scsi_4btoul(cdb->addr);
10802		*len = scsi_2btoul(cdb->length);
10803		break;
10804	}
10805	case READ_12:
10806	case WRITE_12: {
10807		struct scsi_rw_12 *cdb;
10808
10809		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10810
10811		*lba = scsi_4btoul(cdb->addr);
10812		*len = scsi_4btoul(cdb->length);
10813		break;
10814	}
10815	case WRITE_VERIFY_12: {
10816		struct scsi_write_verify_12 *cdb;
10817
10818		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10819
10820		*lba = scsi_4btoul(cdb->addr);
10821		*len = scsi_4btoul(cdb->length);
10822		break;
10823	}
10824	case READ_16:
10825	case WRITE_16: {
10826		struct scsi_rw_16 *cdb;
10827
10828		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10829
10830		*lba = scsi_8btou64(cdb->addr);
10831		*len = scsi_4btoul(cdb->length);
10832		break;
10833	}
10834	case WRITE_VERIFY_16: {
10835		struct scsi_write_verify_16 *cdb;
10836
10837		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10838
10839
10840		*lba = scsi_8btou64(cdb->addr);
10841		*len = scsi_4btoul(cdb->length);
10842		break;
10843	}
10844	case WRITE_SAME_10: {
10845		struct scsi_write_same_10 *cdb;
10846
10847		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10848
10849		*lba = scsi_4btoul(cdb->addr);
10850		*len = scsi_2btoul(cdb->length);
10851		break;
10852	}
10853	case WRITE_SAME_16: {
10854		struct scsi_write_same_16 *cdb;
10855
10856		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10857
10858		*lba = scsi_8btou64(cdb->addr);
10859		*len = scsi_4btoul(cdb->length);
10860		break;
10861	}
10862	case VERIFY_10: {
10863		struct scsi_verify_10 *cdb;
10864
10865		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10866
10867		*lba = scsi_4btoul(cdb->addr);
10868		*len = scsi_2btoul(cdb->length);
10869		break;
10870	}
10871	case VERIFY_12: {
10872		struct scsi_verify_12 *cdb;
10873
10874		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10875
10876		*lba = scsi_4btoul(cdb->addr);
10877		*len = scsi_4btoul(cdb->length);
10878		break;
10879	}
10880	case VERIFY_16: {
10881		struct scsi_verify_16 *cdb;
10882
10883		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10884
10885		*lba = scsi_8btou64(cdb->addr);
10886		*len = scsi_4btoul(cdb->length);
10887		break;
10888	}
10889	case UNMAP: {
10890		*lba = 0;
10891		*len = UINT64_MAX;
10892		break;
10893	}
10894	default:
10895		return (1);
10896		break; /* NOTREACHED */
10897	}
10898
10899	return (0);
10900}
10901
10902static ctl_action
10903ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10904{
10905	uint64_t endlba1, endlba2;
10906
10907	endlba1 = lba1 + len1 - 1;
10908	endlba2 = lba2 + len2 - 1;
10909
10910	if ((endlba1 < lba2)
10911	 || (endlba2 < lba1))
10912		return (CTL_ACTION_PASS);
10913	else
10914		return (CTL_ACTION_BLOCK);
10915}
10916
10917static int
10918ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10919{
10920	struct ctl_ptr_len_flags *ptrlen;
10921	struct scsi_unmap_desc *buf, *end, *range;
10922	uint64_t lba;
10923	uint32_t len;
10924
10925	/* If not UNMAP -- go other way. */
10926	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10927	    io->scsiio.cdb[0] != UNMAP)
10928		return (CTL_ACTION_ERROR);
10929
10930	/* If UNMAP without data -- block and wait for data. */
10931	ptrlen = (struct ctl_ptr_len_flags *)
10932	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10933	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10934	    ptrlen->ptr == NULL)
10935		return (CTL_ACTION_BLOCK);
10936
10937	/* UNMAP with data -- check for collision. */
10938	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10939	end = buf + ptrlen->len / sizeof(*buf);
10940	for (range = buf; range < end; range++) {
10941		lba = scsi_8btou64(range->lba);
10942		len = scsi_4btoul(range->length);
10943		if ((lba < lba2 + len2) && (lba + len > lba2))
10944			return (CTL_ACTION_BLOCK);
10945	}
10946	return (CTL_ACTION_PASS);
10947}
10948
10949static ctl_action
10950ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10951{
10952	uint64_t lba1, lba2;
10953	uint64_t len1, len2;
10954	int retval;
10955
10956	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10957		return (CTL_ACTION_ERROR);
10958
10959	retval = ctl_extent_check_unmap(io2, lba1, len1);
10960	if (retval != CTL_ACTION_ERROR)
10961		return (retval);
10962
10963	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10964		return (CTL_ACTION_ERROR);
10965
10966	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10967}
10968
10969static ctl_action
10970ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10971    union ctl_io *ooa_io)
10972{
10973	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10974	ctl_serialize_action *serialize_row;
10975
10976	/*
10977	 * The initiator attempted multiple untagged commands at the same
10978	 * time.  Can't do that.
10979	 */
10980	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10981	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10982	 && ((pending_io->io_hdr.nexus.targ_port ==
10983	      ooa_io->io_hdr.nexus.targ_port)
10984	  && (pending_io->io_hdr.nexus.initid.id ==
10985	      ooa_io->io_hdr.nexus.initid.id))
10986	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10987		return (CTL_ACTION_OVERLAP);
10988
10989	/*
10990	 * The initiator attempted to send multiple tagged commands with
10991	 * the same ID.  (It's fine if different initiators have the same
10992	 * tag ID.)
10993	 *
10994	 * Even if all of those conditions are true, we don't kill the I/O
10995	 * if the command ahead of us has been aborted.  We won't end up
10996	 * sending it to the FETD, and it's perfectly legal to resend a
10997	 * command with the same tag number as long as the previous
10998	 * instance of this tag number has been aborted somehow.
10999	 */
11000	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11001	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11002	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11003	 && ((pending_io->io_hdr.nexus.targ_port ==
11004	      ooa_io->io_hdr.nexus.targ_port)
11005	  && (pending_io->io_hdr.nexus.initid.id ==
11006	      ooa_io->io_hdr.nexus.initid.id))
11007	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
11008		return (CTL_ACTION_OVERLAP_TAG);
11009
11010	/*
11011	 * If we get a head of queue tag, SAM-3 says that we should
11012	 * immediately execute it.
11013	 *
11014	 * What happens if this command would normally block for some other
11015	 * reason?  e.g. a request sense with a head of queue tag
11016	 * immediately after a write.  Normally that would block, but this
11017	 * will result in its getting executed immediately...
11018	 *
11019	 * We currently return "pass" instead of "skip", so we'll end up
11020	 * going through the rest of the queue to check for overlapped tags.
11021	 *
11022	 * XXX KDM check for other types of blockage first??
11023	 */
11024	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11025		return (CTL_ACTION_PASS);
11026
11027	/*
11028	 * Ordered tags have to block until all items ahead of them
11029	 * have completed.  If we get called with an ordered tag, we always
11030	 * block, if something else is ahead of us in the queue.
11031	 */
11032	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11033		return (CTL_ACTION_BLOCK);
11034
11035	/*
11036	 * Simple tags get blocked until all head of queue and ordered tags
11037	 * ahead of them have completed.  I'm lumping untagged commands in
11038	 * with simple tags here.  XXX KDM is that the right thing to do?
11039	 */
11040	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11041	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11042	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11043	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11044		return (CTL_ACTION_BLOCK);
11045
11046	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11047	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11048
11049	serialize_row = ctl_serialize_table[ooa_entry->seridx];
11050
11051	switch (serialize_row[pending_entry->seridx]) {
11052	case CTL_SER_BLOCK:
11053		return (CTL_ACTION_BLOCK);
11054	case CTL_SER_EXTENT:
11055		return (ctl_extent_check(pending_io, ooa_io));
11056	case CTL_SER_EXTENTOPT:
11057		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11058		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11059			return (ctl_extent_check(pending_io, ooa_io));
11060		/* FALLTHROUGH */
11061	case CTL_SER_PASS:
11062		return (CTL_ACTION_PASS);
11063	case CTL_SER_BLOCKOPT:
11064		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11065		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11066			return (CTL_ACTION_BLOCK);
11067		return (CTL_ACTION_PASS);
11068	case CTL_SER_SKIP:
11069		return (CTL_ACTION_SKIP);
11070	default:
11071		panic("invalid serialization value %d",
11072		      serialize_row[pending_entry->seridx]);
11073	}
11074
11075	return (CTL_ACTION_ERROR);
11076}
11077
11078/*
11079 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11080 * Assumptions:
11081 * - pending_io is generally either incoming, or on the blocked queue
11082 * - starting I/O is the I/O we want to start the check with.
11083 */
11084static ctl_action
11085ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11086	      union ctl_io *starting_io)
11087{
11088	union ctl_io *ooa_io;
11089	ctl_action action;
11090
11091	mtx_assert(&lun->lun_lock, MA_OWNED);
11092
11093	/*
11094	 * Run back along the OOA queue, starting with the current
11095	 * blocked I/O and going through every I/O before it on the
11096	 * queue.  If starting_io is NULL, we'll just end up returning
11097	 * CTL_ACTION_PASS.
11098	 */
11099	for (ooa_io = starting_io; ooa_io != NULL;
11100	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11101	     ooa_links)){
11102
11103		/*
11104		 * This routine just checks to see whether
11105		 * cur_blocked is blocked by ooa_io, which is ahead
11106		 * of it in the queue.  It doesn't queue/dequeue
11107		 * cur_blocked.
11108		 */
11109		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11110		switch (action) {
11111		case CTL_ACTION_BLOCK:
11112		case CTL_ACTION_OVERLAP:
11113		case CTL_ACTION_OVERLAP_TAG:
11114		case CTL_ACTION_SKIP:
11115		case CTL_ACTION_ERROR:
11116			return (action);
11117			break; /* NOTREACHED */
11118		case CTL_ACTION_PASS:
11119			break;
11120		default:
11121			panic("invalid action %d", action);
11122			break;  /* NOTREACHED */
11123		}
11124	}
11125
11126	return (CTL_ACTION_PASS);
11127}
11128
11129/*
11130 * Assumptions:
11131 * - An I/O has just completed, and has been removed from the per-LUN OOA
11132 *   queue, so some items on the blocked queue may now be unblocked.
11133 */
11134static int
11135ctl_check_blocked(struct ctl_lun *lun)
11136{
11137	union ctl_io *cur_blocked, *next_blocked;
11138
11139	mtx_assert(&lun->lun_lock, MA_OWNED);
11140
11141	/*
11142	 * Run forward from the head of the blocked queue, checking each
11143	 * entry against the I/Os prior to it on the OOA queue to see if
11144	 * there is still any blockage.
11145	 *
11146	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11147	 * with our removing a variable on it while it is traversing the
11148	 * list.
11149	 */
11150	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11151	     cur_blocked != NULL; cur_blocked = next_blocked) {
11152		union ctl_io *prev_ooa;
11153		ctl_action action;
11154
11155		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11156							  blocked_links);
11157
11158		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11159						      ctl_ooaq, ooa_links);
11160
11161		/*
11162		 * If cur_blocked happens to be the first item in the OOA
11163		 * queue now, prev_ooa will be NULL, and the action
11164		 * returned will just be CTL_ACTION_PASS.
11165		 */
11166		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11167
11168		switch (action) {
11169		case CTL_ACTION_BLOCK:
11170			/* Nothing to do here, still blocked */
11171			break;
11172		case CTL_ACTION_OVERLAP:
11173		case CTL_ACTION_OVERLAP_TAG:
11174			/*
11175			 * This shouldn't happen!  In theory we've already
11176			 * checked this command for overlap...
11177			 */
11178			break;
11179		case CTL_ACTION_PASS:
11180		case CTL_ACTION_SKIP: {
11181			struct ctl_softc *softc;
11182			const struct ctl_cmd_entry *entry;
11183			uint32_t initidx;
11184			int isc_retval;
11185
11186			/*
11187			 * The skip case shouldn't happen, this transaction
11188			 * should have never made it onto the blocked queue.
11189			 */
11190			/*
11191			 * This I/O is no longer blocked, we can remove it
11192			 * from the blocked queue.  Since this is a TAILQ
11193			 * (doubly linked list), we can do O(1) removals
11194			 * from any place on the list.
11195			 */
11196			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11197				     blocked_links);
11198			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11199
11200			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11201				/*
11202				 * Need to send IO back to original side to
11203				 * run
11204				 */
11205				union ctl_ha_msg msg_info;
11206
11207				msg_info.hdr.original_sc =
11208					cur_blocked->io_hdr.original_sc;
11209				msg_info.hdr.serializing_sc = cur_blocked;
11210				msg_info.hdr.msg_type = CTL_MSG_R2R;
11211				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11212				     &msg_info, sizeof(msg_info), 0)) >
11213				     CTL_HA_STATUS_SUCCESS) {
11214					printf("CTL:Check Blocked error from "
11215					       "ctl_ha_msg_send %d\n",
11216					       isc_retval);
11217				}
11218				break;
11219			}
11220			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11221			softc = control_softc;
11222
11223			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11224
11225			/*
11226			 * Check this I/O for LUN state changes that may
11227			 * have happened while this command was blocked.
11228			 * The LUN state may have been changed by a command
11229			 * ahead of us in the queue, so we need to re-check
11230			 * for any states that can be caused by SCSI
11231			 * commands.
11232			 */
11233			if (ctl_scsiio_lun_check(softc, lun, entry,
11234						 &cur_blocked->scsiio) == 0) {
11235				cur_blocked->io_hdr.flags |=
11236				                      CTL_FLAG_IS_WAS_ON_RTR;
11237				ctl_enqueue_rtr(cur_blocked);
11238			} else
11239				ctl_done(cur_blocked);
11240			break;
11241		}
11242		default:
11243			/*
11244			 * This probably shouldn't happen -- we shouldn't
11245			 * get CTL_ACTION_ERROR, or anything else.
11246			 */
11247			break;
11248		}
11249	}
11250
11251	return (CTL_RETVAL_COMPLETE);
11252}
11253
11254/*
11255 * This routine (with one exception) checks LUN flags that can be set by
11256 * commands ahead of us in the OOA queue.  These flags have to be checked
11257 * when a command initially comes in, and when we pull a command off the
11258 * blocked queue and are preparing to execute it.  The reason we have to
11259 * check these flags for commands on the blocked queue is that the LUN
11260 * state may have been changed by a command ahead of us while we're on the
11261 * blocked queue.
11262 *
11263 * Ordering is somewhat important with these checks, so please pay
11264 * careful attention to the placement of any new checks.
11265 */
11266static int
11267ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11268    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11269{
11270	int retval;
11271	uint32_t residx;
11272
11273	retval = 0;
11274
11275	mtx_assert(&lun->lun_lock, MA_OWNED);
11276
11277	/*
11278	 * If this shelf is a secondary shelf controller, we have to reject
11279	 * any media access commands.
11280	 */
11281#if 0
11282	/* No longer needed for HA */
11283	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11284	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11285		ctl_set_lun_standby(ctsio);
11286		retval = 1;
11287		goto bailout;
11288	}
11289#endif
11290
11291	/*
11292	 * Check for a reservation conflict.  If this command isn't allowed
11293	 * even on reserved LUNs, and if this initiator isn't the one who
11294	 * reserved us, reject the command with a reservation conflict.
11295	 */
11296	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11297	if ((lun->flags & CTL_LUN_RESERVED)
11298	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11299		if (lun->res_idx != residx) {
11300			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11301			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11302			retval = 1;
11303			goto bailout;
11304		}
11305	}
11306
11307	if ((lun->flags & CTL_LUN_PR_RESERVED)
11308	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11309		/*
11310		 * if we aren't registered or it's a res holder type
11311		 * reservation and this isn't the res holder then set a
11312		 * conflict.
11313		 * NOTE: Commands which might be allowed on write exclusive
11314		 * type reservations are checked in the particular command
11315		 * for a conflict. Read and SSU are the only ones.
11316		 */
11317		if (!lun->per_res[residx].registered
11318		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11319			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11320			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11321			retval = 1;
11322			goto bailout;
11323		}
11324
11325	}
11326
11327	if ((lun->flags & CTL_LUN_OFFLINE)
11328	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11329		ctl_set_lun_not_ready(ctsio);
11330		retval = 1;
11331		goto bailout;
11332	}
11333
11334	/*
11335	 * If the LUN is stopped, see if this particular command is allowed
11336	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11337	 */
11338	if ((lun->flags & CTL_LUN_STOPPED)
11339	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11340		/* "Logical unit not ready, initializing cmd. required" */
11341		ctl_set_lun_stopped(ctsio);
11342		retval = 1;
11343		goto bailout;
11344	}
11345
11346	if ((lun->flags & CTL_LUN_INOPERABLE)
11347	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11348		/* "Medium format corrupted" */
11349		ctl_set_medium_format_corrupted(ctsio);
11350		retval = 1;
11351		goto bailout;
11352	}
11353
11354bailout:
11355	return (retval);
11356
11357}
11358
11359static void
11360ctl_failover_io(union ctl_io *io, int have_lock)
11361{
11362	ctl_set_busy(&io->scsiio);
11363	ctl_done(io);
11364}
11365
11366static void
11367ctl_failover(void)
11368{
11369	struct ctl_lun *lun;
11370	struct ctl_softc *ctl_softc;
11371	union ctl_io *next_io, *pending_io;
11372	union ctl_io *io;
11373	int lun_idx;
11374	int i;
11375
11376	ctl_softc = control_softc;
11377
11378	mtx_lock(&ctl_softc->ctl_lock);
11379	/*
11380	 * Remove any cmds from the other SC from the rtr queue.  These
11381	 * will obviously only be for LUNs for which we're the primary.
11382	 * We can't send status or get/send data for these commands.
11383	 * Since they haven't been executed yet, we can just remove them.
11384	 * We'll either abort them or delete them below, depending on
11385	 * which HA mode we're in.
11386	 */
11387#ifdef notyet
11388	mtx_lock(&ctl_softc->queue_lock);
11389	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11390	     io != NULL; io = next_io) {
11391		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11392		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11393			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11394				      ctl_io_hdr, links);
11395	}
11396	mtx_unlock(&ctl_softc->queue_lock);
11397#endif
11398
11399	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11400		lun = ctl_softc->ctl_luns[lun_idx];
11401		if (lun==NULL)
11402			continue;
11403
11404		/*
11405		 * Processor LUNs are primary on both sides.
11406		 * XXX will this always be true?
11407		 */
11408		if (lun->be_lun->lun_type == T_PROCESSOR)
11409			continue;
11410
11411		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11412		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11413			printf("FAILOVER: primary lun %d\n", lun_idx);
11414		        /*
11415			 * Remove all commands from the other SC. First from the
11416			 * blocked queue then from the ooa queue. Once we have
11417			 * removed them. Call ctl_check_blocked to see if there
11418			 * is anything that can run.
11419			 */
11420			for (io = (union ctl_io *)TAILQ_FIRST(
11421			     &lun->blocked_queue); io != NULL; io = next_io) {
11422
11423		        	next_io = (union ctl_io *)TAILQ_NEXT(
11424				    &io->io_hdr, blocked_links);
11425
11426				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11427					TAILQ_REMOVE(&lun->blocked_queue,
11428						     &io->io_hdr,blocked_links);
11429					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11430					TAILQ_REMOVE(&lun->ooa_queue,
11431						     &io->io_hdr, ooa_links);
11432
11433					ctl_free_io(io);
11434				}
11435			}
11436
11437			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11438	     		     io != NULL; io = next_io) {
11439
11440		        	next_io = (union ctl_io *)TAILQ_NEXT(
11441				    &io->io_hdr, ooa_links);
11442
11443				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11444
11445					TAILQ_REMOVE(&lun->ooa_queue,
11446						&io->io_hdr,
11447					     	ooa_links);
11448
11449					ctl_free_io(io);
11450				}
11451			}
11452			ctl_check_blocked(lun);
11453		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11454			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11455
11456			printf("FAILOVER: primary lun %d\n", lun_idx);
11457			/*
11458			 * Abort all commands from the other SC.  We can't
11459			 * send status back for them now.  These should get
11460			 * cleaned up when they are completed or come out
11461			 * for a datamove operation.
11462			 */
11463			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11464	     		     io != NULL; io = next_io) {
11465		        	next_io = (union ctl_io *)TAILQ_NEXT(
11466					&io->io_hdr, ooa_links);
11467
11468				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11469					io->io_hdr.flags |= CTL_FLAG_ABORT;
11470			}
11471		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11472			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11473
11474			printf("FAILOVER: secondary lun %d\n", lun_idx);
11475
11476			lun->flags |= CTL_LUN_PRIMARY_SC;
11477
11478			/*
11479			 * We send all I/O that was sent to this controller
11480			 * and redirected to the other side back with
11481			 * busy status, and have the initiator retry it.
11482			 * Figuring out how much data has been transferred,
11483			 * etc. and picking up where we left off would be
11484			 * very tricky.
11485			 *
11486			 * XXX KDM need to remove I/O from the blocked
11487			 * queue as well!
11488			 */
11489			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11490			     &lun->ooa_queue); pending_io != NULL;
11491			     pending_io = next_io) {
11492
11493				next_io =  (union ctl_io *)TAILQ_NEXT(
11494					&pending_io->io_hdr, ooa_links);
11495
11496				pending_io->io_hdr.flags &=
11497					~CTL_FLAG_SENT_2OTHER_SC;
11498
11499				if (pending_io->io_hdr.flags &
11500				    CTL_FLAG_IO_ACTIVE) {
11501					pending_io->io_hdr.flags |=
11502						CTL_FLAG_FAILOVER;
11503				} else {
11504					ctl_set_busy(&pending_io->scsiio);
11505					ctl_done(pending_io);
11506				}
11507			}
11508
11509			/*
11510			 * Build Unit Attention
11511			 */
11512			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11513				lun->pending_ua[i] |=
11514				                     CTL_UA_ASYM_ACC_CHANGE;
11515			}
11516		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11517			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11518			printf("FAILOVER: secondary lun %d\n", lun_idx);
11519			/*
11520			 * if the first io on the OOA is not on the RtR queue
11521			 * add it.
11522			 */
11523			lun->flags |= CTL_LUN_PRIMARY_SC;
11524
11525			pending_io = (union ctl_io *)TAILQ_FIRST(
11526			    &lun->ooa_queue);
11527			if (pending_io==NULL) {
11528				printf("Nothing on OOA queue\n");
11529				continue;
11530			}
11531
11532			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11533			if ((pending_io->io_hdr.flags &
11534			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11535				pending_io->io_hdr.flags |=
11536				    CTL_FLAG_IS_WAS_ON_RTR;
11537				ctl_enqueue_rtr(pending_io);
11538			}
11539#if 0
11540			else
11541			{
11542				printf("Tag 0x%04x is running\n",
11543				      pending_io->scsiio.tag_num);
11544			}
11545#endif
11546
11547			next_io = (union ctl_io *)TAILQ_NEXT(
11548			    &pending_io->io_hdr, ooa_links);
11549			for (pending_io=next_io; pending_io != NULL;
11550			     pending_io = next_io) {
11551				pending_io->io_hdr.flags &=
11552				    ~CTL_FLAG_SENT_2OTHER_SC;
11553				next_io = (union ctl_io *)TAILQ_NEXT(
11554					&pending_io->io_hdr, ooa_links);
11555				if (pending_io->io_hdr.flags &
11556				    CTL_FLAG_IS_WAS_ON_RTR) {
11557#if 0
11558				        printf("Tag 0x%04x is running\n",
11559				      		pending_io->scsiio.tag_num);
11560#endif
11561					continue;
11562				}
11563
11564				switch (ctl_check_ooa(lun, pending_io,
11565			            (union ctl_io *)TAILQ_PREV(
11566				    &pending_io->io_hdr, ctl_ooaq,
11567				    ooa_links))) {
11568
11569				case CTL_ACTION_BLOCK:
11570					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11571							  &pending_io->io_hdr,
11572							  blocked_links);
11573					pending_io->io_hdr.flags |=
11574					    CTL_FLAG_BLOCKED;
11575					break;
11576				case CTL_ACTION_PASS:
11577				case CTL_ACTION_SKIP:
11578					pending_io->io_hdr.flags |=
11579					    CTL_FLAG_IS_WAS_ON_RTR;
11580					ctl_enqueue_rtr(pending_io);
11581					break;
11582				case CTL_ACTION_OVERLAP:
11583					ctl_set_overlapped_cmd(
11584					    (struct ctl_scsiio *)pending_io);
11585					ctl_done(pending_io);
11586					break;
11587				case CTL_ACTION_OVERLAP_TAG:
11588					ctl_set_overlapped_tag(
11589					    (struct ctl_scsiio *)pending_io,
11590					    pending_io->scsiio.tag_num & 0xff);
11591					ctl_done(pending_io);
11592					break;
11593				case CTL_ACTION_ERROR:
11594				default:
11595					ctl_set_internal_failure(
11596						(struct ctl_scsiio *)pending_io,
11597						0,  // sks_valid
11598						0); //retry count
11599					ctl_done(pending_io);
11600					break;
11601				}
11602			}
11603
11604			/*
11605			 * Build Unit Attention
11606			 */
11607			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11608				lun->pending_ua[i] |=
11609				                     CTL_UA_ASYM_ACC_CHANGE;
11610			}
11611		} else {
11612			panic("Unhandled HA mode failover, LUN flags = %#x, "
11613			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11614		}
11615	}
11616	ctl_pause_rtr = 0;
11617	mtx_unlock(&ctl_softc->ctl_lock);
11618}
11619
11620static int
11621ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11622{
11623	struct ctl_lun *lun;
11624	const struct ctl_cmd_entry *entry;
11625	uint32_t initidx, targ_lun;
11626	int retval;
11627
11628	retval = 0;
11629
11630	lun = NULL;
11631
11632	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11633	if ((targ_lun < CTL_MAX_LUNS)
11634	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11635		lun = ctl_softc->ctl_luns[targ_lun];
11636		/*
11637		 * If the LUN is invalid, pretend that it doesn't exist.
11638		 * It will go away as soon as all pending I/O has been
11639		 * completed.
11640		 */
11641		if (lun->flags & CTL_LUN_DISABLED) {
11642			lun = NULL;
11643		} else {
11644			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11645			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11646				lun->be_lun;
11647			if (lun->be_lun->lun_type == T_PROCESSOR) {
11648				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11649			}
11650
11651			/*
11652			 * Every I/O goes into the OOA queue for a
11653			 * particular LUN, and stays there until completion.
11654			 */
11655			mtx_lock(&lun->lun_lock);
11656			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11657			    ooa_links);
11658		}
11659	} else {
11660		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11661		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11662	}
11663
11664	/* Get command entry and return error if it is unsuppotyed. */
11665	entry = ctl_validate_command(ctsio);
11666	if (entry == NULL) {
11667		if (lun)
11668			mtx_unlock(&lun->lun_lock);
11669		return (retval);
11670	}
11671
11672	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11673	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11674
11675	/*
11676	 * Check to see whether we can send this command to LUNs that don't
11677	 * exist.  This should pretty much only be the case for inquiry
11678	 * and request sense.  Further checks, below, really require having
11679	 * a LUN, so we can't really check the command anymore.  Just put
11680	 * it on the rtr queue.
11681	 */
11682	if (lun == NULL) {
11683		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11684			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11685			ctl_enqueue_rtr((union ctl_io *)ctsio);
11686			return (retval);
11687		}
11688
11689		ctl_set_unsupported_lun(ctsio);
11690		ctl_done((union ctl_io *)ctsio);
11691		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11692		return (retval);
11693	} else {
11694		/*
11695		 * Make sure we support this particular command on this LUN.
11696		 * e.g., we don't support writes to the control LUN.
11697		 */
11698		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11699			mtx_unlock(&lun->lun_lock);
11700			ctl_set_invalid_opcode(ctsio);
11701			ctl_done((union ctl_io *)ctsio);
11702			return (retval);
11703		}
11704	}
11705
11706	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11707
11708#ifdef CTL_WITH_CA
11709	/*
11710	 * If we've got a request sense, it'll clear the contingent
11711	 * allegiance condition.  Otherwise, if we have a CA condition for
11712	 * this initiator, clear it, because it sent down a command other
11713	 * than request sense.
11714	 */
11715	if ((ctsio->cdb[0] != REQUEST_SENSE)
11716	 && (ctl_is_set(lun->have_ca, initidx)))
11717		ctl_clear_mask(lun->have_ca, initidx);
11718#endif
11719
11720	/*
11721	 * If the command has this flag set, it handles its own unit
11722	 * attention reporting, we shouldn't do anything.  Otherwise we
11723	 * check for any pending unit attentions, and send them back to the
11724	 * initiator.  We only do this when a command initially comes in,
11725	 * not when we pull it off the blocked queue.
11726	 *
11727	 * According to SAM-3, section 5.3.2, the order that things get
11728	 * presented back to the host is basically unit attentions caused
11729	 * by some sort of reset event, busy status, reservation conflicts
11730	 * or task set full, and finally any other status.
11731	 *
11732	 * One issue here is that some of the unit attentions we report
11733	 * don't fall into the "reset" category (e.g. "reported luns data
11734	 * has changed").  So reporting it here, before the reservation
11735	 * check, may be technically wrong.  I guess the only thing to do
11736	 * would be to check for and report the reset events here, and then
11737	 * check for the other unit attention types after we check for a
11738	 * reservation conflict.
11739	 *
11740	 * XXX KDM need to fix this
11741	 */
11742	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11743		ctl_ua_type ua_type;
11744
11745		ua_type = lun->pending_ua[initidx];
11746		if (ua_type != CTL_UA_NONE) {
11747			scsi_sense_data_type sense_format;
11748
11749			if (lun != NULL)
11750				sense_format = (lun->flags &
11751				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11752				    SSD_TYPE_FIXED;
11753			else
11754				sense_format = SSD_TYPE_FIXED;
11755
11756			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11757					       sense_format);
11758			if (ua_type != CTL_UA_NONE) {
11759				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11760				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11761						       CTL_AUTOSENSE;
11762				ctsio->sense_len = SSD_FULL_SIZE;
11763				lun->pending_ua[initidx] &= ~ua_type;
11764				mtx_unlock(&lun->lun_lock);
11765				ctl_done((union ctl_io *)ctsio);
11766				return (retval);
11767			}
11768		}
11769	}
11770
11771
11772	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11773		mtx_unlock(&lun->lun_lock);
11774		ctl_done((union ctl_io *)ctsio);
11775		return (retval);
11776	}
11777
11778	/*
11779	 * XXX CHD this is where we want to send IO to other side if
11780	 * this LUN is secondary on this SC. We will need to make a copy
11781	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11782	 * the copy we send as FROM_OTHER.
11783	 * We also need to stuff the address of the original IO so we can
11784	 * find it easily. Something similar will need be done on the other
11785	 * side so when we are done we can find the copy.
11786	 */
11787	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11788		union ctl_ha_msg msg_info;
11789		int isc_retval;
11790
11791		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11792
11793		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11794		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11795#if 0
11796		printf("1. ctsio %p\n", ctsio);
11797#endif
11798		msg_info.hdr.serializing_sc = NULL;
11799		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11800		msg_info.scsi.tag_num = ctsio->tag_num;
11801		msg_info.scsi.tag_type = ctsio->tag_type;
11802		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11803
11804		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11805
11806		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11807		    (void *)&msg_info, sizeof(msg_info), 0)) >
11808		    CTL_HA_STATUS_SUCCESS) {
11809			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11810			       isc_retval);
11811			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11812		} else {
11813#if 0
11814			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11815#endif
11816		}
11817
11818		/*
11819		 * XXX KDM this I/O is off the incoming queue, but hasn't
11820		 * been inserted on any other queue.  We may need to come
11821		 * up with a holding queue while we wait for serialization
11822		 * so that we have an idea of what we're waiting for from
11823		 * the other side.
11824		 */
11825		mtx_unlock(&lun->lun_lock);
11826		return (retval);
11827	}
11828
11829	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11830			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11831			      ctl_ooaq, ooa_links))) {
11832	case CTL_ACTION_BLOCK:
11833		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11834		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11835				  blocked_links);
11836		mtx_unlock(&lun->lun_lock);
11837		return (retval);
11838	case CTL_ACTION_PASS:
11839	case CTL_ACTION_SKIP:
11840		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11841		mtx_unlock(&lun->lun_lock);
11842		ctl_enqueue_rtr((union ctl_io *)ctsio);
11843		break;
11844	case CTL_ACTION_OVERLAP:
11845		mtx_unlock(&lun->lun_lock);
11846		ctl_set_overlapped_cmd(ctsio);
11847		ctl_done((union ctl_io *)ctsio);
11848		break;
11849	case CTL_ACTION_OVERLAP_TAG:
11850		mtx_unlock(&lun->lun_lock);
11851		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11852		ctl_done((union ctl_io *)ctsio);
11853		break;
11854	case CTL_ACTION_ERROR:
11855	default:
11856		mtx_unlock(&lun->lun_lock);
11857		ctl_set_internal_failure(ctsio,
11858					 /*sks_valid*/ 0,
11859					 /*retry_count*/ 0);
11860		ctl_done((union ctl_io *)ctsio);
11861		break;
11862	}
11863	return (retval);
11864}
11865
11866const struct ctl_cmd_entry *
11867ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11868{
11869	const struct ctl_cmd_entry *entry;
11870	int service_action;
11871
11872	entry = &ctl_cmd_table[ctsio->cdb[0]];
11873	if (sa)
11874		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11875	if (entry->flags & CTL_CMD_FLAG_SA5) {
11876		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11877		entry = &((const struct ctl_cmd_entry *)
11878		    entry->execute)[service_action];
11879	}
11880	return (entry);
11881}
11882
11883const struct ctl_cmd_entry *
11884ctl_validate_command(struct ctl_scsiio *ctsio)
11885{
11886	const struct ctl_cmd_entry *entry;
11887	int i, sa;
11888	uint8_t diff;
11889
11890	entry = ctl_get_cmd_entry(ctsio, &sa);
11891	if (entry->execute == NULL) {
11892		if (sa)
11893			ctl_set_invalid_field(ctsio,
11894					      /*sks_valid*/ 1,
11895					      /*command*/ 1,
11896					      /*field*/ 1,
11897					      /*bit_valid*/ 1,
11898					      /*bit*/ 4);
11899		else
11900			ctl_set_invalid_opcode(ctsio);
11901		ctl_done((union ctl_io *)ctsio);
11902		return (NULL);
11903	}
11904	KASSERT(entry->length > 0,
11905	    ("Not defined length for command 0x%02x/0x%02x",
11906	     ctsio->cdb[0], ctsio->cdb[1]));
11907	for (i = 1; i < entry->length; i++) {
11908		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11909		if (diff == 0)
11910			continue;
11911		ctl_set_invalid_field(ctsio,
11912				      /*sks_valid*/ 1,
11913				      /*command*/ 1,
11914				      /*field*/ i,
11915				      /*bit_valid*/ 1,
11916				      /*bit*/ fls(diff) - 1);
11917		ctl_done((union ctl_io *)ctsio);
11918		return (NULL);
11919	}
11920	return (entry);
11921}
11922
11923static int
11924ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11925{
11926
11927	switch (lun_type) {
11928	case T_PROCESSOR:
11929		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11930		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11931			return (0);
11932		break;
11933	case T_DIRECT:
11934		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11935		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11936			return (0);
11937		break;
11938	default:
11939		return (0);
11940	}
11941	return (1);
11942}
11943
11944static int
11945ctl_scsiio(struct ctl_scsiio *ctsio)
11946{
11947	int retval;
11948	const struct ctl_cmd_entry *entry;
11949
11950	retval = CTL_RETVAL_COMPLETE;
11951
11952	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11953
11954	entry = ctl_get_cmd_entry(ctsio, NULL);
11955
11956	/*
11957	 * If this I/O has been aborted, just send it straight to
11958	 * ctl_done() without executing it.
11959	 */
11960	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11961		ctl_done((union ctl_io *)ctsio);
11962		goto bailout;
11963	}
11964
11965	/*
11966	 * All the checks should have been handled by ctl_scsiio_precheck().
11967	 * We should be clear now to just execute the I/O.
11968	 */
11969	retval = entry->execute(ctsio);
11970
11971bailout:
11972	return (retval);
11973}
11974
11975/*
11976 * Since we only implement one target right now, a bus reset simply resets
11977 * our single target.
11978 */
11979static int
11980ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11981{
11982	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11983}
11984
11985static int
11986ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11987		 ctl_ua_type ua_type)
11988{
11989	struct ctl_lun *lun;
11990	int retval;
11991
11992	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11993		union ctl_ha_msg msg_info;
11994
11995		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11996		msg_info.hdr.nexus = io->io_hdr.nexus;
11997		if (ua_type==CTL_UA_TARG_RESET)
11998			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11999		else
12000			msg_info.task.task_action = CTL_TASK_BUS_RESET;
12001		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12002		msg_info.hdr.original_sc = NULL;
12003		msg_info.hdr.serializing_sc = NULL;
12004		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12005		    (void *)&msg_info, sizeof(msg_info), 0)) {
12006		}
12007	}
12008	retval = 0;
12009
12010	mtx_lock(&ctl_softc->ctl_lock);
12011	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
12012		retval += ctl_lun_reset(lun, io, ua_type);
12013	mtx_unlock(&ctl_softc->ctl_lock);
12014
12015	return (retval);
12016}
12017
12018/*
12019 * The LUN should always be set.  The I/O is optional, and is used to
12020 * distinguish between I/Os sent by this initiator, and by other
12021 * initiators.  We set unit attention for initiators other than this one.
12022 * SAM-3 is vague on this point.  It does say that a unit attention should
12023 * be established for other initiators when a LUN is reset (see section
12024 * 5.7.3), but it doesn't specifically say that the unit attention should
12025 * be established for this particular initiator when a LUN is reset.  Here
12026 * is the relevant text, from SAM-3 rev 8:
12027 *
12028 * 5.7.2 When a SCSI initiator port aborts its own tasks
12029 *
12030 * When a SCSI initiator port causes its own task(s) to be aborted, no
12031 * notification that the task(s) have been aborted shall be returned to
12032 * the SCSI initiator port other than the completion response for the
12033 * command or task management function action that caused the task(s) to
12034 * be aborted and notification(s) associated with related effects of the
12035 * action (e.g., a reset unit attention condition).
12036 *
12037 * XXX KDM for now, we're setting unit attention for all initiators.
12038 */
12039static int
12040ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
12041{
12042	union ctl_io *xio;
12043#if 0
12044	uint32_t initindex;
12045#endif
12046	int i;
12047
12048	mtx_lock(&lun->lun_lock);
12049	/*
12050	 * Run through the OOA queue and abort each I/O.
12051	 */
12052#if 0
12053	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12054#endif
12055	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12056	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12057		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
12058	}
12059
12060	/*
12061	 * This version sets unit attention for every
12062	 */
12063#if 0
12064	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12065	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12066		if (initindex == i)
12067			continue;
12068		lun->pending_ua[i] |= ua_type;
12069	}
12070#endif
12071
12072	/*
12073	 * A reset (any kind, really) clears reservations established with
12074	 * RESERVE/RELEASE.  It does not clear reservations established
12075	 * with PERSISTENT RESERVE OUT, but we don't support that at the
12076	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
12077	 * reservations made with the RESERVE/RELEASE commands, because
12078	 * those commands are obsolete in SPC-3.
12079	 */
12080	lun->flags &= ~CTL_LUN_RESERVED;
12081
12082	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12083#ifdef CTL_WITH_CA
12084		ctl_clear_mask(lun->have_ca, i);
12085#endif
12086		lun->pending_ua[i] |= ua_type;
12087	}
12088	mtx_unlock(&lun->lun_lock);
12089
12090	return (0);
12091}
12092
12093static void
12094ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
12095    int other_sc)
12096{
12097	union ctl_io *xio;
12098
12099	mtx_assert(&lun->lun_lock, MA_OWNED);
12100
12101	/*
12102	 * Run through the OOA queue and attempt to find the given I/O.
12103	 * The target port, initiator ID, tag type and tag number have to
12104	 * match the values that we got from the initiator.  If we have an
12105	 * untagged command to abort, simply abort the first untagged command
12106	 * we come to.  We only allow one untagged command at a time of course.
12107	 */
12108	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12109	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12110
12111		if ((targ_port == UINT32_MAX ||
12112		     targ_port == xio->io_hdr.nexus.targ_port) &&
12113		    (init_id == UINT32_MAX ||
12114		     init_id == xio->io_hdr.nexus.initid.id)) {
12115			if (targ_port != xio->io_hdr.nexus.targ_port ||
12116			    init_id != xio->io_hdr.nexus.initid.id)
12117				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12118			xio->io_hdr.flags |= CTL_FLAG_ABORT;
12119			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12120				union ctl_ha_msg msg_info;
12121
12122				msg_info.hdr.nexus = xio->io_hdr.nexus;
12123				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12124				msg_info.task.tag_num = xio->scsiio.tag_num;
12125				msg_info.task.tag_type = xio->scsiio.tag_type;
12126				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12127				msg_info.hdr.original_sc = NULL;
12128				msg_info.hdr.serializing_sc = NULL;
12129				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12130				    (void *)&msg_info, sizeof(msg_info), 0);
12131			}
12132		}
12133	}
12134}
12135
12136static int
12137ctl_abort_task_set(union ctl_io *io)
12138{
12139	struct ctl_softc *softc = control_softc;
12140	struct ctl_lun *lun;
12141	uint32_t targ_lun;
12142
12143	/*
12144	 * Look up the LUN.
12145	 */
12146	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12147	mtx_lock(&softc->ctl_lock);
12148	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12149		lun = softc->ctl_luns[targ_lun];
12150	else {
12151		mtx_unlock(&softc->ctl_lock);
12152		return (1);
12153	}
12154
12155	mtx_lock(&lun->lun_lock);
12156	mtx_unlock(&softc->ctl_lock);
12157	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12158		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12159		    io->io_hdr.nexus.initid.id,
12160		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12161	} else { /* CTL_TASK_CLEAR_TASK_SET */
12162		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12163		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12164	}
12165	mtx_unlock(&lun->lun_lock);
12166	return (0);
12167}
12168
12169static int
12170ctl_i_t_nexus_reset(union ctl_io *io)
12171{
12172	struct ctl_softc *softc = control_softc;
12173	struct ctl_lun *lun;
12174	uint32_t initindex, residx;
12175
12176	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12177	residx = ctl_get_resindex(&io->io_hdr.nexus);
12178	mtx_lock(&softc->ctl_lock);
12179	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12180		mtx_lock(&lun->lun_lock);
12181		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12182		    io->io_hdr.nexus.initid.id,
12183		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12184#ifdef CTL_WITH_CA
12185		ctl_clear_mask(lun->have_ca, initindex);
12186#endif
12187		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
12188			lun->flags &= ~CTL_LUN_RESERVED;
12189		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12190		mtx_unlock(&lun->lun_lock);
12191	}
12192	mtx_unlock(&softc->ctl_lock);
12193	return (0);
12194}
12195
12196static int
12197ctl_abort_task(union ctl_io *io)
12198{
12199	union ctl_io *xio;
12200	struct ctl_lun *lun;
12201	struct ctl_softc *ctl_softc;
12202#if 0
12203	struct sbuf sb;
12204	char printbuf[128];
12205#endif
12206	int found;
12207	uint32_t targ_lun;
12208
12209	ctl_softc = control_softc;
12210	found = 0;
12211
12212	/*
12213	 * Look up the LUN.
12214	 */
12215	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12216	mtx_lock(&ctl_softc->ctl_lock);
12217	if ((targ_lun < CTL_MAX_LUNS)
12218	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12219		lun = ctl_softc->ctl_luns[targ_lun];
12220	else {
12221		mtx_unlock(&ctl_softc->ctl_lock);
12222		return (1);
12223	}
12224
12225#if 0
12226	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12227	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12228#endif
12229
12230	mtx_lock(&lun->lun_lock);
12231	mtx_unlock(&ctl_softc->ctl_lock);
12232	/*
12233	 * Run through the OOA queue and attempt to find the given I/O.
12234	 * The target port, initiator ID, tag type and tag number have to
12235	 * match the values that we got from the initiator.  If we have an
12236	 * untagged command to abort, simply abort the first untagged command
12237	 * we come to.  We only allow one untagged command at a time of course.
12238	 */
12239#if 0
12240	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12241#endif
12242	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12243	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12244#if 0
12245		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12246
12247		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12248			    lun->lun, xio->scsiio.tag_num,
12249			    xio->scsiio.tag_type,
12250			    (xio->io_hdr.blocked_links.tqe_prev
12251			    == NULL) ? "" : " BLOCKED",
12252			    (xio->io_hdr.flags &
12253			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12254			    (xio->io_hdr.flags &
12255			    CTL_FLAG_ABORT) ? " ABORT" : "",
12256			    (xio->io_hdr.flags &
12257			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12258		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12259		sbuf_finish(&sb);
12260		printf("%s\n", sbuf_data(&sb));
12261#endif
12262
12263		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12264		 && (xio->io_hdr.nexus.initid.id ==
12265		     io->io_hdr.nexus.initid.id)) {
12266			/*
12267			 * If the abort says that the task is untagged, the
12268			 * task in the queue must be untagged.  Otherwise,
12269			 * we just check to see whether the tag numbers
12270			 * match.  This is because the QLogic firmware
12271			 * doesn't pass back the tag type in an abort
12272			 * request.
12273			 */
12274#if 0
12275			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12276			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12277			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12278#endif
12279			/*
12280			 * XXX KDM we've got problems with FC, because it
12281			 * doesn't send down a tag type with aborts.  So we
12282			 * can only really go by the tag number...
12283			 * This may cause problems with parallel SCSI.
12284			 * Need to figure that out!!
12285			 */
12286			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12287				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12288				found = 1;
12289				if ((io->io_hdr.flags &
12290				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12291				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12292					union ctl_ha_msg msg_info;
12293
12294					io->io_hdr.flags |=
12295					                CTL_FLAG_SENT_2OTHER_SC;
12296					msg_info.hdr.nexus = io->io_hdr.nexus;
12297					msg_info.task.task_action =
12298						CTL_TASK_ABORT_TASK;
12299					msg_info.task.tag_num =
12300						io->taskio.tag_num;
12301					msg_info.task.tag_type =
12302						io->taskio.tag_type;
12303					msg_info.hdr.msg_type =
12304						CTL_MSG_MANAGE_TASKS;
12305					msg_info.hdr.original_sc = NULL;
12306					msg_info.hdr.serializing_sc = NULL;
12307#if 0
12308					printf("Sent Abort to other side\n");
12309#endif
12310					if (CTL_HA_STATUS_SUCCESS !=
12311					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12312		    				(void *)&msg_info,
12313						sizeof(msg_info), 0)) {
12314					}
12315				}
12316#if 0
12317				printf("ctl_abort_task: found I/O to abort\n");
12318#endif
12319				break;
12320			}
12321		}
12322	}
12323	mtx_unlock(&lun->lun_lock);
12324
12325	if (found == 0) {
12326		/*
12327		 * This isn't really an error.  It's entirely possible for
12328		 * the abort and command completion to cross on the wire.
12329		 * This is more of an informative/diagnostic error.
12330		 */
12331#if 0
12332		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12333		       "%d:%d:%d:%d tag %d type %d\n",
12334		       io->io_hdr.nexus.initid.id,
12335		       io->io_hdr.nexus.targ_port,
12336		       io->io_hdr.nexus.targ_target.id,
12337		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12338		       io->taskio.tag_type);
12339#endif
12340	}
12341	return (0);
12342}
12343
12344static void
12345ctl_run_task(union ctl_io *io)
12346{
12347	struct ctl_softc *ctl_softc = control_softc;
12348	int retval = 1;
12349	const char *task_desc;
12350
12351	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12352
12353	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12354	    ("ctl_run_task: Unextected io_type %d\n",
12355	     io->io_hdr.io_type));
12356
12357	task_desc = ctl_scsi_task_string(&io->taskio);
12358	if (task_desc != NULL) {
12359#ifdef NEEDTOPORT
12360		csevent_log(CSC_CTL | CSC_SHELF_SW |
12361			    CTL_TASK_REPORT,
12362			    csevent_LogType_Trace,
12363			    csevent_Severity_Information,
12364			    csevent_AlertLevel_Green,
12365			    csevent_FRU_Firmware,
12366			    csevent_FRU_Unknown,
12367			    "CTL: received task: %s",task_desc);
12368#endif
12369	} else {
12370#ifdef NEEDTOPORT
12371		csevent_log(CSC_CTL | CSC_SHELF_SW |
12372			    CTL_TASK_REPORT,
12373			    csevent_LogType_Trace,
12374			    csevent_Severity_Information,
12375			    csevent_AlertLevel_Green,
12376			    csevent_FRU_Firmware,
12377			    csevent_FRU_Unknown,
12378			    "CTL: received unknown task "
12379			    "type: %d (%#x)",
12380			    io->taskio.task_action,
12381			    io->taskio.task_action);
12382#endif
12383	}
12384	switch (io->taskio.task_action) {
12385	case CTL_TASK_ABORT_TASK:
12386		retval = ctl_abort_task(io);
12387		break;
12388	case CTL_TASK_ABORT_TASK_SET:
12389	case CTL_TASK_CLEAR_TASK_SET:
12390		retval = ctl_abort_task_set(io);
12391		break;
12392	case CTL_TASK_CLEAR_ACA:
12393		break;
12394	case CTL_TASK_I_T_NEXUS_RESET:
12395		retval = ctl_i_t_nexus_reset(io);
12396		break;
12397	case CTL_TASK_LUN_RESET: {
12398		struct ctl_lun *lun;
12399		uint32_t targ_lun;
12400
12401		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12402		mtx_lock(&ctl_softc->ctl_lock);
12403		if ((targ_lun < CTL_MAX_LUNS)
12404		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12405			lun = ctl_softc->ctl_luns[targ_lun];
12406		else {
12407			mtx_unlock(&ctl_softc->ctl_lock);
12408			retval = 1;
12409			break;
12410		}
12411
12412		if (!(io->io_hdr.flags &
12413		    CTL_FLAG_FROM_OTHER_SC)) {
12414			union ctl_ha_msg msg_info;
12415
12416			io->io_hdr.flags |=
12417				CTL_FLAG_SENT_2OTHER_SC;
12418			msg_info.hdr.msg_type =
12419				CTL_MSG_MANAGE_TASKS;
12420			msg_info.hdr.nexus = io->io_hdr.nexus;
12421			msg_info.task.task_action =
12422				CTL_TASK_LUN_RESET;
12423			msg_info.hdr.original_sc = NULL;
12424			msg_info.hdr.serializing_sc = NULL;
12425			if (CTL_HA_STATUS_SUCCESS !=
12426			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12427			    (void *)&msg_info,
12428			    sizeof(msg_info), 0)) {
12429			}
12430		}
12431
12432		retval = ctl_lun_reset(lun, io,
12433				       CTL_UA_LUN_RESET);
12434		mtx_unlock(&ctl_softc->ctl_lock);
12435		break;
12436	}
12437	case CTL_TASK_TARGET_RESET:
12438		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12439		break;
12440	case CTL_TASK_BUS_RESET:
12441		retval = ctl_bus_reset(ctl_softc, io);
12442		break;
12443	case CTL_TASK_PORT_LOGIN:
12444		break;
12445	case CTL_TASK_PORT_LOGOUT:
12446		break;
12447	default:
12448		printf("ctl_run_task: got unknown task management event %d\n",
12449		       io->taskio.task_action);
12450		break;
12451	}
12452	if (retval == 0)
12453		io->io_hdr.status = CTL_SUCCESS;
12454	else
12455		io->io_hdr.status = CTL_ERROR;
12456	ctl_done(io);
12457}
12458
12459/*
12460 * For HA operation.  Handle commands that come in from the other
12461 * controller.
12462 */
12463static void
12464ctl_handle_isc(union ctl_io *io)
12465{
12466	int free_io;
12467	struct ctl_lun *lun;
12468	struct ctl_softc *ctl_softc;
12469	uint32_t targ_lun;
12470
12471	ctl_softc = control_softc;
12472
12473	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12474	lun = ctl_softc->ctl_luns[targ_lun];
12475
12476	switch (io->io_hdr.msg_type) {
12477	case CTL_MSG_SERIALIZE:
12478		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12479		break;
12480	case CTL_MSG_R2R: {
12481		const struct ctl_cmd_entry *entry;
12482
12483		/*
12484		 * This is only used in SER_ONLY mode.
12485		 */
12486		free_io = 0;
12487		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12488		mtx_lock(&lun->lun_lock);
12489		if (ctl_scsiio_lun_check(ctl_softc, lun,
12490		    entry, (struct ctl_scsiio *)io) != 0) {
12491			mtx_unlock(&lun->lun_lock);
12492			ctl_done(io);
12493			break;
12494		}
12495		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12496		mtx_unlock(&lun->lun_lock);
12497		ctl_enqueue_rtr(io);
12498		break;
12499	}
12500	case CTL_MSG_FINISH_IO:
12501		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12502			free_io = 0;
12503			ctl_done(io);
12504		} else {
12505			free_io = 1;
12506			mtx_lock(&lun->lun_lock);
12507			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12508				     ooa_links);
12509			ctl_check_blocked(lun);
12510			mtx_unlock(&lun->lun_lock);
12511		}
12512		break;
12513	case CTL_MSG_PERS_ACTION:
12514		ctl_hndl_per_res_out_on_other_sc(
12515			(union ctl_ha_msg *)&io->presio.pr_msg);
12516		free_io = 1;
12517		break;
12518	case CTL_MSG_BAD_JUJU:
12519		free_io = 0;
12520		ctl_done(io);
12521		break;
12522	case CTL_MSG_DATAMOVE:
12523		/* Only used in XFER mode */
12524		free_io = 0;
12525		ctl_datamove_remote(io);
12526		break;
12527	case CTL_MSG_DATAMOVE_DONE:
12528		/* Only used in XFER mode */
12529		free_io = 0;
12530		io->scsiio.be_move_done(io);
12531		break;
12532	default:
12533		free_io = 1;
12534		printf("%s: Invalid message type %d\n",
12535		       __func__, io->io_hdr.msg_type);
12536		break;
12537	}
12538	if (free_io)
12539		ctl_free_io(io);
12540
12541}
12542
12543
12544/*
12545 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12546 * there is no match.
12547 */
12548static ctl_lun_error_pattern
12549ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12550{
12551	const struct ctl_cmd_entry *entry;
12552	ctl_lun_error_pattern filtered_pattern, pattern;
12553
12554	pattern = desc->error_pattern;
12555
12556	/*
12557	 * XXX KDM we need more data passed into this function to match a
12558	 * custom pattern, and we actually need to implement custom pattern
12559	 * matching.
12560	 */
12561	if (pattern & CTL_LUN_PAT_CMD)
12562		return (CTL_LUN_PAT_CMD);
12563
12564	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12565		return (CTL_LUN_PAT_ANY);
12566
12567	entry = ctl_get_cmd_entry(ctsio, NULL);
12568
12569	filtered_pattern = entry->pattern & pattern;
12570
12571	/*
12572	 * If the user requested specific flags in the pattern (e.g.
12573	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12574	 * flags.
12575	 *
12576	 * If the user did not specify any flags, it doesn't matter whether
12577	 * or not the command supports the flags.
12578	 */
12579	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12580	     (pattern & ~CTL_LUN_PAT_MASK))
12581		return (CTL_LUN_PAT_NONE);
12582
12583	/*
12584	 * If the user asked for a range check, see if the requested LBA
12585	 * range overlaps with this command's LBA range.
12586	 */
12587	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12588		uint64_t lba1;
12589		uint64_t len1;
12590		ctl_action action;
12591		int retval;
12592
12593		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12594		if (retval != 0)
12595			return (CTL_LUN_PAT_NONE);
12596
12597		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12598					      desc->lba_range.len);
12599		/*
12600		 * A "pass" means that the LBA ranges don't overlap, so
12601		 * this doesn't match the user's range criteria.
12602		 */
12603		if (action == CTL_ACTION_PASS)
12604			return (CTL_LUN_PAT_NONE);
12605	}
12606
12607	return (filtered_pattern);
12608}
12609
12610static void
12611ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12612{
12613	struct ctl_error_desc *desc, *desc2;
12614
12615	mtx_assert(&lun->lun_lock, MA_OWNED);
12616
12617	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12618		ctl_lun_error_pattern pattern;
12619		/*
12620		 * Check to see whether this particular command matches
12621		 * the pattern in the descriptor.
12622		 */
12623		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12624		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12625			continue;
12626
12627		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12628		case CTL_LUN_INJ_ABORTED:
12629			ctl_set_aborted(&io->scsiio);
12630			break;
12631		case CTL_LUN_INJ_MEDIUM_ERR:
12632			ctl_set_medium_error(&io->scsiio);
12633			break;
12634		case CTL_LUN_INJ_UA:
12635			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12636			 * OCCURRED */
12637			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12638			break;
12639		case CTL_LUN_INJ_CUSTOM:
12640			/*
12641			 * We're assuming the user knows what he is doing.
12642			 * Just copy the sense information without doing
12643			 * checks.
12644			 */
12645			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12646			      ctl_min(sizeof(desc->custom_sense),
12647				      sizeof(io->scsiio.sense_data)));
12648			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12649			io->scsiio.sense_len = SSD_FULL_SIZE;
12650			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12651			break;
12652		case CTL_LUN_INJ_NONE:
12653		default:
12654			/*
12655			 * If this is an error injection type we don't know
12656			 * about, clear the continuous flag (if it is set)
12657			 * so it will get deleted below.
12658			 */
12659			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12660			break;
12661		}
12662		/*
12663		 * By default, each error injection action is a one-shot
12664		 */
12665		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12666			continue;
12667
12668		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12669
12670		free(desc, M_CTL);
12671	}
12672}
12673
12674#ifdef CTL_IO_DELAY
12675static void
12676ctl_datamove_timer_wakeup(void *arg)
12677{
12678	union ctl_io *io;
12679
12680	io = (union ctl_io *)arg;
12681
12682	ctl_datamove(io);
12683}
12684#endif /* CTL_IO_DELAY */
12685
12686void
12687ctl_datamove(union ctl_io *io)
12688{
12689	void (*fe_datamove)(union ctl_io *io);
12690
12691	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12692
12693	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12694
12695#ifdef CTL_TIME_IO
12696	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12697		char str[256];
12698		char path_str[64];
12699		struct sbuf sb;
12700
12701		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12702		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12703
12704		sbuf_cat(&sb, path_str);
12705		switch (io->io_hdr.io_type) {
12706		case CTL_IO_SCSI:
12707			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12708			sbuf_printf(&sb, "\n");
12709			sbuf_cat(&sb, path_str);
12710			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12711				    io->scsiio.tag_num, io->scsiio.tag_type);
12712			break;
12713		case CTL_IO_TASK:
12714			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12715				    "Tag Type: %d\n", io->taskio.task_action,
12716				    io->taskio.tag_num, io->taskio.tag_type);
12717			break;
12718		default:
12719			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12720			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12721			break;
12722		}
12723		sbuf_cat(&sb, path_str);
12724		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12725			    (intmax_t)time_uptime - io->io_hdr.start_time);
12726		sbuf_finish(&sb);
12727		printf("%s", sbuf_data(&sb));
12728	}
12729#endif /* CTL_TIME_IO */
12730
12731#ifdef CTL_IO_DELAY
12732	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12733		struct ctl_lun *lun;
12734
12735		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12736
12737		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12738	} else {
12739		struct ctl_lun *lun;
12740
12741		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12742		if ((lun != NULL)
12743		 && (lun->delay_info.datamove_delay > 0)) {
12744			struct callout *callout;
12745
12746			callout = (struct callout *)&io->io_hdr.timer_bytes;
12747			callout_init(callout, /*mpsafe*/ 1);
12748			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12749			callout_reset(callout,
12750				      lun->delay_info.datamove_delay * hz,
12751				      ctl_datamove_timer_wakeup, io);
12752			if (lun->delay_info.datamove_type ==
12753			    CTL_DELAY_TYPE_ONESHOT)
12754				lun->delay_info.datamove_delay = 0;
12755			return;
12756		}
12757	}
12758#endif
12759
12760	/*
12761	 * This command has been aborted.  Set the port status, so we fail
12762	 * the data move.
12763	 */
12764	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12765		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12766		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12767		       io->io_hdr.nexus.targ_port,
12768		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12769		       io->io_hdr.nexus.targ_lun);
12770		io->io_hdr.port_status = 31337;
12771		/*
12772		 * Note that the backend, in this case, will get the
12773		 * callback in its context.  In other cases it may get
12774		 * called in the frontend's interrupt thread context.
12775		 */
12776		io->scsiio.be_move_done(io);
12777		return;
12778	}
12779
12780	/*
12781	 * If we're in XFER mode and this I/O is from the other shelf
12782	 * controller, we need to send the DMA to the other side to
12783	 * actually transfer the data to/from the host.  In serialize only
12784	 * mode the transfer happens below CTL and ctl_datamove() is only
12785	 * called on the machine that originally received the I/O.
12786	 */
12787	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12788	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12789		union ctl_ha_msg msg;
12790		uint32_t sg_entries_sent;
12791		int do_sg_copy;
12792		int i;
12793
12794		memset(&msg, 0, sizeof(msg));
12795		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12796		msg.hdr.original_sc = io->io_hdr.original_sc;
12797		msg.hdr.serializing_sc = io;
12798		msg.hdr.nexus = io->io_hdr.nexus;
12799		msg.dt.flags = io->io_hdr.flags;
12800		/*
12801		 * We convert everything into a S/G list here.  We can't
12802		 * pass by reference, only by value between controllers.
12803		 * So we can't pass a pointer to the S/G list, only as many
12804		 * S/G entries as we can fit in here.  If it's possible for
12805		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12806		 * then we need to break this up into multiple transfers.
12807		 */
12808		if (io->scsiio.kern_sg_entries == 0) {
12809			msg.dt.kern_sg_entries = 1;
12810			/*
12811			 * If this is in cached memory, flush the cache
12812			 * before we send the DMA request to the other
12813			 * controller.  We want to do this in either the
12814			 * read or the write case.  The read case is
12815			 * straightforward.  In the write case, we want to
12816			 * make sure nothing is in the local cache that
12817			 * could overwrite the DMAed data.
12818			 */
12819			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12820				/*
12821				 * XXX KDM use bus_dmamap_sync() here.
12822				 */
12823			}
12824
12825			/*
12826			 * Convert to a physical address if this is a
12827			 * virtual address.
12828			 */
12829			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12830				msg.dt.sg_list[0].addr =
12831					io->scsiio.kern_data_ptr;
12832			} else {
12833				/*
12834				 * XXX KDM use busdma here!
12835				 */
12836#if 0
12837				msg.dt.sg_list[0].addr = (void *)
12838					vtophys(io->scsiio.kern_data_ptr);
12839#endif
12840			}
12841
12842			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12843			do_sg_copy = 0;
12844		} else {
12845			struct ctl_sg_entry *sgl;
12846
12847			do_sg_copy = 1;
12848			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12849			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12850			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12851				/*
12852				 * XXX KDM use bus_dmamap_sync() here.
12853				 */
12854			}
12855		}
12856
12857		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12858		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12859		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12860		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12861		msg.dt.sg_sequence = 0;
12862
12863		/*
12864		 * Loop until we've sent all of the S/G entries.  On the
12865		 * other end, we'll recompose these S/G entries into one
12866		 * contiguous list before passing it to the
12867		 */
12868		for (sg_entries_sent = 0; sg_entries_sent <
12869		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12870			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12871				sizeof(msg.dt.sg_list[0])),
12872				msg.dt.kern_sg_entries - sg_entries_sent);
12873
12874			if (do_sg_copy != 0) {
12875				struct ctl_sg_entry *sgl;
12876				int j;
12877
12878				sgl = (struct ctl_sg_entry *)
12879					io->scsiio.kern_data_ptr;
12880				/*
12881				 * If this is in cached memory, flush the cache
12882				 * before we send the DMA request to the other
12883				 * controller.  We want to do this in either
12884				 * the * read or the write case.  The read
12885				 * case is straightforward.  In the write
12886				 * case, we want to make sure nothing is
12887				 * in the local cache that could overwrite
12888				 * the DMAed data.
12889				 */
12890
12891				for (i = sg_entries_sent, j = 0;
12892				     i < msg.dt.cur_sg_entries; i++, j++) {
12893					if ((io->io_hdr.flags &
12894					     CTL_FLAG_NO_DATASYNC) == 0) {
12895						/*
12896						 * XXX KDM use bus_dmamap_sync()
12897						 */
12898					}
12899					if ((io->io_hdr.flags &
12900					     CTL_FLAG_BUS_ADDR) == 0) {
12901						/*
12902						 * XXX KDM use busdma.
12903						 */
12904#if 0
12905						msg.dt.sg_list[j].addr =(void *)
12906						       vtophys(sgl[i].addr);
12907#endif
12908					} else {
12909						msg.dt.sg_list[j].addr =
12910							sgl[i].addr;
12911					}
12912					msg.dt.sg_list[j].len = sgl[i].len;
12913				}
12914			}
12915
12916			sg_entries_sent += msg.dt.cur_sg_entries;
12917			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12918				msg.dt.sg_last = 1;
12919			else
12920				msg.dt.sg_last = 0;
12921
12922			/*
12923			 * XXX KDM drop and reacquire the lock here?
12924			 */
12925			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12926			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12927				/*
12928				 * XXX do something here.
12929				 */
12930			}
12931
12932			msg.dt.sent_sg_entries = sg_entries_sent;
12933		}
12934		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12935		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12936			ctl_failover_io(io, /*have_lock*/ 0);
12937
12938	} else {
12939
12940		/*
12941		 * Lookup the fe_datamove() function for this particular
12942		 * front end.
12943		 */
12944		fe_datamove =
12945		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12946
12947		fe_datamove(io);
12948	}
12949}
12950
12951static void
12952ctl_send_datamove_done(union ctl_io *io, int have_lock)
12953{
12954	union ctl_ha_msg msg;
12955	int isc_status;
12956
12957	memset(&msg, 0, sizeof(msg));
12958
12959	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12960	msg.hdr.original_sc = io;
12961	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12962	msg.hdr.nexus = io->io_hdr.nexus;
12963	msg.hdr.status = io->io_hdr.status;
12964	msg.scsi.tag_num = io->scsiio.tag_num;
12965	msg.scsi.tag_type = io->scsiio.tag_type;
12966	msg.scsi.scsi_status = io->scsiio.scsi_status;
12967	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12968	       sizeof(io->scsiio.sense_data));
12969	msg.scsi.sense_len = io->scsiio.sense_len;
12970	msg.scsi.sense_residual = io->scsiio.sense_residual;
12971	msg.scsi.fetd_status = io->io_hdr.port_status;
12972	msg.scsi.residual = io->scsiio.residual;
12973	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12974
12975	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12976		ctl_failover_io(io, /*have_lock*/ have_lock);
12977		return;
12978	}
12979
12980	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12981	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12982		/* XXX do something if this fails */
12983	}
12984
12985}
12986
12987/*
12988 * The DMA to the remote side is done, now we need to tell the other side
12989 * we're done so it can continue with its data movement.
12990 */
12991static void
12992ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12993{
12994	union ctl_io *io;
12995
12996	io = rq->context;
12997
12998	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12999		printf("%s: ISC DMA write failed with error %d", __func__,
13000		       rq->ret);
13001		ctl_set_internal_failure(&io->scsiio,
13002					 /*sks_valid*/ 1,
13003					 /*retry_count*/ rq->ret);
13004	}
13005
13006	ctl_dt_req_free(rq);
13007
13008	/*
13009	 * In this case, we had to malloc the memory locally.  Free it.
13010	 */
13011	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13012		int i;
13013		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13014			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13015	}
13016	/*
13017	 * The data is in local and remote memory, so now we need to send
13018	 * status (good or back) back to the other side.
13019	 */
13020	ctl_send_datamove_done(io, /*have_lock*/ 0);
13021}
13022
13023/*
13024 * We've moved the data from the host/controller into local memory.  Now we
13025 * need to push it over to the remote controller's memory.
13026 */
13027static int
13028ctl_datamove_remote_dm_write_cb(union ctl_io *io)
13029{
13030	int retval;
13031
13032	retval = 0;
13033
13034	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
13035					  ctl_datamove_remote_write_cb);
13036
13037	return (retval);
13038}
13039
13040static void
13041ctl_datamove_remote_write(union ctl_io *io)
13042{
13043	int retval;
13044	void (*fe_datamove)(union ctl_io *io);
13045
13046	/*
13047	 * - Get the data from the host/HBA into local memory.
13048	 * - DMA memory from the local controller to the remote controller.
13049	 * - Send status back to the remote controller.
13050	 */
13051
13052	retval = ctl_datamove_remote_sgl_setup(io);
13053	if (retval != 0)
13054		return;
13055
13056	/* Switch the pointer over so the FETD knows what to do */
13057	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13058
13059	/*
13060	 * Use a custom move done callback, since we need to send completion
13061	 * back to the other controller, not to the backend on this side.
13062	 */
13063	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
13064
13065	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13066
13067	fe_datamove(io);
13068
13069	return;
13070
13071}
13072
13073static int
13074ctl_datamove_remote_dm_read_cb(union ctl_io *io)
13075{
13076#if 0
13077	char str[256];
13078	char path_str[64];
13079	struct sbuf sb;
13080#endif
13081
13082	/*
13083	 * In this case, we had to malloc the memory locally.  Free it.
13084	 */
13085	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13086		int i;
13087		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13088			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13089	}
13090
13091#if 0
13092	scsi_path_string(io, path_str, sizeof(path_str));
13093	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13094	sbuf_cat(&sb, path_str);
13095	scsi_command_string(&io->scsiio, NULL, &sb);
13096	sbuf_printf(&sb, "\n");
13097	sbuf_cat(&sb, path_str);
13098	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13099		    io->scsiio.tag_num, io->scsiio.tag_type);
13100	sbuf_cat(&sb, path_str);
13101	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
13102		    io->io_hdr.flags, io->io_hdr.status);
13103	sbuf_finish(&sb);
13104	printk("%s", sbuf_data(&sb));
13105#endif
13106
13107
13108	/*
13109	 * The read is done, now we need to send status (good or bad) back
13110	 * to the other side.
13111	 */
13112	ctl_send_datamove_done(io, /*have_lock*/ 0);
13113
13114	return (0);
13115}
13116
13117static void
13118ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
13119{
13120	union ctl_io *io;
13121	void (*fe_datamove)(union ctl_io *io);
13122
13123	io = rq->context;
13124
13125	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13126		printf("%s: ISC DMA read failed with error %d", __func__,
13127		       rq->ret);
13128		ctl_set_internal_failure(&io->scsiio,
13129					 /*sks_valid*/ 1,
13130					 /*retry_count*/ rq->ret);
13131	}
13132
13133	ctl_dt_req_free(rq);
13134
13135	/* Switch the pointer over so the FETD knows what to do */
13136	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13137
13138	/*
13139	 * Use a custom move done callback, since we need to send completion
13140	 * back to the other controller, not to the backend on this side.
13141	 */
13142	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13143
13144	/* XXX KDM add checks like the ones in ctl_datamove? */
13145
13146	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13147
13148	fe_datamove(io);
13149}
13150
13151static int
13152ctl_datamove_remote_sgl_setup(union ctl_io *io)
13153{
13154	struct ctl_sg_entry *local_sglist, *remote_sglist;
13155	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13156	struct ctl_softc *softc;
13157	int retval;
13158	int i;
13159
13160	retval = 0;
13161	softc = control_softc;
13162
13163	local_sglist = io->io_hdr.local_sglist;
13164	local_dma_sglist = io->io_hdr.local_dma_sglist;
13165	remote_sglist = io->io_hdr.remote_sglist;
13166	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13167
13168	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13169		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13170			local_sglist[i].len = remote_sglist[i].len;
13171
13172			/*
13173			 * XXX Detect the situation where the RS-level I/O
13174			 * redirector on the other side has already read the
13175			 * data off of the AOR RS on this side, and
13176			 * transferred it to remote (mirror) memory on the
13177			 * other side.  Since we already have the data in
13178			 * memory here, we just need to use it.
13179			 *
13180			 * XXX KDM this can probably be removed once we
13181			 * get the cache device code in and take the
13182			 * current AOR implementation out.
13183			 */
13184#ifdef NEEDTOPORT
13185			if ((remote_sglist[i].addr >=
13186			     (void *)vtophys(softc->mirr->addr))
13187			 && (remote_sglist[i].addr <
13188			     ((void *)vtophys(softc->mirr->addr) +
13189			     CacheMirrorOffset))) {
13190				local_sglist[i].addr = remote_sglist[i].addr -
13191					CacheMirrorOffset;
13192				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13193				     CTL_FLAG_DATA_IN)
13194					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13195			} else {
13196				local_sglist[i].addr = remote_sglist[i].addr +
13197					CacheMirrorOffset;
13198			}
13199#endif
13200#if 0
13201			printf("%s: local %p, remote %p, len %d\n",
13202			       __func__, local_sglist[i].addr,
13203			       remote_sglist[i].addr, local_sglist[i].len);
13204#endif
13205		}
13206	} else {
13207		uint32_t len_to_go;
13208
13209		/*
13210		 * In this case, we don't have automatically allocated
13211		 * memory for this I/O on this controller.  This typically
13212		 * happens with internal CTL I/O -- e.g. inquiry, mode
13213		 * sense, etc.  Anything coming from RAIDCore will have
13214		 * a mirror area available.
13215		 */
13216		len_to_go = io->scsiio.kern_data_len;
13217
13218		/*
13219		 * Clear the no datasync flag, we have to use malloced
13220		 * buffers.
13221		 */
13222		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13223
13224		/*
13225		 * The difficult thing here is that the size of the various
13226		 * S/G segments may be different than the size from the
13227		 * remote controller.  That'll make it harder when DMAing
13228		 * the data back to the other side.
13229		 */
13230		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13231		     sizeof(io->io_hdr.remote_sglist[0])) &&
13232		     (len_to_go > 0); i++) {
13233			local_sglist[i].len = ctl_min(len_to_go, 131072);
13234			CTL_SIZE_8B(local_dma_sglist[i].len,
13235				    local_sglist[i].len);
13236			local_sglist[i].addr =
13237				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13238
13239			local_dma_sglist[i].addr = local_sglist[i].addr;
13240
13241			if (local_sglist[i].addr == NULL) {
13242				int j;
13243
13244				printf("malloc failed for %zd bytes!",
13245				       local_dma_sglist[i].len);
13246				for (j = 0; j < i; j++) {
13247					free(local_sglist[j].addr, M_CTL);
13248				}
13249				ctl_set_internal_failure(&io->scsiio,
13250							 /*sks_valid*/ 1,
13251							 /*retry_count*/ 4857);
13252				retval = 1;
13253				goto bailout_error;
13254
13255			}
13256			/* XXX KDM do we need a sync here? */
13257
13258			len_to_go -= local_sglist[i].len;
13259		}
13260		/*
13261		 * Reset the number of S/G entries accordingly.  The
13262		 * original number of S/G entries is available in
13263		 * rem_sg_entries.
13264		 */
13265		io->scsiio.kern_sg_entries = i;
13266
13267#if 0
13268		printf("%s: kern_sg_entries = %d\n", __func__,
13269		       io->scsiio.kern_sg_entries);
13270		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13271			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13272			       local_sglist[i].addr, local_sglist[i].len,
13273			       local_dma_sglist[i].len);
13274#endif
13275	}
13276
13277
13278	return (retval);
13279
13280bailout_error:
13281
13282	ctl_send_datamove_done(io, /*have_lock*/ 0);
13283
13284	return (retval);
13285}
13286
13287static int
13288ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13289			 ctl_ha_dt_cb callback)
13290{
13291	struct ctl_ha_dt_req *rq;
13292	struct ctl_sg_entry *remote_sglist, *local_sglist;
13293	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13294	uint32_t local_used, remote_used, total_used;
13295	int retval;
13296	int i, j;
13297
13298	retval = 0;
13299
13300	rq = ctl_dt_req_alloc();
13301
13302	/*
13303	 * If we failed to allocate the request, and if the DMA didn't fail
13304	 * anyway, set busy status.  This is just a resource allocation
13305	 * failure.
13306	 */
13307	if ((rq == NULL)
13308	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13309		ctl_set_busy(&io->scsiio);
13310
13311	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13312
13313		if (rq != NULL)
13314			ctl_dt_req_free(rq);
13315
13316		/*
13317		 * The data move failed.  We need to return status back
13318		 * to the other controller.  No point in trying to DMA
13319		 * data to the remote controller.
13320		 */
13321
13322		ctl_send_datamove_done(io, /*have_lock*/ 0);
13323
13324		retval = 1;
13325
13326		goto bailout;
13327	}
13328
13329	local_sglist = io->io_hdr.local_sglist;
13330	local_dma_sglist = io->io_hdr.local_dma_sglist;
13331	remote_sglist = io->io_hdr.remote_sglist;
13332	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13333	local_used = 0;
13334	remote_used = 0;
13335	total_used = 0;
13336
13337	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13338		rq->ret = CTL_HA_STATUS_SUCCESS;
13339		rq->context = io;
13340		callback(rq);
13341		goto bailout;
13342	}
13343
13344	/*
13345	 * Pull/push the data over the wire from/to the other controller.
13346	 * This takes into account the possibility that the local and
13347	 * remote sglists may not be identical in terms of the size of
13348	 * the elements and the number of elements.
13349	 *
13350	 * One fundamental assumption here is that the length allocated for
13351	 * both the local and remote sglists is identical.  Otherwise, we've
13352	 * essentially got a coding error of some sort.
13353	 */
13354	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13355		int isc_ret;
13356		uint32_t cur_len, dma_length;
13357		uint8_t *tmp_ptr;
13358
13359		rq->id = CTL_HA_DATA_CTL;
13360		rq->command = command;
13361		rq->context = io;
13362
13363		/*
13364		 * Both pointers should be aligned.  But it is possible
13365		 * that the allocation length is not.  They should both
13366		 * also have enough slack left over at the end, though,
13367		 * to round up to the next 8 byte boundary.
13368		 */
13369		cur_len = ctl_min(local_sglist[i].len - local_used,
13370				  remote_sglist[j].len - remote_used);
13371
13372		/*
13373		 * In this case, we have a size issue and need to decrease
13374		 * the size, except in the case where we actually have less
13375		 * than 8 bytes left.  In that case, we need to increase
13376		 * the DMA length to get the last bit.
13377		 */
13378		if ((cur_len & 0x7) != 0) {
13379			if (cur_len > 0x7) {
13380				cur_len = cur_len - (cur_len & 0x7);
13381				dma_length = cur_len;
13382			} else {
13383				CTL_SIZE_8B(dma_length, cur_len);
13384			}
13385
13386		} else
13387			dma_length = cur_len;
13388
13389		/*
13390		 * If we had to allocate memory for this I/O, instead of using
13391		 * the non-cached mirror memory, we'll need to flush the cache
13392		 * before trying to DMA to the other controller.
13393		 *
13394		 * We could end up doing this multiple times for the same
13395		 * segment if we have a larger local segment than remote
13396		 * segment.  That shouldn't be an issue.
13397		 */
13398		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13399			/*
13400			 * XXX KDM use bus_dmamap_sync() here.
13401			 */
13402		}
13403
13404		rq->size = dma_length;
13405
13406		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13407		tmp_ptr += local_used;
13408
13409		/* Use physical addresses when talking to ISC hardware */
13410		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13411			/* XXX KDM use busdma */
13412#if 0
13413			rq->local = vtophys(tmp_ptr);
13414#endif
13415		} else
13416			rq->local = tmp_ptr;
13417
13418		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13419		tmp_ptr += remote_used;
13420		rq->remote = tmp_ptr;
13421
13422		rq->callback = NULL;
13423
13424		local_used += cur_len;
13425		if (local_used >= local_sglist[i].len) {
13426			i++;
13427			local_used = 0;
13428		}
13429
13430		remote_used += cur_len;
13431		if (remote_used >= remote_sglist[j].len) {
13432			j++;
13433			remote_used = 0;
13434		}
13435		total_used += cur_len;
13436
13437		if (total_used >= io->scsiio.kern_data_len)
13438			rq->callback = callback;
13439
13440		if ((rq->size & 0x7) != 0) {
13441			printf("%s: warning: size %d is not on 8b boundary\n",
13442			       __func__, rq->size);
13443		}
13444		if (((uintptr_t)rq->local & 0x7) != 0) {
13445			printf("%s: warning: local %p not on 8b boundary\n",
13446			       __func__, rq->local);
13447		}
13448		if (((uintptr_t)rq->remote & 0x7) != 0) {
13449			printf("%s: warning: remote %p not on 8b boundary\n",
13450			       __func__, rq->local);
13451		}
13452#if 0
13453		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13454		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13455		       rq->local, rq->remote, rq->size);
13456#endif
13457
13458		isc_ret = ctl_dt_single(rq);
13459		if (isc_ret == CTL_HA_STATUS_WAIT)
13460			continue;
13461
13462		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13463			rq->ret = CTL_HA_STATUS_SUCCESS;
13464		} else {
13465			rq->ret = isc_ret;
13466		}
13467		callback(rq);
13468		goto bailout;
13469	}
13470
13471bailout:
13472	return (retval);
13473
13474}
13475
13476static void
13477ctl_datamove_remote_read(union ctl_io *io)
13478{
13479	int retval;
13480	int i;
13481
13482	/*
13483	 * This will send an error to the other controller in the case of a
13484	 * failure.
13485	 */
13486	retval = ctl_datamove_remote_sgl_setup(io);
13487	if (retval != 0)
13488		return;
13489
13490	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13491					  ctl_datamove_remote_read_cb);
13492	if ((retval != 0)
13493	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13494		/*
13495		 * Make sure we free memory if there was an error..  The
13496		 * ctl_datamove_remote_xfer() function will send the
13497		 * datamove done message, or call the callback with an
13498		 * error if there is a problem.
13499		 */
13500		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13501			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13502	}
13503
13504	return;
13505}
13506
13507/*
13508 * Process a datamove request from the other controller.  This is used for
13509 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13510 * first.  Once that is complete, the data gets DMAed into the remote
13511 * controller's memory.  For reads, we DMA from the remote controller's
13512 * memory into our memory first, and then move it out to the FETD.
13513 */
13514static void
13515ctl_datamove_remote(union ctl_io *io)
13516{
13517	struct ctl_softc *softc;
13518
13519	softc = control_softc;
13520
13521	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13522
13523	/*
13524	 * Note that we look for an aborted I/O here, but don't do some of
13525	 * the other checks that ctl_datamove() normally does.
13526	 * We don't need to run the datamove delay code, since that should
13527	 * have been done if need be on the other controller.
13528	 */
13529	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13530		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13531		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13532		       io->io_hdr.nexus.targ_port,
13533		       io->io_hdr.nexus.targ_target.id,
13534		       io->io_hdr.nexus.targ_lun);
13535		io->io_hdr.port_status = 31338;
13536		ctl_send_datamove_done(io, /*have_lock*/ 0);
13537		return;
13538	}
13539
13540	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13541		ctl_datamove_remote_write(io);
13542	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13543		ctl_datamove_remote_read(io);
13544	} else {
13545		union ctl_ha_msg msg;
13546		struct scsi_sense_data *sense;
13547		uint8_t sks[3];
13548		int retry_count;
13549
13550		memset(&msg, 0, sizeof(msg));
13551
13552		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13553		msg.hdr.status = CTL_SCSI_ERROR;
13554		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13555
13556		retry_count = 4243;
13557
13558		sense = &msg.scsi.sense_data;
13559		sks[0] = SSD_SCS_VALID;
13560		sks[1] = (retry_count >> 8) & 0xff;
13561		sks[2] = retry_count & 0xff;
13562
13563		/* "Internal target failure" */
13564		scsi_set_sense_data(sense,
13565				    /*sense_format*/ SSD_TYPE_NONE,
13566				    /*current_error*/ 1,
13567				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13568				    /*asc*/ 0x44,
13569				    /*ascq*/ 0x00,
13570				    /*type*/ SSD_ELEM_SKS,
13571				    /*size*/ sizeof(sks),
13572				    /*data*/ sks,
13573				    SSD_ELEM_NONE);
13574
13575		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13576		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13577			ctl_failover_io(io, /*have_lock*/ 1);
13578			return;
13579		}
13580
13581		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13582		    CTL_HA_STATUS_SUCCESS) {
13583			/* XXX KDM what to do if this fails? */
13584		}
13585		return;
13586	}
13587
13588}
13589
13590static int
13591ctl_process_done(union ctl_io *io)
13592{
13593	struct ctl_lun *lun;
13594	struct ctl_softc *ctl_softc;
13595	void (*fe_done)(union ctl_io *io);
13596	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13597
13598	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13599
13600	fe_done =
13601	    control_softc->ctl_ports[targ_port]->fe_done;
13602
13603#ifdef CTL_TIME_IO
13604	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13605		char str[256];
13606		char path_str[64];
13607		struct sbuf sb;
13608
13609		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13610		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13611
13612		sbuf_cat(&sb, path_str);
13613		switch (io->io_hdr.io_type) {
13614		case CTL_IO_SCSI:
13615			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13616			sbuf_printf(&sb, "\n");
13617			sbuf_cat(&sb, path_str);
13618			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13619				    io->scsiio.tag_num, io->scsiio.tag_type);
13620			break;
13621		case CTL_IO_TASK:
13622			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13623				    "Tag Type: %d\n", io->taskio.task_action,
13624				    io->taskio.tag_num, io->taskio.tag_type);
13625			break;
13626		default:
13627			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13628			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13629			break;
13630		}
13631		sbuf_cat(&sb, path_str);
13632		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13633			    (intmax_t)time_uptime - io->io_hdr.start_time);
13634		sbuf_finish(&sb);
13635		printf("%s", sbuf_data(&sb));
13636	}
13637#endif /* CTL_TIME_IO */
13638
13639	switch (io->io_hdr.io_type) {
13640	case CTL_IO_SCSI:
13641		break;
13642	case CTL_IO_TASK:
13643		if (bootverbose || verbose > 0)
13644			ctl_io_error_print(io, NULL);
13645		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13646			ctl_free_io(io);
13647		else
13648			fe_done(io);
13649		return (CTL_RETVAL_COMPLETE);
13650		break;
13651	default:
13652		printf("ctl_process_done: invalid io type %d\n",
13653		       io->io_hdr.io_type);
13654		panic("ctl_process_done: invalid io type %d\n",
13655		      io->io_hdr.io_type);
13656		break; /* NOTREACHED */
13657	}
13658
13659	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13660	if (lun == NULL) {
13661		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13662				 io->io_hdr.nexus.targ_mapped_lun));
13663		fe_done(io);
13664		goto bailout;
13665	}
13666	ctl_softc = lun->ctl_softc;
13667
13668	mtx_lock(&lun->lun_lock);
13669
13670	/*
13671	 * Check to see if we have any errors to inject here.  We only
13672	 * inject errors for commands that don't already have errors set.
13673	 */
13674	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13675	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13676		ctl_inject_error(lun, io);
13677
13678	/*
13679	 * XXX KDM how do we treat commands that aren't completed
13680	 * successfully?
13681	 *
13682	 * XXX KDM should we also track I/O latency?
13683	 */
13684	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13685	    io->io_hdr.io_type == CTL_IO_SCSI) {
13686#ifdef CTL_TIME_IO
13687		struct bintime cur_bt;
13688#endif
13689		int type;
13690
13691		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13692		    CTL_FLAG_DATA_IN)
13693			type = CTL_STATS_READ;
13694		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13695		    CTL_FLAG_DATA_OUT)
13696			type = CTL_STATS_WRITE;
13697		else
13698			type = CTL_STATS_NO_IO;
13699
13700		lun->stats.ports[targ_port].bytes[type] +=
13701		    io->scsiio.kern_total_len;
13702		lun->stats.ports[targ_port].operations[type]++;
13703#ifdef CTL_TIME_IO
13704		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13705		   &io->io_hdr.dma_bt);
13706		lun->stats.ports[targ_port].num_dmas[type] +=
13707		    io->io_hdr.num_dmas;
13708		getbintime(&cur_bt);
13709		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13710		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13711#endif
13712	}
13713
13714	/*
13715	 * Remove this from the OOA queue.
13716	 */
13717	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13718
13719	/*
13720	 * Run through the blocked queue on this LUN and see if anything
13721	 * has become unblocked, now that this transaction is done.
13722	 */
13723	ctl_check_blocked(lun);
13724
13725	/*
13726	 * If the LUN has been invalidated, free it if there is nothing
13727	 * left on its OOA queue.
13728	 */
13729	if ((lun->flags & CTL_LUN_INVALID)
13730	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13731		mtx_unlock(&lun->lun_lock);
13732		mtx_lock(&ctl_softc->ctl_lock);
13733		ctl_free_lun(lun);
13734		mtx_unlock(&ctl_softc->ctl_lock);
13735	} else
13736		mtx_unlock(&lun->lun_lock);
13737
13738	/*
13739	 * If this command has been aborted, make sure we set the status
13740	 * properly.  The FETD is responsible for freeing the I/O and doing
13741	 * whatever it needs to do to clean up its state.
13742	 */
13743	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13744		ctl_set_task_aborted(&io->scsiio);
13745
13746	/*
13747	 * We print out status for every task management command.  For SCSI
13748	 * commands, we filter out any unit attention errors; they happen
13749	 * on every boot, and would clutter up the log.  Note:  task
13750	 * management commands aren't printed here, they are printed above,
13751	 * since they should never even make it down here.
13752	 */
13753	switch (io->io_hdr.io_type) {
13754	case CTL_IO_SCSI: {
13755		int error_code, sense_key, asc, ascq;
13756
13757		sense_key = 0;
13758
13759		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13760		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13761			/*
13762			 * Since this is just for printing, no need to
13763			 * show errors here.
13764			 */
13765			scsi_extract_sense_len(&io->scsiio.sense_data,
13766					       io->scsiio.sense_len,
13767					       &error_code,
13768					       &sense_key,
13769					       &asc,
13770					       &ascq,
13771					       /*show_errors*/ 0);
13772		}
13773
13774		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13775		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13776		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13777		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13778
13779			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13780				ctl_softc->skipped_prints++;
13781			} else {
13782				uint32_t skipped_prints;
13783
13784				skipped_prints = ctl_softc->skipped_prints;
13785
13786				ctl_softc->skipped_prints = 0;
13787				ctl_softc->last_print_jiffies = time_uptime;
13788
13789				if (skipped_prints > 0) {
13790#ifdef NEEDTOPORT
13791					csevent_log(CSC_CTL | CSC_SHELF_SW |
13792					    CTL_ERROR_REPORT,
13793					    csevent_LogType_Trace,
13794					    csevent_Severity_Information,
13795					    csevent_AlertLevel_Green,
13796					    csevent_FRU_Firmware,
13797					    csevent_FRU_Unknown,
13798					    "High CTL error volume, %d prints "
13799					    "skipped", skipped_prints);
13800#endif
13801				}
13802				if (bootverbose || verbose > 0)
13803					ctl_io_error_print(io, NULL);
13804			}
13805		}
13806		break;
13807	}
13808	case CTL_IO_TASK:
13809		if (bootverbose || verbose > 0)
13810			ctl_io_error_print(io, NULL);
13811		break;
13812	default:
13813		break;
13814	}
13815
13816	/*
13817	 * Tell the FETD or the other shelf controller we're done with this
13818	 * command.  Note that only SCSI commands get to this point.  Task
13819	 * management commands are completed above.
13820	 *
13821	 * We only send status to the other controller if we're in XFER
13822	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13823	 * received the I/O (from CTL's perspective), and so the status is
13824	 * generated there.
13825	 *
13826	 * XXX KDM if we hold the lock here, we could cause a deadlock
13827	 * if the frontend comes back in in this context to queue
13828	 * something.
13829	 */
13830	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13831	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13832		union ctl_ha_msg msg;
13833
13834		memset(&msg, 0, sizeof(msg));
13835		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13836		msg.hdr.original_sc = io->io_hdr.original_sc;
13837		msg.hdr.nexus = io->io_hdr.nexus;
13838		msg.hdr.status = io->io_hdr.status;
13839		msg.scsi.scsi_status = io->scsiio.scsi_status;
13840		msg.scsi.tag_num = io->scsiio.tag_num;
13841		msg.scsi.tag_type = io->scsiio.tag_type;
13842		msg.scsi.sense_len = io->scsiio.sense_len;
13843		msg.scsi.sense_residual = io->scsiio.sense_residual;
13844		msg.scsi.residual = io->scsiio.residual;
13845		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13846		       sizeof(io->scsiio.sense_data));
13847		/*
13848		 * We copy this whether or not this is an I/O-related
13849		 * command.  Otherwise, we'd have to go and check to see
13850		 * whether it's a read/write command, and it really isn't
13851		 * worth it.
13852		 */
13853		memcpy(&msg.scsi.lbalen,
13854		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13855		       sizeof(msg.scsi.lbalen));
13856
13857		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13858				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13859			/* XXX do something here */
13860		}
13861
13862		ctl_free_io(io);
13863	} else
13864		fe_done(io);
13865
13866bailout:
13867
13868	return (CTL_RETVAL_COMPLETE);
13869}
13870
13871#ifdef CTL_WITH_CA
13872/*
13873 * Front end should call this if it doesn't do autosense.  When the request
13874 * sense comes back in from the initiator, we'll dequeue this and send it.
13875 */
13876int
13877ctl_queue_sense(union ctl_io *io)
13878{
13879	struct ctl_lun *lun;
13880	struct ctl_softc *ctl_softc;
13881	uint32_t initidx, targ_lun;
13882
13883	ctl_softc = control_softc;
13884
13885	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13886
13887	/*
13888	 * LUN lookup will likely move to the ctl_work_thread() once we
13889	 * have our new queueing infrastructure (that doesn't put things on
13890	 * a per-LUN queue initially).  That is so that we can handle
13891	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13892	 * can't deal with that right now.
13893	 */
13894	mtx_lock(&ctl_softc->ctl_lock);
13895
13896	/*
13897	 * If we don't have a LUN for this, just toss the sense
13898	 * information.
13899	 */
13900	targ_lun = io->io_hdr.nexus.targ_lun;
13901	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13902	if ((targ_lun < CTL_MAX_LUNS)
13903	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13904		lun = ctl_softc->ctl_luns[targ_lun];
13905	else
13906		goto bailout;
13907
13908	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13909
13910	mtx_lock(&lun->lun_lock);
13911	/*
13912	 * Already have CA set for this LUN...toss the sense information.
13913	 */
13914	if (ctl_is_set(lun->have_ca, initidx)) {
13915		mtx_unlock(&lun->lun_lock);
13916		goto bailout;
13917	}
13918
13919	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13920	       ctl_min(sizeof(lun->pending_sense[initidx]),
13921	       sizeof(io->scsiio.sense_data)));
13922	ctl_set_mask(lun->have_ca, initidx);
13923	mtx_unlock(&lun->lun_lock);
13924
13925bailout:
13926	mtx_unlock(&ctl_softc->ctl_lock);
13927
13928	ctl_free_io(io);
13929
13930	return (CTL_RETVAL_COMPLETE);
13931}
13932#endif
13933
13934/*
13935 * Primary command inlet from frontend ports.  All SCSI and task I/O
13936 * requests must go through this function.
13937 */
13938int
13939ctl_queue(union ctl_io *io)
13940{
13941	struct ctl_softc *ctl_softc;
13942
13943	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13944
13945	ctl_softc = control_softc;
13946
13947#ifdef CTL_TIME_IO
13948	io->io_hdr.start_time = time_uptime;
13949	getbintime(&io->io_hdr.start_bt);
13950#endif /* CTL_TIME_IO */
13951
13952	/* Map FE-specific LUN ID into global one. */
13953	io->io_hdr.nexus.targ_mapped_lun =
13954	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13955
13956	switch (io->io_hdr.io_type) {
13957	case CTL_IO_SCSI:
13958	case CTL_IO_TASK:
13959		ctl_enqueue_incoming(io);
13960		break;
13961	default:
13962		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13963		return (EINVAL);
13964	}
13965
13966	return (CTL_RETVAL_COMPLETE);
13967}
13968
13969#ifdef CTL_IO_DELAY
13970static void
13971ctl_done_timer_wakeup(void *arg)
13972{
13973	union ctl_io *io;
13974
13975	io = (union ctl_io *)arg;
13976	ctl_done(io);
13977}
13978#endif /* CTL_IO_DELAY */
13979
13980void
13981ctl_done(union ctl_io *io)
13982{
13983	struct ctl_softc *ctl_softc;
13984
13985	ctl_softc = control_softc;
13986
13987	/*
13988	 * Enable this to catch duplicate completion issues.
13989	 */
13990#if 0
13991	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13992		printf("%s: type %d msg %d cdb %x iptl: "
13993		       "%d:%d:%d:%d tag 0x%04x "
13994		       "flag %#x status %x\n",
13995			__func__,
13996			io->io_hdr.io_type,
13997			io->io_hdr.msg_type,
13998			io->scsiio.cdb[0],
13999			io->io_hdr.nexus.initid.id,
14000			io->io_hdr.nexus.targ_port,
14001			io->io_hdr.nexus.targ_target.id,
14002			io->io_hdr.nexus.targ_lun,
14003			(io->io_hdr.io_type ==
14004			CTL_IO_TASK) ?
14005			io->taskio.tag_num :
14006			io->scsiio.tag_num,
14007		        io->io_hdr.flags,
14008			io->io_hdr.status);
14009	} else
14010		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
14011#endif
14012
14013	/*
14014	 * This is an internal copy of an I/O, and should not go through
14015	 * the normal done processing logic.
14016	 */
14017	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
14018		return;
14019
14020	/*
14021	 * We need to send a msg to the serializing shelf to finish the IO
14022	 * as well.  We don't send a finish message to the other shelf if
14023	 * this is a task management command.  Task management commands
14024	 * aren't serialized in the OOA queue, but rather just executed on
14025	 * both shelf controllers for commands that originated on that
14026	 * controller.
14027	 */
14028	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
14029	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
14030		union ctl_ha_msg msg_io;
14031
14032		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
14033		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
14034		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
14035		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
14036		}
14037		/* continue on to finish IO */
14038	}
14039#ifdef CTL_IO_DELAY
14040	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
14041		struct ctl_lun *lun;
14042
14043		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14044
14045		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
14046	} else {
14047		struct ctl_lun *lun;
14048
14049		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14050
14051		if ((lun != NULL)
14052		 && (lun->delay_info.done_delay > 0)) {
14053			struct callout *callout;
14054
14055			callout = (struct callout *)&io->io_hdr.timer_bytes;
14056			callout_init(callout, /*mpsafe*/ 1);
14057			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
14058			callout_reset(callout,
14059				      lun->delay_info.done_delay * hz,
14060				      ctl_done_timer_wakeup, io);
14061			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
14062				lun->delay_info.done_delay = 0;
14063			return;
14064		}
14065	}
14066#endif /* CTL_IO_DELAY */
14067
14068	ctl_enqueue_done(io);
14069}
14070
14071int
14072ctl_isc(struct ctl_scsiio *ctsio)
14073{
14074	struct ctl_lun *lun;
14075	int retval;
14076
14077	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14078
14079	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
14080
14081	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
14082
14083	retval = lun->backend->data_submit((union ctl_io *)ctsio);
14084
14085	return (retval);
14086}
14087
14088
14089static void
14090ctl_work_thread(void *arg)
14091{
14092	struct ctl_thread *thr = (struct ctl_thread *)arg;
14093	struct ctl_softc *softc = thr->ctl_softc;
14094	union ctl_io *io;
14095	int retval;
14096
14097	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
14098
14099	for (;;) {
14100		retval = 0;
14101
14102		/*
14103		 * We handle the queues in this order:
14104		 * - ISC
14105		 * - done queue (to free up resources, unblock other commands)
14106		 * - RtR queue
14107		 * - incoming queue
14108		 *
14109		 * If those queues are empty, we break out of the loop and
14110		 * go to sleep.
14111		 */
14112		mtx_lock(&thr->queue_lock);
14113		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
14114		if (io != NULL) {
14115			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
14116			mtx_unlock(&thr->queue_lock);
14117			ctl_handle_isc(io);
14118			continue;
14119		}
14120		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
14121		if (io != NULL) {
14122			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
14123			/* clear any blocked commands, call fe_done */
14124			mtx_unlock(&thr->queue_lock);
14125			retval = ctl_process_done(io);
14126			continue;
14127		}
14128		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
14129		if (io != NULL) {
14130			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
14131			mtx_unlock(&thr->queue_lock);
14132			if (io->io_hdr.io_type == CTL_IO_TASK)
14133				ctl_run_task(io);
14134			else
14135				ctl_scsiio_precheck(softc, &io->scsiio);
14136			continue;
14137		}
14138		if (!ctl_pause_rtr) {
14139			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
14140			if (io != NULL) {
14141				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
14142				mtx_unlock(&thr->queue_lock);
14143				retval = ctl_scsiio(&io->scsiio);
14144				if (retval != CTL_RETVAL_COMPLETE)
14145					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
14146				continue;
14147			}
14148		}
14149
14150		/* Sleep until we have something to do. */
14151		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
14152	}
14153}
14154
14155static void
14156ctl_lun_thread(void *arg)
14157{
14158	struct ctl_softc *softc = (struct ctl_softc *)arg;
14159	struct ctl_be_lun *be_lun;
14160	int retval;
14161
14162	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14163
14164	for (;;) {
14165		retval = 0;
14166		mtx_lock(&softc->ctl_lock);
14167		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14168		if (be_lun != NULL) {
14169			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14170			mtx_unlock(&softc->ctl_lock);
14171			ctl_create_lun(be_lun);
14172			continue;
14173		}
14174
14175		/* Sleep until we have something to do. */
14176		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14177		    PDROP | PRIBIO, "-", 0);
14178	}
14179}
14180
14181static void
14182ctl_enqueue_incoming(union ctl_io *io)
14183{
14184	struct ctl_softc *softc = control_softc;
14185	struct ctl_thread *thr;
14186	u_int idx;
14187
14188	idx = (io->io_hdr.nexus.targ_port * 127 +
14189	       io->io_hdr.nexus.initid.id) % worker_threads;
14190	thr = &softc->threads[idx];
14191	mtx_lock(&thr->queue_lock);
14192	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14193	mtx_unlock(&thr->queue_lock);
14194	wakeup(thr);
14195}
14196
14197static void
14198ctl_enqueue_rtr(union ctl_io *io)
14199{
14200	struct ctl_softc *softc = control_softc;
14201	struct ctl_thread *thr;
14202
14203	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14204	mtx_lock(&thr->queue_lock);
14205	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14206	mtx_unlock(&thr->queue_lock);
14207	wakeup(thr);
14208}
14209
14210static void
14211ctl_enqueue_done(union ctl_io *io)
14212{
14213	struct ctl_softc *softc = control_softc;
14214	struct ctl_thread *thr;
14215
14216	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14217	mtx_lock(&thr->queue_lock);
14218	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14219	mtx_unlock(&thr->queue_lock);
14220	wakeup(thr);
14221}
14222
14223static void
14224ctl_enqueue_isc(union ctl_io *io)
14225{
14226	struct ctl_softc *softc = control_softc;
14227	struct ctl_thread *thr;
14228
14229	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14230	mtx_lock(&thr->queue_lock);
14231	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14232	mtx_unlock(&thr->queue_lock);
14233	wakeup(thr);
14234}
14235
14236/* Initialization and failover */
14237
14238void
14239ctl_init_isc_msg(void)
14240{
14241	printf("CTL: Still calling this thing\n");
14242}
14243
14244/*
14245 * Init component
14246 * 	Initializes component into configuration defined by bootMode
14247 *	(see hasc-sv.c)
14248 *  	returns hasc_Status:
14249 * 		OK
14250 *		ERROR - fatal error
14251 */
14252static ctl_ha_comp_status
14253ctl_isc_init(struct ctl_ha_component *c)
14254{
14255	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14256
14257	c->status = ret;
14258	return ret;
14259}
14260
14261/* Start component
14262 * 	Starts component in state requested. If component starts successfully,
14263 *	it must set its own state to the requestrd state
14264 *	When requested state is HASC_STATE_HA, the component may refine it
14265 * 	by adding _SLAVE or _MASTER flags.
14266 *	Currently allowed state transitions are:
14267 *	UNKNOWN->HA		- initial startup
14268 *	UNKNOWN->SINGLE - initial startup when no parter detected
14269 *	HA->SINGLE		- failover
14270 * returns ctl_ha_comp_status:
14271 * 		OK	- component successfully started in requested state
14272 *		FAILED  - could not start the requested state, failover may
14273 * 			  be possible
14274 *		ERROR	- fatal error detected, no future startup possible
14275 */
14276static ctl_ha_comp_status
14277ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14278{
14279	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14280
14281	printf("%s: go\n", __func__);
14282
14283	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14284	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14285		ctl_is_single = 0;
14286		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14287		    != CTL_HA_STATUS_SUCCESS) {
14288			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14289			ret = CTL_HA_COMP_STATUS_ERROR;
14290		}
14291	} else if (CTL_HA_STATE_IS_HA(c->state)
14292		&& CTL_HA_STATE_IS_SINGLE(state)){
14293		// HA->SINGLE transition
14294	        ctl_failover();
14295		ctl_is_single = 1;
14296	} else {
14297		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14298		       c->state, state);
14299		ret = CTL_HA_COMP_STATUS_ERROR;
14300	}
14301	if (CTL_HA_STATE_IS_SINGLE(state))
14302		ctl_is_single = 1;
14303
14304	c->state = state;
14305	c->status = ret;
14306	return ret;
14307}
14308
14309/*
14310 * Quiesce component
14311 * The component must clear any error conditions (set status to OK) and
14312 * prepare itself to another Start call
14313 * returns ctl_ha_comp_status:
14314 * 	OK
14315 *	ERROR
14316 */
14317static ctl_ha_comp_status
14318ctl_isc_quiesce(struct ctl_ha_component *c)
14319{
14320	int ret = CTL_HA_COMP_STATUS_OK;
14321
14322	ctl_pause_rtr = 1;
14323	c->status = ret;
14324	return ret;
14325}
14326
14327struct ctl_ha_component ctl_ha_component_ctlisc =
14328{
14329	.name = "CTL ISC",
14330	.state = CTL_HA_STATE_UNKNOWN,
14331	.init = ctl_isc_init,
14332	.start = ctl_isc_start,
14333	.quiesce = ctl_isc_quiesce
14334};
14335
14336/*
14337 *  vim: ts=8
14338 */
14339