ctl.c revision 271940
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 271940 2014-09-22 00:40:20Z 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);
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);
648				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
649				io->io_hdr.flags |=
650					entry->flags & CTL_FLAG_DATA_MASK;
651			}
652			ctl_enqueue_isc(io);
653			break;
654
655		/* Performed on the Originating SC, XFER mode only */
656		case CTL_MSG_DATAMOVE: {
657			struct ctl_sg_entry *sgl;
658			int i, j;
659
660			io = msg_info.hdr.original_sc;
661			if (io == NULL) {
662				printf("%s: original_sc == NULL!\n", __func__);
663				/* XXX KDM do something here */
664				break;
665			}
666			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
667			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
668			/*
669			 * Keep track of this, we need to send it back over
670			 * when the datamove is complete.
671			 */
672			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
673
674			if (msg_info.dt.sg_sequence == 0) {
675				/*
676				 * XXX KDM we use the preallocated S/G list
677				 * here, but we'll need to change this to
678				 * dynamic allocation if we need larger S/G
679				 * lists.
680				 */
681				if (msg_info.dt.kern_sg_entries >
682				    sizeof(io->io_hdr.remote_sglist) /
683				    sizeof(io->io_hdr.remote_sglist[0])) {
684					printf("%s: number of S/G entries "
685					    "needed %u > allocated num %zd\n",
686					    __func__,
687					    msg_info.dt.kern_sg_entries,
688					    sizeof(io->io_hdr.remote_sglist)/
689					    sizeof(io->io_hdr.remote_sglist[0]));
690
691					/*
692					 * XXX KDM send a message back to
693					 * the other side to shut down the
694					 * DMA.  The error will come back
695					 * through via the normal channel.
696					 */
697					break;
698				}
699				sgl = io->io_hdr.remote_sglist;
700				memset(sgl, 0,
701				       sizeof(io->io_hdr.remote_sglist));
702
703				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
704
705				io->scsiio.kern_sg_entries =
706					msg_info.dt.kern_sg_entries;
707				io->scsiio.rem_sg_entries =
708					msg_info.dt.kern_sg_entries;
709				io->scsiio.kern_data_len =
710					msg_info.dt.kern_data_len;
711				io->scsiio.kern_total_len =
712					msg_info.dt.kern_total_len;
713				io->scsiio.kern_data_resid =
714					msg_info.dt.kern_data_resid;
715				io->scsiio.kern_rel_offset =
716					msg_info.dt.kern_rel_offset;
717				/*
718				 * Clear out per-DMA flags.
719				 */
720				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
721				/*
722				 * Add per-DMA flags that are set for this
723				 * particular DMA request.
724				 */
725				io->io_hdr.flags |= msg_info.dt.flags &
726						    CTL_FLAG_RDMA_MASK;
727			} else
728				sgl = (struct ctl_sg_entry *)
729					io->scsiio.kern_data_ptr;
730
731			for (i = msg_info.dt.sent_sg_entries, j = 0;
732			     i < (msg_info.dt.sent_sg_entries +
733			     msg_info.dt.cur_sg_entries); i++, j++) {
734				sgl[i].addr = msg_info.dt.sg_list[j].addr;
735				sgl[i].len = msg_info.dt.sg_list[j].len;
736
737#if 0
738				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
739				       __func__,
740				       msg_info.dt.sg_list[j].addr,
741				       msg_info.dt.sg_list[j].len,
742				       sgl[i].addr, sgl[i].len, j, i);
743#endif
744			}
745#if 0
746			memcpy(&sgl[msg_info.dt.sent_sg_entries],
747			       msg_info.dt.sg_list,
748			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
749#endif
750
751			/*
752			 * If this is the last piece of the I/O, we've got
753			 * the full S/G list.  Queue processing in the thread.
754			 * Otherwise wait for the next piece.
755			 */
756			if (msg_info.dt.sg_last != 0)
757				ctl_enqueue_isc(io);
758			break;
759		}
760		/* Performed on the Serializing (primary) SC, XFER mode only */
761		case CTL_MSG_DATAMOVE_DONE: {
762			if (msg_info.hdr.serializing_sc == NULL) {
763				printf("%s: serializing_sc == NULL!\n",
764				       __func__);
765				/* XXX KDM now what? */
766				break;
767			}
768			/*
769			 * We grab the sense information here in case
770			 * there was a failure, so we can return status
771			 * back to the initiator.
772			 */
773			io = msg_info.hdr.serializing_sc;
774			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
775			io->io_hdr.status = msg_info.hdr.status;
776			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
777			io->scsiio.sense_len = msg_info.scsi.sense_len;
778			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
779			io->io_hdr.port_status = msg_info.scsi.fetd_status;
780			io->scsiio.residual = msg_info.scsi.residual;
781			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
782			       sizeof(io->scsiio.sense_data));
783			ctl_enqueue_isc(io);
784			break;
785		}
786
787		/* Preformed on Originating SC, SER_ONLY mode */
788		case CTL_MSG_R2R:
789			io = msg_info.hdr.original_sc;
790			if (io == NULL) {
791				printf("%s: Major Bummer\n", __func__);
792				return;
793			} else {
794#if 0
795				printf("pOrig %x\n",(int) ctsio);
796#endif
797			}
798			io->io_hdr.msg_type = CTL_MSG_R2R;
799			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
800			ctl_enqueue_isc(io);
801			break;
802
803		/*
804		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
805		 * mode.
806		 * Performed on the Originating (i.e. secondary) SC in XFER
807		 * mode
808		 */
809		case CTL_MSG_FINISH_IO:
810			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
811				ctl_isc_handler_finish_xfer(ctl_softc,
812							    &msg_info);
813			else
814				ctl_isc_handler_finish_ser_only(ctl_softc,
815								&msg_info);
816			break;
817
818		/* Preformed on Originating SC */
819		case CTL_MSG_BAD_JUJU:
820			io = msg_info.hdr.original_sc;
821			if (io == NULL) {
822				printf("%s: Bad JUJU!, original_sc is NULL!\n",
823				       __func__);
824				break;
825			}
826			ctl_copy_sense_data(&msg_info, io);
827			/*
828			 * IO should have already been cleaned up on other
829			 * SC so clear this flag so we won't send a message
830			 * back to finish the IO there.
831			 */
832			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
833			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
834
835			/* io = msg_info.hdr.serializing_sc; */
836			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
837			ctl_enqueue_isc(io);
838			break;
839
840		/* Handle resets sent from the other side */
841		case CTL_MSG_MANAGE_TASKS: {
842			struct ctl_taskio *taskio;
843			taskio = (struct ctl_taskio *)ctl_alloc_io(
844				(void *)ctl_softc->othersc_pool);
845			if (taskio == NULL) {
846				printf("ctl_isc_event_handler: can't allocate "
847				       "ctl_io!\n");
848				/* Bad Juju */
849				/* should I just call the proper reset func
850				   here??? */
851				goto bailout;
852			}
853			ctl_zero_io((union ctl_io *)taskio);
854			taskio->io_hdr.io_type = CTL_IO_TASK;
855			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
856			taskio->io_hdr.nexus = msg_info.hdr.nexus;
857			taskio->task_action = msg_info.task.task_action;
858			taskio->tag_num = msg_info.task.tag_num;
859			taskio->tag_type = msg_info.task.tag_type;
860#ifdef CTL_TIME_IO
861			taskio->io_hdr.start_time = time_uptime;
862			getbintime(&taskio->io_hdr.start_bt);
863#if 0
864			cs_prof_gettime(&taskio->io_hdr.start_ticks);
865#endif
866#endif /* CTL_TIME_IO */
867			ctl_run_task((union ctl_io *)taskio);
868			break;
869		}
870		/* Persistent Reserve action which needs attention */
871		case CTL_MSG_PERS_ACTION:
872			presio = (struct ctl_prio *)ctl_alloc_io(
873				(void *)ctl_softc->othersc_pool);
874			if (presio == NULL) {
875				printf("ctl_isc_event_handler: can't allocate "
876				       "ctl_io!\n");
877				/* Bad Juju */
878				/* Need to set busy and send msg back */
879				goto bailout;
880			}
881			ctl_zero_io((union ctl_io *)presio);
882			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
883			presio->pr_msg = msg_info.pr;
884			ctl_enqueue_isc((union ctl_io *)presio);
885			break;
886		case CTL_MSG_SYNC_FE:
887			rcv_sync_msg = 1;
888			break;
889		case CTL_MSG_APS_LOCK: {
890			// It's quicker to execute this then to
891			// queue it.
892			struct ctl_lun *lun;
893			struct ctl_page_index *page_index;
894			struct copan_aps_subpage *current_sp;
895			uint32_t targ_lun;
896
897			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
898			lun = ctl_softc->ctl_luns[targ_lun];
899			mtx_lock(&lun->lun_lock);
900			page_index = &lun->mode_pages.index[index_to_aps_page];
901			current_sp = (struct copan_aps_subpage *)
902				     (page_index->page_data +
903				     (page_index->page_len * CTL_PAGE_CURRENT));
904
905			current_sp->lock_active = msg_info.aps.lock_flag;
906			mtx_unlock(&lun->lun_lock);
907		        break;
908		}
909		default:
910		        printf("How did I get here?\n");
911		}
912	} else if (event == CTL_HA_EVT_MSG_SENT) {
913		if (param != CTL_HA_STATUS_SUCCESS) {
914			printf("Bad status from ctl_ha_msg_send status %d\n",
915			       param);
916		}
917		return;
918	} else if (event == CTL_HA_EVT_DISCONNECT) {
919		printf("CTL: Got a disconnect from Isc\n");
920		return;
921	} else {
922		printf("ctl_isc_event_handler: Unknown event %d\n", event);
923		return;
924	}
925
926bailout:
927	return;
928}
929
930static void
931ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
932{
933	struct scsi_sense_data *sense;
934
935	sense = &dest->scsiio.sense_data;
936	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
937	dest->scsiio.scsi_status = src->scsi.scsi_status;
938	dest->scsiio.sense_len = src->scsi.sense_len;
939	dest->io_hdr.status = src->hdr.status;
940}
941
942static int
943ctl_init(void)
944{
945	struct ctl_softc *softc;
946	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
947	struct ctl_port *port;
948        uint8_t sc_id =0;
949	int i, error, retval;
950	//int isc_retval;
951
952	retval = 0;
953	ctl_pause_rtr = 0;
954        rcv_sync_msg = 0;
955
956	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
957			       M_WAITOK | M_ZERO);
958	softc = control_softc;
959
960	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
961			      "cam/ctl");
962
963	softc->dev->si_drv1 = softc;
964
965	/*
966	 * By default, return a "bad LUN" peripheral qualifier for unknown
967	 * LUNs.  The user can override this default using the tunable or
968	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
969	 */
970	softc->inquiry_pq_no_lun = 1;
971	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
972			  &softc->inquiry_pq_no_lun);
973	sysctl_ctx_init(&softc->sysctl_ctx);
974	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
975		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
976		CTLFLAG_RD, 0, "CAM Target Layer");
977
978	if (softc->sysctl_tree == NULL) {
979		printf("%s: unable to allocate sysctl tree\n", __func__);
980		destroy_dev(softc->dev);
981		free(control_softc, M_DEVBUF);
982		control_softc = NULL;
983		return (ENOMEM);
984	}
985
986	SYSCTL_ADD_INT(&softc->sysctl_ctx,
987		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
988		       "inquiry_pq_no_lun", CTLFLAG_RW,
989		       &softc->inquiry_pq_no_lun, 0,
990		       "Report no lun possible for invalid LUNs");
991
992	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
993	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
994	softc->open_count = 0;
995
996	/*
997	 * Default to actually sending a SYNCHRONIZE CACHE command down to
998	 * the drive.
999	 */
1000	softc->flags = CTL_FLAG_REAL_SYNC;
1001
1002	/*
1003	 * In Copan's HA scheme, the "master" and "slave" roles are
1004	 * figured out through the slot the controller is in.  Although it
1005	 * is an active/active system, someone has to be in charge.
1006 	 */
1007#ifdef NEEDTOPORT
1008        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1009#endif
1010
1011        if (sc_id == 0) {
1012		softc->flags |= CTL_FLAG_MASTER_SHELF;
1013		persis_offset = 0;
1014	} else
1015		persis_offset = CTL_MAX_INITIATORS;
1016
1017	/*
1018	 * XXX KDM need to figure out where we want to get our target ID
1019	 * and WWID.  Is it different on each port?
1020	 */
1021	softc->target.id = 0;
1022	softc->target.wwid[0] = 0x12345678;
1023	softc->target.wwid[1] = 0x87654321;
1024	STAILQ_INIT(&softc->lun_list);
1025	STAILQ_INIT(&softc->pending_lun_queue);
1026	STAILQ_INIT(&softc->fe_list);
1027	STAILQ_INIT(&softc->port_list);
1028	STAILQ_INIT(&softc->be_list);
1029	STAILQ_INIT(&softc->io_pools);
1030	ctl_tpc_init(softc);
1031
1032	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1033			    &internal_pool)!= 0){
1034		printf("ctl: can't allocate %d entry internal pool, "
1035		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1036		return (ENOMEM);
1037	}
1038
1039	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1040			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1041		printf("ctl: can't allocate %d entry emergency pool, "
1042		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1043		ctl_pool_free(internal_pool);
1044		return (ENOMEM);
1045	}
1046
1047	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1048	                    &other_pool) != 0)
1049	{
1050		printf("ctl: can't allocate %d entry other SC pool, "
1051		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1052		ctl_pool_free(internal_pool);
1053		ctl_pool_free(emergency_pool);
1054		return (ENOMEM);
1055	}
1056
1057	softc->internal_pool = internal_pool;
1058	softc->emergency_pool = emergency_pool;
1059	softc->othersc_pool = other_pool;
1060
1061	if (worker_threads <= 0)
1062		worker_threads = max(1, mp_ncpus / 4);
1063	if (worker_threads > CTL_MAX_THREADS)
1064		worker_threads = CTL_MAX_THREADS;
1065
1066	for (i = 0; i < worker_threads; i++) {
1067		struct ctl_thread *thr = &softc->threads[i];
1068
1069		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1070		thr->ctl_softc = softc;
1071		STAILQ_INIT(&thr->incoming_queue);
1072		STAILQ_INIT(&thr->rtr_queue);
1073		STAILQ_INIT(&thr->done_queue);
1074		STAILQ_INIT(&thr->isc_queue);
1075
1076		error = kproc_kthread_add(ctl_work_thread, thr,
1077		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1078		if (error != 0) {
1079			printf("error creating CTL work thread!\n");
1080			ctl_pool_free(internal_pool);
1081			ctl_pool_free(emergency_pool);
1082			ctl_pool_free(other_pool);
1083			return (error);
1084		}
1085	}
1086	error = kproc_kthread_add(ctl_lun_thread, softc,
1087	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1088	if (error != 0) {
1089		printf("error creating CTL lun thread!\n");
1090		ctl_pool_free(internal_pool);
1091		ctl_pool_free(emergency_pool);
1092		ctl_pool_free(other_pool);
1093		return (error);
1094	}
1095	if (bootverbose)
1096		printf("ctl: CAM Target Layer loaded\n");
1097
1098	/*
1099	 * Initialize the ioctl front end.
1100	 */
1101	ctl_frontend_register(&ioctl_frontend);
1102	port = &softc->ioctl_info.port;
1103	port->frontend = &ioctl_frontend;
1104	sprintf(softc->ioctl_info.port_name, "ioctl");
1105	port->port_type = CTL_PORT_IOCTL;
1106	port->num_requested_ctl_io = 100;
1107	port->port_name = softc->ioctl_info.port_name;
1108	port->port_online = ctl_ioctl_online;
1109	port->port_offline = ctl_ioctl_offline;
1110	port->onoff_arg = &softc->ioctl_info;
1111	port->lun_enable = ctl_ioctl_lun_enable;
1112	port->lun_disable = ctl_ioctl_lun_disable;
1113	port->targ_lun_arg = &softc->ioctl_info;
1114	port->fe_datamove = ctl_ioctl_datamove;
1115	port->fe_done = ctl_ioctl_done;
1116	port->max_targets = 15;
1117	port->max_target_id = 15;
1118
1119	if (ctl_port_register(&softc->ioctl_info.port,
1120	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1121		printf("ctl: ioctl front end registration failed, will "
1122		       "continue anyway\n");
1123	}
1124
1125#ifdef CTL_IO_DELAY
1126	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1127		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1128		       sizeof(struct callout), CTL_TIMER_BYTES);
1129		return (EINVAL);
1130	}
1131#endif /* CTL_IO_DELAY */
1132
1133	return (0);
1134}
1135
1136void
1137ctl_shutdown(void)
1138{
1139	struct ctl_softc *softc;
1140	struct ctl_lun *lun, *next_lun;
1141	struct ctl_io_pool *pool;
1142
1143	softc = (struct ctl_softc *)control_softc;
1144
1145	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1146		printf("ctl: ioctl front end deregistration failed\n");
1147
1148	mtx_lock(&softc->ctl_lock);
1149
1150	/*
1151	 * Free up each LUN.
1152	 */
1153	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1154		next_lun = STAILQ_NEXT(lun, links);
1155		ctl_free_lun(lun);
1156	}
1157
1158	mtx_unlock(&softc->ctl_lock);
1159
1160	ctl_frontend_deregister(&ioctl_frontend);
1161
1162	/*
1163	 * This will rip the rug out from under any FETDs or anyone else
1164	 * that has a pool allocated.  Since we increment our module
1165	 * refcount any time someone outside the main CTL module allocates
1166	 * a pool, we shouldn't have any problems here.  The user won't be
1167	 * able to unload the CTL module until client modules have
1168	 * successfully unloaded.
1169	 */
1170	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1171		ctl_pool_free(pool);
1172
1173#if 0
1174	ctl_shutdown_thread(softc->work_thread);
1175	mtx_destroy(&softc->queue_lock);
1176#endif
1177
1178	ctl_tpc_shutdown(softc);
1179	mtx_destroy(&softc->pool_lock);
1180	mtx_destroy(&softc->ctl_lock);
1181
1182	destroy_dev(softc->dev);
1183
1184	sysctl_ctx_free(&softc->sysctl_ctx);
1185
1186	free(control_softc, M_DEVBUF);
1187	control_softc = NULL;
1188
1189	if (bootverbose)
1190		printf("ctl: CAM Target Layer unloaded\n");
1191}
1192
1193static int
1194ctl_module_event_handler(module_t mod, int what, void *arg)
1195{
1196
1197	switch (what) {
1198	case MOD_LOAD:
1199		return (ctl_init());
1200	case MOD_UNLOAD:
1201		return (EBUSY);
1202	default:
1203		return (EOPNOTSUPP);
1204	}
1205}
1206
1207/*
1208 * XXX KDM should we do some access checks here?  Bump a reference count to
1209 * prevent a CTL module from being unloaded while someone has it open?
1210 */
1211static int
1212ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1213{
1214	return (0);
1215}
1216
1217static int
1218ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1219{
1220	return (0);
1221}
1222
1223int
1224ctl_port_enable(ctl_port_type port_type)
1225{
1226	struct ctl_softc *softc;
1227	struct ctl_port *port;
1228
1229	if (ctl_is_single == 0) {
1230		union ctl_ha_msg msg_info;
1231		int isc_retval;
1232
1233#if 0
1234		printf("%s: HA mode, synchronizing frontend enable\n",
1235		        __func__);
1236#endif
1237		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1238	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1239		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1240			printf("Sync msg send error retval %d\n", isc_retval);
1241		}
1242		if (!rcv_sync_msg) {
1243			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1244			        sizeof(msg_info), 1);
1245		}
1246#if 0
1247        	printf("CTL:Frontend Enable\n");
1248	} else {
1249		printf("%s: single mode, skipping frontend synchronization\n",
1250		        __func__);
1251#endif
1252	}
1253
1254	softc = control_softc;
1255
1256	STAILQ_FOREACH(port, &softc->port_list, links) {
1257		if (port_type & port->port_type)
1258		{
1259#if 0
1260			printf("port %d\n", port->targ_port);
1261#endif
1262			ctl_port_online(port);
1263		}
1264	}
1265
1266	return (0);
1267}
1268
1269int
1270ctl_port_disable(ctl_port_type port_type)
1271{
1272	struct ctl_softc *softc;
1273	struct ctl_port *port;
1274
1275	softc = control_softc;
1276
1277	STAILQ_FOREACH(port, &softc->port_list, links) {
1278		if (port_type & port->port_type)
1279			ctl_port_offline(port);
1280	}
1281
1282	return (0);
1283}
1284
1285/*
1286 * Returns 0 for success, 1 for failure.
1287 * Currently the only failure mode is if there aren't enough entries
1288 * allocated.  So, in case of a failure, look at num_entries_dropped,
1289 * reallocate and try again.
1290 */
1291int
1292ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1293	      int *num_entries_filled, int *num_entries_dropped,
1294	      ctl_port_type port_type, int no_virtual)
1295{
1296	struct ctl_softc *softc;
1297	struct ctl_port *port;
1298	int entries_dropped, entries_filled;
1299	int retval;
1300	int i;
1301
1302	softc = control_softc;
1303
1304	retval = 0;
1305	entries_filled = 0;
1306	entries_dropped = 0;
1307
1308	i = 0;
1309	mtx_lock(&softc->ctl_lock);
1310	STAILQ_FOREACH(port, &softc->port_list, links) {
1311		struct ctl_port_entry *entry;
1312
1313		if ((port->port_type & port_type) == 0)
1314			continue;
1315
1316		if ((no_virtual != 0)
1317		 && (port->virtual_port != 0))
1318			continue;
1319
1320		if (entries_filled >= num_entries_alloced) {
1321			entries_dropped++;
1322			continue;
1323		}
1324		entry = &entries[i];
1325
1326		entry->port_type = port->port_type;
1327		strlcpy(entry->port_name, port->port_name,
1328			sizeof(entry->port_name));
1329		entry->physical_port = port->physical_port;
1330		entry->virtual_port = port->virtual_port;
1331		entry->wwnn = port->wwnn;
1332		entry->wwpn = port->wwpn;
1333
1334		i++;
1335		entries_filled++;
1336	}
1337
1338	mtx_unlock(&softc->ctl_lock);
1339
1340	if (entries_dropped > 0)
1341		retval = 1;
1342
1343	*num_entries_dropped = entries_dropped;
1344	*num_entries_filled = entries_filled;
1345
1346	return (retval);
1347}
1348
1349static void
1350ctl_ioctl_online(void *arg)
1351{
1352	struct ctl_ioctl_info *ioctl_info;
1353
1354	ioctl_info = (struct ctl_ioctl_info *)arg;
1355
1356	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1357}
1358
1359static void
1360ctl_ioctl_offline(void *arg)
1361{
1362	struct ctl_ioctl_info *ioctl_info;
1363
1364	ioctl_info = (struct ctl_ioctl_info *)arg;
1365
1366	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1367}
1368
1369/*
1370 * Remove an initiator by port number and initiator ID.
1371 * Returns 0 for success, -1 for failure.
1372 */
1373int
1374ctl_remove_initiator(struct ctl_port *port, int iid)
1375{
1376	struct ctl_softc *softc = control_softc;
1377
1378	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1379
1380	if (iid > CTL_MAX_INIT_PER_PORT) {
1381		printf("%s: initiator ID %u > maximun %u!\n",
1382		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1383		return (-1);
1384	}
1385
1386	mtx_lock(&softc->ctl_lock);
1387	port->wwpn_iid[iid].in_use--;
1388	port->wwpn_iid[iid].last_use = time_uptime;
1389	mtx_unlock(&softc->ctl_lock);
1390
1391	return (0);
1392}
1393
1394/*
1395 * Add an initiator to the initiator map.
1396 * Returns iid for success, < 0 for failure.
1397 */
1398int
1399ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1400{
1401	struct ctl_softc *softc = control_softc;
1402	time_t best_time;
1403	int i, best;
1404
1405	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1406
1407	if (iid >= CTL_MAX_INIT_PER_PORT) {
1408		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1409		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1410		free(name, M_CTL);
1411		return (-1);
1412	}
1413
1414	mtx_lock(&softc->ctl_lock);
1415
1416	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1417		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1418			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1419				iid = i;
1420				break;
1421			}
1422			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1423			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1424				iid = i;
1425				break;
1426			}
1427		}
1428	}
1429
1430	if (iid < 0) {
1431		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1432			if (port->wwpn_iid[i].in_use == 0 &&
1433			    port->wwpn_iid[i].wwpn == 0 &&
1434			    port->wwpn_iid[i].name == NULL) {
1435				iid = i;
1436				break;
1437			}
1438		}
1439	}
1440
1441	if (iid < 0) {
1442		best = -1;
1443		best_time = INT32_MAX;
1444		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1445			if (port->wwpn_iid[i].in_use == 0) {
1446				if (port->wwpn_iid[i].last_use < best_time) {
1447					best = i;
1448					best_time = port->wwpn_iid[i].last_use;
1449				}
1450			}
1451		}
1452		iid = best;
1453	}
1454
1455	if (iid < 0) {
1456		mtx_unlock(&softc->ctl_lock);
1457		free(name, M_CTL);
1458		return (-2);
1459	}
1460
1461	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1462		/*
1463		 * This is not an error yet.
1464		 */
1465		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1466#if 0
1467			printf("%s: port %d iid %u WWPN %#jx arrived"
1468			    " again\n", __func__, port->targ_port,
1469			    iid, (uintmax_t)wwpn);
1470#endif
1471			goto take;
1472		}
1473		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1474		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1475#if 0
1476			printf("%s: port %d iid %u name '%s' arrived"
1477			    " again\n", __func__, port->targ_port,
1478			    iid, name);
1479#endif
1480			goto take;
1481		}
1482
1483		/*
1484		 * This is an error, but what do we do about it?  The
1485		 * driver is telling us we have a new WWPN for this
1486		 * initiator ID, so we pretty much need to use it.
1487		 */
1488		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1489		    " but WWPN %#jx '%s' is still at that address\n",
1490		    __func__, port->targ_port, iid, wwpn, name,
1491		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1492		    port->wwpn_iid[iid].name);
1493
1494		/*
1495		 * XXX KDM clear have_ca and ua_pending on each LUN for
1496		 * this initiator.
1497		 */
1498	}
1499take:
1500	free(port->wwpn_iid[iid].name, M_CTL);
1501	port->wwpn_iid[iid].name = name;
1502	port->wwpn_iid[iid].wwpn = wwpn;
1503	port->wwpn_iid[iid].in_use++;
1504	mtx_unlock(&softc->ctl_lock);
1505
1506	return (iid);
1507}
1508
1509static int
1510ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1511{
1512	int len;
1513
1514	switch (port->port_type) {
1515	case CTL_PORT_FC:
1516	{
1517		struct scsi_transportid_fcp *id =
1518		    (struct scsi_transportid_fcp *)buf;
1519		if (port->wwpn_iid[iid].wwpn == 0)
1520			return (0);
1521		memset(id, 0, sizeof(*id));
1522		id->format_protocol = SCSI_PROTO_FC;
1523		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1524		return (sizeof(*id));
1525	}
1526	case CTL_PORT_ISCSI:
1527	{
1528		struct scsi_transportid_iscsi_port *id =
1529		    (struct scsi_transportid_iscsi_port *)buf;
1530		if (port->wwpn_iid[iid].name == NULL)
1531			return (0);
1532		memset(id, 0, 256);
1533		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1534		    SCSI_PROTO_ISCSI;
1535		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1536		len = roundup2(min(len, 252), 4);
1537		scsi_ulto2b(len, id->additional_length);
1538		return (sizeof(*id) + len);
1539	}
1540	case CTL_PORT_SAS:
1541	{
1542		struct scsi_transportid_sas *id =
1543		    (struct scsi_transportid_sas *)buf;
1544		if (port->wwpn_iid[iid].wwpn == 0)
1545			return (0);
1546		memset(id, 0, sizeof(*id));
1547		id->format_protocol = SCSI_PROTO_SAS;
1548		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1549		return (sizeof(*id));
1550	}
1551	default:
1552	{
1553		struct scsi_transportid_spi *id =
1554		    (struct scsi_transportid_spi *)buf;
1555		memset(id, 0, sizeof(*id));
1556		id->format_protocol = SCSI_PROTO_SPI;
1557		scsi_ulto2b(iid, id->scsi_addr);
1558		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1559		return (sizeof(*id));
1560	}
1561	}
1562}
1563
1564static int
1565ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1566{
1567	return (0);
1568}
1569
1570static int
1571ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1572{
1573	return (0);
1574}
1575
1576/*
1577 * Data movement routine for the CTL ioctl frontend port.
1578 */
1579static int
1580ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1581{
1582	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1583	struct ctl_sg_entry ext_entry, kern_entry;
1584	int ext_sglen, ext_sg_entries, kern_sg_entries;
1585	int ext_sg_start, ext_offset;
1586	int len_to_copy, len_copied;
1587	int kern_watermark, ext_watermark;
1588	int ext_sglist_malloced;
1589	int i, j;
1590
1591	ext_sglist_malloced = 0;
1592	ext_sg_start = 0;
1593	ext_offset = 0;
1594
1595	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1596
1597	/*
1598	 * If this flag is set, fake the data transfer.
1599	 */
1600	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1601		ctsio->ext_data_filled = ctsio->ext_data_len;
1602		goto bailout;
1603	}
1604
1605	/*
1606	 * To simplify things here, if we have a single buffer, stick it in
1607	 * a S/G entry and just make it a single entry S/G list.
1608	 */
1609	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1610		int len_seen;
1611
1612		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1613
1614		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1615							   M_WAITOK);
1616		ext_sglist_malloced = 1;
1617		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1618				   ext_sglen) != 0) {
1619			ctl_set_internal_failure(ctsio,
1620						 /*sks_valid*/ 0,
1621						 /*retry_count*/ 0);
1622			goto bailout;
1623		}
1624		ext_sg_entries = ctsio->ext_sg_entries;
1625		len_seen = 0;
1626		for (i = 0; i < ext_sg_entries; i++) {
1627			if ((len_seen + ext_sglist[i].len) >=
1628			     ctsio->ext_data_filled) {
1629				ext_sg_start = i;
1630				ext_offset = ctsio->ext_data_filled - len_seen;
1631				break;
1632			}
1633			len_seen += ext_sglist[i].len;
1634		}
1635	} else {
1636		ext_sglist = &ext_entry;
1637		ext_sglist->addr = ctsio->ext_data_ptr;
1638		ext_sglist->len = ctsio->ext_data_len;
1639		ext_sg_entries = 1;
1640		ext_sg_start = 0;
1641		ext_offset = ctsio->ext_data_filled;
1642	}
1643
1644	if (ctsio->kern_sg_entries > 0) {
1645		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1646		kern_sg_entries = ctsio->kern_sg_entries;
1647	} else {
1648		kern_sglist = &kern_entry;
1649		kern_sglist->addr = ctsio->kern_data_ptr;
1650		kern_sglist->len = ctsio->kern_data_len;
1651		kern_sg_entries = 1;
1652	}
1653
1654
1655	kern_watermark = 0;
1656	ext_watermark = ext_offset;
1657	len_copied = 0;
1658	for (i = ext_sg_start, j = 0;
1659	     i < ext_sg_entries && j < kern_sg_entries;) {
1660		uint8_t *ext_ptr, *kern_ptr;
1661
1662		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1663				      kern_sglist[j].len - kern_watermark);
1664
1665		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1666		ext_ptr = ext_ptr + ext_watermark;
1667		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1668			/*
1669			 * XXX KDM fix this!
1670			 */
1671			panic("need to implement bus address support");
1672#if 0
1673			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1674#endif
1675		} else
1676			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1677		kern_ptr = kern_ptr + kern_watermark;
1678
1679		kern_watermark += len_to_copy;
1680		ext_watermark += len_to_copy;
1681
1682		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1683		     CTL_FLAG_DATA_IN) {
1684			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1685					 "bytes to user\n", len_to_copy));
1686			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1687					 "to %p\n", kern_ptr, ext_ptr));
1688			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1689				ctl_set_internal_failure(ctsio,
1690							 /*sks_valid*/ 0,
1691							 /*retry_count*/ 0);
1692				goto bailout;
1693			}
1694		} else {
1695			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1696					 "bytes from user\n", len_to_copy));
1697			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1698					 "to %p\n", ext_ptr, kern_ptr));
1699			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1700				ctl_set_internal_failure(ctsio,
1701							 /*sks_valid*/ 0,
1702							 /*retry_count*/0);
1703				goto bailout;
1704			}
1705		}
1706
1707		len_copied += len_to_copy;
1708
1709		if (ext_sglist[i].len == ext_watermark) {
1710			i++;
1711			ext_watermark = 0;
1712		}
1713
1714		if (kern_sglist[j].len == kern_watermark) {
1715			j++;
1716			kern_watermark = 0;
1717		}
1718	}
1719
1720	ctsio->ext_data_filled += len_copied;
1721
1722	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1723			 "kern_sg_entries: %d\n", ext_sg_entries,
1724			 kern_sg_entries));
1725	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1726			 "kern_data_len = %d\n", ctsio->ext_data_len,
1727			 ctsio->kern_data_len));
1728
1729
1730	/* XXX KDM set residual?? */
1731bailout:
1732
1733	if (ext_sglist_malloced != 0)
1734		free(ext_sglist, M_CTL);
1735
1736	return (CTL_RETVAL_COMPLETE);
1737}
1738
1739/*
1740 * Serialize a command that went down the "wrong" side, and so was sent to
1741 * this controller for execution.  The logic is a little different than the
1742 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1743 * sent back to the other side, but in the success case, we execute the
1744 * command on this side (XFER mode) or tell the other side to execute it
1745 * (SER_ONLY mode).
1746 */
1747static int
1748ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1749{
1750	struct ctl_softc *ctl_softc;
1751	union ctl_ha_msg msg_info;
1752	struct ctl_lun *lun;
1753	int retval = 0;
1754	uint32_t targ_lun;
1755
1756	ctl_softc = control_softc;
1757
1758	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1759	lun = ctl_softc->ctl_luns[targ_lun];
1760	if (lun==NULL)
1761	{
1762		/*
1763		 * Why isn't LUN defined? The other side wouldn't
1764		 * send a cmd if the LUN is undefined.
1765		 */
1766		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1767
1768		/* "Logical unit not supported" */
1769		ctl_set_sense_data(&msg_info.scsi.sense_data,
1770				   lun,
1771				   /*sense_format*/SSD_TYPE_NONE,
1772				   /*current_error*/ 1,
1773				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1774				   /*asc*/ 0x25,
1775				   /*ascq*/ 0x00,
1776				   SSD_ELEM_NONE);
1777
1778		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1779		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1780		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1781		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1782		msg_info.hdr.serializing_sc = NULL;
1783		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1784	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1785				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1786		}
1787		return(1);
1788
1789	}
1790
1791	mtx_lock(&lun->lun_lock);
1792    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1793
1794	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1795		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1796		 ooa_links))) {
1797	case CTL_ACTION_BLOCK:
1798		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1799		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1800				  blocked_links);
1801		break;
1802	case CTL_ACTION_PASS:
1803	case CTL_ACTION_SKIP:
1804		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1805			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1806			ctl_enqueue_rtr((union ctl_io *)ctsio);
1807		} else {
1808
1809			/* send msg back to other side */
1810			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1811			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1812			msg_info.hdr.msg_type = CTL_MSG_R2R;
1813#if 0
1814			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1815#endif
1816		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1817			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1818			}
1819		}
1820		break;
1821	case CTL_ACTION_OVERLAP:
1822		/* OVERLAPPED COMMANDS ATTEMPTED */
1823		ctl_set_sense_data(&msg_info.scsi.sense_data,
1824				   lun,
1825				   /*sense_format*/SSD_TYPE_NONE,
1826				   /*current_error*/ 1,
1827				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1828				   /*asc*/ 0x4E,
1829				   /*ascq*/ 0x00,
1830				   SSD_ELEM_NONE);
1831
1832		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1833		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1834		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1835		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1836		msg_info.hdr.serializing_sc = NULL;
1837		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1838#if 0
1839		printf("BAD JUJU:Major Bummer Overlap\n");
1840#endif
1841		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1842		retval = 1;
1843		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1844		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1845		}
1846		break;
1847	case CTL_ACTION_OVERLAP_TAG:
1848		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1849		ctl_set_sense_data(&msg_info.scsi.sense_data,
1850				   lun,
1851				   /*sense_format*/SSD_TYPE_NONE,
1852				   /*current_error*/ 1,
1853				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1854				   /*asc*/ 0x4D,
1855				   /*ascq*/ ctsio->tag_num & 0xff,
1856				   SSD_ELEM_NONE);
1857
1858		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1859		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1860		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1861		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1862		msg_info.hdr.serializing_sc = NULL;
1863		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1864#if 0
1865		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1866#endif
1867		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1868		retval = 1;
1869		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1870		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1871		}
1872		break;
1873	case CTL_ACTION_ERROR:
1874	default:
1875		/* "Internal target failure" */
1876		ctl_set_sense_data(&msg_info.scsi.sense_data,
1877				   lun,
1878				   /*sense_format*/SSD_TYPE_NONE,
1879				   /*current_error*/ 1,
1880				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1881				   /*asc*/ 0x44,
1882				   /*ascq*/ 0x00,
1883				   SSD_ELEM_NONE);
1884
1885		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1886		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1887		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1888		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1889		msg_info.hdr.serializing_sc = NULL;
1890		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1891#if 0
1892		printf("BAD JUJU:Major Bummer HW Error\n");
1893#endif
1894		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1895		retval = 1;
1896		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1897		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1898		}
1899		break;
1900	}
1901	mtx_unlock(&lun->lun_lock);
1902	return (retval);
1903}
1904
1905static int
1906ctl_ioctl_submit_wait(union ctl_io *io)
1907{
1908	struct ctl_fe_ioctl_params params;
1909	ctl_fe_ioctl_state last_state;
1910	int done, retval;
1911
1912	retval = 0;
1913
1914	bzero(&params, sizeof(params));
1915
1916	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1917	cv_init(&params.sem, "ctlioccv");
1918	params.state = CTL_IOCTL_INPROG;
1919	last_state = params.state;
1920
1921	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1922
1923	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1924
1925	/* This shouldn't happen */
1926	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1927		return (retval);
1928
1929	done = 0;
1930
1931	do {
1932		mtx_lock(&params.ioctl_mtx);
1933		/*
1934		 * Check the state here, and don't sleep if the state has
1935		 * already changed (i.e. wakeup has already occured, but we
1936		 * weren't waiting yet).
1937		 */
1938		if (params.state == last_state) {
1939			/* XXX KDM cv_wait_sig instead? */
1940			cv_wait(&params.sem, &params.ioctl_mtx);
1941		}
1942		last_state = params.state;
1943
1944		switch (params.state) {
1945		case CTL_IOCTL_INPROG:
1946			/* Why did we wake up? */
1947			/* XXX KDM error here? */
1948			mtx_unlock(&params.ioctl_mtx);
1949			break;
1950		case CTL_IOCTL_DATAMOVE:
1951			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1952
1953			/*
1954			 * change last_state back to INPROG to avoid
1955			 * deadlock on subsequent data moves.
1956			 */
1957			params.state = last_state = CTL_IOCTL_INPROG;
1958
1959			mtx_unlock(&params.ioctl_mtx);
1960			ctl_ioctl_do_datamove(&io->scsiio);
1961			/*
1962			 * Note that in some cases, most notably writes,
1963			 * this will queue the I/O and call us back later.
1964			 * In other cases, generally reads, this routine
1965			 * will immediately call back and wake us up,
1966			 * probably using our own context.
1967			 */
1968			io->scsiio.be_move_done(io);
1969			break;
1970		case CTL_IOCTL_DONE:
1971			mtx_unlock(&params.ioctl_mtx);
1972			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1973			done = 1;
1974			break;
1975		default:
1976			mtx_unlock(&params.ioctl_mtx);
1977			/* XXX KDM error here? */
1978			break;
1979		}
1980	} while (done == 0);
1981
1982	mtx_destroy(&params.ioctl_mtx);
1983	cv_destroy(&params.sem);
1984
1985	return (CTL_RETVAL_COMPLETE);
1986}
1987
1988static void
1989ctl_ioctl_datamove(union ctl_io *io)
1990{
1991	struct ctl_fe_ioctl_params *params;
1992
1993	params = (struct ctl_fe_ioctl_params *)
1994		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1995
1996	mtx_lock(&params->ioctl_mtx);
1997	params->state = CTL_IOCTL_DATAMOVE;
1998	cv_broadcast(&params->sem);
1999	mtx_unlock(&params->ioctl_mtx);
2000}
2001
2002static void
2003ctl_ioctl_done(union ctl_io *io)
2004{
2005	struct ctl_fe_ioctl_params *params;
2006
2007	params = (struct ctl_fe_ioctl_params *)
2008		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2009
2010	mtx_lock(&params->ioctl_mtx);
2011	params->state = CTL_IOCTL_DONE;
2012	cv_broadcast(&params->sem);
2013	mtx_unlock(&params->ioctl_mtx);
2014}
2015
2016static void
2017ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2018{
2019	struct ctl_fe_ioctl_startstop_info *sd_info;
2020
2021	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2022
2023	sd_info->hs_info.status = metatask->status;
2024	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2025	sd_info->hs_info.luns_complete =
2026		metatask->taskinfo.startstop.luns_complete;
2027	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2028
2029	cv_broadcast(&sd_info->sem);
2030}
2031
2032static void
2033ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2034{
2035	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2036
2037	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2038
2039	mtx_lock(fe_bbr_info->lock);
2040	fe_bbr_info->bbr_info->status = metatask->status;
2041	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2042	fe_bbr_info->wakeup_done = 1;
2043	mtx_unlock(fe_bbr_info->lock);
2044
2045	cv_broadcast(&fe_bbr_info->sem);
2046}
2047
2048/*
2049 * Returns 0 for success, errno for failure.
2050 */
2051static int
2052ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2053		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2054{
2055	union ctl_io *io;
2056	int retval;
2057
2058	retval = 0;
2059
2060	mtx_lock(&lun->lun_lock);
2061	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2062	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2063	     ooa_links)) {
2064		struct ctl_ooa_entry *entry;
2065
2066		/*
2067		 * If we've got more than we can fit, just count the
2068		 * remaining entries.
2069		 */
2070		if (*cur_fill_num >= ooa_hdr->alloc_num)
2071			continue;
2072
2073		entry = &kern_entries[*cur_fill_num];
2074
2075		entry->tag_num = io->scsiio.tag_num;
2076		entry->lun_num = lun->lun;
2077#ifdef CTL_TIME_IO
2078		entry->start_bt = io->io_hdr.start_bt;
2079#endif
2080		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2081		entry->cdb_len = io->scsiio.cdb_len;
2082		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2083			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2084
2085		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2086			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2087
2088		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2089			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2090
2091		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2092			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2093
2094		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2095			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2096	}
2097	mtx_unlock(&lun->lun_lock);
2098
2099	return (retval);
2100}
2101
2102static void *
2103ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2104		 size_t error_str_len)
2105{
2106	void *kptr;
2107
2108	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2109
2110	if (copyin(user_addr, kptr, len) != 0) {
2111		snprintf(error_str, error_str_len, "Error copying %d bytes "
2112			 "from user address %p to kernel address %p", len,
2113			 user_addr, kptr);
2114		free(kptr, M_CTL);
2115		return (NULL);
2116	}
2117
2118	return (kptr);
2119}
2120
2121static void
2122ctl_free_args(int num_args, struct ctl_be_arg *args)
2123{
2124	int i;
2125
2126	if (args == NULL)
2127		return;
2128
2129	for (i = 0; i < num_args; i++) {
2130		free(args[i].kname, M_CTL);
2131		free(args[i].kvalue, M_CTL);
2132	}
2133
2134	free(args, M_CTL);
2135}
2136
2137static struct ctl_be_arg *
2138ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2139		char *error_str, size_t error_str_len)
2140{
2141	struct ctl_be_arg *args;
2142	int i;
2143
2144	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2145				error_str, error_str_len);
2146
2147	if (args == NULL)
2148		goto bailout;
2149
2150	for (i = 0; i < num_args; i++) {
2151		args[i].kname = NULL;
2152		args[i].kvalue = NULL;
2153	}
2154
2155	for (i = 0; i < num_args; i++) {
2156		uint8_t *tmpptr;
2157
2158		args[i].kname = ctl_copyin_alloc(args[i].name,
2159			args[i].namelen, error_str, error_str_len);
2160		if (args[i].kname == NULL)
2161			goto bailout;
2162
2163		if (args[i].kname[args[i].namelen - 1] != '\0') {
2164			snprintf(error_str, error_str_len, "Argument %d "
2165				 "name is not NUL-terminated", i);
2166			goto bailout;
2167		}
2168
2169		if (args[i].flags & CTL_BEARG_RD) {
2170			tmpptr = ctl_copyin_alloc(args[i].value,
2171				args[i].vallen, error_str, error_str_len);
2172			if (tmpptr == NULL)
2173				goto bailout;
2174			if ((args[i].flags & CTL_BEARG_ASCII)
2175			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2176				snprintf(error_str, error_str_len, "Argument "
2177				    "%d value is not NUL-terminated", i);
2178				goto bailout;
2179			}
2180			args[i].kvalue = tmpptr;
2181		} else {
2182			args[i].kvalue = malloc(args[i].vallen,
2183			    M_CTL, M_WAITOK | M_ZERO);
2184		}
2185	}
2186
2187	return (args);
2188bailout:
2189
2190	ctl_free_args(num_args, args);
2191
2192	return (NULL);
2193}
2194
2195static void
2196ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2197{
2198	int i;
2199
2200	for (i = 0; i < num_args; i++) {
2201		if (args[i].flags & CTL_BEARG_WR)
2202			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2203	}
2204}
2205
2206/*
2207 * Escape characters that are illegal or not recommended in XML.
2208 */
2209int
2210ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2211{
2212	int retval;
2213
2214	retval = 0;
2215
2216	for (; *str; str++) {
2217		switch (*str) {
2218		case '&':
2219			retval = sbuf_printf(sb, "&amp;");
2220			break;
2221		case '>':
2222			retval = sbuf_printf(sb, "&gt;");
2223			break;
2224		case '<':
2225			retval = sbuf_printf(sb, "&lt;");
2226			break;
2227		default:
2228			retval = sbuf_putc(sb, *str);
2229			break;
2230		}
2231
2232		if (retval != 0)
2233			break;
2234
2235	}
2236
2237	return (retval);
2238}
2239
2240static int
2241ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2242	  struct thread *td)
2243{
2244	struct ctl_softc *softc;
2245	int retval;
2246
2247	softc = control_softc;
2248
2249	retval = 0;
2250
2251	switch (cmd) {
2252	case CTL_IO: {
2253		union ctl_io *io;
2254		void *pool_tmp;
2255
2256		/*
2257		 * If we haven't been "enabled", don't allow any SCSI I/O
2258		 * to this FETD.
2259		 */
2260		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2261			retval = EPERM;
2262			break;
2263		}
2264
2265		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2266		if (io == NULL) {
2267			printf("ctl_ioctl: can't allocate ctl_io!\n");
2268			retval = ENOSPC;
2269			break;
2270		}
2271
2272		/*
2273		 * Need to save the pool reference so it doesn't get
2274		 * spammed by the user's ctl_io.
2275		 */
2276		pool_tmp = io->io_hdr.pool;
2277
2278		memcpy(io, (void *)addr, sizeof(*io));
2279
2280		io->io_hdr.pool = pool_tmp;
2281		/*
2282		 * No status yet, so make sure the status is set properly.
2283		 */
2284		io->io_hdr.status = CTL_STATUS_NONE;
2285
2286		/*
2287		 * The user sets the initiator ID, target and LUN IDs.
2288		 */
2289		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2290		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2291		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2292		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2293			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2294
2295		retval = ctl_ioctl_submit_wait(io);
2296
2297		if (retval != 0) {
2298			ctl_free_io(io);
2299			break;
2300		}
2301
2302		memcpy((void *)addr, io, sizeof(*io));
2303
2304		/* return this to our pool */
2305		ctl_free_io(io);
2306
2307		break;
2308	}
2309	case CTL_ENABLE_PORT:
2310	case CTL_DISABLE_PORT:
2311	case CTL_SET_PORT_WWNS: {
2312		struct ctl_port *port;
2313		struct ctl_port_entry *entry;
2314
2315		entry = (struct ctl_port_entry *)addr;
2316
2317		mtx_lock(&softc->ctl_lock);
2318		STAILQ_FOREACH(port, &softc->port_list, links) {
2319			int action, done;
2320
2321			action = 0;
2322			done = 0;
2323
2324			if ((entry->port_type == CTL_PORT_NONE)
2325			 && (entry->targ_port == port->targ_port)) {
2326				/*
2327				 * If the user only wants to enable or
2328				 * disable or set WWNs on a specific port,
2329				 * do the operation and we're done.
2330				 */
2331				action = 1;
2332				done = 1;
2333			} else if (entry->port_type & port->port_type) {
2334				/*
2335				 * Compare the user's type mask with the
2336				 * particular frontend type to see if we
2337				 * have a match.
2338				 */
2339				action = 1;
2340				done = 0;
2341
2342				/*
2343				 * Make sure the user isn't trying to set
2344				 * WWNs on multiple ports at the same time.
2345				 */
2346				if (cmd == CTL_SET_PORT_WWNS) {
2347					printf("%s: Can't set WWNs on "
2348					       "multiple ports\n", __func__);
2349					retval = EINVAL;
2350					break;
2351				}
2352			}
2353			if (action != 0) {
2354				/*
2355				 * XXX KDM we have to drop the lock here,
2356				 * because the online/offline operations
2357				 * can potentially block.  We need to
2358				 * reference count the frontends so they
2359				 * can't go away,
2360				 */
2361				mtx_unlock(&softc->ctl_lock);
2362
2363				if (cmd == CTL_ENABLE_PORT) {
2364					struct ctl_lun *lun;
2365
2366					STAILQ_FOREACH(lun, &softc->lun_list,
2367						       links) {
2368						port->lun_enable(port->targ_lun_arg,
2369						    lun->target,
2370						    lun->lun);
2371					}
2372
2373					ctl_port_online(port);
2374				} else if (cmd == CTL_DISABLE_PORT) {
2375					struct ctl_lun *lun;
2376
2377					ctl_port_offline(port);
2378
2379					STAILQ_FOREACH(lun, &softc->lun_list,
2380						       links) {
2381						port->lun_disable(
2382						    port->targ_lun_arg,
2383						    lun->target,
2384						    lun->lun);
2385					}
2386				}
2387
2388				mtx_lock(&softc->ctl_lock);
2389
2390				if (cmd == CTL_SET_PORT_WWNS)
2391					ctl_port_set_wwns(port,
2392					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2393					    1 : 0, entry->wwnn,
2394					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2395					    1 : 0, entry->wwpn);
2396			}
2397			if (done != 0)
2398				break;
2399		}
2400		mtx_unlock(&softc->ctl_lock);
2401		break;
2402	}
2403	case CTL_GET_PORT_LIST: {
2404		struct ctl_port *port;
2405		struct ctl_port_list *list;
2406		int i;
2407
2408		list = (struct ctl_port_list *)addr;
2409
2410		if (list->alloc_len != (list->alloc_num *
2411		    sizeof(struct ctl_port_entry))) {
2412			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2413			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2414			       "%zu\n", __func__, list->alloc_len,
2415			       list->alloc_num, sizeof(struct ctl_port_entry));
2416			retval = EINVAL;
2417			break;
2418		}
2419		list->fill_len = 0;
2420		list->fill_num = 0;
2421		list->dropped_num = 0;
2422		i = 0;
2423		mtx_lock(&softc->ctl_lock);
2424		STAILQ_FOREACH(port, &softc->port_list, links) {
2425			struct ctl_port_entry entry, *list_entry;
2426
2427			if (list->fill_num >= list->alloc_num) {
2428				list->dropped_num++;
2429				continue;
2430			}
2431
2432			entry.port_type = port->port_type;
2433			strlcpy(entry.port_name, port->port_name,
2434				sizeof(entry.port_name));
2435			entry.targ_port = port->targ_port;
2436			entry.physical_port = port->physical_port;
2437			entry.virtual_port = port->virtual_port;
2438			entry.wwnn = port->wwnn;
2439			entry.wwpn = port->wwpn;
2440			if (port->status & CTL_PORT_STATUS_ONLINE)
2441				entry.online = 1;
2442			else
2443				entry.online = 0;
2444
2445			list_entry = &list->entries[i];
2446
2447			retval = copyout(&entry, list_entry, sizeof(entry));
2448			if (retval != 0) {
2449				printf("%s: CTL_GET_PORT_LIST: copyout "
2450				       "returned %d\n", __func__, retval);
2451				break;
2452			}
2453			i++;
2454			list->fill_num++;
2455			list->fill_len += sizeof(entry);
2456		}
2457		mtx_unlock(&softc->ctl_lock);
2458
2459		/*
2460		 * If this is non-zero, we had a copyout fault, so there's
2461		 * probably no point in attempting to set the status inside
2462		 * the structure.
2463		 */
2464		if (retval != 0)
2465			break;
2466
2467		if (list->dropped_num > 0)
2468			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2469		else
2470			list->status = CTL_PORT_LIST_OK;
2471		break;
2472	}
2473	case CTL_DUMP_OOA: {
2474		struct ctl_lun *lun;
2475		union ctl_io *io;
2476		char printbuf[128];
2477		struct sbuf sb;
2478
2479		mtx_lock(&softc->ctl_lock);
2480		printf("Dumping OOA queues:\n");
2481		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2482			mtx_lock(&lun->lun_lock);
2483			for (io = (union ctl_io *)TAILQ_FIRST(
2484			     &lun->ooa_queue); io != NULL;
2485			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2486			     ooa_links)) {
2487				sbuf_new(&sb, printbuf, sizeof(printbuf),
2488					 SBUF_FIXEDLEN);
2489				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2490					    (intmax_t)lun->lun,
2491					    io->scsiio.tag_num,
2492					    (io->io_hdr.flags &
2493					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2494					    (io->io_hdr.flags &
2495					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2496					    (io->io_hdr.flags &
2497					    CTL_FLAG_ABORT) ? " ABORT" : "",
2498			                    (io->io_hdr.flags &
2499		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2500				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2501				sbuf_finish(&sb);
2502				printf("%s\n", sbuf_data(&sb));
2503			}
2504			mtx_unlock(&lun->lun_lock);
2505		}
2506		printf("OOA queues dump done\n");
2507		mtx_unlock(&softc->ctl_lock);
2508		break;
2509	}
2510	case CTL_GET_OOA: {
2511		struct ctl_lun *lun;
2512		struct ctl_ooa *ooa_hdr;
2513		struct ctl_ooa_entry *entries;
2514		uint32_t cur_fill_num;
2515
2516		ooa_hdr = (struct ctl_ooa *)addr;
2517
2518		if ((ooa_hdr->alloc_len == 0)
2519		 || (ooa_hdr->alloc_num == 0)) {
2520			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2521			       "must be non-zero\n", __func__,
2522			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2523			retval = EINVAL;
2524			break;
2525		}
2526
2527		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2528		    sizeof(struct ctl_ooa_entry))) {
2529			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2530			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2531			       __func__, ooa_hdr->alloc_len,
2532			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2533			retval = EINVAL;
2534			break;
2535		}
2536
2537		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2538		if (entries == NULL) {
2539			printf("%s: could not allocate %d bytes for OOA "
2540			       "dump\n", __func__, ooa_hdr->alloc_len);
2541			retval = ENOMEM;
2542			break;
2543		}
2544
2545		mtx_lock(&softc->ctl_lock);
2546		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2547		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2548		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2549			mtx_unlock(&softc->ctl_lock);
2550			free(entries, M_CTL);
2551			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2552			       __func__, (uintmax_t)ooa_hdr->lun_num);
2553			retval = EINVAL;
2554			break;
2555		}
2556
2557		cur_fill_num = 0;
2558
2559		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2560			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2561				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2562					ooa_hdr, entries);
2563				if (retval != 0)
2564					break;
2565			}
2566			if (retval != 0) {
2567				mtx_unlock(&softc->ctl_lock);
2568				free(entries, M_CTL);
2569				break;
2570			}
2571		} else {
2572			lun = softc->ctl_luns[ooa_hdr->lun_num];
2573
2574			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2575						    entries);
2576		}
2577		mtx_unlock(&softc->ctl_lock);
2578
2579		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2580		ooa_hdr->fill_len = ooa_hdr->fill_num *
2581			sizeof(struct ctl_ooa_entry);
2582		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2583		if (retval != 0) {
2584			printf("%s: error copying out %d bytes for OOA dump\n",
2585			       __func__, ooa_hdr->fill_len);
2586		}
2587
2588		getbintime(&ooa_hdr->cur_bt);
2589
2590		if (cur_fill_num > ooa_hdr->alloc_num) {
2591			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2592			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2593		} else {
2594			ooa_hdr->dropped_num = 0;
2595			ooa_hdr->status = CTL_OOA_OK;
2596		}
2597
2598		free(entries, M_CTL);
2599		break;
2600	}
2601	case CTL_CHECK_OOA: {
2602		union ctl_io *io;
2603		struct ctl_lun *lun;
2604		struct ctl_ooa_info *ooa_info;
2605
2606
2607		ooa_info = (struct ctl_ooa_info *)addr;
2608
2609		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2610			ooa_info->status = CTL_OOA_INVALID_LUN;
2611			break;
2612		}
2613		mtx_lock(&softc->ctl_lock);
2614		lun = softc->ctl_luns[ooa_info->lun_id];
2615		if (lun == NULL) {
2616			mtx_unlock(&softc->ctl_lock);
2617			ooa_info->status = CTL_OOA_INVALID_LUN;
2618			break;
2619		}
2620		mtx_lock(&lun->lun_lock);
2621		mtx_unlock(&softc->ctl_lock);
2622		ooa_info->num_entries = 0;
2623		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2624		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2625		     &io->io_hdr, ooa_links)) {
2626			ooa_info->num_entries++;
2627		}
2628		mtx_unlock(&lun->lun_lock);
2629
2630		ooa_info->status = CTL_OOA_SUCCESS;
2631
2632		break;
2633	}
2634	case CTL_HARD_START:
2635	case CTL_HARD_STOP: {
2636		struct ctl_fe_ioctl_startstop_info ss_info;
2637		struct cfi_metatask *metatask;
2638		struct mtx hs_mtx;
2639
2640		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2641
2642		cv_init(&ss_info.sem, "hard start/stop cv" );
2643
2644		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2645		if (metatask == NULL) {
2646			retval = ENOMEM;
2647			mtx_destroy(&hs_mtx);
2648			break;
2649		}
2650
2651		if (cmd == CTL_HARD_START)
2652			metatask->tasktype = CFI_TASK_STARTUP;
2653		else
2654			metatask->tasktype = CFI_TASK_SHUTDOWN;
2655
2656		metatask->callback = ctl_ioctl_hard_startstop_callback;
2657		metatask->callback_arg = &ss_info;
2658
2659		cfi_action(metatask);
2660
2661		/* Wait for the callback */
2662		mtx_lock(&hs_mtx);
2663		cv_wait_sig(&ss_info.sem, &hs_mtx);
2664		mtx_unlock(&hs_mtx);
2665
2666		/*
2667		 * All information has been copied from the metatask by the
2668		 * time cv_broadcast() is called, so we free the metatask here.
2669		 */
2670		cfi_free_metatask(metatask);
2671
2672		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2673
2674		mtx_destroy(&hs_mtx);
2675		break;
2676	}
2677	case CTL_BBRREAD: {
2678		struct ctl_bbrread_info *bbr_info;
2679		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2680		struct mtx bbr_mtx;
2681		struct cfi_metatask *metatask;
2682
2683		bbr_info = (struct ctl_bbrread_info *)addr;
2684
2685		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2686
2687		bzero(&bbr_mtx, sizeof(bbr_mtx));
2688		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2689
2690		fe_bbr_info.bbr_info = bbr_info;
2691		fe_bbr_info.lock = &bbr_mtx;
2692
2693		cv_init(&fe_bbr_info.sem, "BBR read cv");
2694		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2695
2696		if (metatask == NULL) {
2697			mtx_destroy(&bbr_mtx);
2698			cv_destroy(&fe_bbr_info.sem);
2699			retval = ENOMEM;
2700			break;
2701		}
2702		metatask->tasktype = CFI_TASK_BBRREAD;
2703		metatask->callback = ctl_ioctl_bbrread_callback;
2704		metatask->callback_arg = &fe_bbr_info;
2705		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2706		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2707		metatask->taskinfo.bbrread.len = bbr_info->len;
2708
2709		cfi_action(metatask);
2710
2711		mtx_lock(&bbr_mtx);
2712		while (fe_bbr_info.wakeup_done == 0)
2713			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2714		mtx_unlock(&bbr_mtx);
2715
2716		bbr_info->status = metatask->status;
2717		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2718		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2719		memcpy(&bbr_info->sense_data,
2720		       &metatask->taskinfo.bbrread.sense_data,
2721		       ctl_min(sizeof(bbr_info->sense_data),
2722			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2723
2724		cfi_free_metatask(metatask);
2725
2726		mtx_destroy(&bbr_mtx);
2727		cv_destroy(&fe_bbr_info.sem);
2728
2729		break;
2730	}
2731	case CTL_DELAY_IO: {
2732		struct ctl_io_delay_info *delay_info;
2733#ifdef CTL_IO_DELAY
2734		struct ctl_lun *lun;
2735#endif /* CTL_IO_DELAY */
2736
2737		delay_info = (struct ctl_io_delay_info *)addr;
2738
2739#ifdef CTL_IO_DELAY
2740		mtx_lock(&softc->ctl_lock);
2741
2742		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2743		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2744			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2745		} else {
2746			lun = softc->ctl_luns[delay_info->lun_id];
2747			mtx_lock(&lun->lun_lock);
2748
2749			delay_info->status = CTL_DELAY_STATUS_OK;
2750
2751			switch (delay_info->delay_type) {
2752			case CTL_DELAY_TYPE_CONT:
2753				break;
2754			case CTL_DELAY_TYPE_ONESHOT:
2755				break;
2756			default:
2757				delay_info->status =
2758					CTL_DELAY_STATUS_INVALID_TYPE;
2759				break;
2760			}
2761
2762			switch (delay_info->delay_loc) {
2763			case CTL_DELAY_LOC_DATAMOVE:
2764				lun->delay_info.datamove_type =
2765					delay_info->delay_type;
2766				lun->delay_info.datamove_delay =
2767					delay_info->delay_secs;
2768				break;
2769			case CTL_DELAY_LOC_DONE:
2770				lun->delay_info.done_type =
2771					delay_info->delay_type;
2772				lun->delay_info.done_delay =
2773					delay_info->delay_secs;
2774				break;
2775			default:
2776				delay_info->status =
2777					CTL_DELAY_STATUS_INVALID_LOC;
2778				break;
2779			}
2780			mtx_unlock(&lun->lun_lock);
2781		}
2782
2783		mtx_unlock(&softc->ctl_lock);
2784#else
2785		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2786#endif /* CTL_IO_DELAY */
2787		break;
2788	}
2789	case CTL_REALSYNC_SET: {
2790		int *syncstate;
2791
2792		syncstate = (int *)addr;
2793
2794		mtx_lock(&softc->ctl_lock);
2795		switch (*syncstate) {
2796		case 0:
2797			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2798			break;
2799		case 1:
2800			softc->flags |= CTL_FLAG_REAL_SYNC;
2801			break;
2802		default:
2803			retval = EINVAL;
2804			break;
2805		}
2806		mtx_unlock(&softc->ctl_lock);
2807		break;
2808	}
2809	case CTL_REALSYNC_GET: {
2810		int *syncstate;
2811
2812		syncstate = (int*)addr;
2813
2814		mtx_lock(&softc->ctl_lock);
2815		if (softc->flags & CTL_FLAG_REAL_SYNC)
2816			*syncstate = 1;
2817		else
2818			*syncstate = 0;
2819		mtx_unlock(&softc->ctl_lock);
2820
2821		break;
2822	}
2823	case CTL_SETSYNC:
2824	case CTL_GETSYNC: {
2825		struct ctl_sync_info *sync_info;
2826		struct ctl_lun *lun;
2827
2828		sync_info = (struct ctl_sync_info *)addr;
2829
2830		mtx_lock(&softc->ctl_lock);
2831		lun = softc->ctl_luns[sync_info->lun_id];
2832		if (lun == NULL) {
2833			mtx_unlock(&softc->ctl_lock);
2834			sync_info->status = CTL_GS_SYNC_NO_LUN;
2835		}
2836		/*
2837		 * Get or set the sync interval.  We're not bounds checking
2838		 * in the set case, hopefully the user won't do something
2839		 * silly.
2840		 */
2841		mtx_lock(&lun->lun_lock);
2842		mtx_unlock(&softc->ctl_lock);
2843		if (cmd == CTL_GETSYNC)
2844			sync_info->sync_interval = lun->sync_interval;
2845		else
2846			lun->sync_interval = sync_info->sync_interval;
2847		mtx_unlock(&lun->lun_lock);
2848
2849		sync_info->status = CTL_GS_SYNC_OK;
2850
2851		break;
2852	}
2853	case CTL_GETSTATS: {
2854		struct ctl_stats *stats;
2855		struct ctl_lun *lun;
2856		int i;
2857
2858		stats = (struct ctl_stats *)addr;
2859
2860		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2861		     stats->alloc_len) {
2862			stats->status = CTL_SS_NEED_MORE_SPACE;
2863			stats->num_luns = softc->num_luns;
2864			break;
2865		}
2866		/*
2867		 * XXX KDM no locking here.  If the LUN list changes,
2868		 * things can blow up.
2869		 */
2870		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2871		     i++, lun = STAILQ_NEXT(lun, links)) {
2872			retval = copyout(&lun->stats, &stats->lun_stats[i],
2873					 sizeof(lun->stats));
2874			if (retval != 0)
2875				break;
2876		}
2877		stats->num_luns = softc->num_luns;
2878		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2879				 softc->num_luns;
2880		stats->status = CTL_SS_OK;
2881#ifdef CTL_TIME_IO
2882		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2883#else
2884		stats->flags = CTL_STATS_FLAG_NONE;
2885#endif
2886		getnanouptime(&stats->timestamp);
2887		break;
2888	}
2889	case CTL_ERROR_INJECT: {
2890		struct ctl_error_desc *err_desc, *new_err_desc;
2891		struct ctl_lun *lun;
2892
2893		err_desc = (struct ctl_error_desc *)addr;
2894
2895		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2896				      M_WAITOK | M_ZERO);
2897		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2898
2899		mtx_lock(&softc->ctl_lock);
2900		lun = softc->ctl_luns[err_desc->lun_id];
2901		if (lun == NULL) {
2902			mtx_unlock(&softc->ctl_lock);
2903			free(new_err_desc, M_CTL);
2904			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2905			       __func__, (uintmax_t)err_desc->lun_id);
2906			retval = EINVAL;
2907			break;
2908		}
2909		mtx_lock(&lun->lun_lock);
2910		mtx_unlock(&softc->ctl_lock);
2911
2912		/*
2913		 * We could do some checking here to verify the validity
2914		 * of the request, but given the complexity of error
2915		 * injection requests, the checking logic would be fairly
2916		 * complex.
2917		 *
2918		 * For now, if the request is invalid, it just won't get
2919		 * executed and might get deleted.
2920		 */
2921		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2922
2923		/*
2924		 * XXX KDM check to make sure the serial number is unique,
2925		 * in case we somehow manage to wrap.  That shouldn't
2926		 * happen for a very long time, but it's the right thing to
2927		 * do.
2928		 */
2929		new_err_desc->serial = lun->error_serial;
2930		err_desc->serial = lun->error_serial;
2931		lun->error_serial++;
2932
2933		mtx_unlock(&lun->lun_lock);
2934		break;
2935	}
2936	case CTL_ERROR_INJECT_DELETE: {
2937		struct ctl_error_desc *delete_desc, *desc, *desc2;
2938		struct ctl_lun *lun;
2939		int delete_done;
2940
2941		delete_desc = (struct ctl_error_desc *)addr;
2942		delete_done = 0;
2943
2944		mtx_lock(&softc->ctl_lock);
2945		lun = softc->ctl_luns[delete_desc->lun_id];
2946		if (lun == NULL) {
2947			mtx_unlock(&softc->ctl_lock);
2948			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2949			       __func__, (uintmax_t)delete_desc->lun_id);
2950			retval = EINVAL;
2951			break;
2952		}
2953		mtx_lock(&lun->lun_lock);
2954		mtx_unlock(&softc->ctl_lock);
2955		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2956			if (desc->serial != delete_desc->serial)
2957				continue;
2958
2959			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2960				      links);
2961			free(desc, M_CTL);
2962			delete_done = 1;
2963		}
2964		mtx_unlock(&lun->lun_lock);
2965		if (delete_done == 0) {
2966			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2967			       "error serial %ju on LUN %u\n", __func__,
2968			       delete_desc->serial, delete_desc->lun_id);
2969			retval = EINVAL;
2970			break;
2971		}
2972		break;
2973	}
2974	case CTL_DUMP_STRUCTS: {
2975		int i, j, k, idx;
2976		struct ctl_port *port;
2977		struct ctl_frontend *fe;
2978
2979		mtx_lock(&softc->ctl_lock);
2980		printf("CTL Persistent Reservation information start:\n");
2981		for (i = 0; i < CTL_MAX_LUNS; i++) {
2982			struct ctl_lun *lun;
2983
2984			lun = softc->ctl_luns[i];
2985
2986			if ((lun == NULL)
2987			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2988				continue;
2989
2990			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2991				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2992					idx = j * CTL_MAX_INIT_PER_PORT + k;
2993					if (lun->per_res[idx].registered == 0)
2994						continue;
2995					printf("  LUN %d port %d iid %d key "
2996					       "%#jx\n", i, j, k,
2997					       (uintmax_t)scsi_8btou64(
2998					       lun->per_res[idx].res_key.key));
2999				}
3000			}
3001		}
3002		printf("CTL Persistent Reservation information end\n");
3003		printf("CTL Ports:\n");
3004		STAILQ_FOREACH(port, &softc->port_list, links) {
3005			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3006			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3007			       port->frontend->name, port->port_type,
3008			       port->physical_port, port->virtual_port,
3009			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3010			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3011				if (port->wwpn_iid[j].in_use == 0 &&
3012				    port->wwpn_iid[j].wwpn == 0 &&
3013				    port->wwpn_iid[j].name == NULL)
3014					continue;
3015
3016				printf("    iid %u use %d WWPN %#jx '%s'\n",
3017				    j, port->wwpn_iid[j].in_use,
3018				    (uintmax_t)port->wwpn_iid[j].wwpn,
3019				    port->wwpn_iid[j].name);
3020			}
3021		}
3022		printf("CTL Port information end\n");
3023		mtx_unlock(&softc->ctl_lock);
3024		/*
3025		 * XXX KDM calling this without a lock.  We'd likely want
3026		 * to drop the lock before calling the frontend's dump
3027		 * routine anyway.
3028		 */
3029		printf("CTL Frontends:\n");
3030		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3031			printf("  Frontend '%s'\n", fe->name);
3032			if (fe->fe_dump != NULL)
3033				fe->fe_dump();
3034		}
3035		printf("CTL Frontend information end\n");
3036		break;
3037	}
3038	case CTL_LUN_REQ: {
3039		struct ctl_lun_req *lun_req;
3040		struct ctl_backend_driver *backend;
3041
3042		lun_req = (struct ctl_lun_req *)addr;
3043
3044		backend = ctl_backend_find(lun_req->backend);
3045		if (backend == NULL) {
3046			lun_req->status = CTL_LUN_ERROR;
3047			snprintf(lun_req->error_str,
3048				 sizeof(lun_req->error_str),
3049				 "Backend \"%s\" not found.",
3050				 lun_req->backend);
3051			break;
3052		}
3053		if (lun_req->num_be_args > 0) {
3054			lun_req->kern_be_args = ctl_copyin_args(
3055				lun_req->num_be_args,
3056				lun_req->be_args,
3057				lun_req->error_str,
3058				sizeof(lun_req->error_str));
3059			if (lun_req->kern_be_args == NULL) {
3060				lun_req->status = CTL_LUN_ERROR;
3061				break;
3062			}
3063		}
3064
3065		retval = backend->ioctl(dev, cmd, addr, flag, td);
3066
3067		if (lun_req->num_be_args > 0) {
3068			ctl_copyout_args(lun_req->num_be_args,
3069				      lun_req->kern_be_args);
3070			ctl_free_args(lun_req->num_be_args,
3071				      lun_req->kern_be_args);
3072		}
3073		break;
3074	}
3075	case CTL_LUN_LIST: {
3076		struct sbuf *sb;
3077		struct ctl_lun *lun;
3078		struct ctl_lun_list *list;
3079		struct ctl_option *opt;
3080
3081		list = (struct ctl_lun_list *)addr;
3082
3083		/*
3084		 * Allocate a fixed length sbuf here, based on the length
3085		 * of the user's buffer.  We could allocate an auto-extending
3086		 * buffer, and then tell the user how much larger our
3087		 * amount of data is than his buffer, but that presents
3088		 * some problems:
3089		 *
3090		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3091		 *     we can't hold a lock while calling them with an
3092		 *     auto-extending buffer.
3093 		 *
3094		 * 2.  There is not currently a LUN reference counting
3095		 *     mechanism, outside of outstanding transactions on
3096		 *     the LUN's OOA queue.  So a LUN could go away on us
3097		 *     while we're getting the LUN number, backend-specific
3098		 *     information, etc.  Thus, given the way things
3099		 *     currently work, we need to hold the CTL lock while
3100		 *     grabbing LUN information.
3101		 *
3102		 * So, from the user's standpoint, the best thing to do is
3103		 * allocate what he thinks is a reasonable buffer length,
3104		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3105		 * double the buffer length and try again.  (And repeat
3106		 * that until he succeeds.)
3107		 */
3108		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3109		if (sb == NULL) {
3110			list->status = CTL_LUN_LIST_ERROR;
3111			snprintf(list->error_str, sizeof(list->error_str),
3112				 "Unable to allocate %d bytes for LUN list",
3113				 list->alloc_len);
3114			break;
3115		}
3116
3117		sbuf_printf(sb, "<ctllunlist>\n");
3118
3119		mtx_lock(&softc->ctl_lock);
3120		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3121			mtx_lock(&lun->lun_lock);
3122			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3123					     (uintmax_t)lun->lun);
3124
3125			/*
3126			 * Bail out as soon as we see that we've overfilled
3127			 * the buffer.
3128			 */
3129			if (retval != 0)
3130				break;
3131
3132			retval = sbuf_printf(sb, "\t<backend_type>%s"
3133					     "</backend_type>\n",
3134					     (lun->backend == NULL) ?  "none" :
3135					     lun->backend->name);
3136
3137			if (retval != 0)
3138				break;
3139
3140			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3141					     lun->be_lun->lun_type);
3142
3143			if (retval != 0)
3144				break;
3145
3146			if (lun->backend == NULL) {
3147				retval = sbuf_printf(sb, "</lun>\n");
3148				if (retval != 0)
3149					break;
3150				continue;
3151			}
3152
3153			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3154					     (lun->be_lun->maxlba > 0) ?
3155					     lun->be_lun->maxlba + 1 : 0);
3156
3157			if (retval != 0)
3158				break;
3159
3160			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3161					     lun->be_lun->blocksize);
3162
3163			if (retval != 0)
3164				break;
3165
3166			retval = sbuf_printf(sb, "\t<serial_number>");
3167
3168			if (retval != 0)
3169				break;
3170
3171			retval = ctl_sbuf_printf_esc(sb,
3172						     lun->be_lun->serial_num);
3173
3174			if (retval != 0)
3175				break;
3176
3177			retval = sbuf_printf(sb, "</serial_number>\n");
3178
3179			if (retval != 0)
3180				break;
3181
3182			retval = sbuf_printf(sb, "\t<device_id>");
3183
3184			if (retval != 0)
3185				break;
3186
3187			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3188
3189			if (retval != 0)
3190				break;
3191
3192			retval = sbuf_printf(sb, "</device_id>\n");
3193
3194			if (retval != 0)
3195				break;
3196
3197			if (lun->backend->lun_info != NULL) {
3198				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3199				if (retval != 0)
3200					break;
3201			}
3202			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3203				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3204				    opt->name, opt->value, opt->name);
3205				if (retval != 0)
3206					break;
3207			}
3208
3209			retval = sbuf_printf(sb, "</lun>\n");
3210
3211			if (retval != 0)
3212				break;
3213			mtx_unlock(&lun->lun_lock);
3214		}
3215		if (lun != NULL)
3216			mtx_unlock(&lun->lun_lock);
3217		mtx_unlock(&softc->ctl_lock);
3218
3219		if ((retval != 0)
3220		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3221			retval = 0;
3222			sbuf_delete(sb);
3223			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3224			snprintf(list->error_str, sizeof(list->error_str),
3225				 "Out of space, %d bytes is too small",
3226				 list->alloc_len);
3227			break;
3228		}
3229
3230		sbuf_finish(sb);
3231
3232		retval = copyout(sbuf_data(sb), list->lun_xml,
3233				 sbuf_len(sb) + 1);
3234
3235		list->fill_len = sbuf_len(sb) + 1;
3236		list->status = CTL_LUN_LIST_OK;
3237		sbuf_delete(sb);
3238		break;
3239	}
3240	case CTL_ISCSI: {
3241		struct ctl_iscsi *ci;
3242		struct ctl_frontend *fe;
3243
3244		ci = (struct ctl_iscsi *)addr;
3245
3246		fe = ctl_frontend_find("iscsi");
3247		if (fe == NULL) {
3248			ci->status = CTL_ISCSI_ERROR;
3249			snprintf(ci->error_str, sizeof(ci->error_str),
3250			    "Frontend \"iscsi\" not found.");
3251			break;
3252		}
3253
3254		retval = fe->ioctl(dev, cmd, addr, flag, td);
3255		break;
3256	}
3257	case CTL_PORT_REQ: {
3258		struct ctl_req *req;
3259		struct ctl_frontend *fe;
3260
3261		req = (struct ctl_req *)addr;
3262
3263		fe = ctl_frontend_find(req->driver);
3264		if (fe == NULL) {
3265			req->status = CTL_LUN_ERROR;
3266			snprintf(req->error_str, sizeof(req->error_str),
3267			    "Frontend \"%s\" not found.", req->driver);
3268			break;
3269		}
3270		if (req->num_args > 0) {
3271			req->kern_args = ctl_copyin_args(req->num_args,
3272			    req->args, req->error_str, sizeof(req->error_str));
3273			if (req->kern_args == NULL) {
3274				req->status = CTL_LUN_ERROR;
3275				break;
3276			}
3277		}
3278
3279		retval = fe->ioctl(dev, cmd, addr, flag, td);
3280
3281		if (req->num_args > 0) {
3282			ctl_copyout_args(req->num_args, req->kern_args);
3283			ctl_free_args(req->num_args, req->kern_args);
3284		}
3285		break;
3286	}
3287	case CTL_PORT_LIST: {
3288		struct sbuf *sb;
3289		struct ctl_port *port;
3290		struct ctl_lun_list *list;
3291		struct ctl_option *opt;
3292
3293		list = (struct ctl_lun_list *)addr;
3294
3295		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3296		if (sb == NULL) {
3297			list->status = CTL_LUN_LIST_ERROR;
3298			snprintf(list->error_str, sizeof(list->error_str),
3299				 "Unable to allocate %d bytes for LUN list",
3300				 list->alloc_len);
3301			break;
3302		}
3303
3304		sbuf_printf(sb, "<ctlportlist>\n");
3305
3306		mtx_lock(&softc->ctl_lock);
3307		STAILQ_FOREACH(port, &softc->port_list, links) {
3308			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3309					     (uintmax_t)port->targ_port);
3310
3311			/*
3312			 * Bail out as soon as we see that we've overfilled
3313			 * the buffer.
3314			 */
3315			if (retval != 0)
3316				break;
3317
3318			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3319			    "</frontend_type>\n", port->frontend->name);
3320			if (retval != 0)
3321				break;
3322
3323			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3324					     port->port_type);
3325			if (retval != 0)
3326				break;
3327
3328			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3329			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3330			if (retval != 0)
3331				break;
3332
3333			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3334			    port->port_name);
3335			if (retval != 0)
3336				break;
3337
3338			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3339			    port->physical_port);
3340			if (retval != 0)
3341				break;
3342
3343			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3344			    port->virtual_port);
3345			if (retval != 0)
3346				break;
3347
3348			retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3349			    (uintmax_t)port->wwnn);
3350			if (retval != 0)
3351				break;
3352
3353			retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3354			    (uintmax_t)port->wwpn);
3355			if (retval != 0)
3356				break;
3357
3358			if (port->port_info != NULL) {
3359				retval = port->port_info(port->onoff_arg, sb);
3360				if (retval != 0)
3361					break;
3362			}
3363			STAILQ_FOREACH(opt, &port->options, links) {
3364				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3365				    opt->name, opt->value, opt->name);
3366				if (retval != 0)
3367					break;
3368			}
3369
3370			retval = sbuf_printf(sb, "</targ_port>\n");
3371			if (retval != 0)
3372				break;
3373		}
3374		mtx_unlock(&softc->ctl_lock);
3375
3376		if ((retval != 0)
3377		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3378			retval = 0;
3379			sbuf_delete(sb);
3380			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3381			snprintf(list->error_str, sizeof(list->error_str),
3382				 "Out of space, %d bytes is too small",
3383				 list->alloc_len);
3384			break;
3385		}
3386
3387		sbuf_finish(sb);
3388
3389		retval = copyout(sbuf_data(sb), list->lun_xml,
3390				 sbuf_len(sb) + 1);
3391
3392		list->fill_len = sbuf_len(sb) + 1;
3393		list->status = CTL_LUN_LIST_OK;
3394		sbuf_delete(sb);
3395		break;
3396	}
3397	default: {
3398		/* XXX KDM should we fix this? */
3399#if 0
3400		struct ctl_backend_driver *backend;
3401		unsigned int type;
3402		int found;
3403
3404		found = 0;
3405
3406		/*
3407		 * We encode the backend type as the ioctl type for backend
3408		 * ioctls.  So parse it out here, and then search for a
3409		 * backend of this type.
3410		 */
3411		type = _IOC_TYPE(cmd);
3412
3413		STAILQ_FOREACH(backend, &softc->be_list, links) {
3414			if (backend->type == type) {
3415				found = 1;
3416				break;
3417			}
3418		}
3419		if (found == 0) {
3420			printf("ctl: unknown ioctl command %#lx or backend "
3421			       "%d\n", cmd, type);
3422			retval = EINVAL;
3423			break;
3424		}
3425		retval = backend->ioctl(dev, cmd, addr, flag, td);
3426#endif
3427		retval = ENOTTY;
3428		break;
3429	}
3430	}
3431	return (retval);
3432}
3433
3434uint32_t
3435ctl_get_initindex(struct ctl_nexus *nexus)
3436{
3437	if (nexus->targ_port < CTL_MAX_PORTS)
3438		return (nexus->initid.id +
3439			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3440	else
3441		return (nexus->initid.id +
3442		       ((nexus->targ_port - CTL_MAX_PORTS) *
3443			CTL_MAX_INIT_PER_PORT));
3444}
3445
3446uint32_t
3447ctl_get_resindex(struct ctl_nexus *nexus)
3448{
3449	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3450}
3451
3452uint32_t
3453ctl_port_idx(int port_num)
3454{
3455	if (port_num < CTL_MAX_PORTS)
3456		return(port_num);
3457	else
3458		return(port_num - CTL_MAX_PORTS);
3459}
3460
3461static uint32_t
3462ctl_map_lun(int port_num, uint32_t lun_id)
3463{
3464	struct ctl_port *port;
3465
3466	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3467	if (port == NULL)
3468		return (UINT32_MAX);
3469	if (port->lun_map == NULL)
3470		return (lun_id);
3471	return (port->lun_map(port->targ_lun_arg, lun_id));
3472}
3473
3474static uint32_t
3475ctl_map_lun_back(int port_num, uint32_t lun_id)
3476{
3477	struct ctl_port *port;
3478	uint32_t i;
3479
3480	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3481	if (port->lun_map == NULL)
3482		return (lun_id);
3483	for (i = 0; i < CTL_MAX_LUNS; i++) {
3484		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3485			return (i);
3486	}
3487	return (UINT32_MAX);
3488}
3489
3490/*
3491 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3492 * that are a power of 2.
3493 */
3494int
3495ctl_ffz(uint32_t *mask, uint32_t size)
3496{
3497	uint32_t num_chunks, num_pieces;
3498	int i, j;
3499
3500	num_chunks = (size >> 5);
3501	if (num_chunks == 0)
3502		num_chunks++;
3503	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3504
3505	for (i = 0; i < num_chunks; i++) {
3506		for (j = 0; j < num_pieces; j++) {
3507			if ((mask[i] & (1 << j)) == 0)
3508				return ((i << 5) + j);
3509		}
3510	}
3511
3512	return (-1);
3513}
3514
3515int
3516ctl_set_mask(uint32_t *mask, uint32_t bit)
3517{
3518	uint32_t chunk, piece;
3519
3520	chunk = bit >> 5;
3521	piece = bit % (sizeof(uint32_t) * 8);
3522
3523	if ((mask[chunk] & (1 << piece)) != 0)
3524		return (-1);
3525	else
3526		mask[chunk] |= (1 << piece);
3527
3528	return (0);
3529}
3530
3531int
3532ctl_clear_mask(uint32_t *mask, uint32_t bit)
3533{
3534	uint32_t chunk, piece;
3535
3536	chunk = bit >> 5;
3537	piece = bit % (sizeof(uint32_t) * 8);
3538
3539	if ((mask[chunk] & (1 << piece)) == 0)
3540		return (-1);
3541	else
3542		mask[chunk] &= ~(1 << piece);
3543
3544	return (0);
3545}
3546
3547int
3548ctl_is_set(uint32_t *mask, uint32_t bit)
3549{
3550	uint32_t chunk, piece;
3551
3552	chunk = bit >> 5;
3553	piece = bit % (sizeof(uint32_t) * 8);
3554
3555	if ((mask[chunk] & (1 << piece)) == 0)
3556		return (0);
3557	else
3558		return (1);
3559}
3560
3561#ifdef unused
3562/*
3563 * The bus, target and lun are optional, they can be filled in later.
3564 * can_wait is used to determine whether we can wait on the malloc or not.
3565 */
3566union ctl_io*
3567ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3568	      uint32_t targ_lun, int can_wait)
3569{
3570	union ctl_io *io;
3571
3572	if (can_wait)
3573		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3574	else
3575		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3576
3577	if (io != NULL) {
3578		io->io_hdr.io_type = io_type;
3579		io->io_hdr.targ_port = targ_port;
3580		/*
3581		 * XXX KDM this needs to change/go away.  We need to move
3582		 * to a preallocated pool of ctl_scsiio structures.
3583		 */
3584		io->io_hdr.nexus.targ_target.id = targ_target;
3585		io->io_hdr.nexus.targ_lun = targ_lun;
3586	}
3587
3588	return (io);
3589}
3590
3591void
3592ctl_kfree_io(union ctl_io *io)
3593{
3594	free(io, M_CTL);
3595}
3596#endif /* unused */
3597
3598/*
3599 * ctl_softc, pool_type, total_ctl_io are passed in.
3600 * npool is passed out.
3601 */
3602int
3603ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3604		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3605{
3606	uint32_t i;
3607	union ctl_io *cur_io, *next_io;
3608	struct ctl_io_pool *pool;
3609	int retval;
3610
3611	retval = 0;
3612
3613	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3614					    M_NOWAIT | M_ZERO);
3615	if (pool == NULL) {
3616		retval = ENOMEM;
3617		goto bailout;
3618	}
3619
3620	pool->type = pool_type;
3621	pool->ctl_softc = ctl_softc;
3622
3623	mtx_lock(&ctl_softc->pool_lock);
3624	pool->id = ctl_softc->cur_pool_id++;
3625	mtx_unlock(&ctl_softc->pool_lock);
3626
3627	pool->flags = CTL_POOL_FLAG_NONE;
3628	pool->refcount = 1;		/* Reference for validity. */
3629	STAILQ_INIT(&pool->free_queue);
3630
3631	/*
3632	 * XXX KDM other options here:
3633	 * - allocate a page at a time
3634	 * - allocate one big chunk of memory.
3635	 * Page allocation might work well, but would take a little more
3636	 * tracking.
3637	 */
3638	for (i = 0; i < total_ctl_io; i++) {
3639		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3640						M_NOWAIT);
3641		if (cur_io == NULL) {
3642			retval = ENOMEM;
3643			break;
3644		}
3645		cur_io->io_hdr.pool = pool;
3646		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3647		pool->total_ctl_io++;
3648		pool->free_ctl_io++;
3649	}
3650
3651	if (retval != 0) {
3652		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3653		     cur_io != NULL; cur_io = next_io) {
3654			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3655							      links);
3656			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3657				      ctl_io_hdr, links);
3658			free(cur_io, M_CTLIO);
3659		}
3660
3661		free(pool, M_CTL);
3662		goto bailout;
3663	}
3664	mtx_lock(&ctl_softc->pool_lock);
3665	ctl_softc->num_pools++;
3666	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3667	/*
3668	 * Increment our usage count if this is an external consumer, so we
3669	 * can't get unloaded until the external consumer (most likely a
3670	 * FETD) unloads and frees his pool.
3671	 *
3672	 * XXX KDM will this increment the caller's module use count, or
3673	 * mine?
3674	 */
3675#if 0
3676	if ((pool_type != CTL_POOL_EMERGENCY)
3677	 && (pool_type != CTL_POOL_INTERNAL)
3678	 && (pool_type != CTL_POOL_4OTHERSC))
3679		MOD_INC_USE_COUNT;
3680#endif
3681
3682	mtx_unlock(&ctl_softc->pool_lock);
3683
3684	*npool = pool;
3685
3686bailout:
3687
3688	return (retval);
3689}
3690
3691static int
3692ctl_pool_acquire(struct ctl_io_pool *pool)
3693{
3694
3695	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3696
3697	if (pool->flags & CTL_POOL_FLAG_INVALID)
3698		return (EINVAL);
3699
3700	pool->refcount++;
3701
3702	return (0);
3703}
3704
3705static void
3706ctl_pool_release(struct ctl_io_pool *pool)
3707{
3708	struct ctl_softc *ctl_softc = pool->ctl_softc;
3709	union ctl_io *io;
3710
3711	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3712
3713	if (--pool->refcount != 0)
3714		return;
3715
3716	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3717		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3718			      links);
3719		free(io, M_CTLIO);
3720	}
3721
3722	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3723	ctl_softc->num_pools--;
3724
3725	/*
3726	 * XXX KDM will this decrement the caller's usage count or mine?
3727	 */
3728#if 0
3729	if ((pool->type != CTL_POOL_EMERGENCY)
3730	 && (pool->type != CTL_POOL_INTERNAL)
3731	 && (pool->type != CTL_POOL_4OTHERSC))
3732		MOD_DEC_USE_COUNT;
3733#endif
3734
3735	free(pool, M_CTL);
3736}
3737
3738void
3739ctl_pool_free(struct ctl_io_pool *pool)
3740{
3741	struct ctl_softc *ctl_softc;
3742
3743	if (pool == NULL)
3744		return;
3745
3746	ctl_softc = pool->ctl_softc;
3747	mtx_lock(&ctl_softc->pool_lock);
3748	pool->flags |= CTL_POOL_FLAG_INVALID;
3749	ctl_pool_release(pool);
3750	mtx_unlock(&ctl_softc->pool_lock);
3751}
3752
3753/*
3754 * This routine does not block (except for spinlocks of course).
3755 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3756 * possible.
3757 */
3758union ctl_io *
3759ctl_alloc_io(void *pool_ref)
3760{
3761	union ctl_io *io;
3762	struct ctl_softc *ctl_softc;
3763	struct ctl_io_pool *pool, *npool;
3764	struct ctl_io_pool *emergency_pool;
3765
3766	pool = (struct ctl_io_pool *)pool_ref;
3767
3768	if (pool == NULL) {
3769		printf("%s: pool is NULL\n", __func__);
3770		return (NULL);
3771	}
3772
3773	emergency_pool = NULL;
3774
3775	ctl_softc = pool->ctl_softc;
3776
3777	mtx_lock(&ctl_softc->pool_lock);
3778	/*
3779	 * First, try to get the io structure from the user's pool.
3780	 */
3781	if (ctl_pool_acquire(pool) == 0) {
3782		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3783		if (io != NULL) {
3784			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3785			pool->total_allocated++;
3786			pool->free_ctl_io--;
3787			mtx_unlock(&ctl_softc->pool_lock);
3788			return (io);
3789		} else
3790			ctl_pool_release(pool);
3791	}
3792	/*
3793	 * If he doesn't have any io structures left, search for an
3794	 * emergency pool and grab one from there.
3795	 */
3796	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3797		if (npool->type != CTL_POOL_EMERGENCY)
3798			continue;
3799
3800		if (ctl_pool_acquire(npool) != 0)
3801			continue;
3802
3803		emergency_pool = npool;
3804
3805		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3806		if (io != NULL) {
3807			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3808			npool->total_allocated++;
3809			npool->free_ctl_io--;
3810			mtx_unlock(&ctl_softc->pool_lock);
3811			return (io);
3812		} else
3813			ctl_pool_release(npool);
3814	}
3815
3816	/* Drop the spinlock before we malloc */
3817	mtx_unlock(&ctl_softc->pool_lock);
3818
3819	/*
3820	 * The emergency pool (if it exists) didn't have one, so try an
3821	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3822	 */
3823	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3824	if (io != NULL) {
3825		/*
3826		 * If the emergency pool exists but is empty, add this
3827		 * ctl_io to its list when it gets freed.
3828		 */
3829		if (emergency_pool != NULL) {
3830			mtx_lock(&ctl_softc->pool_lock);
3831			if (ctl_pool_acquire(emergency_pool) == 0) {
3832				io->io_hdr.pool = emergency_pool;
3833				emergency_pool->total_ctl_io++;
3834				/*
3835				 * Need to bump this, otherwise
3836				 * total_allocated and total_freed won't
3837				 * match when we no longer have anything
3838				 * outstanding.
3839				 */
3840				emergency_pool->total_allocated++;
3841			}
3842			mtx_unlock(&ctl_softc->pool_lock);
3843		} else
3844			io->io_hdr.pool = NULL;
3845	}
3846
3847	return (io);
3848}
3849
3850void
3851ctl_free_io(union ctl_io *io)
3852{
3853	if (io == NULL)
3854		return;
3855
3856	/*
3857	 * If this ctl_io has a pool, return it to that pool.
3858	 */
3859	if (io->io_hdr.pool != NULL) {
3860		struct ctl_io_pool *pool;
3861
3862		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3863		mtx_lock(&pool->ctl_softc->pool_lock);
3864		io->io_hdr.io_type = 0xff;
3865		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3866		pool->total_freed++;
3867		pool->free_ctl_io++;
3868		ctl_pool_release(pool);
3869		mtx_unlock(&pool->ctl_softc->pool_lock);
3870	} else {
3871		/*
3872		 * Otherwise, just free it.  We probably malloced it and
3873		 * the emergency pool wasn't available.
3874		 */
3875		free(io, M_CTLIO);
3876	}
3877
3878}
3879
3880void
3881ctl_zero_io(union ctl_io *io)
3882{
3883	void *pool_ref;
3884
3885	if (io == NULL)
3886		return;
3887
3888	/*
3889	 * May need to preserve linked list pointers at some point too.
3890	 */
3891	pool_ref = io->io_hdr.pool;
3892
3893	memset(io, 0, sizeof(*io));
3894
3895	io->io_hdr.pool = pool_ref;
3896}
3897
3898/*
3899 * This routine is currently used for internal copies of ctl_ios that need
3900 * to persist for some reason after we've already returned status to the
3901 * FETD.  (Thus the flag set.)
3902 *
3903 * XXX XXX
3904 * Note that this makes a blind copy of all fields in the ctl_io, except
3905 * for the pool reference.  This includes any memory that has been
3906 * allocated!  That memory will no longer be valid after done has been
3907 * called, so this would be VERY DANGEROUS for command that actually does
3908 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3909 * start and stop commands, which don't transfer any data, so this is not a
3910 * problem.  If it is used for anything else, the caller would also need to
3911 * allocate data buffer space and this routine would need to be modified to
3912 * copy the data buffer(s) as well.
3913 */
3914void
3915ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3916{
3917	void *pool_ref;
3918
3919	if ((src == NULL)
3920	 || (dest == NULL))
3921		return;
3922
3923	/*
3924	 * May need to preserve linked list pointers at some point too.
3925	 */
3926	pool_ref = dest->io_hdr.pool;
3927
3928	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3929
3930	dest->io_hdr.pool = pool_ref;
3931	/*
3932	 * We need to know that this is an internal copy, and doesn't need
3933	 * to get passed back to the FETD that allocated it.
3934	 */
3935	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3936}
3937
3938#ifdef NEEDTOPORT
3939static void
3940ctl_update_power_subpage(struct copan_power_subpage *page)
3941{
3942	int num_luns, num_partitions, config_type;
3943	struct ctl_softc *softc;
3944	cs_BOOL_t aor_present, shelf_50pct_power;
3945	cs_raidset_personality_t rs_type;
3946	int max_active_luns;
3947
3948	softc = control_softc;
3949
3950	/* subtract out the processor LUN */
3951	num_luns = softc->num_luns - 1;
3952	/*
3953	 * Default to 7 LUNs active, which was the only number we allowed
3954	 * in the past.
3955	 */
3956	max_active_luns = 7;
3957
3958	num_partitions = config_GetRsPartitionInfo();
3959	config_type = config_GetConfigType();
3960	shelf_50pct_power = config_GetShelfPowerMode();
3961	aor_present = config_IsAorRsPresent();
3962
3963	rs_type = ddb_GetRsRaidType(1);
3964	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3965	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3966		EPRINT(0, "Unsupported RS type %d!", rs_type);
3967	}
3968
3969
3970	page->total_luns = num_luns;
3971
3972	switch (config_type) {
3973	case 40:
3974		/*
3975		 * In a 40 drive configuration, it doesn't matter what DC
3976		 * cards we have, whether we have AOR enabled or not,
3977		 * partitioning or not, or what type of RAIDset we have.
3978		 * In that scenario, we can power up every LUN we present
3979		 * to the user.
3980		 */
3981		max_active_luns = num_luns;
3982
3983		break;
3984	case 64:
3985		if (shelf_50pct_power == CS_FALSE) {
3986			/* 25% power */
3987			if (aor_present == CS_TRUE) {
3988				if (rs_type ==
3989				     CS_RAIDSET_PERSONALITY_RAID5) {
3990					max_active_luns = 7;
3991				} else if (rs_type ==
3992					 CS_RAIDSET_PERSONALITY_RAID1){
3993					max_active_luns = 14;
3994				} else {
3995					/* XXX KDM now what?? */
3996				}
3997			} else {
3998				if (rs_type ==
3999				     CS_RAIDSET_PERSONALITY_RAID5) {
4000					max_active_luns = 8;
4001				} else if (rs_type ==
4002					 CS_RAIDSET_PERSONALITY_RAID1){
4003					max_active_luns = 16;
4004				} else {
4005					/* XXX KDM now what?? */
4006				}
4007			}
4008		} else {
4009			/* 50% power */
4010			/*
4011			 * With 50% power in a 64 drive configuration, we
4012			 * can power all LUNs we present.
4013			 */
4014			max_active_luns = num_luns;
4015		}
4016		break;
4017	case 112:
4018		if (shelf_50pct_power == CS_FALSE) {
4019			/* 25% power */
4020			if (aor_present == CS_TRUE) {
4021				if (rs_type ==
4022				     CS_RAIDSET_PERSONALITY_RAID5) {
4023					max_active_luns = 7;
4024				} else if (rs_type ==
4025					 CS_RAIDSET_PERSONALITY_RAID1){
4026					max_active_luns = 14;
4027				} else {
4028					/* XXX KDM now what?? */
4029				}
4030			} else {
4031				if (rs_type ==
4032				     CS_RAIDSET_PERSONALITY_RAID5) {
4033					max_active_luns = 8;
4034				} else if (rs_type ==
4035					 CS_RAIDSET_PERSONALITY_RAID1){
4036					max_active_luns = 16;
4037				} else {
4038					/* XXX KDM now what?? */
4039				}
4040			}
4041		} else {
4042			/* 50% power */
4043			if (aor_present == CS_TRUE) {
4044				if (rs_type ==
4045				     CS_RAIDSET_PERSONALITY_RAID5) {
4046					max_active_luns = 14;
4047				} else if (rs_type ==
4048					 CS_RAIDSET_PERSONALITY_RAID1){
4049					/*
4050					 * We're assuming here that disk
4051					 * caching is enabled, and so we're
4052					 * able to power up half of each
4053					 * LUN, and cache all writes.
4054					 */
4055					max_active_luns = num_luns;
4056				} else {
4057					/* XXX KDM now what?? */
4058				}
4059			} else {
4060				if (rs_type ==
4061				     CS_RAIDSET_PERSONALITY_RAID5) {
4062					max_active_luns = 15;
4063				} else if (rs_type ==
4064					 CS_RAIDSET_PERSONALITY_RAID1){
4065					max_active_luns = 30;
4066				} else {
4067					/* XXX KDM now what?? */
4068				}
4069			}
4070		}
4071		break;
4072	default:
4073		/*
4074		 * In this case, we have an unknown configuration, so we
4075		 * just use the default from above.
4076		 */
4077		break;
4078	}
4079
4080	page->max_active_luns = max_active_luns;
4081#if 0
4082	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4083	       page->total_luns, page->max_active_luns);
4084#endif
4085}
4086#endif /* NEEDTOPORT */
4087
4088/*
4089 * This routine could be used in the future to load default and/or saved
4090 * mode page parameters for a particuar lun.
4091 */
4092static int
4093ctl_init_page_index(struct ctl_lun *lun)
4094{
4095	int i;
4096	struct ctl_page_index *page_index;
4097	struct ctl_softc *softc;
4098	const char *value;
4099
4100	memcpy(&lun->mode_pages.index, page_index_template,
4101	       sizeof(page_index_template));
4102
4103	softc = lun->ctl_softc;
4104
4105	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4106
4107		page_index = &lun->mode_pages.index[i];
4108		/*
4109		 * If this is a disk-only mode page, there's no point in
4110		 * setting it up.  For some pages, we have to have some
4111		 * basic information about the disk in order to calculate the
4112		 * mode page data.
4113		 */
4114		if ((lun->be_lun->lun_type != T_DIRECT)
4115		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4116			continue;
4117
4118		switch (page_index->page_code & SMPH_PC_MASK) {
4119		case SMS_FORMAT_DEVICE_PAGE: {
4120			struct scsi_format_page *format_page;
4121
4122			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4123				panic("subpage is incorrect!");
4124
4125			/*
4126			 * Sectors per track are set above.  Bytes per
4127			 * sector need to be set here on a per-LUN basis.
4128			 */
4129			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4130			       &format_page_default,
4131			       sizeof(format_page_default));
4132			memcpy(&lun->mode_pages.format_page[
4133			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4134			       sizeof(format_page_changeable));
4135			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4136			       &format_page_default,
4137			       sizeof(format_page_default));
4138			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4139			       &format_page_default,
4140			       sizeof(format_page_default));
4141
4142			format_page = &lun->mode_pages.format_page[
4143				CTL_PAGE_CURRENT];
4144			scsi_ulto2b(lun->be_lun->blocksize,
4145				    format_page->bytes_per_sector);
4146
4147			format_page = &lun->mode_pages.format_page[
4148				CTL_PAGE_DEFAULT];
4149			scsi_ulto2b(lun->be_lun->blocksize,
4150				    format_page->bytes_per_sector);
4151
4152			format_page = &lun->mode_pages.format_page[
4153				CTL_PAGE_SAVED];
4154			scsi_ulto2b(lun->be_lun->blocksize,
4155				    format_page->bytes_per_sector);
4156
4157			page_index->page_data =
4158				(uint8_t *)lun->mode_pages.format_page;
4159			break;
4160		}
4161		case SMS_RIGID_DISK_PAGE: {
4162			struct scsi_rigid_disk_page *rigid_disk_page;
4163			uint32_t sectors_per_cylinder;
4164			uint64_t cylinders;
4165#ifndef	__XSCALE__
4166			int shift;
4167#endif /* !__XSCALE__ */
4168
4169			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4170				panic("invalid subpage value %d",
4171				      page_index->subpage);
4172
4173			/*
4174			 * Rotation rate and sectors per track are set
4175			 * above.  We calculate the cylinders here based on
4176			 * capacity.  Due to the number of heads and
4177			 * sectors per track we're using, smaller arrays
4178			 * may turn out to have 0 cylinders.  Linux and
4179			 * FreeBSD don't pay attention to these mode pages
4180			 * to figure out capacity, but Solaris does.  It
4181			 * seems to deal with 0 cylinders just fine, and
4182			 * works out a fake geometry based on the capacity.
4183			 */
4184			memcpy(&lun->mode_pages.rigid_disk_page[
4185			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4186			       sizeof(rigid_disk_page_default));
4187			memcpy(&lun->mode_pages.rigid_disk_page[
4188			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4189			       sizeof(rigid_disk_page_changeable));
4190			memcpy(&lun->mode_pages.rigid_disk_page[
4191			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4192			       sizeof(rigid_disk_page_default));
4193			memcpy(&lun->mode_pages.rigid_disk_page[
4194			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4195			       sizeof(rigid_disk_page_default));
4196
4197			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4198				CTL_DEFAULT_HEADS;
4199
4200			/*
4201			 * The divide method here will be more accurate,
4202			 * probably, but results in floating point being
4203			 * used in the kernel on i386 (__udivdi3()).  On the
4204			 * XScale, though, __udivdi3() is implemented in
4205			 * software.
4206			 *
4207			 * The shift method for cylinder calculation is
4208			 * accurate if sectors_per_cylinder is a power of
4209			 * 2.  Otherwise it might be slightly off -- you
4210			 * might have a bit of a truncation problem.
4211			 */
4212#ifdef	__XSCALE__
4213			cylinders = (lun->be_lun->maxlba + 1) /
4214				sectors_per_cylinder;
4215#else
4216			for (shift = 31; shift > 0; shift--) {
4217				if (sectors_per_cylinder & (1 << shift))
4218					break;
4219			}
4220			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4221#endif
4222
4223			/*
4224			 * We've basically got 3 bytes, or 24 bits for the
4225			 * cylinder size in the mode page.  If we're over,
4226			 * just round down to 2^24.
4227			 */
4228			if (cylinders > 0xffffff)
4229				cylinders = 0xffffff;
4230
4231			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4232				CTL_PAGE_CURRENT];
4233			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4234
4235			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4236				CTL_PAGE_DEFAULT];
4237			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4238
4239			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4240				CTL_PAGE_SAVED];
4241			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4242
4243			page_index->page_data =
4244				(uint8_t *)lun->mode_pages.rigid_disk_page;
4245			break;
4246		}
4247		case SMS_CACHING_PAGE: {
4248			struct scsi_caching_page *caching_page;
4249
4250			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4251				panic("invalid subpage value %d",
4252				      page_index->subpage);
4253			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4254			       &caching_page_default,
4255			       sizeof(caching_page_default));
4256			memcpy(&lun->mode_pages.caching_page[
4257			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4258			       sizeof(caching_page_changeable));
4259			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4260			       &caching_page_default,
4261			       sizeof(caching_page_default));
4262			caching_page = &lun->mode_pages.caching_page[
4263			    CTL_PAGE_SAVED];
4264			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4265			if (value != NULL && strcmp(value, "off") == 0)
4266				caching_page->flags1 &= ~SCP_WCE;
4267			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4268			if (value != NULL && strcmp(value, "off") == 0)
4269				caching_page->flags1 |= SCP_RCD;
4270			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4271			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4272			       sizeof(caching_page_default));
4273			page_index->page_data =
4274				(uint8_t *)lun->mode_pages.caching_page;
4275			break;
4276		}
4277		case SMS_CONTROL_MODE_PAGE: {
4278			struct scsi_control_page *control_page;
4279
4280			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4281				panic("invalid subpage value %d",
4282				      page_index->subpage);
4283
4284			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4285			       &control_page_default,
4286			       sizeof(control_page_default));
4287			memcpy(&lun->mode_pages.control_page[
4288			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4289			       sizeof(control_page_changeable));
4290			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4291			       &control_page_default,
4292			       sizeof(control_page_default));
4293			control_page = &lun->mode_pages.control_page[
4294			    CTL_PAGE_SAVED];
4295			value = ctl_get_opt(&lun->be_lun->options, "reordering");
4296			if (value != NULL && strcmp(value, "unrestricted") == 0) {
4297				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4298				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4299			}
4300			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4301			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4302			       sizeof(control_page_default));
4303			page_index->page_data =
4304				(uint8_t *)lun->mode_pages.control_page;
4305			break;
4306
4307		}
4308		case SMS_VENDOR_SPECIFIC_PAGE:{
4309			switch (page_index->subpage) {
4310			case PWR_SUBPAGE_CODE: {
4311				struct copan_power_subpage *current_page,
4312							   *saved_page;
4313
4314				memcpy(&lun->mode_pages.power_subpage[
4315				       CTL_PAGE_CURRENT],
4316				       &power_page_default,
4317				       sizeof(power_page_default));
4318				memcpy(&lun->mode_pages.power_subpage[
4319				       CTL_PAGE_CHANGEABLE],
4320				       &power_page_changeable,
4321				       sizeof(power_page_changeable));
4322				memcpy(&lun->mode_pages.power_subpage[
4323				       CTL_PAGE_DEFAULT],
4324				       &power_page_default,
4325				       sizeof(power_page_default));
4326				memcpy(&lun->mode_pages.power_subpage[
4327				       CTL_PAGE_SAVED],
4328				       &power_page_default,
4329				       sizeof(power_page_default));
4330				page_index->page_data =
4331				    (uint8_t *)lun->mode_pages.power_subpage;
4332
4333				current_page = (struct copan_power_subpage *)
4334					(page_index->page_data +
4335					 (page_index->page_len *
4336					  CTL_PAGE_CURRENT));
4337			        saved_page = (struct copan_power_subpage *)
4338				        (page_index->page_data +
4339					 (page_index->page_len *
4340					  CTL_PAGE_SAVED));
4341				break;
4342			}
4343			case APS_SUBPAGE_CODE: {
4344				struct copan_aps_subpage *current_page,
4345							 *saved_page;
4346
4347				// This gets set multiple times but
4348				// it should always be the same. It's
4349				// only done during init so who cares.
4350				index_to_aps_page = i;
4351
4352				memcpy(&lun->mode_pages.aps_subpage[
4353				       CTL_PAGE_CURRENT],
4354				       &aps_page_default,
4355				       sizeof(aps_page_default));
4356				memcpy(&lun->mode_pages.aps_subpage[
4357				       CTL_PAGE_CHANGEABLE],
4358				       &aps_page_changeable,
4359				       sizeof(aps_page_changeable));
4360				memcpy(&lun->mode_pages.aps_subpage[
4361				       CTL_PAGE_DEFAULT],
4362				       &aps_page_default,
4363				       sizeof(aps_page_default));
4364				memcpy(&lun->mode_pages.aps_subpage[
4365				       CTL_PAGE_SAVED],
4366				       &aps_page_default,
4367				       sizeof(aps_page_default));
4368				page_index->page_data =
4369					(uint8_t *)lun->mode_pages.aps_subpage;
4370
4371				current_page = (struct copan_aps_subpage *)
4372					(page_index->page_data +
4373					 (page_index->page_len *
4374					  CTL_PAGE_CURRENT));
4375				saved_page = (struct copan_aps_subpage *)
4376					(page_index->page_data +
4377					 (page_index->page_len *
4378					  CTL_PAGE_SAVED));
4379				break;
4380			}
4381			case DBGCNF_SUBPAGE_CODE: {
4382				struct copan_debugconf_subpage *current_page,
4383							       *saved_page;
4384
4385				memcpy(&lun->mode_pages.debugconf_subpage[
4386				       CTL_PAGE_CURRENT],
4387				       &debugconf_page_default,
4388				       sizeof(debugconf_page_default));
4389				memcpy(&lun->mode_pages.debugconf_subpage[
4390				       CTL_PAGE_CHANGEABLE],
4391				       &debugconf_page_changeable,
4392				       sizeof(debugconf_page_changeable));
4393				memcpy(&lun->mode_pages.debugconf_subpage[
4394				       CTL_PAGE_DEFAULT],
4395				       &debugconf_page_default,
4396				       sizeof(debugconf_page_default));
4397				memcpy(&lun->mode_pages.debugconf_subpage[
4398				       CTL_PAGE_SAVED],
4399				       &debugconf_page_default,
4400				       sizeof(debugconf_page_default));
4401				page_index->page_data =
4402					(uint8_t *)lun->mode_pages.debugconf_subpage;
4403
4404				current_page = (struct copan_debugconf_subpage *)
4405					(page_index->page_data +
4406					 (page_index->page_len *
4407					  CTL_PAGE_CURRENT));
4408				saved_page = (struct copan_debugconf_subpage *)
4409					(page_index->page_data +
4410					 (page_index->page_len *
4411					  CTL_PAGE_SAVED));
4412				break;
4413			}
4414			default:
4415				panic("invalid subpage value %d",
4416				      page_index->subpage);
4417				break;
4418			}
4419   			break;
4420		}
4421		default:
4422			panic("invalid page value %d",
4423			      page_index->page_code & SMPH_PC_MASK);
4424			break;
4425    	}
4426	}
4427
4428	return (CTL_RETVAL_COMPLETE);
4429}
4430
4431/*
4432 * LUN allocation.
4433 *
4434 * Requirements:
4435 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4436 *   wants us to allocate the LUN and he can block.
4437 * - ctl_softc is always set
4438 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4439 *
4440 * Returns 0 for success, non-zero (errno) for failure.
4441 */
4442static int
4443ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4444	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4445{
4446	struct ctl_lun *nlun, *lun;
4447	struct ctl_port *port;
4448	struct scsi_vpd_id_descriptor *desc;
4449	struct scsi_vpd_id_t10 *t10id;
4450	const char *eui, *naa, *scsiname, *vendor;
4451	int lun_number, i, lun_malloced;
4452	int devidlen, idlen1, idlen2 = 0, len;
4453
4454	if (be_lun == NULL)
4455		return (EINVAL);
4456
4457	/*
4458	 * We currently only support Direct Access or Processor LUN types.
4459	 */
4460	switch (be_lun->lun_type) {
4461	case T_DIRECT:
4462		break;
4463	case T_PROCESSOR:
4464		break;
4465	case T_SEQUENTIAL:
4466	case T_CHANGER:
4467	default:
4468		be_lun->lun_config_status(be_lun->be_lun,
4469					  CTL_LUN_CONFIG_FAILURE);
4470		break;
4471	}
4472	if (ctl_lun == NULL) {
4473		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4474		lun_malloced = 1;
4475	} else {
4476		lun_malloced = 0;
4477		lun = ctl_lun;
4478	}
4479
4480	memset(lun, 0, sizeof(*lun));
4481	if (lun_malloced)
4482		lun->flags = CTL_LUN_MALLOCED;
4483
4484	/* Generate LUN ID. */
4485	devidlen = max(CTL_DEVID_MIN_LEN,
4486	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4487	idlen1 = sizeof(*t10id) + devidlen;
4488	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4489	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4490	if (scsiname != NULL) {
4491		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4492		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4493	}
4494	eui = ctl_get_opt(&be_lun->options, "eui");
4495	if (eui != NULL) {
4496		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4497	}
4498	naa = ctl_get_opt(&be_lun->options, "naa");
4499	if (naa != NULL) {
4500		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4501	}
4502	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4503	    M_CTL, M_WAITOK | M_ZERO);
4504	lun->lun_devid->len = len;
4505	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4506	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4507	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4508	desc->length = idlen1;
4509	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4510	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4511	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4512		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4513	} else {
4514		strncpy(t10id->vendor, vendor,
4515		    min(sizeof(t10id->vendor), strlen(vendor)));
4516	}
4517	strncpy((char *)t10id->vendor_spec_id,
4518	    (char *)be_lun->device_id, devidlen);
4519	if (scsiname != NULL) {
4520		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4521		    desc->length);
4522		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4523		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4524		    SVPD_ID_TYPE_SCSI_NAME;
4525		desc->length = idlen2;
4526		strlcpy(desc->identifier, scsiname, idlen2);
4527	}
4528	if (eui != NULL) {
4529		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4530		    desc->length);
4531		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4532		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4533		    SVPD_ID_TYPE_EUI64;
4534		desc->length = 8;
4535		scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4536	}
4537	if (naa != NULL) {
4538		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4539		    desc->length);
4540		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4541		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4542		    SVPD_ID_TYPE_NAA;
4543		desc->length = 8;
4544		scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4545	}
4546
4547	mtx_lock(&ctl_softc->ctl_lock);
4548	/*
4549	 * See if the caller requested a particular LUN number.  If so, see
4550	 * if it is available.  Otherwise, allocate the first available LUN.
4551	 */
4552	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4553		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4554		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4555			mtx_unlock(&ctl_softc->ctl_lock);
4556			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4557				printf("ctl: requested LUN ID %d is higher "
4558				       "than CTL_MAX_LUNS - 1 (%d)\n",
4559				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4560			} else {
4561				/*
4562				 * XXX KDM return an error, or just assign
4563				 * another LUN ID in this case??
4564				 */
4565				printf("ctl: requested LUN ID %d is already "
4566				       "in use\n", be_lun->req_lun_id);
4567			}
4568			if (lun->flags & CTL_LUN_MALLOCED)
4569				free(lun, M_CTL);
4570			be_lun->lun_config_status(be_lun->be_lun,
4571						  CTL_LUN_CONFIG_FAILURE);
4572			return (ENOSPC);
4573		}
4574		lun_number = be_lun->req_lun_id;
4575	} else {
4576		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4577		if (lun_number == -1) {
4578			mtx_unlock(&ctl_softc->ctl_lock);
4579			printf("ctl: can't allocate LUN on target %ju, out of "
4580			       "LUNs\n", (uintmax_t)target_id.id);
4581			if (lun->flags & CTL_LUN_MALLOCED)
4582				free(lun, M_CTL);
4583			be_lun->lun_config_status(be_lun->be_lun,
4584						  CTL_LUN_CONFIG_FAILURE);
4585			return (ENOSPC);
4586		}
4587	}
4588	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4589
4590	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4591	lun->target = target_id;
4592	lun->lun = lun_number;
4593	lun->be_lun = be_lun;
4594	/*
4595	 * The processor LUN is always enabled.  Disk LUNs come on line
4596	 * disabled, and must be enabled by the backend.
4597	 */
4598	lun->flags |= CTL_LUN_DISABLED;
4599	lun->backend = be_lun->be;
4600	be_lun->ctl_lun = lun;
4601	be_lun->lun_id = lun_number;
4602	atomic_add_int(&be_lun->be->num_luns, 1);
4603	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4604		lun->flags |= CTL_LUN_STOPPED;
4605
4606	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4607		lun->flags |= CTL_LUN_INOPERABLE;
4608
4609	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4610		lun->flags |= CTL_LUN_PRIMARY_SC;
4611
4612	lun->ctl_softc = ctl_softc;
4613	TAILQ_INIT(&lun->ooa_queue);
4614	TAILQ_INIT(&lun->blocked_queue);
4615	STAILQ_INIT(&lun->error_list);
4616	ctl_tpc_lun_init(lun);
4617
4618	/*
4619	 * Initialize the mode page index.
4620	 */
4621	ctl_init_page_index(lun);
4622
4623	/*
4624	 * Set the poweron UA for all initiators on this LUN only.
4625	 */
4626	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4627		lun->pending_ua[i] = CTL_UA_POWERON;
4628
4629	/*
4630	 * Now, before we insert this lun on the lun list, set the lun
4631	 * inventory changed UA for all other luns.
4632	 */
4633	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4634		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4635			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4636		}
4637	}
4638
4639	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4640
4641	ctl_softc->ctl_luns[lun_number] = lun;
4642
4643	ctl_softc->num_luns++;
4644
4645	/* Setup statistics gathering */
4646	lun->stats.device_type = be_lun->lun_type;
4647	lun->stats.lun_number = lun_number;
4648	if (lun->stats.device_type == T_DIRECT)
4649		lun->stats.blocksize = be_lun->blocksize;
4650	else
4651		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4652	for (i = 0;i < CTL_MAX_PORTS;i++)
4653		lun->stats.ports[i].targ_port = i;
4654
4655	mtx_unlock(&ctl_softc->ctl_lock);
4656
4657	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4658
4659	/*
4660	 * Run through each registered FETD and bring it online if it isn't
4661	 * already.  Enable the target ID if it hasn't been enabled, and
4662	 * enable this particular LUN.
4663	 */
4664	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4665		int retval;
4666
4667		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4668		if (retval != 0) {
4669			printf("ctl_alloc_lun: FETD %s port %d returned error "
4670			       "%d for lun_enable on target %ju lun %d\n",
4671			       port->port_name, port->targ_port, retval,
4672			       (uintmax_t)target_id.id, lun_number);
4673		} else
4674			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4675	}
4676	return (0);
4677}
4678
4679/*
4680 * Delete a LUN.
4681 * Assumptions:
4682 * - LUN has already been marked invalid and any pending I/O has been taken
4683 *   care of.
4684 */
4685static int
4686ctl_free_lun(struct ctl_lun *lun)
4687{
4688	struct ctl_softc *softc;
4689#if 0
4690	struct ctl_port *port;
4691#endif
4692	struct ctl_lun *nlun;
4693	int i;
4694
4695	softc = lun->ctl_softc;
4696
4697	mtx_assert(&softc->ctl_lock, MA_OWNED);
4698
4699	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4700
4701	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4702
4703	softc->ctl_luns[lun->lun] = NULL;
4704
4705	if (!TAILQ_EMPTY(&lun->ooa_queue))
4706		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4707
4708	softc->num_luns--;
4709
4710	/*
4711	 * XXX KDM this scheme only works for a single target/multiple LUN
4712	 * setup.  It needs to be revamped for a multiple target scheme.
4713	 *
4714	 * XXX KDM this results in port->lun_disable() getting called twice,
4715	 * once when ctl_disable_lun() is called, and a second time here.
4716	 * We really need to re-think the LUN disable semantics.  There
4717	 * should probably be several steps/levels to LUN removal:
4718	 *  - disable
4719	 *  - invalidate
4720	 *  - free
4721 	 *
4722	 * Right now we only have a disable method when communicating to
4723	 * the front end ports, at least for individual LUNs.
4724	 */
4725#if 0
4726	STAILQ_FOREACH(port, &softc->port_list, links) {
4727		int retval;
4728
4729		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4730					 lun->lun);
4731		if (retval != 0) {
4732			printf("ctl_free_lun: FETD %s port %d returned error "
4733			       "%d for lun_disable on target %ju lun %jd\n",
4734			       port->port_name, port->targ_port, retval,
4735			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4736		}
4737
4738		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4739			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4740
4741			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4742			if (retval != 0) {
4743				printf("ctl_free_lun: FETD %s port %d "
4744				       "returned error %d for targ_disable on "
4745				       "target %ju\n", port->port_name,
4746				       port->targ_port, retval,
4747				       (uintmax_t)lun->target.id);
4748			} else
4749				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4750
4751			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4752				continue;
4753
4754#if 0
4755			port->port_offline(port->onoff_arg);
4756			port->status &= ~CTL_PORT_STATUS_ONLINE;
4757#endif
4758		}
4759	}
4760#endif
4761
4762	/*
4763	 * Tell the backend to free resources, if this LUN has a backend.
4764	 */
4765	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4766	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4767
4768	ctl_tpc_lun_shutdown(lun);
4769	mtx_destroy(&lun->lun_lock);
4770	free(lun->lun_devid, M_CTL);
4771	if (lun->flags & CTL_LUN_MALLOCED)
4772		free(lun, M_CTL);
4773
4774	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4775		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4776			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4777		}
4778	}
4779
4780	return (0);
4781}
4782
4783static void
4784ctl_create_lun(struct ctl_be_lun *be_lun)
4785{
4786	struct ctl_softc *ctl_softc;
4787
4788	ctl_softc = control_softc;
4789
4790	/*
4791	 * ctl_alloc_lun() should handle all potential failure cases.
4792	 */
4793	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4794}
4795
4796int
4797ctl_add_lun(struct ctl_be_lun *be_lun)
4798{
4799	struct ctl_softc *ctl_softc = control_softc;
4800
4801	mtx_lock(&ctl_softc->ctl_lock);
4802	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4803	mtx_unlock(&ctl_softc->ctl_lock);
4804	wakeup(&ctl_softc->pending_lun_queue);
4805
4806	return (0);
4807}
4808
4809int
4810ctl_enable_lun(struct ctl_be_lun *be_lun)
4811{
4812	struct ctl_softc *ctl_softc;
4813	struct ctl_port *port, *nport;
4814	struct ctl_lun *lun;
4815	int retval;
4816
4817	ctl_softc = control_softc;
4818
4819	lun = (struct ctl_lun *)be_lun->ctl_lun;
4820
4821	mtx_lock(&ctl_softc->ctl_lock);
4822	mtx_lock(&lun->lun_lock);
4823	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4824		/*
4825		 * eh?  Why did we get called if the LUN is already
4826		 * enabled?
4827		 */
4828		mtx_unlock(&lun->lun_lock);
4829		mtx_unlock(&ctl_softc->ctl_lock);
4830		return (0);
4831	}
4832	lun->flags &= ~CTL_LUN_DISABLED;
4833	mtx_unlock(&lun->lun_lock);
4834
4835	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4836		nport = STAILQ_NEXT(port, links);
4837
4838		/*
4839		 * Drop the lock while we call the FETD's enable routine.
4840		 * This can lead to a callback into CTL (at least in the
4841		 * case of the internal initiator frontend.
4842		 */
4843		mtx_unlock(&ctl_softc->ctl_lock);
4844		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4845		mtx_lock(&ctl_softc->ctl_lock);
4846		if (retval != 0) {
4847			printf("%s: FETD %s port %d returned error "
4848			       "%d for lun_enable on target %ju lun %jd\n",
4849			       __func__, port->port_name, port->targ_port, retval,
4850			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4851		}
4852#if 0
4853		 else {
4854            /* NOTE:  TODO:  why does lun enable affect port status? */
4855			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4856		}
4857#endif
4858	}
4859
4860	mtx_unlock(&ctl_softc->ctl_lock);
4861
4862	return (0);
4863}
4864
4865int
4866ctl_disable_lun(struct ctl_be_lun *be_lun)
4867{
4868	struct ctl_softc *ctl_softc;
4869	struct ctl_port *port;
4870	struct ctl_lun *lun;
4871	int retval;
4872
4873	ctl_softc = control_softc;
4874
4875	lun = (struct ctl_lun *)be_lun->ctl_lun;
4876
4877	mtx_lock(&ctl_softc->ctl_lock);
4878	mtx_lock(&lun->lun_lock);
4879	if (lun->flags & CTL_LUN_DISABLED) {
4880		mtx_unlock(&lun->lun_lock);
4881		mtx_unlock(&ctl_softc->ctl_lock);
4882		return (0);
4883	}
4884	lun->flags |= CTL_LUN_DISABLED;
4885	mtx_unlock(&lun->lun_lock);
4886
4887	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4888		mtx_unlock(&ctl_softc->ctl_lock);
4889		/*
4890		 * Drop the lock before we call the frontend's disable
4891		 * routine, to avoid lock order reversals.
4892		 *
4893		 * XXX KDM what happens if the frontend list changes while
4894		 * we're traversing it?  It's unlikely, but should be handled.
4895		 */
4896		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4897					 lun->lun);
4898		mtx_lock(&ctl_softc->ctl_lock);
4899		if (retval != 0) {
4900			printf("ctl_alloc_lun: FETD %s port %d returned error "
4901			       "%d for lun_disable on target %ju lun %jd\n",
4902			       port->port_name, port->targ_port, retval,
4903			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4904		}
4905	}
4906
4907	mtx_unlock(&ctl_softc->ctl_lock);
4908
4909	return (0);
4910}
4911
4912int
4913ctl_start_lun(struct ctl_be_lun *be_lun)
4914{
4915	struct ctl_softc *ctl_softc;
4916	struct ctl_lun *lun;
4917
4918	ctl_softc = control_softc;
4919
4920	lun = (struct ctl_lun *)be_lun->ctl_lun;
4921
4922	mtx_lock(&lun->lun_lock);
4923	lun->flags &= ~CTL_LUN_STOPPED;
4924	mtx_unlock(&lun->lun_lock);
4925
4926	return (0);
4927}
4928
4929int
4930ctl_stop_lun(struct ctl_be_lun *be_lun)
4931{
4932	struct ctl_softc *ctl_softc;
4933	struct ctl_lun *lun;
4934
4935	ctl_softc = control_softc;
4936
4937	lun = (struct ctl_lun *)be_lun->ctl_lun;
4938
4939	mtx_lock(&lun->lun_lock);
4940	lun->flags |= CTL_LUN_STOPPED;
4941	mtx_unlock(&lun->lun_lock);
4942
4943	return (0);
4944}
4945
4946int
4947ctl_lun_offline(struct ctl_be_lun *be_lun)
4948{
4949	struct ctl_softc *ctl_softc;
4950	struct ctl_lun *lun;
4951
4952	ctl_softc = control_softc;
4953
4954	lun = (struct ctl_lun *)be_lun->ctl_lun;
4955
4956	mtx_lock(&lun->lun_lock);
4957	lun->flags |= CTL_LUN_OFFLINE;
4958	mtx_unlock(&lun->lun_lock);
4959
4960	return (0);
4961}
4962
4963int
4964ctl_lun_online(struct ctl_be_lun *be_lun)
4965{
4966	struct ctl_softc *ctl_softc;
4967	struct ctl_lun *lun;
4968
4969	ctl_softc = control_softc;
4970
4971	lun = (struct ctl_lun *)be_lun->ctl_lun;
4972
4973	mtx_lock(&lun->lun_lock);
4974	lun->flags &= ~CTL_LUN_OFFLINE;
4975	mtx_unlock(&lun->lun_lock);
4976
4977	return (0);
4978}
4979
4980int
4981ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4982{
4983	struct ctl_softc *ctl_softc;
4984	struct ctl_lun *lun;
4985
4986	ctl_softc = control_softc;
4987
4988	lun = (struct ctl_lun *)be_lun->ctl_lun;
4989
4990	mtx_lock(&lun->lun_lock);
4991
4992	/*
4993	 * The LUN needs to be disabled before it can be marked invalid.
4994	 */
4995	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4996		mtx_unlock(&lun->lun_lock);
4997		return (-1);
4998	}
4999	/*
5000	 * Mark the LUN invalid.
5001	 */
5002	lun->flags |= CTL_LUN_INVALID;
5003
5004	/*
5005	 * If there is nothing in the OOA queue, go ahead and free the LUN.
5006	 * If we have something in the OOA queue, we'll free it when the
5007	 * last I/O completes.
5008	 */
5009	if (TAILQ_EMPTY(&lun->ooa_queue)) {
5010		mtx_unlock(&lun->lun_lock);
5011		mtx_lock(&ctl_softc->ctl_lock);
5012		ctl_free_lun(lun);
5013		mtx_unlock(&ctl_softc->ctl_lock);
5014	} else
5015		mtx_unlock(&lun->lun_lock);
5016
5017	return (0);
5018}
5019
5020int
5021ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5022{
5023	struct ctl_softc *ctl_softc;
5024	struct ctl_lun *lun;
5025
5026	ctl_softc = control_softc;
5027	lun = (struct ctl_lun *)be_lun->ctl_lun;
5028
5029	mtx_lock(&lun->lun_lock);
5030	lun->flags |= CTL_LUN_INOPERABLE;
5031	mtx_unlock(&lun->lun_lock);
5032
5033	return (0);
5034}
5035
5036int
5037ctl_lun_operable(struct ctl_be_lun *be_lun)
5038{
5039	struct ctl_softc *ctl_softc;
5040	struct ctl_lun *lun;
5041
5042	ctl_softc = control_softc;
5043	lun = (struct ctl_lun *)be_lun->ctl_lun;
5044
5045	mtx_lock(&lun->lun_lock);
5046	lun->flags &= ~CTL_LUN_INOPERABLE;
5047	mtx_unlock(&lun->lun_lock);
5048
5049	return (0);
5050}
5051
5052int
5053ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5054		   int lock)
5055{
5056	struct ctl_softc *softc;
5057	struct ctl_lun *lun;
5058	struct copan_aps_subpage *current_sp;
5059	struct ctl_page_index *page_index;
5060	int i;
5061
5062	softc = control_softc;
5063
5064	mtx_lock(&softc->ctl_lock);
5065
5066	lun = (struct ctl_lun *)be_lun->ctl_lun;
5067	mtx_lock(&lun->lun_lock);
5068
5069	page_index = NULL;
5070	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5071		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5072		     APS_PAGE_CODE)
5073			continue;
5074
5075		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5076			continue;
5077		page_index = &lun->mode_pages.index[i];
5078	}
5079
5080	if (page_index == NULL) {
5081		mtx_unlock(&lun->lun_lock);
5082		mtx_unlock(&softc->ctl_lock);
5083		printf("%s: APS subpage not found for lun %ju!\n", __func__,
5084		       (uintmax_t)lun->lun);
5085		return (1);
5086	}
5087#if 0
5088	if ((softc->aps_locked_lun != 0)
5089	 && (softc->aps_locked_lun != lun->lun)) {
5090		printf("%s: attempt to lock LUN %llu when %llu is already "
5091		       "locked\n");
5092		mtx_unlock(&lun->lun_lock);
5093		mtx_unlock(&softc->ctl_lock);
5094		return (1);
5095	}
5096#endif
5097
5098	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5099		(page_index->page_len * CTL_PAGE_CURRENT));
5100
5101	if (lock != 0) {
5102		current_sp->lock_active = APS_LOCK_ACTIVE;
5103		softc->aps_locked_lun = lun->lun;
5104	} else {
5105		current_sp->lock_active = 0;
5106		softc->aps_locked_lun = 0;
5107	}
5108
5109
5110	/*
5111	 * If we're in HA mode, try to send the lock message to the other
5112	 * side.
5113	 */
5114	if (ctl_is_single == 0) {
5115		int isc_retval;
5116		union ctl_ha_msg lock_msg;
5117
5118		lock_msg.hdr.nexus = *nexus;
5119		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5120		if (lock != 0)
5121			lock_msg.aps.lock_flag = 1;
5122		else
5123			lock_msg.aps.lock_flag = 0;
5124		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5125					 sizeof(lock_msg), 0);
5126		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5127			printf("%s: APS (lock=%d) error returned from "
5128			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5129			mtx_unlock(&lun->lun_lock);
5130			mtx_unlock(&softc->ctl_lock);
5131			return (1);
5132		}
5133	}
5134
5135	mtx_unlock(&lun->lun_lock);
5136	mtx_unlock(&softc->ctl_lock);
5137
5138	return (0);
5139}
5140
5141void
5142ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5143{
5144	struct ctl_lun *lun;
5145	struct ctl_softc *softc;
5146	int i;
5147
5148	softc = control_softc;
5149
5150	lun = (struct ctl_lun *)be_lun->ctl_lun;
5151
5152	mtx_lock(&lun->lun_lock);
5153
5154	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5155		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5156
5157	mtx_unlock(&lun->lun_lock);
5158}
5159
5160/*
5161 * Backend "memory move is complete" callback for requests that never
5162 * make it down to say RAIDCore's configuration code.
5163 */
5164int
5165ctl_config_move_done(union ctl_io *io)
5166{
5167	int retval;
5168
5169	retval = CTL_RETVAL_COMPLETE;
5170
5171
5172	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5173	/*
5174	 * XXX KDM this shouldn't happen, but what if it does?
5175	 */
5176	if (io->io_hdr.io_type != CTL_IO_SCSI)
5177		panic("I/O type isn't CTL_IO_SCSI!");
5178
5179	if ((io->io_hdr.port_status == 0)
5180	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5181	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5182		io->io_hdr.status = CTL_SUCCESS;
5183	else if ((io->io_hdr.port_status != 0)
5184	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5185	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5186		/*
5187		 * For hardware error sense keys, the sense key
5188		 * specific value is defined to be a retry count,
5189		 * but we use it to pass back an internal FETD
5190		 * error code.  XXX KDM  Hopefully the FETD is only
5191		 * using 16 bits for an error code, since that's
5192		 * all the space we have in the sks field.
5193		 */
5194		ctl_set_internal_failure(&io->scsiio,
5195					 /*sks_valid*/ 1,
5196					 /*retry_count*/
5197					 io->io_hdr.port_status);
5198		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5199			free(io->scsiio.kern_data_ptr, M_CTL);
5200		ctl_done(io);
5201		goto bailout;
5202	}
5203
5204	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5205	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5206	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5207		/*
5208		 * XXX KDM just assuming a single pointer here, and not a
5209		 * S/G list.  If we start using S/G lists for config data,
5210		 * we'll need to know how to clean them up here as well.
5211		 */
5212		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5213			free(io->scsiio.kern_data_ptr, M_CTL);
5214		/* Hopefully the user has already set the status... */
5215		ctl_done(io);
5216	} else {
5217		/*
5218		 * XXX KDM now we need to continue data movement.  Some
5219		 * options:
5220		 * - call ctl_scsiio() again?  We don't do this for data
5221		 *   writes, because for those at least we know ahead of
5222		 *   time where the write will go and how long it is.  For
5223		 *   config writes, though, that information is largely
5224		 *   contained within the write itself, thus we need to
5225		 *   parse out the data again.
5226		 *
5227		 * - Call some other function once the data is in?
5228		 */
5229
5230		/*
5231		 * XXX KDM call ctl_scsiio() again for now, and check flag
5232		 * bits to see whether we're allocated or not.
5233		 */
5234		retval = ctl_scsiio(&io->scsiio);
5235	}
5236bailout:
5237	return (retval);
5238}
5239
5240/*
5241 * This gets called by a backend driver when it is done with a
5242 * data_submit method.
5243 */
5244void
5245ctl_data_submit_done(union ctl_io *io)
5246{
5247	/*
5248	 * If the IO_CONT flag is set, we need to call the supplied
5249	 * function to continue processing the I/O, instead of completing
5250	 * the I/O just yet.
5251	 *
5252	 * If there is an error, though, we don't want to keep processing.
5253	 * Instead, just send status back to the initiator.
5254	 */
5255	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5256	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5257	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5258	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5259		io->scsiio.io_cont(io);
5260		return;
5261	}
5262	ctl_done(io);
5263}
5264
5265/*
5266 * This gets called by a backend driver when it is done with a
5267 * configuration write.
5268 */
5269void
5270ctl_config_write_done(union ctl_io *io)
5271{
5272	uint8_t *buf;
5273
5274	/*
5275	 * If the IO_CONT flag is set, we need to call the supplied
5276	 * function to continue processing the I/O, instead of completing
5277	 * the I/O just yet.
5278	 *
5279	 * If there is an error, though, we don't want to keep processing.
5280	 * Instead, just send status back to the initiator.
5281	 */
5282	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5283	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5284	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5285		io->scsiio.io_cont(io);
5286		return;
5287	}
5288	/*
5289	 * Since a configuration write can be done for commands that actually
5290	 * have data allocated, like write buffer, and commands that have
5291	 * no data, like start/stop unit, we need to check here.
5292	 */
5293	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5294		buf = io->scsiio.kern_data_ptr;
5295	else
5296		buf = NULL;
5297	ctl_done(io);
5298	if (buf)
5299		free(buf, M_CTL);
5300}
5301
5302/*
5303 * SCSI release command.
5304 */
5305int
5306ctl_scsi_release(struct ctl_scsiio *ctsio)
5307{
5308	int length, longid, thirdparty_id, resv_id;
5309	struct ctl_softc *ctl_softc;
5310	struct ctl_lun *lun;
5311
5312	length = 0;
5313	resv_id = 0;
5314
5315	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5316
5317	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5318	ctl_softc = control_softc;
5319
5320	switch (ctsio->cdb[0]) {
5321	case RELEASE_10: {
5322		struct scsi_release_10 *cdb;
5323
5324		cdb = (struct scsi_release_10 *)ctsio->cdb;
5325
5326		if (cdb->byte2 & SR10_LONGID)
5327			longid = 1;
5328		else
5329			thirdparty_id = cdb->thirdparty_id;
5330
5331		resv_id = cdb->resv_id;
5332		length = scsi_2btoul(cdb->length);
5333		break;
5334	}
5335	}
5336
5337
5338	/*
5339	 * XXX KDM right now, we only support LUN reservation.  We don't
5340	 * support 3rd party reservations, or extent reservations, which
5341	 * might actually need the parameter list.  If we've gotten this
5342	 * far, we've got a LUN reservation.  Anything else got kicked out
5343	 * above.  So, according to SPC, ignore the length.
5344	 */
5345	length = 0;
5346
5347	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5348	 && (length > 0)) {
5349		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5350		ctsio->kern_data_len = length;
5351		ctsio->kern_total_len = length;
5352		ctsio->kern_data_resid = 0;
5353		ctsio->kern_rel_offset = 0;
5354		ctsio->kern_sg_entries = 0;
5355		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5356		ctsio->be_move_done = ctl_config_move_done;
5357		ctl_datamove((union ctl_io *)ctsio);
5358
5359		return (CTL_RETVAL_COMPLETE);
5360	}
5361
5362	if (length > 0)
5363		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5364
5365	mtx_lock(&lun->lun_lock);
5366
5367	/*
5368	 * According to SPC, it is not an error for an intiator to attempt
5369	 * to release a reservation on a LUN that isn't reserved, or that
5370	 * is reserved by another initiator.  The reservation can only be
5371	 * released, though, by the initiator who made it or by one of
5372	 * several reset type events.
5373	 */
5374	if (lun->flags & CTL_LUN_RESERVED) {
5375		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5376		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5377		 && (ctsio->io_hdr.nexus.targ_target.id ==
5378		     lun->rsv_nexus.targ_target.id)) {
5379			lun->flags &= ~CTL_LUN_RESERVED;
5380		}
5381	}
5382
5383	mtx_unlock(&lun->lun_lock);
5384
5385	ctsio->scsi_status = SCSI_STATUS_OK;
5386	ctsio->io_hdr.status = CTL_SUCCESS;
5387
5388	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5389		free(ctsio->kern_data_ptr, M_CTL);
5390		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5391	}
5392
5393	ctl_done((union ctl_io *)ctsio);
5394	return (CTL_RETVAL_COMPLETE);
5395}
5396
5397int
5398ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5399{
5400	int extent, thirdparty, longid;
5401	int resv_id, length;
5402	uint64_t thirdparty_id;
5403	struct ctl_softc *ctl_softc;
5404	struct ctl_lun *lun;
5405
5406	extent = 0;
5407	thirdparty = 0;
5408	longid = 0;
5409	resv_id = 0;
5410	length = 0;
5411	thirdparty_id = 0;
5412
5413	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5414
5415	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5416	ctl_softc = control_softc;
5417
5418	switch (ctsio->cdb[0]) {
5419	case RESERVE_10: {
5420		struct scsi_reserve_10 *cdb;
5421
5422		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5423
5424		if (cdb->byte2 & SR10_LONGID)
5425			longid = 1;
5426		else
5427			thirdparty_id = cdb->thirdparty_id;
5428
5429		resv_id = cdb->resv_id;
5430		length = scsi_2btoul(cdb->length);
5431		break;
5432	}
5433	}
5434
5435	/*
5436	 * XXX KDM right now, we only support LUN reservation.  We don't
5437	 * support 3rd party reservations, or extent reservations, which
5438	 * might actually need the parameter list.  If we've gotten this
5439	 * far, we've got a LUN reservation.  Anything else got kicked out
5440	 * above.  So, according to SPC, ignore the length.
5441	 */
5442	length = 0;
5443
5444	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5445	 && (length > 0)) {
5446		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5447		ctsio->kern_data_len = length;
5448		ctsio->kern_total_len = length;
5449		ctsio->kern_data_resid = 0;
5450		ctsio->kern_rel_offset = 0;
5451		ctsio->kern_sg_entries = 0;
5452		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5453		ctsio->be_move_done = ctl_config_move_done;
5454		ctl_datamove((union ctl_io *)ctsio);
5455
5456		return (CTL_RETVAL_COMPLETE);
5457	}
5458
5459	if (length > 0)
5460		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5461
5462	mtx_lock(&lun->lun_lock);
5463	if (lun->flags & CTL_LUN_RESERVED) {
5464		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5465		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5466		 || (ctsio->io_hdr.nexus.targ_target.id !=
5467		     lun->rsv_nexus.targ_target.id)) {
5468			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5469			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5470			goto bailout;
5471		}
5472	}
5473
5474	lun->flags |= CTL_LUN_RESERVED;
5475	lun->rsv_nexus = ctsio->io_hdr.nexus;
5476
5477	ctsio->scsi_status = SCSI_STATUS_OK;
5478	ctsio->io_hdr.status = CTL_SUCCESS;
5479
5480bailout:
5481	mtx_unlock(&lun->lun_lock);
5482
5483	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5484		free(ctsio->kern_data_ptr, M_CTL);
5485		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5486	}
5487
5488	ctl_done((union ctl_io *)ctsio);
5489	return (CTL_RETVAL_COMPLETE);
5490}
5491
5492int
5493ctl_start_stop(struct ctl_scsiio *ctsio)
5494{
5495	struct scsi_start_stop_unit *cdb;
5496	struct ctl_lun *lun;
5497	struct ctl_softc *ctl_softc;
5498	int retval;
5499
5500	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5501
5502	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5503	ctl_softc = control_softc;
5504	retval = 0;
5505
5506	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5507
5508	/*
5509	 * XXX KDM
5510	 * We don't support the immediate bit on a stop unit.  In order to
5511	 * do that, we would need to code up a way to know that a stop is
5512	 * pending, and hold off any new commands until it completes, one
5513	 * way or another.  Then we could accept or reject those commands
5514	 * depending on its status.  We would almost need to do the reverse
5515	 * of what we do below for an immediate start -- return the copy of
5516	 * the ctl_io to the FETD with status to send to the host (and to
5517	 * free the copy!) and then free the original I/O once the stop
5518	 * actually completes.  That way, the OOA queue mechanism can work
5519	 * to block commands that shouldn't proceed.  Another alternative
5520	 * would be to put the copy in the queue in place of the original,
5521	 * and return the original back to the caller.  That could be
5522	 * slightly safer..
5523	 */
5524	if ((cdb->byte2 & SSS_IMMED)
5525	 && ((cdb->how & SSS_START) == 0)) {
5526		ctl_set_invalid_field(ctsio,
5527				      /*sks_valid*/ 1,
5528				      /*command*/ 1,
5529				      /*field*/ 1,
5530				      /*bit_valid*/ 1,
5531				      /*bit*/ 0);
5532		ctl_done((union ctl_io *)ctsio);
5533		return (CTL_RETVAL_COMPLETE);
5534	}
5535
5536	if ((lun->flags & CTL_LUN_PR_RESERVED)
5537	 && ((cdb->how & SSS_START)==0)) {
5538		uint32_t residx;
5539
5540		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5541		if (!lun->per_res[residx].registered
5542		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5543
5544			ctl_set_reservation_conflict(ctsio);
5545			ctl_done((union ctl_io *)ctsio);
5546			return (CTL_RETVAL_COMPLETE);
5547		}
5548	}
5549
5550	/*
5551	 * If there is no backend on this device, we can't start or stop
5552	 * it.  In theory we shouldn't get any start/stop commands in the
5553	 * first place at this level if the LUN doesn't have a backend.
5554	 * That should get stopped by the command decode code.
5555	 */
5556	if (lun->backend == NULL) {
5557		ctl_set_invalid_opcode(ctsio);
5558		ctl_done((union ctl_io *)ctsio);
5559		return (CTL_RETVAL_COMPLETE);
5560	}
5561
5562	/*
5563	 * XXX KDM Copan-specific offline behavior.
5564	 * Figure out a reasonable way to port this?
5565	 */
5566#ifdef NEEDTOPORT
5567	mtx_lock(&lun->lun_lock);
5568
5569	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5570	 && (lun->flags & CTL_LUN_OFFLINE)) {
5571		/*
5572		 * If the LUN is offline, and the on/offline bit isn't set,
5573		 * reject the start or stop.  Otherwise, let it through.
5574		 */
5575		mtx_unlock(&lun->lun_lock);
5576		ctl_set_lun_not_ready(ctsio);
5577		ctl_done((union ctl_io *)ctsio);
5578	} else {
5579		mtx_unlock(&lun->lun_lock);
5580#endif /* NEEDTOPORT */
5581		/*
5582		 * This could be a start or a stop when we're online,
5583		 * or a stop/offline or start/online.  A start or stop when
5584		 * we're offline is covered in the case above.
5585		 */
5586		/*
5587		 * In the non-immediate case, we send the request to
5588		 * the backend and return status to the user when
5589		 * it is done.
5590		 *
5591		 * In the immediate case, we allocate a new ctl_io
5592		 * to hold a copy of the request, and send that to
5593		 * the backend.  We then set good status on the
5594		 * user's request and return it immediately.
5595		 */
5596		if (cdb->byte2 & SSS_IMMED) {
5597			union ctl_io *new_io;
5598
5599			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5600			if (new_io == NULL) {
5601				ctl_set_busy(ctsio);
5602				ctl_done((union ctl_io *)ctsio);
5603			} else {
5604				ctl_copy_io((union ctl_io *)ctsio,
5605					    new_io);
5606				retval = lun->backend->config_write(new_io);
5607				ctl_set_success(ctsio);
5608				ctl_done((union ctl_io *)ctsio);
5609			}
5610		} else {
5611			retval = lun->backend->config_write(
5612				(union ctl_io *)ctsio);
5613		}
5614#ifdef NEEDTOPORT
5615	}
5616#endif
5617	return (retval);
5618}
5619
5620/*
5621 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5622 * we don't really do anything with the LBA and length fields if the user
5623 * passes them in.  Instead we'll just flush out the cache for the entire
5624 * LUN.
5625 */
5626int
5627ctl_sync_cache(struct ctl_scsiio *ctsio)
5628{
5629	struct ctl_lun *lun;
5630	struct ctl_softc *ctl_softc;
5631	uint64_t starting_lba;
5632	uint32_t block_count;
5633	int retval;
5634
5635	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5636
5637	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5638	ctl_softc = control_softc;
5639	retval = 0;
5640
5641	switch (ctsio->cdb[0]) {
5642	case SYNCHRONIZE_CACHE: {
5643		struct scsi_sync_cache *cdb;
5644		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5645
5646		starting_lba = scsi_4btoul(cdb->begin_lba);
5647		block_count = scsi_2btoul(cdb->lb_count);
5648		break;
5649	}
5650	case SYNCHRONIZE_CACHE_16: {
5651		struct scsi_sync_cache_16 *cdb;
5652		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5653
5654		starting_lba = scsi_8btou64(cdb->begin_lba);
5655		block_count = scsi_4btoul(cdb->lb_count);
5656		break;
5657	}
5658	default:
5659		ctl_set_invalid_opcode(ctsio);
5660		ctl_done((union ctl_io *)ctsio);
5661		goto bailout;
5662		break; /* NOTREACHED */
5663	}
5664
5665	/*
5666	 * We check the LBA and length, but don't do anything with them.
5667	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5668	 * get flushed.  This check will just help satisfy anyone who wants
5669	 * to see an error for an out of range LBA.
5670	 */
5671	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5672		ctl_set_lba_out_of_range(ctsio);
5673		ctl_done((union ctl_io *)ctsio);
5674		goto bailout;
5675	}
5676
5677	/*
5678	 * If this LUN has no backend, we can't flush the cache anyway.
5679	 */
5680	if (lun->backend == NULL) {
5681		ctl_set_invalid_opcode(ctsio);
5682		ctl_done((union ctl_io *)ctsio);
5683		goto bailout;
5684	}
5685
5686	/*
5687	 * Check to see whether we're configured to send the SYNCHRONIZE
5688	 * CACHE command directly to the back end.
5689	 */
5690	mtx_lock(&lun->lun_lock);
5691	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5692	 && (++(lun->sync_count) >= lun->sync_interval)) {
5693		lun->sync_count = 0;
5694		mtx_unlock(&lun->lun_lock);
5695		retval = lun->backend->config_write((union ctl_io *)ctsio);
5696	} else {
5697		mtx_unlock(&lun->lun_lock);
5698		ctl_set_success(ctsio);
5699		ctl_done((union ctl_io *)ctsio);
5700	}
5701
5702bailout:
5703
5704	return (retval);
5705}
5706
5707int
5708ctl_format(struct ctl_scsiio *ctsio)
5709{
5710	struct scsi_format *cdb;
5711	struct ctl_lun *lun;
5712	struct ctl_softc *ctl_softc;
5713	int length, defect_list_len;
5714
5715	CTL_DEBUG_PRINT(("ctl_format\n"));
5716
5717	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5718	ctl_softc = control_softc;
5719
5720	cdb = (struct scsi_format *)ctsio->cdb;
5721
5722	length = 0;
5723	if (cdb->byte2 & SF_FMTDATA) {
5724		if (cdb->byte2 & SF_LONGLIST)
5725			length = sizeof(struct scsi_format_header_long);
5726		else
5727			length = sizeof(struct scsi_format_header_short);
5728	}
5729
5730	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5731	 && (length > 0)) {
5732		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5733		ctsio->kern_data_len = length;
5734		ctsio->kern_total_len = length;
5735		ctsio->kern_data_resid = 0;
5736		ctsio->kern_rel_offset = 0;
5737		ctsio->kern_sg_entries = 0;
5738		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5739		ctsio->be_move_done = ctl_config_move_done;
5740		ctl_datamove((union ctl_io *)ctsio);
5741
5742		return (CTL_RETVAL_COMPLETE);
5743	}
5744
5745	defect_list_len = 0;
5746
5747	if (cdb->byte2 & SF_FMTDATA) {
5748		if (cdb->byte2 & SF_LONGLIST) {
5749			struct scsi_format_header_long *header;
5750
5751			header = (struct scsi_format_header_long *)
5752				ctsio->kern_data_ptr;
5753
5754			defect_list_len = scsi_4btoul(header->defect_list_len);
5755			if (defect_list_len != 0) {
5756				ctl_set_invalid_field(ctsio,
5757						      /*sks_valid*/ 1,
5758						      /*command*/ 0,
5759						      /*field*/ 2,
5760						      /*bit_valid*/ 0,
5761						      /*bit*/ 0);
5762				goto bailout;
5763			}
5764		} else {
5765			struct scsi_format_header_short *header;
5766
5767			header = (struct scsi_format_header_short *)
5768				ctsio->kern_data_ptr;
5769
5770			defect_list_len = scsi_2btoul(header->defect_list_len);
5771			if (defect_list_len != 0) {
5772				ctl_set_invalid_field(ctsio,
5773						      /*sks_valid*/ 1,
5774						      /*command*/ 0,
5775						      /*field*/ 2,
5776						      /*bit_valid*/ 0,
5777						      /*bit*/ 0);
5778				goto bailout;
5779			}
5780		}
5781	}
5782
5783	/*
5784	 * The format command will clear out the "Medium format corrupted"
5785	 * status if set by the configuration code.  That status is really
5786	 * just a way to notify the host that we have lost the media, and
5787	 * get them to issue a command that will basically make them think
5788	 * they're blowing away the media.
5789	 */
5790	mtx_lock(&lun->lun_lock);
5791	lun->flags &= ~CTL_LUN_INOPERABLE;
5792	mtx_unlock(&lun->lun_lock);
5793
5794	ctsio->scsi_status = SCSI_STATUS_OK;
5795	ctsio->io_hdr.status = CTL_SUCCESS;
5796bailout:
5797
5798	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5799		free(ctsio->kern_data_ptr, M_CTL);
5800		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5801	}
5802
5803	ctl_done((union ctl_io *)ctsio);
5804	return (CTL_RETVAL_COMPLETE);
5805}
5806
5807int
5808ctl_read_buffer(struct ctl_scsiio *ctsio)
5809{
5810	struct scsi_read_buffer *cdb;
5811	struct ctl_lun *lun;
5812	int buffer_offset, len;
5813	static uint8_t descr[4];
5814	static uint8_t echo_descr[4] = { 0 };
5815
5816	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5817
5818	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5819	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5820
5821	if (lun->flags & CTL_LUN_PR_RESERVED) {
5822		uint32_t residx;
5823
5824		/*
5825		 * XXX KDM need a lock here.
5826		 */
5827		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5828		if ((lun->res_type == SPR_TYPE_EX_AC
5829		  && residx != lun->pr_res_idx)
5830		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5831		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5832		  && !lun->per_res[residx].registered)) {
5833			ctl_set_reservation_conflict(ctsio);
5834			ctl_done((union ctl_io *)ctsio);
5835			return (CTL_RETVAL_COMPLETE);
5836	        }
5837	}
5838
5839	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5840	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5841	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5842		ctl_set_invalid_field(ctsio,
5843				      /*sks_valid*/ 1,
5844				      /*command*/ 1,
5845				      /*field*/ 1,
5846				      /*bit_valid*/ 1,
5847				      /*bit*/ 4);
5848		ctl_done((union ctl_io *)ctsio);
5849		return (CTL_RETVAL_COMPLETE);
5850	}
5851
5852	len = scsi_3btoul(cdb->length);
5853	buffer_offset = scsi_3btoul(cdb->offset);
5854
5855	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5856		ctl_set_invalid_field(ctsio,
5857				      /*sks_valid*/ 1,
5858				      /*command*/ 1,
5859				      /*field*/ 6,
5860				      /*bit_valid*/ 0,
5861				      /*bit*/ 0);
5862		ctl_done((union ctl_io *)ctsio);
5863		return (CTL_RETVAL_COMPLETE);
5864	}
5865
5866	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5867		descr[0] = 0;
5868		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5869		ctsio->kern_data_ptr = descr;
5870		len = min(len, sizeof(descr));
5871	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5872		ctsio->kern_data_ptr = echo_descr;
5873		len = min(len, sizeof(echo_descr));
5874	} else
5875		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5876	ctsio->kern_data_len = len;
5877	ctsio->kern_total_len = len;
5878	ctsio->kern_data_resid = 0;
5879	ctsio->kern_rel_offset = 0;
5880	ctsio->kern_sg_entries = 0;
5881	ctsio->be_move_done = ctl_config_move_done;
5882	ctl_datamove((union ctl_io *)ctsio);
5883
5884	return (CTL_RETVAL_COMPLETE);
5885}
5886
5887int
5888ctl_write_buffer(struct ctl_scsiio *ctsio)
5889{
5890	struct scsi_write_buffer *cdb;
5891	struct ctl_lun *lun;
5892	int buffer_offset, len;
5893
5894	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5895
5896	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5897	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5898
5899	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5900		ctl_set_invalid_field(ctsio,
5901				      /*sks_valid*/ 1,
5902				      /*command*/ 1,
5903				      /*field*/ 1,
5904				      /*bit_valid*/ 1,
5905				      /*bit*/ 4);
5906		ctl_done((union ctl_io *)ctsio);
5907		return (CTL_RETVAL_COMPLETE);
5908	}
5909
5910	len = scsi_3btoul(cdb->length);
5911	buffer_offset = scsi_3btoul(cdb->offset);
5912
5913	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5914		ctl_set_invalid_field(ctsio,
5915				      /*sks_valid*/ 1,
5916				      /*command*/ 1,
5917				      /*field*/ 6,
5918				      /*bit_valid*/ 0,
5919				      /*bit*/ 0);
5920		ctl_done((union ctl_io *)ctsio);
5921		return (CTL_RETVAL_COMPLETE);
5922	}
5923
5924	/*
5925	 * If we've got a kernel request that hasn't been malloced yet,
5926	 * malloc it and tell the caller the data buffer is here.
5927	 */
5928	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5929		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5930		ctsio->kern_data_len = len;
5931		ctsio->kern_total_len = len;
5932		ctsio->kern_data_resid = 0;
5933		ctsio->kern_rel_offset = 0;
5934		ctsio->kern_sg_entries = 0;
5935		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5936		ctsio->be_move_done = ctl_config_move_done;
5937		ctl_datamove((union ctl_io *)ctsio);
5938
5939		return (CTL_RETVAL_COMPLETE);
5940	}
5941
5942	ctl_done((union ctl_io *)ctsio);
5943
5944	return (CTL_RETVAL_COMPLETE);
5945}
5946
5947int
5948ctl_write_same(struct ctl_scsiio *ctsio)
5949{
5950	struct ctl_lun *lun;
5951	struct ctl_lba_len_flags *lbalen;
5952	uint64_t lba;
5953	uint32_t num_blocks;
5954	int len, retval;
5955	uint8_t byte2;
5956
5957	retval = CTL_RETVAL_COMPLETE;
5958
5959	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5960
5961	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5962
5963	switch (ctsio->cdb[0]) {
5964	case WRITE_SAME_10: {
5965		struct scsi_write_same_10 *cdb;
5966
5967		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5968
5969		lba = scsi_4btoul(cdb->addr);
5970		num_blocks = scsi_2btoul(cdb->length);
5971		byte2 = cdb->byte2;
5972		break;
5973	}
5974	case WRITE_SAME_16: {
5975		struct scsi_write_same_16 *cdb;
5976
5977		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5978
5979		lba = scsi_8btou64(cdb->addr);
5980		num_blocks = scsi_4btoul(cdb->length);
5981		byte2 = cdb->byte2;
5982		break;
5983	}
5984	default:
5985		/*
5986		 * We got a command we don't support.  This shouldn't
5987		 * happen, commands should be filtered out above us.
5988		 */
5989		ctl_set_invalid_opcode(ctsio);
5990		ctl_done((union ctl_io *)ctsio);
5991
5992		return (CTL_RETVAL_COMPLETE);
5993		break; /* NOTREACHED */
5994	}
5995
5996	/* NDOB flag can be used only together with UNMAP */
5997	if ((byte2 & (SWS_NDOB | SWS_UNMAP)) == SWS_NDOB) {
5998		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5999		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
6000		ctl_done((union ctl_io *)ctsio);
6001		return (CTL_RETVAL_COMPLETE);
6002	}
6003
6004	/*
6005	 * The first check is to make sure we're in bounds, the second
6006	 * check is to catch wrap-around problems.  If the lba + num blocks
6007	 * is less than the lba, then we've wrapped around and the block
6008	 * range is invalid anyway.
6009	 */
6010	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6011	 || ((lba + num_blocks) < lba)) {
6012		ctl_set_lba_out_of_range(ctsio);
6013		ctl_done((union ctl_io *)ctsio);
6014		return (CTL_RETVAL_COMPLETE);
6015	}
6016
6017	/* Zero number of blocks means "to the last logical block" */
6018	if (num_blocks == 0) {
6019		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
6020			ctl_set_invalid_field(ctsio,
6021					      /*sks_valid*/ 0,
6022					      /*command*/ 1,
6023					      /*field*/ 0,
6024					      /*bit_valid*/ 0,
6025					      /*bit*/ 0);
6026			ctl_done((union ctl_io *)ctsio);
6027			return (CTL_RETVAL_COMPLETE);
6028		}
6029		num_blocks = (lun->be_lun->maxlba + 1) - lba;
6030	}
6031
6032	len = lun->be_lun->blocksize;
6033
6034	/*
6035	 * If we've got a kernel request that hasn't been malloced yet,
6036	 * malloc it and tell the caller the data buffer is here.
6037	 */
6038	if ((byte2 & SWS_NDOB) == 0 &&
6039	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6040		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6041		ctsio->kern_data_len = len;
6042		ctsio->kern_total_len = len;
6043		ctsio->kern_data_resid = 0;
6044		ctsio->kern_rel_offset = 0;
6045		ctsio->kern_sg_entries = 0;
6046		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6047		ctsio->be_move_done = ctl_config_move_done;
6048		ctl_datamove((union ctl_io *)ctsio);
6049
6050		return (CTL_RETVAL_COMPLETE);
6051	}
6052
6053	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6054	lbalen->lba = lba;
6055	lbalen->len = num_blocks;
6056	lbalen->flags = byte2;
6057	retval = lun->backend->config_write((union ctl_io *)ctsio);
6058
6059	return (retval);
6060}
6061
6062int
6063ctl_unmap(struct ctl_scsiio *ctsio)
6064{
6065	struct ctl_lun *lun;
6066	struct scsi_unmap *cdb;
6067	struct ctl_ptr_len_flags *ptrlen;
6068	struct scsi_unmap_header *hdr;
6069	struct scsi_unmap_desc *buf, *end, *range;
6070	uint64_t lba;
6071	uint32_t num_blocks;
6072	int len, retval;
6073	uint8_t byte2;
6074
6075	retval = CTL_RETVAL_COMPLETE;
6076
6077	CTL_DEBUG_PRINT(("ctl_unmap\n"));
6078
6079	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6080	cdb = (struct scsi_unmap *)ctsio->cdb;
6081
6082	len = scsi_2btoul(cdb->length);
6083	byte2 = cdb->byte2;
6084
6085	/*
6086	 * If we've got a kernel request that hasn't been malloced yet,
6087	 * malloc it and tell the caller the data buffer is here.
6088	 */
6089	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6090		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6091		ctsio->kern_data_len = len;
6092		ctsio->kern_total_len = len;
6093		ctsio->kern_data_resid = 0;
6094		ctsio->kern_rel_offset = 0;
6095		ctsio->kern_sg_entries = 0;
6096		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6097		ctsio->be_move_done = ctl_config_move_done;
6098		ctl_datamove((union ctl_io *)ctsio);
6099
6100		return (CTL_RETVAL_COMPLETE);
6101	}
6102
6103	len = ctsio->kern_total_len - ctsio->kern_data_resid;
6104	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6105	if (len < sizeof (*hdr) ||
6106	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6107	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6108	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6109		ctl_set_invalid_field(ctsio,
6110				      /*sks_valid*/ 0,
6111				      /*command*/ 0,
6112				      /*field*/ 0,
6113				      /*bit_valid*/ 0,
6114				      /*bit*/ 0);
6115		ctl_done((union ctl_io *)ctsio);
6116		return (CTL_RETVAL_COMPLETE);
6117	}
6118	len = scsi_2btoul(hdr->desc_length);
6119	buf = (struct scsi_unmap_desc *)(hdr + 1);
6120	end = buf + len / sizeof(*buf);
6121
6122	for (range = buf; range < end; range++) {
6123		lba = scsi_8btou64(range->lba);
6124		num_blocks = scsi_4btoul(range->length);
6125		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6126		 || ((lba + num_blocks) < lba)) {
6127			ctl_set_lba_out_of_range(ctsio);
6128			ctl_done((union ctl_io *)ctsio);
6129			return (CTL_RETVAL_COMPLETE);
6130		}
6131	}
6132
6133	mtx_lock(&lun->lun_lock);
6134	ptrlen = (struct ctl_ptr_len_flags *)
6135	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6136	ptrlen->ptr = (void *)buf;
6137	ptrlen->len = len;
6138	ptrlen->flags = byte2;
6139	ctl_check_blocked(lun);
6140	mtx_unlock(&lun->lun_lock);
6141
6142	retval = lun->backend->config_write((union ctl_io *)ctsio);
6143	return (retval);
6144}
6145
6146/*
6147 * Note that this function currently doesn't actually do anything inside
6148 * CTL to enforce things if the DQue bit is turned on.
6149 *
6150 * Also note that this function can't be used in the default case, because
6151 * the DQue bit isn't set in the changeable mask for the control mode page
6152 * anyway.  This is just here as an example for how to implement a page
6153 * handler, and a placeholder in case we want to allow the user to turn
6154 * tagged queueing on and off.
6155 *
6156 * The D_SENSE bit handling is functional, however, and will turn
6157 * descriptor sense on and off for a given LUN.
6158 */
6159int
6160ctl_control_page_handler(struct ctl_scsiio *ctsio,
6161			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6162{
6163	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6164	struct ctl_lun *lun;
6165	struct ctl_softc *softc;
6166	int set_ua;
6167	uint32_t initidx;
6168
6169	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6170	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6171	set_ua = 0;
6172
6173	user_cp = (struct scsi_control_page *)page_ptr;
6174	current_cp = (struct scsi_control_page *)
6175		(page_index->page_data + (page_index->page_len *
6176		CTL_PAGE_CURRENT));
6177	saved_cp = (struct scsi_control_page *)
6178		(page_index->page_data + (page_index->page_len *
6179		CTL_PAGE_SAVED));
6180
6181	softc = control_softc;
6182
6183	mtx_lock(&lun->lun_lock);
6184	if (((current_cp->rlec & SCP_DSENSE) == 0)
6185	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6186		/*
6187		 * Descriptor sense is currently turned off and the user
6188		 * wants to turn it on.
6189		 */
6190		current_cp->rlec |= SCP_DSENSE;
6191		saved_cp->rlec |= SCP_DSENSE;
6192		lun->flags |= CTL_LUN_SENSE_DESC;
6193		set_ua = 1;
6194	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6195		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6196		/*
6197		 * Descriptor sense is currently turned on, and the user
6198		 * wants to turn it off.
6199		 */
6200		current_cp->rlec &= ~SCP_DSENSE;
6201		saved_cp->rlec &= ~SCP_DSENSE;
6202		lun->flags &= ~CTL_LUN_SENSE_DESC;
6203		set_ua = 1;
6204	}
6205	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6206	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6207		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6208		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6209		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6210		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6211		set_ua = 1;
6212	}
6213	if (set_ua != 0) {
6214		int i;
6215		/*
6216		 * Let other initiators know that the mode
6217		 * parameters for this LUN have changed.
6218		 */
6219		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6220			if (i == initidx)
6221				continue;
6222
6223			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6224		}
6225	}
6226	mtx_unlock(&lun->lun_lock);
6227
6228	return (0);
6229}
6230
6231int
6232ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6233		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6234{
6235	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6236	struct ctl_lun *lun;
6237	int set_ua;
6238	uint32_t initidx;
6239
6240	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6241	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6242	set_ua = 0;
6243
6244	user_cp = (struct scsi_caching_page *)page_ptr;
6245	current_cp = (struct scsi_caching_page *)
6246		(page_index->page_data + (page_index->page_len *
6247		CTL_PAGE_CURRENT));
6248	saved_cp = (struct scsi_caching_page *)
6249		(page_index->page_data + (page_index->page_len *
6250		CTL_PAGE_SAVED));
6251
6252	mtx_lock(&lun->lun_lock);
6253	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6254	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6255		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6256		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6257		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6258		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6259		set_ua = 1;
6260	}
6261	if (set_ua != 0) {
6262		int i;
6263		/*
6264		 * Let other initiators know that the mode
6265		 * parameters for this LUN have changed.
6266		 */
6267		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6268			if (i == initidx)
6269				continue;
6270
6271			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6272		}
6273	}
6274	mtx_unlock(&lun->lun_lock);
6275
6276	return (0);
6277}
6278
6279int
6280ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6281		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6282{
6283	return (0);
6284}
6285
6286int
6287ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6288			   struct ctl_page_index *page_index, int pc)
6289{
6290	struct copan_power_subpage *page;
6291
6292	page = (struct copan_power_subpage *)page_index->page_data +
6293		(page_index->page_len * pc);
6294
6295	switch (pc) {
6296	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6297		/*
6298		 * We don't update the changable bits for this page.
6299		 */
6300		break;
6301	case SMS_PAGE_CTRL_CURRENT >> 6:
6302	case SMS_PAGE_CTRL_DEFAULT >> 6:
6303	case SMS_PAGE_CTRL_SAVED >> 6:
6304#ifdef NEEDTOPORT
6305		ctl_update_power_subpage(page);
6306#endif
6307		break;
6308	default:
6309#ifdef NEEDTOPORT
6310		EPRINT(0, "Invalid PC %d!!", pc);
6311#endif
6312		break;
6313	}
6314	return (0);
6315}
6316
6317
6318int
6319ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6320		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6321{
6322	struct copan_aps_subpage *user_sp;
6323	struct copan_aps_subpage *current_sp;
6324	union ctl_modepage_info *modepage_info;
6325	struct ctl_softc *softc;
6326	struct ctl_lun *lun;
6327	int retval;
6328
6329	retval = CTL_RETVAL_COMPLETE;
6330	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6331		     (page_index->page_len * CTL_PAGE_CURRENT));
6332	softc = control_softc;
6333	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6334
6335	user_sp = (struct copan_aps_subpage *)page_ptr;
6336
6337	modepage_info = (union ctl_modepage_info *)
6338		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6339
6340	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6341	modepage_info->header.subpage = page_index->subpage;
6342	modepage_info->aps.lock_active = user_sp->lock_active;
6343
6344	mtx_lock(&softc->ctl_lock);
6345
6346	/*
6347	 * If there is a request to lock the LUN and another LUN is locked
6348	 * this is an error. If the requested LUN is already locked ignore
6349	 * the request. If no LUN is locked attempt to lock it.
6350	 * if there is a request to unlock the LUN and the LUN is currently
6351	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6352	 * if another LUN is locked or no LUN is locked.
6353	 */
6354	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6355		if (softc->aps_locked_lun == lun->lun) {
6356			/*
6357			 * This LUN is already locked, so we're done.
6358			 */
6359			retval = CTL_RETVAL_COMPLETE;
6360		} else if (softc->aps_locked_lun == 0) {
6361			/*
6362			 * No one has the lock, pass the request to the
6363			 * backend.
6364			 */
6365			retval = lun->backend->config_write(
6366				(union ctl_io *)ctsio);
6367		} else {
6368			/*
6369			 * Someone else has the lock, throw out the request.
6370			 */
6371			ctl_set_already_locked(ctsio);
6372			free(ctsio->kern_data_ptr, M_CTL);
6373			ctl_done((union ctl_io *)ctsio);
6374
6375			/*
6376			 * Set the return value so that ctl_do_mode_select()
6377			 * won't try to complete the command.  We already
6378			 * completed it here.
6379			 */
6380			retval = CTL_RETVAL_ERROR;
6381		}
6382	} else if (softc->aps_locked_lun == lun->lun) {
6383		/*
6384		 * This LUN is locked, so pass the unlock request to the
6385		 * backend.
6386		 */
6387		retval = lun->backend->config_write((union ctl_io *)ctsio);
6388	}
6389	mtx_unlock(&softc->ctl_lock);
6390
6391	return (retval);
6392}
6393
6394int
6395ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6396				struct ctl_page_index *page_index,
6397				uint8_t *page_ptr)
6398{
6399	uint8_t *c;
6400	int i;
6401
6402	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6403	ctl_time_io_secs =
6404		(c[0] << 8) |
6405		(c[1] << 0) |
6406		0;
6407	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6408	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6409	printf("page data:");
6410	for (i=0; i<8; i++)
6411		printf(" %.2x",page_ptr[i]);
6412	printf("\n");
6413	return (0);
6414}
6415
6416int
6417ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6418			       struct ctl_page_index *page_index,
6419			       int pc)
6420{
6421	struct copan_debugconf_subpage *page;
6422
6423	page = (struct copan_debugconf_subpage *)page_index->page_data +
6424		(page_index->page_len * pc);
6425
6426	switch (pc) {
6427	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6428	case SMS_PAGE_CTRL_DEFAULT >> 6:
6429	case SMS_PAGE_CTRL_SAVED >> 6:
6430		/*
6431		 * We don't update the changable or default bits for this page.
6432		 */
6433		break;
6434	case SMS_PAGE_CTRL_CURRENT >> 6:
6435		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6436		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6437		break;
6438	default:
6439#ifdef NEEDTOPORT
6440		EPRINT(0, "Invalid PC %d!!", pc);
6441#endif /* NEEDTOPORT */
6442		break;
6443	}
6444	return (0);
6445}
6446
6447
6448static int
6449ctl_do_mode_select(union ctl_io *io)
6450{
6451	struct scsi_mode_page_header *page_header;
6452	struct ctl_page_index *page_index;
6453	struct ctl_scsiio *ctsio;
6454	int control_dev, page_len;
6455	int page_len_offset, page_len_size;
6456	union ctl_modepage_info *modepage_info;
6457	struct ctl_lun *lun;
6458	int *len_left, *len_used;
6459	int retval, i;
6460
6461	ctsio = &io->scsiio;
6462	page_index = NULL;
6463	page_len = 0;
6464	retval = CTL_RETVAL_COMPLETE;
6465
6466	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6467
6468	if (lun->be_lun->lun_type != T_DIRECT)
6469		control_dev = 1;
6470	else
6471		control_dev = 0;
6472
6473	modepage_info = (union ctl_modepage_info *)
6474		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6475	len_left = &modepage_info->header.len_left;
6476	len_used = &modepage_info->header.len_used;
6477
6478do_next_page:
6479
6480	page_header = (struct scsi_mode_page_header *)
6481		(ctsio->kern_data_ptr + *len_used);
6482
6483	if (*len_left == 0) {
6484		free(ctsio->kern_data_ptr, M_CTL);
6485		ctl_set_success(ctsio);
6486		ctl_done((union ctl_io *)ctsio);
6487		return (CTL_RETVAL_COMPLETE);
6488	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6489
6490		free(ctsio->kern_data_ptr, M_CTL);
6491		ctl_set_param_len_error(ctsio);
6492		ctl_done((union ctl_io *)ctsio);
6493		return (CTL_RETVAL_COMPLETE);
6494
6495	} else if ((page_header->page_code & SMPH_SPF)
6496		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
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
6504
6505	/*
6506	 * XXX KDM should we do something with the block descriptor?
6507	 */
6508	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6509
6510		if ((control_dev != 0)
6511		 && (lun->mode_pages.index[i].page_flags &
6512		     CTL_PAGE_FLAG_DISK_ONLY))
6513			continue;
6514
6515		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6516		    (page_header->page_code & SMPH_PC_MASK))
6517			continue;
6518
6519		/*
6520		 * If neither page has a subpage code, then we've got a
6521		 * match.
6522		 */
6523		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6524		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6525			page_index = &lun->mode_pages.index[i];
6526			page_len = page_header->page_length;
6527			break;
6528		}
6529
6530		/*
6531		 * If both pages have subpages, then the subpage numbers
6532		 * have to match.
6533		 */
6534		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6535		  && (page_header->page_code & SMPH_SPF)) {
6536			struct scsi_mode_page_header_sp *sph;
6537
6538			sph = (struct scsi_mode_page_header_sp *)page_header;
6539
6540			if (lun->mode_pages.index[i].subpage ==
6541			    sph->subpage) {
6542				page_index = &lun->mode_pages.index[i];
6543				page_len = scsi_2btoul(sph->page_length);
6544				break;
6545			}
6546		}
6547	}
6548
6549	/*
6550	 * If we couldn't find the page, or if we don't have a mode select
6551	 * handler for it, send back an error to the user.
6552	 */
6553	if ((page_index == NULL)
6554	 || (page_index->select_handler == NULL)) {
6555		ctl_set_invalid_field(ctsio,
6556				      /*sks_valid*/ 1,
6557				      /*command*/ 0,
6558				      /*field*/ *len_used,
6559				      /*bit_valid*/ 0,
6560				      /*bit*/ 0);
6561		free(ctsio->kern_data_ptr, M_CTL);
6562		ctl_done((union ctl_io *)ctsio);
6563		return (CTL_RETVAL_COMPLETE);
6564	}
6565
6566	if (page_index->page_code & SMPH_SPF) {
6567		page_len_offset = 2;
6568		page_len_size = 2;
6569	} else {
6570		page_len_size = 1;
6571		page_len_offset = 1;
6572	}
6573
6574	/*
6575	 * If the length the initiator gives us isn't the one we specify in
6576	 * the mode page header, or if they didn't specify enough data in
6577	 * the CDB to avoid truncating this page, kick out the request.
6578	 */
6579	if ((page_len != (page_index->page_len - page_len_offset -
6580			  page_len_size))
6581	 || (*len_left < page_index->page_len)) {
6582
6583
6584		ctl_set_invalid_field(ctsio,
6585				      /*sks_valid*/ 1,
6586				      /*command*/ 0,
6587				      /*field*/ *len_used + page_len_offset,
6588				      /*bit_valid*/ 0,
6589				      /*bit*/ 0);
6590		free(ctsio->kern_data_ptr, M_CTL);
6591		ctl_done((union ctl_io *)ctsio);
6592		return (CTL_RETVAL_COMPLETE);
6593	}
6594
6595	/*
6596	 * Run through the mode page, checking to make sure that the bits
6597	 * the user changed are actually legal for him to change.
6598	 */
6599	for (i = 0; i < page_index->page_len; i++) {
6600		uint8_t *user_byte, *change_mask, *current_byte;
6601		int bad_bit;
6602		int j;
6603
6604		user_byte = (uint8_t *)page_header + i;
6605		change_mask = page_index->page_data +
6606			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6607		current_byte = page_index->page_data +
6608			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6609
6610		/*
6611		 * Check to see whether the user set any bits in this byte
6612		 * that he is not allowed to set.
6613		 */
6614		if ((*user_byte & ~(*change_mask)) ==
6615		    (*current_byte & ~(*change_mask)))
6616			continue;
6617
6618		/*
6619		 * Go through bit by bit to determine which one is illegal.
6620		 */
6621		bad_bit = 0;
6622		for (j = 7; j >= 0; j--) {
6623			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6624			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6625				bad_bit = i;
6626				break;
6627			}
6628		}
6629		ctl_set_invalid_field(ctsio,
6630				      /*sks_valid*/ 1,
6631				      /*command*/ 0,
6632				      /*field*/ *len_used + i,
6633				      /*bit_valid*/ 1,
6634				      /*bit*/ bad_bit);
6635		free(ctsio->kern_data_ptr, M_CTL);
6636		ctl_done((union ctl_io *)ctsio);
6637		return (CTL_RETVAL_COMPLETE);
6638	}
6639
6640	/*
6641	 * Decrement these before we call the page handler, since we may
6642	 * end up getting called back one way or another before the handler
6643	 * returns to this context.
6644	 */
6645	*len_left -= page_index->page_len;
6646	*len_used += page_index->page_len;
6647
6648	retval = page_index->select_handler(ctsio, page_index,
6649					    (uint8_t *)page_header);
6650
6651	/*
6652	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6653	 * wait until this queued command completes to finish processing
6654	 * the mode page.  If it returns anything other than
6655	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6656	 * already set the sense information, freed the data pointer, and
6657	 * completed the io for us.
6658	 */
6659	if (retval != CTL_RETVAL_COMPLETE)
6660		goto bailout_no_done;
6661
6662	/*
6663	 * If the initiator sent us more than one page, parse the next one.
6664	 */
6665	if (*len_left > 0)
6666		goto do_next_page;
6667
6668	ctl_set_success(ctsio);
6669	free(ctsio->kern_data_ptr, M_CTL);
6670	ctl_done((union ctl_io *)ctsio);
6671
6672bailout_no_done:
6673
6674	return (CTL_RETVAL_COMPLETE);
6675
6676}
6677
6678int
6679ctl_mode_select(struct ctl_scsiio *ctsio)
6680{
6681	int param_len, pf, sp;
6682	int header_size, bd_len;
6683	int len_left, len_used;
6684	struct ctl_page_index *page_index;
6685	struct ctl_lun *lun;
6686	int control_dev, page_len;
6687	union ctl_modepage_info *modepage_info;
6688	int retval;
6689
6690	pf = 0;
6691	sp = 0;
6692	page_len = 0;
6693	len_used = 0;
6694	len_left = 0;
6695	retval = 0;
6696	bd_len = 0;
6697	page_index = NULL;
6698
6699	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6700
6701	if (lun->be_lun->lun_type != T_DIRECT)
6702		control_dev = 1;
6703	else
6704		control_dev = 0;
6705
6706	switch (ctsio->cdb[0]) {
6707	case MODE_SELECT_6: {
6708		struct scsi_mode_select_6 *cdb;
6709
6710		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6711
6712		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6713		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6714
6715		param_len = cdb->length;
6716		header_size = sizeof(struct scsi_mode_header_6);
6717		break;
6718	}
6719	case MODE_SELECT_10: {
6720		struct scsi_mode_select_10 *cdb;
6721
6722		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6723
6724		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6725		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6726
6727		param_len = scsi_2btoul(cdb->length);
6728		header_size = sizeof(struct scsi_mode_header_10);
6729		break;
6730	}
6731	default:
6732		ctl_set_invalid_opcode(ctsio);
6733		ctl_done((union ctl_io *)ctsio);
6734		return (CTL_RETVAL_COMPLETE);
6735		break; /* NOTREACHED */
6736	}
6737
6738	/*
6739	 * From SPC-3:
6740	 * "A parameter list length of zero indicates that the Data-Out Buffer
6741	 * shall be empty. This condition shall not be considered as an error."
6742	 */
6743	if (param_len == 0) {
6744		ctl_set_success(ctsio);
6745		ctl_done((union ctl_io *)ctsio);
6746		return (CTL_RETVAL_COMPLETE);
6747	}
6748
6749	/*
6750	 * Since we'll hit this the first time through, prior to
6751	 * allocation, we don't need to free a data buffer here.
6752	 */
6753	if (param_len < header_size) {
6754		ctl_set_param_len_error(ctsio);
6755		ctl_done((union ctl_io *)ctsio);
6756		return (CTL_RETVAL_COMPLETE);
6757	}
6758
6759	/*
6760	 * Allocate the data buffer and grab the user's data.  In theory,
6761	 * we shouldn't have to sanity check the parameter list length here
6762	 * because the maximum size is 64K.  We should be able to malloc
6763	 * that much without too many problems.
6764	 */
6765	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6766		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6767		ctsio->kern_data_len = param_len;
6768		ctsio->kern_total_len = param_len;
6769		ctsio->kern_data_resid = 0;
6770		ctsio->kern_rel_offset = 0;
6771		ctsio->kern_sg_entries = 0;
6772		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6773		ctsio->be_move_done = ctl_config_move_done;
6774		ctl_datamove((union ctl_io *)ctsio);
6775
6776		return (CTL_RETVAL_COMPLETE);
6777	}
6778
6779	switch (ctsio->cdb[0]) {
6780	case MODE_SELECT_6: {
6781		struct scsi_mode_header_6 *mh6;
6782
6783		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6784		bd_len = mh6->blk_desc_len;
6785		break;
6786	}
6787	case MODE_SELECT_10: {
6788		struct scsi_mode_header_10 *mh10;
6789
6790		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6791		bd_len = scsi_2btoul(mh10->blk_desc_len);
6792		break;
6793	}
6794	default:
6795		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6796		break;
6797	}
6798
6799	if (param_len < (header_size + bd_len)) {
6800		free(ctsio->kern_data_ptr, M_CTL);
6801		ctl_set_param_len_error(ctsio);
6802		ctl_done((union ctl_io *)ctsio);
6803		return (CTL_RETVAL_COMPLETE);
6804	}
6805
6806	/*
6807	 * Set the IO_CONT flag, so that if this I/O gets passed to
6808	 * ctl_config_write_done(), it'll get passed back to
6809	 * ctl_do_mode_select() for further processing, or completion if
6810	 * we're all done.
6811	 */
6812	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6813	ctsio->io_cont = ctl_do_mode_select;
6814
6815	modepage_info = (union ctl_modepage_info *)
6816		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6817
6818	memset(modepage_info, 0, sizeof(*modepage_info));
6819
6820	len_left = param_len - header_size - bd_len;
6821	len_used = header_size + bd_len;
6822
6823	modepage_info->header.len_left = len_left;
6824	modepage_info->header.len_used = len_used;
6825
6826	return (ctl_do_mode_select((union ctl_io *)ctsio));
6827}
6828
6829int
6830ctl_mode_sense(struct ctl_scsiio *ctsio)
6831{
6832	struct ctl_lun *lun;
6833	int pc, page_code, dbd, llba, subpage;
6834	int alloc_len, page_len, header_len, total_len;
6835	struct scsi_mode_block_descr *block_desc;
6836	struct ctl_page_index *page_index;
6837	int control_dev;
6838
6839	dbd = 0;
6840	llba = 0;
6841	block_desc = NULL;
6842	page_index = NULL;
6843
6844	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6845
6846	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6847
6848	if (lun->be_lun->lun_type != T_DIRECT)
6849		control_dev = 1;
6850	else
6851		control_dev = 0;
6852
6853	if (lun->flags & CTL_LUN_PR_RESERVED) {
6854		uint32_t residx;
6855
6856		/*
6857		 * XXX KDM need a lock here.
6858		 */
6859		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6860		if ((lun->res_type == SPR_TYPE_EX_AC
6861		  && residx != lun->pr_res_idx)
6862		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6863		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6864		  && !lun->per_res[residx].registered)) {
6865			ctl_set_reservation_conflict(ctsio);
6866			ctl_done((union ctl_io *)ctsio);
6867			return (CTL_RETVAL_COMPLETE);
6868		}
6869	}
6870
6871	switch (ctsio->cdb[0]) {
6872	case MODE_SENSE_6: {
6873		struct scsi_mode_sense_6 *cdb;
6874
6875		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6876
6877		header_len = sizeof(struct scsi_mode_hdr_6);
6878		if (cdb->byte2 & SMS_DBD)
6879			dbd = 1;
6880		else
6881			header_len += sizeof(struct scsi_mode_block_descr);
6882
6883		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6884		page_code = cdb->page & SMS_PAGE_CODE;
6885		subpage = cdb->subpage;
6886		alloc_len = cdb->length;
6887		break;
6888	}
6889	case MODE_SENSE_10: {
6890		struct scsi_mode_sense_10 *cdb;
6891
6892		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6893
6894		header_len = sizeof(struct scsi_mode_hdr_10);
6895
6896		if (cdb->byte2 & SMS_DBD)
6897			dbd = 1;
6898		else
6899			header_len += sizeof(struct scsi_mode_block_descr);
6900		if (cdb->byte2 & SMS10_LLBAA)
6901			llba = 1;
6902		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6903		page_code = cdb->page & SMS_PAGE_CODE;
6904		subpage = cdb->subpage;
6905		alloc_len = scsi_2btoul(cdb->length);
6906		break;
6907	}
6908	default:
6909		ctl_set_invalid_opcode(ctsio);
6910		ctl_done((union ctl_io *)ctsio);
6911		return (CTL_RETVAL_COMPLETE);
6912		break; /* NOTREACHED */
6913	}
6914
6915	/*
6916	 * We have to make a first pass through to calculate the size of
6917	 * the pages that match the user's query.  Then we allocate enough
6918	 * memory to hold it, and actually copy the data into the buffer.
6919	 */
6920	switch (page_code) {
6921	case SMS_ALL_PAGES_PAGE: {
6922		int i;
6923
6924		page_len = 0;
6925
6926		/*
6927		 * At the moment, values other than 0 and 0xff here are
6928		 * reserved according to SPC-3.
6929		 */
6930		if ((subpage != SMS_SUBPAGE_PAGE_0)
6931		 && (subpage != SMS_SUBPAGE_ALL)) {
6932			ctl_set_invalid_field(ctsio,
6933					      /*sks_valid*/ 1,
6934					      /*command*/ 1,
6935					      /*field*/ 3,
6936					      /*bit_valid*/ 0,
6937					      /*bit*/ 0);
6938			ctl_done((union ctl_io *)ctsio);
6939			return (CTL_RETVAL_COMPLETE);
6940		}
6941
6942		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6943			if ((control_dev != 0)
6944			 && (lun->mode_pages.index[i].page_flags &
6945			     CTL_PAGE_FLAG_DISK_ONLY))
6946				continue;
6947
6948			/*
6949			 * We don't use this subpage if the user didn't
6950			 * request all subpages.
6951			 */
6952			if ((lun->mode_pages.index[i].subpage != 0)
6953			 && (subpage == SMS_SUBPAGE_PAGE_0))
6954				continue;
6955
6956#if 0
6957			printf("found page %#x len %d\n",
6958			       lun->mode_pages.index[i].page_code &
6959			       SMPH_PC_MASK,
6960			       lun->mode_pages.index[i].page_len);
6961#endif
6962			page_len += lun->mode_pages.index[i].page_len;
6963		}
6964		break;
6965	}
6966	default: {
6967		int i;
6968
6969		page_len = 0;
6970
6971		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6972			/* Look for the right page code */
6973			if ((lun->mode_pages.index[i].page_code &
6974			     SMPH_PC_MASK) != page_code)
6975				continue;
6976
6977			/* Look for the right subpage or the subpage wildcard*/
6978			if ((lun->mode_pages.index[i].subpage != subpage)
6979			 && (subpage != SMS_SUBPAGE_ALL))
6980				continue;
6981
6982			/* Make sure the page is supported for this dev type */
6983			if ((control_dev != 0)
6984			 && (lun->mode_pages.index[i].page_flags &
6985			     CTL_PAGE_FLAG_DISK_ONLY))
6986				continue;
6987
6988#if 0
6989			printf("found page %#x len %d\n",
6990			       lun->mode_pages.index[i].page_code &
6991			       SMPH_PC_MASK,
6992			       lun->mode_pages.index[i].page_len);
6993#endif
6994
6995			page_len += lun->mode_pages.index[i].page_len;
6996		}
6997
6998		if (page_len == 0) {
6999			ctl_set_invalid_field(ctsio,
7000					      /*sks_valid*/ 1,
7001					      /*command*/ 1,
7002					      /*field*/ 2,
7003					      /*bit_valid*/ 1,
7004					      /*bit*/ 5);
7005			ctl_done((union ctl_io *)ctsio);
7006			return (CTL_RETVAL_COMPLETE);
7007		}
7008		break;
7009	}
7010	}
7011
7012	total_len = header_len + page_len;
7013#if 0
7014	printf("header_len = %d, page_len = %d, total_len = %d\n",
7015	       header_len, page_len, total_len);
7016#endif
7017
7018	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7019	ctsio->kern_sg_entries = 0;
7020	ctsio->kern_data_resid = 0;
7021	ctsio->kern_rel_offset = 0;
7022	if (total_len < alloc_len) {
7023		ctsio->residual = alloc_len - total_len;
7024		ctsio->kern_data_len = total_len;
7025		ctsio->kern_total_len = total_len;
7026	} else {
7027		ctsio->residual = 0;
7028		ctsio->kern_data_len = alloc_len;
7029		ctsio->kern_total_len = alloc_len;
7030	}
7031
7032	switch (ctsio->cdb[0]) {
7033	case MODE_SENSE_6: {
7034		struct scsi_mode_hdr_6 *header;
7035
7036		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7037
7038		header->datalen = ctl_min(total_len - 1, 254);
7039		if (control_dev == 0)
7040			header->dev_specific = 0x10; /* DPOFUA */
7041		if (dbd)
7042			header->block_descr_len = 0;
7043		else
7044			header->block_descr_len =
7045				sizeof(struct scsi_mode_block_descr);
7046		block_desc = (struct scsi_mode_block_descr *)&header[1];
7047		break;
7048	}
7049	case MODE_SENSE_10: {
7050		struct scsi_mode_hdr_10 *header;
7051		int datalen;
7052
7053		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7054
7055		datalen = ctl_min(total_len - 2, 65533);
7056		scsi_ulto2b(datalen, header->datalen);
7057		if (control_dev == 0)
7058			header->dev_specific = 0x10; /* DPOFUA */
7059		if (dbd)
7060			scsi_ulto2b(0, header->block_descr_len);
7061		else
7062			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7063				    header->block_descr_len);
7064		block_desc = (struct scsi_mode_block_descr *)&header[1];
7065		break;
7066	}
7067	default:
7068		panic("invalid CDB type %#x", ctsio->cdb[0]);
7069		break; /* NOTREACHED */
7070	}
7071
7072	/*
7073	 * If we've got a disk, use its blocksize in the block
7074	 * descriptor.  Otherwise, just set it to 0.
7075	 */
7076	if (dbd == 0) {
7077		if (control_dev == 0)
7078			scsi_ulto3b(lun->be_lun->blocksize,
7079				    block_desc->block_len);
7080		else
7081			scsi_ulto3b(0, block_desc->block_len);
7082	}
7083
7084	switch (page_code) {
7085	case SMS_ALL_PAGES_PAGE: {
7086		int i, data_used;
7087
7088		data_used = header_len;
7089		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7090			struct ctl_page_index *page_index;
7091
7092			page_index = &lun->mode_pages.index[i];
7093
7094			if ((control_dev != 0)
7095			 && (page_index->page_flags &
7096			    CTL_PAGE_FLAG_DISK_ONLY))
7097				continue;
7098
7099			/*
7100			 * We don't use this subpage if the user didn't
7101			 * request all subpages.  We already checked (above)
7102			 * to make sure the user only specified a subpage
7103			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7104			 */
7105			if ((page_index->subpage != 0)
7106			 && (subpage == SMS_SUBPAGE_PAGE_0))
7107				continue;
7108
7109			/*
7110			 * Call the handler, if it exists, to update the
7111			 * page to the latest values.
7112			 */
7113			if (page_index->sense_handler != NULL)
7114				page_index->sense_handler(ctsio, page_index,pc);
7115
7116			memcpy(ctsio->kern_data_ptr + data_used,
7117			       page_index->page_data +
7118			       (page_index->page_len * pc),
7119			       page_index->page_len);
7120			data_used += page_index->page_len;
7121		}
7122		break;
7123	}
7124	default: {
7125		int i, data_used;
7126
7127		data_used = header_len;
7128
7129		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7130			struct ctl_page_index *page_index;
7131
7132			page_index = &lun->mode_pages.index[i];
7133
7134			/* Look for the right page code */
7135			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7136				continue;
7137
7138			/* Look for the right subpage or the subpage wildcard*/
7139			if ((page_index->subpage != subpage)
7140			 && (subpage != SMS_SUBPAGE_ALL))
7141				continue;
7142
7143			/* Make sure the page is supported for this dev type */
7144			if ((control_dev != 0)
7145			 && (page_index->page_flags &
7146			     CTL_PAGE_FLAG_DISK_ONLY))
7147				continue;
7148
7149			/*
7150			 * Call the handler, if it exists, to update the
7151			 * page to the latest values.
7152			 */
7153			if (page_index->sense_handler != NULL)
7154				page_index->sense_handler(ctsio, page_index,pc);
7155
7156			memcpy(ctsio->kern_data_ptr + data_used,
7157			       page_index->page_data +
7158			       (page_index->page_len * pc),
7159			       page_index->page_len);
7160			data_used += page_index->page_len;
7161		}
7162		break;
7163	}
7164	}
7165
7166	ctsio->scsi_status = SCSI_STATUS_OK;
7167
7168	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7169	ctsio->be_move_done = ctl_config_move_done;
7170	ctl_datamove((union ctl_io *)ctsio);
7171
7172	return (CTL_RETVAL_COMPLETE);
7173}
7174
7175int
7176ctl_read_capacity(struct ctl_scsiio *ctsio)
7177{
7178	struct scsi_read_capacity *cdb;
7179	struct scsi_read_capacity_data *data;
7180	struct ctl_lun *lun;
7181	uint32_t lba;
7182
7183	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7184
7185	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7186
7187	lba = scsi_4btoul(cdb->addr);
7188	if (((cdb->pmi & SRC_PMI) == 0)
7189	 && (lba != 0)) {
7190		ctl_set_invalid_field(/*ctsio*/ ctsio,
7191				      /*sks_valid*/ 1,
7192				      /*command*/ 1,
7193				      /*field*/ 2,
7194				      /*bit_valid*/ 0,
7195				      /*bit*/ 0);
7196		ctl_done((union ctl_io *)ctsio);
7197		return (CTL_RETVAL_COMPLETE);
7198	}
7199
7200	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7201
7202	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7203	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7204	ctsio->residual = 0;
7205	ctsio->kern_data_len = sizeof(*data);
7206	ctsio->kern_total_len = sizeof(*data);
7207	ctsio->kern_data_resid = 0;
7208	ctsio->kern_rel_offset = 0;
7209	ctsio->kern_sg_entries = 0;
7210
7211	/*
7212	 * If the maximum LBA is greater than 0xfffffffe, the user must
7213	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7214	 * serivce action set.
7215	 */
7216	if (lun->be_lun->maxlba > 0xfffffffe)
7217		scsi_ulto4b(0xffffffff, data->addr);
7218	else
7219		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7220
7221	/*
7222	 * XXX KDM this may not be 512 bytes...
7223	 */
7224	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7225
7226	ctsio->scsi_status = SCSI_STATUS_OK;
7227
7228	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7229	ctsio->be_move_done = ctl_config_move_done;
7230	ctl_datamove((union ctl_io *)ctsio);
7231
7232	return (CTL_RETVAL_COMPLETE);
7233}
7234
7235int
7236ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7237{
7238	struct scsi_read_capacity_16 *cdb;
7239	struct scsi_read_capacity_data_long *data;
7240	struct ctl_lun *lun;
7241	uint64_t lba;
7242	uint32_t alloc_len;
7243
7244	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7245
7246	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7247
7248	alloc_len = scsi_4btoul(cdb->alloc_len);
7249	lba = scsi_8btou64(cdb->addr);
7250
7251	if ((cdb->reladr & SRC16_PMI)
7252	 && (lba != 0)) {
7253		ctl_set_invalid_field(/*ctsio*/ ctsio,
7254				      /*sks_valid*/ 1,
7255				      /*command*/ 1,
7256				      /*field*/ 2,
7257				      /*bit_valid*/ 0,
7258				      /*bit*/ 0);
7259		ctl_done((union ctl_io *)ctsio);
7260		return (CTL_RETVAL_COMPLETE);
7261	}
7262
7263	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7264
7265	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7266	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7267
7268	if (sizeof(*data) < alloc_len) {
7269		ctsio->residual = alloc_len - sizeof(*data);
7270		ctsio->kern_data_len = sizeof(*data);
7271		ctsio->kern_total_len = sizeof(*data);
7272	} else {
7273		ctsio->residual = 0;
7274		ctsio->kern_data_len = alloc_len;
7275		ctsio->kern_total_len = alloc_len;
7276	}
7277	ctsio->kern_data_resid = 0;
7278	ctsio->kern_rel_offset = 0;
7279	ctsio->kern_sg_entries = 0;
7280
7281	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7282	/* XXX KDM this may not be 512 bytes... */
7283	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7284	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7285	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7286	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7287		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7288
7289	ctsio->scsi_status = SCSI_STATUS_OK;
7290
7291	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7292	ctsio->be_move_done = ctl_config_move_done;
7293	ctl_datamove((union ctl_io *)ctsio);
7294
7295	return (CTL_RETVAL_COMPLETE);
7296}
7297
7298int
7299ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7300{
7301	struct scsi_maintenance_in *cdb;
7302	int retval;
7303	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7304	int num_target_port_groups, num_target_ports, single;
7305	struct ctl_lun *lun;
7306	struct ctl_softc *softc;
7307	struct ctl_port *port;
7308	struct scsi_target_group_data *rtg_ptr;
7309	struct scsi_target_group_data_extended *rtg_ext_ptr;
7310	struct scsi_target_port_group_descriptor *tpg_desc;
7311
7312	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7313
7314	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7315	softc = control_softc;
7316	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7317
7318	retval = CTL_RETVAL_COMPLETE;
7319
7320	switch (cdb->byte2 & STG_PDF_MASK) {
7321	case STG_PDF_LENGTH:
7322		ext = 0;
7323		break;
7324	case STG_PDF_EXTENDED:
7325		ext = 1;
7326		break;
7327	default:
7328		ctl_set_invalid_field(/*ctsio*/ ctsio,
7329				      /*sks_valid*/ 1,
7330				      /*command*/ 1,
7331				      /*field*/ 2,
7332				      /*bit_valid*/ 1,
7333				      /*bit*/ 5);
7334		ctl_done((union ctl_io *)ctsio);
7335		return(retval);
7336	}
7337
7338	single = ctl_is_single;
7339	if (single)
7340		num_target_port_groups = 1;
7341	else
7342		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7343	num_target_ports = 0;
7344	mtx_lock(&softc->ctl_lock);
7345	STAILQ_FOREACH(port, &softc->port_list, links) {
7346		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7347			continue;
7348		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7349			continue;
7350		num_target_ports++;
7351	}
7352	mtx_unlock(&softc->ctl_lock);
7353
7354	if (ext)
7355		total_len = sizeof(struct scsi_target_group_data_extended);
7356	else
7357		total_len = sizeof(struct scsi_target_group_data);
7358	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7359		num_target_port_groups +
7360	    sizeof(struct scsi_target_port_descriptor) *
7361		num_target_ports * num_target_port_groups;
7362
7363	alloc_len = scsi_4btoul(cdb->length);
7364
7365	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7366
7367	ctsio->kern_sg_entries = 0;
7368
7369	if (total_len < alloc_len) {
7370		ctsio->residual = alloc_len - total_len;
7371		ctsio->kern_data_len = total_len;
7372		ctsio->kern_total_len = total_len;
7373	} else {
7374		ctsio->residual = 0;
7375		ctsio->kern_data_len = alloc_len;
7376		ctsio->kern_total_len = alloc_len;
7377	}
7378	ctsio->kern_data_resid = 0;
7379	ctsio->kern_rel_offset = 0;
7380
7381	if (ext) {
7382		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7383		    ctsio->kern_data_ptr;
7384		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7385		rtg_ext_ptr->format_type = 0x10;
7386		rtg_ext_ptr->implicit_transition_time = 0;
7387		tpg_desc = &rtg_ext_ptr->groups[0];
7388	} else {
7389		rtg_ptr = (struct scsi_target_group_data *)
7390		    ctsio->kern_data_ptr;
7391		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7392		tpg_desc = &rtg_ptr->groups[0];
7393	}
7394
7395	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7396	mtx_lock(&softc->ctl_lock);
7397	for (g = 0; g < num_target_port_groups; g++) {
7398		if (g == pg)
7399			tpg_desc->pref_state = TPG_PRIMARY |
7400			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7401		else
7402			tpg_desc->pref_state =
7403			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7404		tpg_desc->support = TPG_AO_SUP;
7405		if (!single)
7406			tpg_desc->support |= TPG_AN_SUP;
7407		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7408		tpg_desc->status = TPG_IMPLICIT;
7409		pc = 0;
7410		STAILQ_FOREACH(port, &softc->port_list, links) {
7411			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7412				continue;
7413			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7414			    CTL_MAX_LUNS)
7415				continue;
7416			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7417			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7418			    relative_target_port_identifier);
7419			pc++;
7420		}
7421		tpg_desc->target_port_count = pc;
7422		tpg_desc = (struct scsi_target_port_group_descriptor *)
7423		    &tpg_desc->descriptors[pc];
7424	}
7425	mtx_unlock(&softc->ctl_lock);
7426
7427	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7428	ctsio->be_move_done = ctl_config_move_done;
7429
7430	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7431			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7432			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7433			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7434			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7435
7436	ctl_datamove((union ctl_io *)ctsio);
7437	return(retval);
7438}
7439
7440int
7441ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7442{
7443	struct ctl_lun *lun;
7444	struct scsi_report_supported_opcodes *cdb;
7445	const struct ctl_cmd_entry *entry, *sentry;
7446	struct scsi_report_supported_opcodes_all *all;
7447	struct scsi_report_supported_opcodes_descr *descr;
7448	struct scsi_report_supported_opcodes_one *one;
7449	int retval;
7450	int alloc_len, total_len;
7451	int opcode, service_action, i, j, num;
7452
7453	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7454
7455	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7456	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7457
7458	retval = CTL_RETVAL_COMPLETE;
7459
7460	opcode = cdb->requested_opcode;
7461	service_action = scsi_2btoul(cdb->requested_service_action);
7462	switch (cdb->options & RSO_OPTIONS_MASK) {
7463	case RSO_OPTIONS_ALL:
7464		num = 0;
7465		for (i = 0; i < 256; i++) {
7466			entry = &ctl_cmd_table[i];
7467			if (entry->flags & CTL_CMD_FLAG_SA5) {
7468				for (j = 0; j < 32; j++) {
7469					sentry = &((const struct ctl_cmd_entry *)
7470					    entry->execute)[j];
7471					if (ctl_cmd_applicable(
7472					    lun->be_lun->lun_type, sentry))
7473						num++;
7474				}
7475			} else {
7476				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7477				    entry))
7478					num++;
7479			}
7480		}
7481		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7482		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7483		break;
7484	case RSO_OPTIONS_OC:
7485		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7486			ctl_set_invalid_field(/*ctsio*/ ctsio,
7487					      /*sks_valid*/ 1,
7488					      /*command*/ 1,
7489					      /*field*/ 2,
7490					      /*bit_valid*/ 1,
7491					      /*bit*/ 2);
7492			ctl_done((union ctl_io *)ctsio);
7493			return (CTL_RETVAL_COMPLETE);
7494		}
7495		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7496		break;
7497	case RSO_OPTIONS_OC_SA:
7498		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7499		    service_action >= 32) {
7500			ctl_set_invalid_field(/*ctsio*/ ctsio,
7501					      /*sks_valid*/ 1,
7502					      /*command*/ 1,
7503					      /*field*/ 2,
7504					      /*bit_valid*/ 1,
7505					      /*bit*/ 2);
7506			ctl_done((union ctl_io *)ctsio);
7507			return (CTL_RETVAL_COMPLETE);
7508		}
7509		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7510		break;
7511	default:
7512		ctl_set_invalid_field(/*ctsio*/ ctsio,
7513				      /*sks_valid*/ 1,
7514				      /*command*/ 1,
7515				      /*field*/ 2,
7516				      /*bit_valid*/ 1,
7517				      /*bit*/ 2);
7518		ctl_done((union ctl_io *)ctsio);
7519		return (CTL_RETVAL_COMPLETE);
7520	}
7521
7522	alloc_len = scsi_4btoul(cdb->length);
7523
7524	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7525
7526	ctsio->kern_sg_entries = 0;
7527
7528	if (total_len < alloc_len) {
7529		ctsio->residual = alloc_len - total_len;
7530		ctsio->kern_data_len = total_len;
7531		ctsio->kern_total_len = total_len;
7532	} else {
7533		ctsio->residual = 0;
7534		ctsio->kern_data_len = alloc_len;
7535		ctsio->kern_total_len = alloc_len;
7536	}
7537	ctsio->kern_data_resid = 0;
7538	ctsio->kern_rel_offset = 0;
7539
7540	switch (cdb->options & RSO_OPTIONS_MASK) {
7541	case RSO_OPTIONS_ALL:
7542		all = (struct scsi_report_supported_opcodes_all *)
7543		    ctsio->kern_data_ptr;
7544		num = 0;
7545		for (i = 0; i < 256; i++) {
7546			entry = &ctl_cmd_table[i];
7547			if (entry->flags & CTL_CMD_FLAG_SA5) {
7548				for (j = 0; j < 32; j++) {
7549					sentry = &((const struct ctl_cmd_entry *)
7550					    entry->execute)[j];
7551					if (!ctl_cmd_applicable(
7552					    lun->be_lun->lun_type, sentry))
7553						continue;
7554					descr = &all->descr[num++];
7555					descr->opcode = i;
7556					scsi_ulto2b(j, descr->service_action);
7557					descr->flags = RSO_SERVACTV;
7558					scsi_ulto2b(sentry->length,
7559					    descr->cdb_length);
7560				}
7561			} else {
7562				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7563				    entry))
7564					continue;
7565				descr = &all->descr[num++];
7566				descr->opcode = i;
7567				scsi_ulto2b(0, descr->service_action);
7568				descr->flags = 0;
7569				scsi_ulto2b(entry->length, descr->cdb_length);
7570			}
7571		}
7572		scsi_ulto4b(
7573		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7574		    all->length);
7575		break;
7576	case RSO_OPTIONS_OC:
7577		one = (struct scsi_report_supported_opcodes_one *)
7578		    ctsio->kern_data_ptr;
7579		entry = &ctl_cmd_table[opcode];
7580		goto fill_one;
7581	case RSO_OPTIONS_OC_SA:
7582		one = (struct scsi_report_supported_opcodes_one *)
7583		    ctsio->kern_data_ptr;
7584		entry = &ctl_cmd_table[opcode];
7585		entry = &((const struct ctl_cmd_entry *)
7586		    entry->execute)[service_action];
7587fill_one:
7588		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7589			one->support = 3;
7590			scsi_ulto2b(entry->length, one->cdb_length);
7591			one->cdb_usage[0] = opcode;
7592			memcpy(&one->cdb_usage[1], entry->usage,
7593			    entry->length - 1);
7594		} else
7595			one->support = 1;
7596		break;
7597	}
7598
7599	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7600	ctsio->be_move_done = ctl_config_move_done;
7601
7602	ctl_datamove((union ctl_io *)ctsio);
7603	return(retval);
7604}
7605
7606int
7607ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7608{
7609	struct ctl_lun *lun;
7610	struct scsi_report_supported_tmf *cdb;
7611	struct scsi_report_supported_tmf_data *data;
7612	int retval;
7613	int alloc_len, total_len;
7614
7615	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7616
7617	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7618	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7619
7620	retval = CTL_RETVAL_COMPLETE;
7621
7622	total_len = sizeof(struct scsi_report_supported_tmf_data);
7623	alloc_len = scsi_4btoul(cdb->length);
7624
7625	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7626
7627	ctsio->kern_sg_entries = 0;
7628
7629	if (total_len < alloc_len) {
7630		ctsio->residual = alloc_len - total_len;
7631		ctsio->kern_data_len = total_len;
7632		ctsio->kern_total_len = total_len;
7633	} else {
7634		ctsio->residual = 0;
7635		ctsio->kern_data_len = alloc_len;
7636		ctsio->kern_total_len = alloc_len;
7637	}
7638	ctsio->kern_data_resid = 0;
7639	ctsio->kern_rel_offset = 0;
7640
7641	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7642	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7643	data->byte2 |= RST_ITNRS;
7644
7645	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7646	ctsio->be_move_done = ctl_config_move_done;
7647
7648	ctl_datamove((union ctl_io *)ctsio);
7649	return (retval);
7650}
7651
7652int
7653ctl_report_timestamp(struct ctl_scsiio *ctsio)
7654{
7655	struct ctl_lun *lun;
7656	struct scsi_report_timestamp *cdb;
7657	struct scsi_report_timestamp_data *data;
7658	struct timeval tv;
7659	int64_t timestamp;
7660	int retval;
7661	int alloc_len, total_len;
7662
7663	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7664
7665	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7666	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7667
7668	retval = CTL_RETVAL_COMPLETE;
7669
7670	total_len = sizeof(struct scsi_report_timestamp_data);
7671	alloc_len = scsi_4btoul(cdb->length);
7672
7673	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7674
7675	ctsio->kern_sg_entries = 0;
7676
7677	if (total_len < alloc_len) {
7678		ctsio->residual = alloc_len - total_len;
7679		ctsio->kern_data_len = total_len;
7680		ctsio->kern_total_len = total_len;
7681	} else {
7682		ctsio->residual = 0;
7683		ctsio->kern_data_len = alloc_len;
7684		ctsio->kern_total_len = alloc_len;
7685	}
7686	ctsio->kern_data_resid = 0;
7687	ctsio->kern_rel_offset = 0;
7688
7689	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7690	scsi_ulto2b(sizeof(*data) - 2, data->length);
7691	data->origin = RTS_ORIG_OUTSIDE;
7692	getmicrotime(&tv);
7693	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7694	scsi_ulto4b(timestamp >> 16, data->timestamp);
7695	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7696
7697	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7698	ctsio->be_move_done = ctl_config_move_done;
7699
7700	ctl_datamove((union ctl_io *)ctsio);
7701	return (retval);
7702}
7703
7704int
7705ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7706{
7707	struct scsi_per_res_in *cdb;
7708	int alloc_len, total_len = 0;
7709	/* struct scsi_per_res_in_rsrv in_data; */
7710	struct ctl_lun *lun;
7711	struct ctl_softc *softc;
7712
7713	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7714
7715	softc = control_softc;
7716
7717	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7718
7719	alloc_len = scsi_2btoul(cdb->length);
7720
7721	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7722
7723retry:
7724	mtx_lock(&lun->lun_lock);
7725	switch (cdb->action) {
7726	case SPRI_RK: /* read keys */
7727		total_len = sizeof(struct scsi_per_res_in_keys) +
7728			lun->pr_key_count *
7729			sizeof(struct scsi_per_res_key);
7730		break;
7731	case SPRI_RR: /* read reservation */
7732		if (lun->flags & CTL_LUN_PR_RESERVED)
7733			total_len = sizeof(struct scsi_per_res_in_rsrv);
7734		else
7735			total_len = sizeof(struct scsi_per_res_in_header);
7736		break;
7737	case SPRI_RC: /* report capabilities */
7738		total_len = sizeof(struct scsi_per_res_cap);
7739		break;
7740	case SPRI_RS: /* read full status */
7741		total_len = sizeof(struct scsi_per_res_in_header) +
7742		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7743		    lun->pr_key_count;
7744		break;
7745	default:
7746		panic("Invalid PR type %x", cdb->action);
7747	}
7748	mtx_unlock(&lun->lun_lock);
7749
7750	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7751
7752	if (total_len < alloc_len) {
7753		ctsio->residual = alloc_len - total_len;
7754		ctsio->kern_data_len = total_len;
7755		ctsio->kern_total_len = total_len;
7756	} else {
7757		ctsio->residual = 0;
7758		ctsio->kern_data_len = alloc_len;
7759		ctsio->kern_total_len = alloc_len;
7760	}
7761
7762	ctsio->kern_data_resid = 0;
7763	ctsio->kern_rel_offset = 0;
7764	ctsio->kern_sg_entries = 0;
7765
7766	mtx_lock(&lun->lun_lock);
7767	switch (cdb->action) {
7768	case SPRI_RK: { // read keys
7769        struct scsi_per_res_in_keys *res_keys;
7770		int i, key_count;
7771
7772		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7773
7774		/*
7775		 * We had to drop the lock to allocate our buffer, which
7776		 * leaves time for someone to come in with another
7777		 * persistent reservation.  (That is unlikely, though,
7778		 * since this should be the only persistent reservation
7779		 * command active right now.)
7780		 */
7781		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7782		    (lun->pr_key_count *
7783		     sizeof(struct scsi_per_res_key)))){
7784			mtx_unlock(&lun->lun_lock);
7785			free(ctsio->kern_data_ptr, M_CTL);
7786			printf("%s: reservation length changed, retrying\n",
7787			       __func__);
7788			goto retry;
7789		}
7790
7791		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7792
7793		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7794			     lun->pr_key_count, res_keys->header.length);
7795
7796		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7797			if (!lun->per_res[i].registered)
7798				continue;
7799
7800			/*
7801			 * We used lun->pr_key_count to calculate the
7802			 * size to allocate.  If it turns out the number of
7803			 * initiators with the registered flag set is
7804			 * larger than that (i.e. they haven't been kept in
7805			 * sync), we've got a problem.
7806			 */
7807			if (key_count >= lun->pr_key_count) {
7808#ifdef NEEDTOPORT
7809				csevent_log(CSC_CTL | CSC_SHELF_SW |
7810					    CTL_PR_ERROR,
7811					    csevent_LogType_Fault,
7812					    csevent_AlertLevel_Yellow,
7813					    csevent_FRU_ShelfController,
7814					    csevent_FRU_Firmware,
7815				        csevent_FRU_Unknown,
7816					    "registered keys %d >= key "
7817					    "count %d", key_count,
7818					    lun->pr_key_count);
7819#endif
7820				key_count++;
7821				continue;
7822			}
7823			memcpy(res_keys->keys[key_count].key,
7824			       lun->per_res[i].res_key.key,
7825			       ctl_min(sizeof(res_keys->keys[key_count].key),
7826			       sizeof(lun->per_res[i].res_key)));
7827			key_count++;
7828		}
7829		break;
7830	}
7831	case SPRI_RR: { // read reservation
7832		struct scsi_per_res_in_rsrv *res;
7833		int tmp_len, header_only;
7834
7835		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7836
7837		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7838
7839		if (lun->flags & CTL_LUN_PR_RESERVED)
7840		{
7841			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7842			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7843				    res->header.length);
7844			header_only = 0;
7845		} else {
7846			tmp_len = sizeof(struct scsi_per_res_in_header);
7847			scsi_ulto4b(0, res->header.length);
7848			header_only = 1;
7849		}
7850
7851		/*
7852		 * We had to drop the lock to allocate our buffer, which
7853		 * leaves time for someone to come in with another
7854		 * persistent reservation.  (That is unlikely, though,
7855		 * since this should be the only persistent reservation
7856		 * command active right now.)
7857		 */
7858		if (tmp_len != total_len) {
7859			mtx_unlock(&lun->lun_lock);
7860			free(ctsio->kern_data_ptr, M_CTL);
7861			printf("%s: reservation status changed, retrying\n",
7862			       __func__);
7863			goto retry;
7864		}
7865
7866		/*
7867		 * No reservation held, so we're done.
7868		 */
7869		if (header_only != 0)
7870			break;
7871
7872		/*
7873		 * If the registration is an All Registrants type, the key
7874		 * is 0, since it doesn't really matter.
7875		 */
7876		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7877			memcpy(res->data.reservation,
7878			       &lun->per_res[lun->pr_res_idx].res_key,
7879			       sizeof(struct scsi_per_res_key));
7880		}
7881		res->data.scopetype = lun->res_type;
7882		break;
7883	}
7884	case SPRI_RC:     //report capabilities
7885	{
7886		struct scsi_per_res_cap *res_cap;
7887		uint16_t type_mask;
7888
7889		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7890		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7891		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7892		type_mask = SPRI_TM_WR_EX_AR |
7893			    SPRI_TM_EX_AC_RO |
7894			    SPRI_TM_WR_EX_RO |
7895			    SPRI_TM_EX_AC |
7896			    SPRI_TM_WR_EX |
7897			    SPRI_TM_EX_AC_AR;
7898		scsi_ulto2b(type_mask, res_cap->type_mask);
7899		break;
7900	}
7901	case SPRI_RS: { // read full status
7902		struct scsi_per_res_in_full *res_status;
7903		struct scsi_per_res_in_full_desc *res_desc;
7904		struct ctl_port *port;
7905		int i, len;
7906
7907		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7908
7909		/*
7910		 * We had to drop the lock to allocate our buffer, which
7911		 * leaves time for someone to come in with another
7912		 * persistent reservation.  (That is unlikely, though,
7913		 * since this should be the only persistent reservation
7914		 * command active right now.)
7915		 */
7916		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7917		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7918		     lun->pr_key_count)){
7919			mtx_unlock(&lun->lun_lock);
7920			free(ctsio->kern_data_ptr, M_CTL);
7921			printf("%s: reservation length changed, retrying\n",
7922			       __func__);
7923			goto retry;
7924		}
7925
7926		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7927
7928		res_desc = &res_status->desc[0];
7929		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7930			if (!lun->per_res[i].registered)
7931				continue;
7932
7933			memcpy(&res_desc->res_key, &lun->per_res[i].res_key.key,
7934			    sizeof(res_desc->res_key));
7935			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7936			    (lun->pr_res_idx == i ||
7937			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7938				res_desc->flags = SPRI_FULL_R_HOLDER;
7939				res_desc->scopetype = lun->res_type;
7940			}
7941			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7942			    res_desc->rel_trgt_port_id);
7943			len = 0;
7944			port = softc->ctl_ports[
7945			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7946			if (port != NULL)
7947				len = ctl_create_iid(port,
7948				    i % CTL_MAX_INIT_PER_PORT,
7949				    res_desc->transport_id);
7950			scsi_ulto4b(len, res_desc->additional_length);
7951			res_desc = (struct scsi_per_res_in_full_desc *)
7952			    &res_desc->transport_id[len];
7953		}
7954		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7955		    res_status->header.length);
7956		break;
7957	}
7958	default:
7959		/*
7960		 * This is a bug, because we just checked for this above,
7961		 * and should have returned an error.
7962		 */
7963		panic("Invalid PR type %x", cdb->action);
7964		break; /* NOTREACHED */
7965	}
7966	mtx_unlock(&lun->lun_lock);
7967
7968	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7969	ctsio->be_move_done = ctl_config_move_done;
7970
7971	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7972			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7973			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7974			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7975			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7976
7977	ctl_datamove((union ctl_io *)ctsio);
7978
7979	return (CTL_RETVAL_COMPLETE);
7980}
7981
7982/*
7983 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7984 * it should return.
7985 */
7986static int
7987ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7988		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7989		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7990		struct scsi_per_res_out_parms* param)
7991{
7992	union ctl_ha_msg persis_io;
7993	int retval, i;
7994	int isc_retval;
7995
7996	retval = 0;
7997
7998	mtx_lock(&lun->lun_lock);
7999	if (sa_res_key == 0) {
8000		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8001			/* validate scope and type */
8002			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8003			     SPR_LU_SCOPE) {
8004				mtx_unlock(&lun->lun_lock);
8005				ctl_set_invalid_field(/*ctsio*/ ctsio,
8006						      /*sks_valid*/ 1,
8007						      /*command*/ 1,
8008						      /*field*/ 2,
8009						      /*bit_valid*/ 1,
8010						      /*bit*/ 4);
8011				ctl_done((union ctl_io *)ctsio);
8012				return (1);
8013			}
8014
8015		        if (type>8 || type==2 || type==4 || type==0) {
8016				mtx_unlock(&lun->lun_lock);
8017				ctl_set_invalid_field(/*ctsio*/ ctsio,
8018       	           				      /*sks_valid*/ 1,
8019						      /*command*/ 1,
8020						      /*field*/ 2,
8021						      /*bit_valid*/ 1,
8022						      /*bit*/ 0);
8023				ctl_done((union ctl_io *)ctsio);
8024				return (1);
8025		        }
8026
8027			/* temporarily unregister this nexus */
8028			lun->per_res[residx].registered = 0;
8029
8030			/*
8031			 * Unregister everybody else and build UA for
8032			 * them
8033			 */
8034			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8035				if (lun->per_res[i].registered == 0)
8036					continue;
8037
8038				if (!persis_offset
8039				 && i <CTL_MAX_INITIATORS)
8040					lun->pending_ua[i] |=
8041						CTL_UA_REG_PREEMPT;
8042				else if (persis_offset
8043				      && i >= persis_offset)
8044					lun->pending_ua[i-persis_offset] |=
8045						CTL_UA_REG_PREEMPT;
8046				lun->per_res[i].registered = 0;
8047				memset(&lun->per_res[i].res_key, 0,
8048				       sizeof(struct scsi_per_res_key));
8049			}
8050			lun->per_res[residx].registered = 1;
8051			lun->pr_key_count = 1;
8052			lun->res_type = type;
8053			if (lun->res_type != SPR_TYPE_WR_EX_AR
8054			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8055				lun->pr_res_idx = residx;
8056
8057			/* send msg to other side */
8058			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8059			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8060			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8061			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8062			persis_io.pr.pr_info.res_type = type;
8063			memcpy(persis_io.pr.pr_info.sa_res_key,
8064			       param->serv_act_res_key,
8065			       sizeof(param->serv_act_res_key));
8066			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8067			     &persis_io, sizeof(persis_io), 0)) >
8068			     CTL_HA_STATUS_SUCCESS) {
8069				printf("CTL:Persis Out error returned "
8070				       "from ctl_ha_msg_send %d\n",
8071				       isc_retval);
8072			}
8073		} else {
8074			/* not all registrants */
8075			mtx_unlock(&lun->lun_lock);
8076			free(ctsio->kern_data_ptr, M_CTL);
8077			ctl_set_invalid_field(ctsio,
8078					      /*sks_valid*/ 1,
8079					      /*command*/ 0,
8080					      /*field*/ 8,
8081					      /*bit_valid*/ 0,
8082					      /*bit*/ 0);
8083			ctl_done((union ctl_io *)ctsio);
8084			return (1);
8085		}
8086	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8087		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8088		int found = 0;
8089
8090		if (res_key == sa_res_key) {
8091			/* special case */
8092			/*
8093			 * The spec implies this is not good but doesn't
8094			 * say what to do. There are two choices either
8095			 * generate a res conflict or check condition
8096			 * with illegal field in parameter data. Since
8097			 * that is what is done when the sa_res_key is
8098			 * zero I'll take that approach since this has
8099			 * to do with the sa_res_key.
8100			 */
8101			mtx_unlock(&lun->lun_lock);
8102			free(ctsio->kern_data_ptr, M_CTL);
8103			ctl_set_invalid_field(ctsio,
8104					      /*sks_valid*/ 1,
8105					      /*command*/ 0,
8106					      /*field*/ 8,
8107					      /*bit_valid*/ 0,
8108					      /*bit*/ 0);
8109			ctl_done((union ctl_io *)ctsio);
8110			return (1);
8111		}
8112
8113		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8114			if (lun->per_res[i].registered
8115			 && memcmp(param->serv_act_res_key,
8116			    lun->per_res[i].res_key.key,
8117			    sizeof(struct scsi_per_res_key)) != 0)
8118				continue;
8119
8120			found = 1;
8121			lun->per_res[i].registered = 0;
8122			memset(&lun->per_res[i].res_key, 0,
8123			       sizeof(struct scsi_per_res_key));
8124			lun->pr_key_count--;
8125
8126			if (!persis_offset && i < CTL_MAX_INITIATORS)
8127				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8128			else if (persis_offset && i >= persis_offset)
8129				lun->pending_ua[i-persis_offset] |=
8130					CTL_UA_REG_PREEMPT;
8131		}
8132		if (!found) {
8133			mtx_unlock(&lun->lun_lock);
8134			free(ctsio->kern_data_ptr, M_CTL);
8135			ctl_set_reservation_conflict(ctsio);
8136			ctl_done((union ctl_io *)ctsio);
8137			return (CTL_RETVAL_COMPLETE);
8138		}
8139		/* send msg to other side */
8140		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8141		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8142		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8143		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8144		persis_io.pr.pr_info.res_type = type;
8145		memcpy(persis_io.pr.pr_info.sa_res_key,
8146		       param->serv_act_res_key,
8147		       sizeof(param->serv_act_res_key));
8148		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8149		     &persis_io, sizeof(persis_io), 0)) >
8150		     CTL_HA_STATUS_SUCCESS) {
8151			printf("CTL:Persis Out error returned from "
8152			       "ctl_ha_msg_send %d\n", isc_retval);
8153		}
8154	} else {
8155		/* Reserved but not all registrants */
8156		/* sa_res_key is res holder */
8157		if (memcmp(param->serv_act_res_key,
8158                   lun->per_res[lun->pr_res_idx].res_key.key,
8159                   sizeof(struct scsi_per_res_key)) == 0) {
8160			/* validate scope and type */
8161			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8162			     SPR_LU_SCOPE) {
8163				mtx_unlock(&lun->lun_lock);
8164				ctl_set_invalid_field(/*ctsio*/ ctsio,
8165						      /*sks_valid*/ 1,
8166						      /*command*/ 1,
8167						      /*field*/ 2,
8168						      /*bit_valid*/ 1,
8169						      /*bit*/ 4);
8170				ctl_done((union ctl_io *)ctsio);
8171				return (1);
8172			}
8173
8174			if (type>8 || type==2 || type==4 || type==0) {
8175				mtx_unlock(&lun->lun_lock);
8176				ctl_set_invalid_field(/*ctsio*/ ctsio,
8177						      /*sks_valid*/ 1,
8178						      /*command*/ 1,
8179						      /*field*/ 2,
8180						      /*bit_valid*/ 1,
8181						      /*bit*/ 0);
8182				ctl_done((union ctl_io *)ctsio);
8183				return (1);
8184			}
8185
8186			/*
8187			 * Do the following:
8188			 * if sa_res_key != res_key remove all
8189			 * registrants w/sa_res_key and generate UA
8190			 * for these registrants(Registrations
8191			 * Preempted) if it wasn't an exclusive
8192			 * reservation generate UA(Reservations
8193			 * Preempted) for all other registered nexuses
8194			 * if the type has changed. Establish the new
8195			 * reservation and holder. If res_key and
8196			 * sa_res_key are the same do the above
8197			 * except don't unregister the res holder.
8198			 */
8199
8200			/*
8201			 * Temporarily unregister so it won't get
8202			 * removed or UA generated
8203			 */
8204			lun->per_res[residx].registered = 0;
8205			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8206				if (lun->per_res[i].registered == 0)
8207					continue;
8208
8209				if (memcmp(param->serv_act_res_key,
8210				    lun->per_res[i].res_key.key,
8211				    sizeof(struct scsi_per_res_key)) == 0) {
8212					lun->per_res[i].registered = 0;
8213					memset(&lun->per_res[i].res_key,
8214					       0,
8215					       sizeof(struct scsi_per_res_key));
8216					lun->pr_key_count--;
8217
8218					if (!persis_offset
8219					 && i < CTL_MAX_INITIATORS)
8220						lun->pending_ua[i] |=
8221							CTL_UA_REG_PREEMPT;
8222					else if (persis_offset
8223					      && i >= persis_offset)
8224						lun->pending_ua[i-persis_offset] |=
8225						  CTL_UA_REG_PREEMPT;
8226				} else if (type != lun->res_type
8227					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8228					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8229						if (!persis_offset
8230						 && i < CTL_MAX_INITIATORS)
8231							lun->pending_ua[i] |=
8232							CTL_UA_RES_RELEASE;
8233						else if (persis_offset
8234						      && i >= persis_offset)
8235							lun->pending_ua[
8236							i-persis_offset] |=
8237							CTL_UA_RES_RELEASE;
8238				}
8239			}
8240			lun->per_res[residx].registered = 1;
8241			lun->res_type = type;
8242			if (lun->res_type != SPR_TYPE_WR_EX_AR
8243			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8244				lun->pr_res_idx = residx;
8245			else
8246				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8247
8248			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8249			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8250			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8251			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8252			persis_io.pr.pr_info.res_type = type;
8253			memcpy(persis_io.pr.pr_info.sa_res_key,
8254			       param->serv_act_res_key,
8255			       sizeof(param->serv_act_res_key));
8256			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8257			     &persis_io, sizeof(persis_io), 0)) >
8258			     CTL_HA_STATUS_SUCCESS) {
8259				printf("CTL:Persis Out error returned "
8260				       "from ctl_ha_msg_send %d\n",
8261				       isc_retval);
8262			}
8263		} else {
8264			/*
8265			 * sa_res_key is not the res holder just
8266			 * remove registrants
8267			 */
8268			int found=0;
8269
8270			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8271				if (memcmp(param->serv_act_res_key,
8272				    lun->per_res[i].res_key.key,
8273				    sizeof(struct scsi_per_res_key)) != 0)
8274					continue;
8275
8276				found = 1;
8277				lun->per_res[i].registered = 0;
8278				memset(&lun->per_res[i].res_key, 0,
8279				       sizeof(struct scsi_per_res_key));
8280				lun->pr_key_count--;
8281
8282				if (!persis_offset
8283				 && i < CTL_MAX_INITIATORS)
8284					lun->pending_ua[i] |=
8285						CTL_UA_REG_PREEMPT;
8286				else if (persis_offset
8287				      && i >= persis_offset)
8288					lun->pending_ua[i-persis_offset] |=
8289						CTL_UA_REG_PREEMPT;
8290			}
8291
8292			if (!found) {
8293				mtx_unlock(&lun->lun_lock);
8294				free(ctsio->kern_data_ptr, M_CTL);
8295				ctl_set_reservation_conflict(ctsio);
8296				ctl_done((union ctl_io *)ctsio);
8297		        	return (1);
8298			}
8299			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8300			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8301			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8302			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8303			persis_io.pr.pr_info.res_type = type;
8304			memcpy(persis_io.pr.pr_info.sa_res_key,
8305			       param->serv_act_res_key,
8306			       sizeof(param->serv_act_res_key));
8307			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8308			     &persis_io, sizeof(persis_io), 0)) >
8309			     CTL_HA_STATUS_SUCCESS) {
8310				printf("CTL:Persis Out error returned "
8311				       "from ctl_ha_msg_send %d\n",
8312				isc_retval);
8313			}
8314		}
8315	}
8316
8317	lun->PRGeneration++;
8318	mtx_unlock(&lun->lun_lock);
8319
8320	return (retval);
8321}
8322
8323static void
8324ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8325{
8326	int i;
8327
8328	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8329	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8330	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8331		   msg->pr.pr_info.sa_res_key,
8332		   sizeof(struct scsi_per_res_key)) != 0) {
8333		uint64_t sa_res_key;
8334		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8335
8336		if (sa_res_key == 0) {
8337			/* temporarily unregister this nexus */
8338			lun->per_res[msg->pr.pr_info.residx].registered = 0;
8339
8340			/*
8341			 * Unregister everybody else and build UA for
8342			 * them
8343			 */
8344			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8345				if (lun->per_res[i].registered == 0)
8346					continue;
8347
8348				if (!persis_offset
8349				 && i < CTL_MAX_INITIATORS)
8350					lun->pending_ua[i] |=
8351						CTL_UA_REG_PREEMPT;
8352				else if (persis_offset && i >= persis_offset)
8353					lun->pending_ua[i - persis_offset] |=
8354						CTL_UA_REG_PREEMPT;
8355				lun->per_res[i].registered = 0;
8356				memset(&lun->per_res[i].res_key, 0,
8357				       sizeof(struct scsi_per_res_key));
8358			}
8359
8360			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8361			lun->pr_key_count = 1;
8362			lun->res_type = msg->pr.pr_info.res_type;
8363			if (lun->res_type != SPR_TYPE_WR_EX_AR
8364			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8365				lun->pr_res_idx = msg->pr.pr_info.residx;
8366		} else {
8367		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8368				if (memcmp(msg->pr.pr_info.sa_res_key,
8369		                   lun->per_res[i].res_key.key,
8370		                   sizeof(struct scsi_per_res_key)) != 0)
8371					continue;
8372
8373				lun->per_res[i].registered = 0;
8374				memset(&lun->per_res[i].res_key, 0,
8375				       sizeof(struct scsi_per_res_key));
8376				lun->pr_key_count--;
8377
8378				if (!persis_offset
8379				 && i < persis_offset)
8380					lun->pending_ua[i] |=
8381						CTL_UA_REG_PREEMPT;
8382				else if (persis_offset
8383				      && i >= persis_offset)
8384					lun->pending_ua[i - persis_offset] |=
8385						CTL_UA_REG_PREEMPT;
8386			}
8387		}
8388	} else {
8389		/*
8390		 * Temporarily unregister so it won't get removed
8391		 * or UA generated
8392		 */
8393		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8394		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8395			if (lun->per_res[i].registered == 0)
8396				continue;
8397
8398			if (memcmp(msg->pr.pr_info.sa_res_key,
8399	                   lun->per_res[i].res_key.key,
8400	                   sizeof(struct scsi_per_res_key)) == 0) {
8401				lun->per_res[i].registered = 0;
8402				memset(&lun->per_res[i].res_key, 0,
8403				       sizeof(struct scsi_per_res_key));
8404				lun->pr_key_count--;
8405				if (!persis_offset
8406				 && i < CTL_MAX_INITIATORS)
8407					lun->pending_ua[i] |=
8408						CTL_UA_REG_PREEMPT;
8409				else if (persis_offset
8410				      && i >= persis_offset)
8411					lun->pending_ua[i - persis_offset] |=
8412						CTL_UA_REG_PREEMPT;
8413			} else if (msg->pr.pr_info.res_type != lun->res_type
8414				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8415				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8416					if (!persis_offset
8417					 && i < persis_offset)
8418						lun->pending_ua[i] |=
8419							CTL_UA_RES_RELEASE;
8420					else if (persis_offset
8421					      && i >= persis_offset)
8422					lun->pending_ua[i - persis_offset] |=
8423						CTL_UA_RES_RELEASE;
8424			}
8425		}
8426		lun->per_res[msg->pr.pr_info.residx].registered = 1;
8427		lun->res_type = msg->pr.pr_info.res_type;
8428		if (lun->res_type != SPR_TYPE_WR_EX_AR
8429		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8430			lun->pr_res_idx = msg->pr.pr_info.residx;
8431		else
8432			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8433	}
8434	lun->PRGeneration++;
8435
8436}
8437
8438
8439int
8440ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8441{
8442	int retval;
8443	int isc_retval;
8444	u_int32_t param_len;
8445	struct scsi_per_res_out *cdb;
8446	struct ctl_lun *lun;
8447	struct scsi_per_res_out_parms* param;
8448	struct ctl_softc *softc;
8449	uint32_t residx;
8450	uint64_t res_key, sa_res_key;
8451	uint8_t type;
8452	union ctl_ha_msg persis_io;
8453	int    i;
8454
8455	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8456
8457	retval = CTL_RETVAL_COMPLETE;
8458
8459	softc = control_softc;
8460
8461	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8462	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8463
8464	/*
8465	 * We only support whole-LUN scope.  The scope & type are ignored for
8466	 * register, register and ignore existing key and clear.
8467	 * We sometimes ignore scope and type on preempts too!!
8468	 * Verify reservation type here as well.
8469	 */
8470	type = cdb->scope_type & SPR_TYPE_MASK;
8471	if ((cdb->action == SPRO_RESERVE)
8472	 || (cdb->action == SPRO_RELEASE)) {
8473		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8474			ctl_set_invalid_field(/*ctsio*/ ctsio,
8475					      /*sks_valid*/ 1,
8476					      /*command*/ 1,
8477					      /*field*/ 2,
8478					      /*bit_valid*/ 1,
8479					      /*bit*/ 4);
8480			ctl_done((union ctl_io *)ctsio);
8481			return (CTL_RETVAL_COMPLETE);
8482		}
8483
8484		if (type>8 || type==2 || type==4 || type==0) {
8485			ctl_set_invalid_field(/*ctsio*/ ctsio,
8486					      /*sks_valid*/ 1,
8487					      /*command*/ 1,
8488					      /*field*/ 2,
8489					      /*bit_valid*/ 1,
8490					      /*bit*/ 0);
8491			ctl_done((union ctl_io *)ctsio);
8492			return (CTL_RETVAL_COMPLETE);
8493		}
8494	}
8495
8496	param_len = scsi_4btoul(cdb->length);
8497
8498	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8499		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8500		ctsio->kern_data_len = param_len;
8501		ctsio->kern_total_len = param_len;
8502		ctsio->kern_data_resid = 0;
8503		ctsio->kern_rel_offset = 0;
8504		ctsio->kern_sg_entries = 0;
8505		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8506		ctsio->be_move_done = ctl_config_move_done;
8507		ctl_datamove((union ctl_io *)ctsio);
8508
8509		return (CTL_RETVAL_COMPLETE);
8510	}
8511
8512	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8513
8514	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8515	res_key = scsi_8btou64(param->res_key.key);
8516	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8517
8518	/*
8519	 * Validate the reservation key here except for SPRO_REG_IGNO
8520	 * This must be done for all other service actions
8521	 */
8522	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8523		mtx_lock(&lun->lun_lock);
8524		if (lun->per_res[residx].registered) {
8525		    if (memcmp(param->res_key.key,
8526			       lun->per_res[residx].res_key.key,
8527			       ctl_min(sizeof(param->res_key),
8528			       sizeof(lun->per_res[residx].res_key))) != 0) {
8529				/*
8530				 * The current key passed in doesn't match
8531				 * the one the initiator previously
8532				 * registered.
8533				 */
8534				mtx_unlock(&lun->lun_lock);
8535				free(ctsio->kern_data_ptr, M_CTL);
8536				ctl_set_reservation_conflict(ctsio);
8537				ctl_done((union ctl_io *)ctsio);
8538				return (CTL_RETVAL_COMPLETE);
8539			}
8540		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8541			/*
8542			 * We are not registered
8543			 */
8544			mtx_unlock(&lun->lun_lock);
8545			free(ctsio->kern_data_ptr, M_CTL);
8546			ctl_set_reservation_conflict(ctsio);
8547			ctl_done((union ctl_io *)ctsio);
8548			return (CTL_RETVAL_COMPLETE);
8549		} else if (res_key != 0) {
8550			/*
8551			 * We are not registered and trying to register but
8552			 * the register key isn't zero.
8553			 */
8554			mtx_unlock(&lun->lun_lock);
8555			free(ctsio->kern_data_ptr, M_CTL);
8556			ctl_set_reservation_conflict(ctsio);
8557			ctl_done((union ctl_io *)ctsio);
8558			return (CTL_RETVAL_COMPLETE);
8559		}
8560		mtx_unlock(&lun->lun_lock);
8561	}
8562
8563	switch (cdb->action & SPRO_ACTION_MASK) {
8564	case SPRO_REGISTER:
8565	case SPRO_REG_IGNO: {
8566
8567#if 0
8568		printf("Registration received\n");
8569#endif
8570
8571		/*
8572		 * We don't support any of these options, as we report in
8573		 * the read capabilities request (see
8574		 * ctl_persistent_reserve_in(), above).
8575		 */
8576		if ((param->flags & SPR_SPEC_I_PT)
8577		 || (param->flags & SPR_ALL_TG_PT)
8578		 || (param->flags & SPR_APTPL)) {
8579			int bit_ptr;
8580
8581			if (param->flags & SPR_APTPL)
8582				bit_ptr = 0;
8583			else if (param->flags & SPR_ALL_TG_PT)
8584				bit_ptr = 2;
8585			else /* SPR_SPEC_I_PT */
8586				bit_ptr = 3;
8587
8588			free(ctsio->kern_data_ptr, M_CTL);
8589			ctl_set_invalid_field(ctsio,
8590					      /*sks_valid*/ 1,
8591					      /*command*/ 0,
8592					      /*field*/ 20,
8593					      /*bit_valid*/ 1,
8594					      /*bit*/ bit_ptr);
8595			ctl_done((union ctl_io *)ctsio);
8596			return (CTL_RETVAL_COMPLETE);
8597		}
8598
8599		mtx_lock(&lun->lun_lock);
8600
8601		/*
8602		 * The initiator wants to clear the
8603		 * key/unregister.
8604		 */
8605		if (sa_res_key == 0) {
8606			if ((res_key == 0
8607			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8608			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8609			  && !lun->per_res[residx].registered)) {
8610				mtx_unlock(&lun->lun_lock);
8611				goto done;
8612			}
8613
8614			lun->per_res[residx].registered = 0;
8615			memset(&lun->per_res[residx].res_key,
8616			       0, sizeof(lun->per_res[residx].res_key));
8617			lun->pr_key_count--;
8618
8619			if (residx == lun->pr_res_idx) {
8620				lun->flags &= ~CTL_LUN_PR_RESERVED;
8621				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8622
8623				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8624				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8625				 && lun->pr_key_count) {
8626					/*
8627					 * If the reservation is a registrants
8628					 * only type we need to generate a UA
8629					 * for other registered inits.  The
8630					 * sense code should be RESERVATIONS
8631					 * RELEASED
8632					 */
8633
8634					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8635						if (lun->per_res[
8636						    i+persis_offset].registered
8637						    == 0)
8638							continue;
8639						lun->pending_ua[i] |=
8640							CTL_UA_RES_RELEASE;
8641					}
8642				}
8643				lun->res_type = 0;
8644			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8645				if (lun->pr_key_count==0) {
8646					lun->flags &= ~CTL_LUN_PR_RESERVED;
8647					lun->res_type = 0;
8648					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8649				}
8650			}
8651			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8652			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8653			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8654			persis_io.pr.pr_info.residx = residx;
8655			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8656			     &persis_io, sizeof(persis_io), 0 )) >
8657			     CTL_HA_STATUS_SUCCESS) {
8658				printf("CTL:Persis Out error returned from "
8659				       "ctl_ha_msg_send %d\n", isc_retval);
8660			}
8661		} else /* sa_res_key != 0 */ {
8662
8663			/*
8664			 * If we aren't registered currently then increment
8665			 * the key count and set the registered flag.
8666			 */
8667			if (!lun->per_res[residx].registered) {
8668				lun->pr_key_count++;
8669				lun->per_res[residx].registered = 1;
8670			}
8671
8672			memcpy(&lun->per_res[residx].res_key,
8673			       param->serv_act_res_key,
8674			       ctl_min(sizeof(param->serv_act_res_key),
8675			       sizeof(lun->per_res[residx].res_key)));
8676
8677			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8678			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8679			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8680			persis_io.pr.pr_info.residx = residx;
8681			memcpy(persis_io.pr.pr_info.sa_res_key,
8682			       param->serv_act_res_key,
8683			       sizeof(param->serv_act_res_key));
8684			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8685			     &persis_io, sizeof(persis_io), 0)) >
8686			     CTL_HA_STATUS_SUCCESS) {
8687				printf("CTL:Persis Out error returned from "
8688				       "ctl_ha_msg_send %d\n", isc_retval);
8689			}
8690		}
8691		lun->PRGeneration++;
8692		mtx_unlock(&lun->lun_lock);
8693
8694		break;
8695	}
8696	case SPRO_RESERVE:
8697#if 0
8698                printf("Reserve executed type %d\n", type);
8699#endif
8700		mtx_lock(&lun->lun_lock);
8701		if (lun->flags & CTL_LUN_PR_RESERVED) {
8702			/*
8703			 * if this isn't the reservation holder and it's
8704			 * not a "all registrants" type or if the type is
8705			 * different then we have a conflict
8706			 */
8707			if ((lun->pr_res_idx != residx
8708			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8709			 || lun->res_type != type) {
8710				mtx_unlock(&lun->lun_lock);
8711				free(ctsio->kern_data_ptr, M_CTL);
8712				ctl_set_reservation_conflict(ctsio);
8713				ctl_done((union ctl_io *)ctsio);
8714				return (CTL_RETVAL_COMPLETE);
8715			}
8716			mtx_unlock(&lun->lun_lock);
8717		} else /* create a reservation */ {
8718			/*
8719			 * If it's not an "all registrants" type record
8720			 * reservation holder
8721			 */
8722			if (type != SPR_TYPE_WR_EX_AR
8723			 && type != SPR_TYPE_EX_AC_AR)
8724				lun->pr_res_idx = residx; /* Res holder */
8725			else
8726				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8727
8728			lun->flags |= CTL_LUN_PR_RESERVED;
8729			lun->res_type = type;
8730
8731			mtx_unlock(&lun->lun_lock);
8732
8733			/* send msg to other side */
8734			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8735			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8736			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8737			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8738			persis_io.pr.pr_info.res_type = type;
8739			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8740			     &persis_io, sizeof(persis_io), 0)) >
8741			     CTL_HA_STATUS_SUCCESS) {
8742				printf("CTL:Persis Out error returned from "
8743				       "ctl_ha_msg_send %d\n", isc_retval);
8744			}
8745		}
8746		break;
8747
8748	case SPRO_RELEASE:
8749		mtx_lock(&lun->lun_lock);
8750		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8751			/* No reservation exists return good status */
8752			mtx_unlock(&lun->lun_lock);
8753			goto done;
8754		}
8755		/*
8756		 * Is this nexus a reservation holder?
8757		 */
8758		if (lun->pr_res_idx != residx
8759		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8760			/*
8761			 * not a res holder return good status but
8762			 * do nothing
8763			 */
8764			mtx_unlock(&lun->lun_lock);
8765			goto done;
8766		}
8767
8768		if (lun->res_type != type) {
8769			mtx_unlock(&lun->lun_lock);
8770			free(ctsio->kern_data_ptr, M_CTL);
8771			ctl_set_illegal_pr_release(ctsio);
8772			ctl_done((union ctl_io *)ctsio);
8773			return (CTL_RETVAL_COMPLETE);
8774		}
8775
8776		/* okay to release */
8777		lun->flags &= ~CTL_LUN_PR_RESERVED;
8778		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8779		lun->res_type = 0;
8780
8781		/*
8782		 * if this isn't an exclusive access
8783		 * res generate UA for all other
8784		 * registrants.
8785		 */
8786		if (type != SPR_TYPE_EX_AC
8787		 && type != SPR_TYPE_WR_EX) {
8788			/*
8789			 * temporarily unregister so we don't generate UA
8790			 */
8791			lun->per_res[residx].registered = 0;
8792
8793			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8794				if (lun->per_res[i+persis_offset].registered
8795				    == 0)
8796					continue;
8797				lun->pending_ua[i] |=
8798					CTL_UA_RES_RELEASE;
8799			}
8800
8801			lun->per_res[residx].registered = 1;
8802		}
8803		mtx_unlock(&lun->lun_lock);
8804		/* Send msg to other side */
8805		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8806		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8807		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8808		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8809		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8810			printf("CTL:Persis Out error returned from "
8811			       "ctl_ha_msg_send %d\n", isc_retval);
8812		}
8813		break;
8814
8815	case SPRO_CLEAR:
8816		/* send msg to other side */
8817
8818		mtx_lock(&lun->lun_lock);
8819		lun->flags &= ~CTL_LUN_PR_RESERVED;
8820		lun->res_type = 0;
8821		lun->pr_key_count = 0;
8822		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8823
8824
8825		memset(&lun->per_res[residx].res_key,
8826		       0, sizeof(lun->per_res[residx].res_key));
8827		lun->per_res[residx].registered = 0;
8828
8829		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8830			if (lun->per_res[i].registered) {
8831				if (!persis_offset && i < CTL_MAX_INITIATORS)
8832					lun->pending_ua[i] |=
8833						CTL_UA_RES_PREEMPT;
8834				else if (persis_offset && i >= persis_offset)
8835					lun->pending_ua[i-persis_offset] |=
8836					    CTL_UA_RES_PREEMPT;
8837
8838				memset(&lun->per_res[i].res_key,
8839				       0, sizeof(struct scsi_per_res_key));
8840				lun->per_res[i].registered = 0;
8841			}
8842		lun->PRGeneration++;
8843		mtx_unlock(&lun->lun_lock);
8844		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8845		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8846		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8847		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8848		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8849			printf("CTL:Persis Out error returned from "
8850			       "ctl_ha_msg_send %d\n", isc_retval);
8851		}
8852		break;
8853
8854	case SPRO_PREEMPT: {
8855		int nretval;
8856
8857		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8858					  residx, ctsio, cdb, param);
8859		if (nretval != 0)
8860			return (CTL_RETVAL_COMPLETE);
8861		break;
8862	}
8863	default:
8864		panic("Invalid PR type %x", cdb->action);
8865	}
8866
8867done:
8868	free(ctsio->kern_data_ptr, M_CTL);
8869	ctl_set_success(ctsio);
8870	ctl_done((union ctl_io *)ctsio);
8871
8872	return (retval);
8873}
8874
8875/*
8876 * This routine is for handling a message from the other SC pertaining to
8877 * persistent reserve out. All the error checking will have been done
8878 * so only perorming the action need be done here to keep the two
8879 * in sync.
8880 */
8881static void
8882ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8883{
8884	struct ctl_lun *lun;
8885	struct ctl_softc *softc;
8886	int i;
8887	uint32_t targ_lun;
8888
8889	softc = control_softc;
8890
8891	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8892	lun = softc->ctl_luns[targ_lun];
8893	mtx_lock(&lun->lun_lock);
8894	switch(msg->pr.pr_info.action) {
8895	case CTL_PR_REG_KEY:
8896		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8897			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8898			lun->pr_key_count++;
8899		}
8900		lun->PRGeneration++;
8901		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8902		       msg->pr.pr_info.sa_res_key,
8903		       sizeof(struct scsi_per_res_key));
8904		break;
8905
8906	case CTL_PR_UNREG_KEY:
8907		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8908		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8909		       0, sizeof(struct scsi_per_res_key));
8910		lun->pr_key_count--;
8911
8912		/* XXX Need to see if the reservation has been released */
8913		/* if so do we need to generate UA? */
8914		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8915			lun->flags &= ~CTL_LUN_PR_RESERVED;
8916			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8917
8918			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8919			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8920			 && lun->pr_key_count) {
8921				/*
8922				 * If the reservation is a registrants
8923				 * only type we need to generate a UA
8924				 * for other registered inits.  The
8925				 * sense code should be RESERVATIONS
8926				 * RELEASED
8927				 */
8928
8929				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8930					if (lun->per_res[i+
8931					    persis_offset].registered == 0)
8932						continue;
8933
8934					lun->pending_ua[i] |=
8935						CTL_UA_RES_RELEASE;
8936				}
8937			}
8938			lun->res_type = 0;
8939		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8940			if (lun->pr_key_count==0) {
8941				lun->flags &= ~CTL_LUN_PR_RESERVED;
8942				lun->res_type = 0;
8943				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8944			}
8945		}
8946		lun->PRGeneration++;
8947		break;
8948
8949	case CTL_PR_RESERVE:
8950		lun->flags |= CTL_LUN_PR_RESERVED;
8951		lun->res_type = msg->pr.pr_info.res_type;
8952		lun->pr_res_idx = msg->pr.pr_info.residx;
8953
8954		break;
8955
8956	case CTL_PR_RELEASE:
8957		/*
8958		 * if this isn't an exclusive access res generate UA for all
8959		 * other registrants.
8960		 */
8961		if (lun->res_type != SPR_TYPE_EX_AC
8962		 && lun->res_type != SPR_TYPE_WR_EX) {
8963			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8964				if (lun->per_res[i+persis_offset].registered)
8965					lun->pending_ua[i] |=
8966						CTL_UA_RES_RELEASE;
8967		}
8968
8969		lun->flags &= ~CTL_LUN_PR_RESERVED;
8970		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8971		lun->res_type = 0;
8972		break;
8973
8974	case CTL_PR_PREEMPT:
8975		ctl_pro_preempt_other(lun, msg);
8976		break;
8977	case CTL_PR_CLEAR:
8978		lun->flags &= ~CTL_LUN_PR_RESERVED;
8979		lun->res_type = 0;
8980		lun->pr_key_count = 0;
8981		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8982
8983		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8984			if (lun->per_res[i].registered == 0)
8985				continue;
8986			if (!persis_offset
8987			 && i < CTL_MAX_INITIATORS)
8988				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
8989			else if (persis_offset
8990			      && i >= persis_offset)
8991				lun->pending_ua[i-persis_offset] |=
8992					CTL_UA_RES_PREEMPT;
8993			memset(&lun->per_res[i].res_key, 0,
8994			       sizeof(struct scsi_per_res_key));
8995			lun->per_res[i].registered = 0;
8996		}
8997		lun->PRGeneration++;
8998		break;
8999	}
9000
9001	mtx_unlock(&lun->lun_lock);
9002}
9003
9004int
9005ctl_read_write(struct ctl_scsiio *ctsio)
9006{
9007	struct ctl_lun *lun;
9008	struct ctl_lba_len_flags *lbalen;
9009	uint64_t lba;
9010	uint32_t num_blocks;
9011	int flags, retval;
9012	int isread;
9013
9014	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9015
9016	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
9017
9018	flags = 0;
9019	retval = CTL_RETVAL_COMPLETE;
9020
9021	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
9022	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
9023	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
9024		uint32_t residx;
9025
9026		/*
9027		 * XXX KDM need a lock here.
9028		 */
9029		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9030		if ((lun->res_type == SPR_TYPE_EX_AC
9031		  && residx != lun->pr_res_idx)
9032		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
9033		   || lun->res_type == SPR_TYPE_EX_AC_AR)
9034		  && !lun->per_res[residx].registered)) {
9035			ctl_set_reservation_conflict(ctsio);
9036			ctl_done((union ctl_io *)ctsio);
9037			return (CTL_RETVAL_COMPLETE);
9038	        }
9039	}
9040
9041	switch (ctsio->cdb[0]) {
9042	case READ_6:
9043	case WRITE_6: {
9044		struct scsi_rw_6 *cdb;
9045
9046		cdb = (struct scsi_rw_6 *)ctsio->cdb;
9047
9048		lba = scsi_3btoul(cdb->addr);
9049		/* only 5 bits are valid in the most significant address byte */
9050		lba &= 0x1fffff;
9051		num_blocks = cdb->length;
9052		/*
9053		 * This is correct according to SBC-2.
9054		 */
9055		if (num_blocks == 0)
9056			num_blocks = 256;
9057		break;
9058	}
9059	case READ_10:
9060	case WRITE_10: {
9061		struct scsi_rw_10 *cdb;
9062
9063		cdb = (struct scsi_rw_10 *)ctsio->cdb;
9064		if (cdb->byte2 & SRW10_FUA)
9065			flags |= CTL_LLF_FUA;
9066		if (cdb->byte2 & SRW10_DPO)
9067			flags |= CTL_LLF_DPO;
9068		lba = scsi_4btoul(cdb->addr);
9069		num_blocks = scsi_2btoul(cdb->length);
9070		break;
9071	}
9072	case WRITE_VERIFY_10: {
9073		struct scsi_write_verify_10 *cdb;
9074
9075		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9076		flags |= CTL_LLF_FUA;
9077		if (cdb->byte2 & SWV_DPO)
9078			flags |= CTL_LLF_DPO;
9079		lba = scsi_4btoul(cdb->addr);
9080		num_blocks = scsi_2btoul(cdb->length);
9081		break;
9082	}
9083	case READ_12:
9084	case WRITE_12: {
9085		struct scsi_rw_12 *cdb;
9086
9087		cdb = (struct scsi_rw_12 *)ctsio->cdb;
9088		if (cdb->byte2 & SRW12_FUA)
9089			flags |= CTL_LLF_FUA;
9090		if (cdb->byte2 & SRW12_DPO)
9091			flags |= CTL_LLF_DPO;
9092		lba = scsi_4btoul(cdb->addr);
9093		num_blocks = scsi_4btoul(cdb->length);
9094		break;
9095	}
9096	case WRITE_VERIFY_12: {
9097		struct scsi_write_verify_12 *cdb;
9098
9099		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9100		flags |= CTL_LLF_FUA;
9101		if (cdb->byte2 & SWV_DPO)
9102			flags |= CTL_LLF_DPO;
9103		lba = scsi_4btoul(cdb->addr);
9104		num_blocks = scsi_4btoul(cdb->length);
9105		break;
9106	}
9107	case READ_16:
9108	case WRITE_16: {
9109		struct scsi_rw_16 *cdb;
9110
9111		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9112		if (cdb->byte2 & SRW12_FUA)
9113			flags |= CTL_LLF_FUA;
9114		if (cdb->byte2 & SRW12_DPO)
9115			flags |= CTL_LLF_DPO;
9116		lba = scsi_8btou64(cdb->addr);
9117		num_blocks = scsi_4btoul(cdb->length);
9118		break;
9119	}
9120	case WRITE_VERIFY_16: {
9121		struct scsi_write_verify_16 *cdb;
9122
9123		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9124		flags |= CTL_LLF_FUA;
9125		if (cdb->byte2 & SWV_DPO)
9126			flags |= CTL_LLF_DPO;
9127		lba = scsi_8btou64(cdb->addr);
9128		num_blocks = scsi_4btoul(cdb->length);
9129		break;
9130	}
9131	default:
9132		/*
9133		 * We got a command we don't support.  This shouldn't
9134		 * happen, commands should be filtered out above us.
9135		 */
9136		ctl_set_invalid_opcode(ctsio);
9137		ctl_done((union ctl_io *)ctsio);
9138
9139		return (CTL_RETVAL_COMPLETE);
9140		break; /* NOTREACHED */
9141	}
9142
9143	/*
9144	 * The first check is to make sure we're in bounds, the second
9145	 * check is to catch wrap-around problems.  If the lba + num blocks
9146	 * is less than the lba, then we've wrapped around and the block
9147	 * range is invalid anyway.
9148	 */
9149	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9150	 || ((lba + num_blocks) < lba)) {
9151		ctl_set_lba_out_of_range(ctsio);
9152		ctl_done((union ctl_io *)ctsio);
9153		return (CTL_RETVAL_COMPLETE);
9154	}
9155
9156	/*
9157	 * According to SBC-3, a transfer length of 0 is not an error.
9158	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9159	 * translates to 256 blocks for those commands.
9160	 */
9161	if (num_blocks == 0) {
9162		ctl_set_success(ctsio);
9163		ctl_done((union ctl_io *)ctsio);
9164		return (CTL_RETVAL_COMPLETE);
9165	}
9166
9167	/* Set FUA and/or DPO if caches are disabled. */
9168	if (isread) {
9169		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9170		    SCP_RCD) != 0)
9171			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9172	} else {
9173		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9174		    SCP_WCE) == 0)
9175			flags |= CTL_LLF_FUA;
9176	}
9177
9178	lbalen = (struct ctl_lba_len_flags *)
9179	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9180	lbalen->lba = lba;
9181	lbalen->len = num_blocks;
9182	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9183
9184	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9185	ctsio->kern_rel_offset = 0;
9186
9187	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9188
9189	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9190
9191	return (retval);
9192}
9193
9194static int
9195ctl_cnw_cont(union ctl_io *io)
9196{
9197	struct ctl_scsiio *ctsio;
9198	struct ctl_lun *lun;
9199	struct ctl_lba_len_flags *lbalen;
9200	int retval;
9201
9202	ctsio = &io->scsiio;
9203	ctsio->io_hdr.status = CTL_STATUS_NONE;
9204	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9205	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9206	lbalen = (struct ctl_lba_len_flags *)
9207	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9208	lbalen->flags &= ~CTL_LLF_COMPARE;
9209	lbalen->flags |= CTL_LLF_WRITE;
9210
9211	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9212	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9213	return (retval);
9214}
9215
9216int
9217ctl_cnw(struct ctl_scsiio *ctsio)
9218{
9219	struct ctl_lun *lun;
9220	struct ctl_lba_len_flags *lbalen;
9221	uint64_t lba;
9222	uint32_t num_blocks;
9223	int flags, retval;
9224
9225	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9226
9227	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9228
9229	flags = 0;
9230	retval = CTL_RETVAL_COMPLETE;
9231
9232	switch (ctsio->cdb[0]) {
9233	case COMPARE_AND_WRITE: {
9234		struct scsi_compare_and_write *cdb;
9235
9236		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9237		if (cdb->byte2 & SRW10_FUA)
9238			flags |= CTL_LLF_FUA;
9239		if (cdb->byte2 & SRW10_DPO)
9240			flags |= CTL_LLF_DPO;
9241		lba = scsi_8btou64(cdb->addr);
9242		num_blocks = cdb->length;
9243		break;
9244	}
9245	default:
9246		/*
9247		 * We got a command we don't support.  This shouldn't
9248		 * happen, commands should be filtered out above us.
9249		 */
9250		ctl_set_invalid_opcode(ctsio);
9251		ctl_done((union ctl_io *)ctsio);
9252
9253		return (CTL_RETVAL_COMPLETE);
9254		break; /* NOTREACHED */
9255	}
9256
9257	/*
9258	 * The first check is to make sure we're in bounds, the second
9259	 * check is to catch wrap-around problems.  If the lba + num blocks
9260	 * is less than the lba, then we've wrapped around and the block
9261	 * range is invalid anyway.
9262	 */
9263	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9264	 || ((lba + num_blocks) < lba)) {
9265		ctl_set_lba_out_of_range(ctsio);
9266		ctl_done((union ctl_io *)ctsio);
9267		return (CTL_RETVAL_COMPLETE);
9268	}
9269
9270	/*
9271	 * According to SBC-3, a transfer length of 0 is not an error.
9272	 */
9273	if (num_blocks == 0) {
9274		ctl_set_success(ctsio);
9275		ctl_done((union ctl_io *)ctsio);
9276		return (CTL_RETVAL_COMPLETE);
9277	}
9278
9279	/* Set FUA if write cache is disabled. */
9280	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9281	    SCP_WCE) == 0)
9282		flags |= CTL_LLF_FUA;
9283
9284	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9285	ctsio->kern_rel_offset = 0;
9286
9287	/*
9288	 * Set the IO_CONT flag, so that if this I/O gets passed to
9289	 * ctl_data_submit_done(), it'll get passed back to
9290	 * ctl_ctl_cnw_cont() for further processing.
9291	 */
9292	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9293	ctsio->io_cont = ctl_cnw_cont;
9294
9295	lbalen = (struct ctl_lba_len_flags *)
9296	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9297	lbalen->lba = lba;
9298	lbalen->len = num_blocks;
9299	lbalen->flags = CTL_LLF_COMPARE | flags;
9300
9301	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9302	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9303	return (retval);
9304}
9305
9306int
9307ctl_verify(struct ctl_scsiio *ctsio)
9308{
9309	struct ctl_lun *lun;
9310	struct ctl_lba_len_flags *lbalen;
9311	uint64_t lba;
9312	uint32_t num_blocks;
9313	int bytchk, flags;
9314	int retval;
9315
9316	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9317
9318	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9319
9320	bytchk = 0;
9321	flags = CTL_LLF_FUA;
9322	retval = CTL_RETVAL_COMPLETE;
9323
9324	switch (ctsio->cdb[0]) {
9325	case VERIFY_10: {
9326		struct scsi_verify_10 *cdb;
9327
9328		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9329		if (cdb->byte2 & SVFY_BYTCHK)
9330			bytchk = 1;
9331		if (cdb->byte2 & SVFY_DPO)
9332			flags |= CTL_LLF_DPO;
9333		lba = scsi_4btoul(cdb->addr);
9334		num_blocks = scsi_2btoul(cdb->length);
9335		break;
9336	}
9337	case VERIFY_12: {
9338		struct scsi_verify_12 *cdb;
9339
9340		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9341		if (cdb->byte2 & SVFY_BYTCHK)
9342			bytchk = 1;
9343		if (cdb->byte2 & SVFY_DPO)
9344			flags |= CTL_LLF_DPO;
9345		lba = scsi_4btoul(cdb->addr);
9346		num_blocks = scsi_4btoul(cdb->length);
9347		break;
9348	}
9349	case VERIFY_16: {
9350		struct scsi_rw_16 *cdb;
9351
9352		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9353		if (cdb->byte2 & SVFY_BYTCHK)
9354			bytchk = 1;
9355		if (cdb->byte2 & SVFY_DPO)
9356			flags |= CTL_LLF_DPO;
9357		lba = scsi_8btou64(cdb->addr);
9358		num_blocks = scsi_4btoul(cdb->length);
9359		break;
9360	}
9361	default:
9362		/*
9363		 * We got a command we don't support.  This shouldn't
9364		 * happen, commands should be filtered out above us.
9365		 */
9366		ctl_set_invalid_opcode(ctsio);
9367		ctl_done((union ctl_io *)ctsio);
9368		return (CTL_RETVAL_COMPLETE);
9369	}
9370
9371	/*
9372	 * The first check is to make sure we're in bounds, the second
9373	 * check is to catch wrap-around problems.  If the lba + num blocks
9374	 * is less than the lba, then we've wrapped around and the block
9375	 * range is invalid anyway.
9376	 */
9377	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9378	 || ((lba + num_blocks) < lba)) {
9379		ctl_set_lba_out_of_range(ctsio);
9380		ctl_done((union ctl_io *)ctsio);
9381		return (CTL_RETVAL_COMPLETE);
9382	}
9383
9384	/*
9385	 * According to SBC-3, a transfer length of 0 is not an error.
9386	 */
9387	if (num_blocks == 0) {
9388		ctl_set_success(ctsio);
9389		ctl_done((union ctl_io *)ctsio);
9390		return (CTL_RETVAL_COMPLETE);
9391	}
9392
9393	lbalen = (struct ctl_lba_len_flags *)
9394	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9395	lbalen->lba = lba;
9396	lbalen->len = num_blocks;
9397	if (bytchk) {
9398		lbalen->flags = CTL_LLF_COMPARE | flags;
9399		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9400	} else {
9401		lbalen->flags = CTL_LLF_VERIFY | flags;
9402		ctsio->kern_total_len = 0;
9403	}
9404	ctsio->kern_rel_offset = 0;
9405
9406	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9407	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9408	return (retval);
9409}
9410
9411int
9412ctl_report_luns(struct ctl_scsiio *ctsio)
9413{
9414	struct scsi_report_luns *cdb;
9415	struct scsi_report_luns_data *lun_data;
9416	struct ctl_lun *lun, *request_lun;
9417	int num_luns, retval;
9418	uint32_t alloc_len, lun_datalen;
9419	int num_filled, well_known;
9420	uint32_t initidx, targ_lun_id, lun_id;
9421
9422	retval = CTL_RETVAL_COMPLETE;
9423	well_known = 0;
9424
9425	cdb = (struct scsi_report_luns *)ctsio->cdb;
9426
9427	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9428
9429	mtx_lock(&control_softc->ctl_lock);
9430	num_luns = control_softc->num_luns;
9431	mtx_unlock(&control_softc->ctl_lock);
9432
9433	switch (cdb->select_report) {
9434	case RPL_REPORT_DEFAULT:
9435	case RPL_REPORT_ALL:
9436		break;
9437	case RPL_REPORT_WELLKNOWN:
9438		well_known = 1;
9439		num_luns = 0;
9440		break;
9441	default:
9442		ctl_set_invalid_field(ctsio,
9443				      /*sks_valid*/ 1,
9444				      /*command*/ 1,
9445				      /*field*/ 2,
9446				      /*bit_valid*/ 0,
9447				      /*bit*/ 0);
9448		ctl_done((union ctl_io *)ctsio);
9449		return (retval);
9450		break; /* NOTREACHED */
9451	}
9452
9453	alloc_len = scsi_4btoul(cdb->length);
9454	/*
9455	 * The initiator has to allocate at least 16 bytes for this request,
9456	 * so he can at least get the header and the first LUN.  Otherwise
9457	 * we reject the request (per SPC-3 rev 14, section 6.21).
9458	 */
9459	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9460	    sizeof(struct scsi_report_luns_lundata))) {
9461		ctl_set_invalid_field(ctsio,
9462				      /*sks_valid*/ 1,
9463				      /*command*/ 1,
9464				      /*field*/ 6,
9465				      /*bit_valid*/ 0,
9466				      /*bit*/ 0);
9467		ctl_done((union ctl_io *)ctsio);
9468		return (retval);
9469	}
9470
9471	request_lun = (struct ctl_lun *)
9472		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9473
9474	lun_datalen = sizeof(*lun_data) +
9475		(num_luns * sizeof(struct scsi_report_luns_lundata));
9476
9477	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9478	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9479	ctsio->kern_sg_entries = 0;
9480
9481	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9482
9483	mtx_lock(&control_softc->ctl_lock);
9484	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9485		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9486		if (lun_id >= CTL_MAX_LUNS)
9487			continue;
9488		lun = control_softc->ctl_luns[lun_id];
9489		if (lun == NULL)
9490			continue;
9491
9492		if (targ_lun_id <= 0xff) {
9493			/*
9494			 * Peripheral addressing method, bus number 0.
9495			 */
9496			lun_data->luns[num_filled].lundata[0] =
9497				RPL_LUNDATA_ATYP_PERIPH;
9498			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9499			num_filled++;
9500		} else if (targ_lun_id <= 0x3fff) {
9501			/*
9502			 * Flat addressing method.
9503			 */
9504			lun_data->luns[num_filled].lundata[0] =
9505				RPL_LUNDATA_ATYP_FLAT |
9506				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9507#ifdef OLDCTLHEADERS
9508				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9509				(targ_lun_id & SRLD_BUS_LUN_MASK);
9510#endif
9511			lun_data->luns[num_filled].lundata[1] =
9512#ifdef OLDCTLHEADERS
9513				targ_lun_id >> SRLD_BUS_LUN_BITS;
9514#endif
9515				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9516			num_filled++;
9517		} else {
9518			printf("ctl_report_luns: bogus LUN number %jd, "
9519			       "skipping\n", (intmax_t)targ_lun_id);
9520		}
9521		/*
9522		 * According to SPC-3, rev 14 section 6.21:
9523		 *
9524		 * "The execution of a REPORT LUNS command to any valid and
9525		 * installed logical unit shall clear the REPORTED LUNS DATA
9526		 * HAS CHANGED unit attention condition for all logical
9527		 * units of that target with respect to the requesting
9528		 * initiator. A valid and installed logical unit is one
9529		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9530		 * INQUIRY data (see 6.4.2)."
9531		 *
9532		 * If request_lun is NULL, the LUN this report luns command
9533		 * was issued to is either disabled or doesn't exist. In that
9534		 * case, we shouldn't clear any pending lun change unit
9535		 * attention.
9536		 */
9537		if (request_lun != NULL) {
9538			mtx_lock(&lun->lun_lock);
9539			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9540			mtx_unlock(&lun->lun_lock);
9541		}
9542	}
9543	mtx_unlock(&control_softc->ctl_lock);
9544
9545	/*
9546	 * It's quite possible that we've returned fewer LUNs than we allocated
9547	 * space for.  Trim it.
9548	 */
9549	lun_datalen = sizeof(*lun_data) +
9550		(num_filled * sizeof(struct scsi_report_luns_lundata));
9551
9552	if (lun_datalen < alloc_len) {
9553		ctsio->residual = alloc_len - lun_datalen;
9554		ctsio->kern_data_len = lun_datalen;
9555		ctsio->kern_total_len = lun_datalen;
9556	} else {
9557		ctsio->residual = 0;
9558		ctsio->kern_data_len = alloc_len;
9559		ctsio->kern_total_len = alloc_len;
9560	}
9561	ctsio->kern_data_resid = 0;
9562	ctsio->kern_rel_offset = 0;
9563	ctsio->kern_sg_entries = 0;
9564
9565	/*
9566	 * We set this to the actual data length, regardless of how much
9567	 * space we actually have to return results.  If the user looks at
9568	 * this value, he'll know whether or not he allocated enough space
9569	 * and reissue the command if necessary.  We don't support well
9570	 * known logical units, so if the user asks for that, return none.
9571	 */
9572	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9573
9574	/*
9575	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9576	 * this request.
9577	 */
9578	ctsio->scsi_status = SCSI_STATUS_OK;
9579
9580	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9581	ctsio->be_move_done = ctl_config_move_done;
9582	ctl_datamove((union ctl_io *)ctsio);
9583
9584	return (retval);
9585}
9586
9587int
9588ctl_request_sense(struct ctl_scsiio *ctsio)
9589{
9590	struct scsi_request_sense *cdb;
9591	struct scsi_sense_data *sense_ptr;
9592	struct ctl_lun *lun;
9593	uint32_t initidx;
9594	int have_error;
9595	scsi_sense_data_type sense_format;
9596
9597	cdb = (struct scsi_request_sense *)ctsio->cdb;
9598
9599	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9600
9601	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9602
9603	/*
9604	 * Determine which sense format the user wants.
9605	 */
9606	if (cdb->byte2 & SRS_DESC)
9607		sense_format = SSD_TYPE_DESC;
9608	else
9609		sense_format = SSD_TYPE_FIXED;
9610
9611	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9612	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9613	ctsio->kern_sg_entries = 0;
9614
9615	/*
9616	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9617	 * larger than the largest allowed value for the length field in the
9618	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9619	 */
9620	ctsio->residual = 0;
9621	ctsio->kern_data_len = cdb->length;
9622	ctsio->kern_total_len = cdb->length;
9623
9624	ctsio->kern_data_resid = 0;
9625	ctsio->kern_rel_offset = 0;
9626	ctsio->kern_sg_entries = 0;
9627
9628	/*
9629	 * If we don't have a LUN, we don't have any pending sense.
9630	 */
9631	if (lun == NULL)
9632		goto no_sense;
9633
9634	have_error = 0;
9635	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9636	/*
9637	 * Check for pending sense, and then for pending unit attentions.
9638	 * Pending sense gets returned first, then pending unit attentions.
9639	 */
9640	mtx_lock(&lun->lun_lock);
9641#ifdef CTL_WITH_CA
9642	if (ctl_is_set(lun->have_ca, initidx)) {
9643		scsi_sense_data_type stored_format;
9644
9645		/*
9646		 * Check to see which sense format was used for the stored
9647		 * sense data.
9648		 */
9649		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9650
9651		/*
9652		 * If the user requested a different sense format than the
9653		 * one we stored, then we need to convert it to the other
9654		 * format.  If we're going from descriptor to fixed format
9655		 * sense data, we may lose things in translation, depending
9656		 * on what options were used.
9657		 *
9658		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9659		 * for some reason we'll just copy it out as-is.
9660		 */
9661		if ((stored_format == SSD_TYPE_FIXED)
9662		 && (sense_format == SSD_TYPE_DESC))
9663			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9664			    &lun->pending_sense[initidx],
9665			    (struct scsi_sense_data_desc *)sense_ptr);
9666		else if ((stored_format == SSD_TYPE_DESC)
9667		      && (sense_format == SSD_TYPE_FIXED))
9668			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9669			    &lun->pending_sense[initidx],
9670			    (struct scsi_sense_data_fixed *)sense_ptr);
9671		else
9672			memcpy(sense_ptr, &lun->pending_sense[initidx],
9673			       ctl_min(sizeof(*sense_ptr),
9674			       sizeof(lun->pending_sense[initidx])));
9675
9676		ctl_clear_mask(lun->have_ca, initidx);
9677		have_error = 1;
9678	} else
9679#endif
9680	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9681		ctl_ua_type ua_type;
9682
9683		ua_type = ctl_build_ua(lun->pending_ua[initidx],
9684				       sense_ptr, sense_format);
9685		if (ua_type != CTL_UA_NONE) {
9686			have_error = 1;
9687			/* We're reporting this UA, so clear it */
9688			lun->pending_ua[initidx] &= ~ua_type;
9689		}
9690	}
9691	mtx_unlock(&lun->lun_lock);
9692
9693	/*
9694	 * We already have a pending error, return it.
9695	 */
9696	if (have_error != 0) {
9697		/*
9698		 * We report the SCSI status as OK, since the status of the
9699		 * request sense command itself is OK.
9700		 */
9701		ctsio->scsi_status = SCSI_STATUS_OK;
9702
9703		/*
9704		 * We report 0 for the sense length, because we aren't doing
9705		 * autosense in this case.  We're reporting sense as
9706		 * parameter data.
9707		 */
9708		ctsio->sense_len = 0;
9709		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9710		ctsio->be_move_done = ctl_config_move_done;
9711		ctl_datamove((union ctl_io *)ctsio);
9712
9713		return (CTL_RETVAL_COMPLETE);
9714	}
9715
9716no_sense:
9717
9718	/*
9719	 * No sense information to report, so we report that everything is
9720	 * okay.
9721	 */
9722	ctl_set_sense_data(sense_ptr,
9723			   lun,
9724			   sense_format,
9725			   /*current_error*/ 1,
9726			   /*sense_key*/ SSD_KEY_NO_SENSE,
9727			   /*asc*/ 0x00,
9728			   /*ascq*/ 0x00,
9729			   SSD_ELEM_NONE);
9730
9731	ctsio->scsi_status = SCSI_STATUS_OK;
9732
9733	/*
9734	 * We report 0 for the sense length, because we aren't doing
9735	 * autosense in this case.  We're reporting sense as parameter data.
9736	 */
9737	ctsio->sense_len = 0;
9738	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9739	ctsio->be_move_done = ctl_config_move_done;
9740	ctl_datamove((union ctl_io *)ctsio);
9741
9742	return (CTL_RETVAL_COMPLETE);
9743}
9744
9745int
9746ctl_tur(struct ctl_scsiio *ctsio)
9747{
9748	struct ctl_lun *lun;
9749
9750	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9751
9752	CTL_DEBUG_PRINT(("ctl_tur\n"));
9753
9754	if (lun == NULL)
9755		return (EINVAL);
9756
9757	ctsio->scsi_status = SCSI_STATUS_OK;
9758	ctsio->io_hdr.status = CTL_SUCCESS;
9759
9760	ctl_done((union ctl_io *)ctsio);
9761
9762	return (CTL_RETVAL_COMPLETE);
9763}
9764
9765#ifdef notyet
9766static int
9767ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9768{
9769
9770}
9771#endif
9772
9773static int
9774ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9775{
9776	struct scsi_vpd_supported_pages *pages;
9777	int sup_page_size;
9778	struct ctl_lun *lun;
9779
9780	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9781
9782	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9783	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9784	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9785	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9786	ctsio->kern_sg_entries = 0;
9787
9788	if (sup_page_size < alloc_len) {
9789		ctsio->residual = alloc_len - sup_page_size;
9790		ctsio->kern_data_len = sup_page_size;
9791		ctsio->kern_total_len = sup_page_size;
9792	} else {
9793		ctsio->residual = 0;
9794		ctsio->kern_data_len = alloc_len;
9795		ctsio->kern_total_len = alloc_len;
9796	}
9797	ctsio->kern_data_resid = 0;
9798	ctsio->kern_rel_offset = 0;
9799	ctsio->kern_sg_entries = 0;
9800
9801	/*
9802	 * The control device is always connected.  The disk device, on the
9803	 * other hand, may not be online all the time.  Need to change this
9804	 * to figure out whether the disk device is actually online or not.
9805	 */
9806	if (lun != NULL)
9807		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9808				lun->be_lun->lun_type;
9809	else
9810		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9811
9812	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9813	/* Supported VPD pages */
9814	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9815	/* Serial Number */
9816	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9817	/* Device Identification */
9818	pages->page_list[2] = SVPD_DEVICE_ID;
9819	/* Extended INQUIRY Data */
9820	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9821	/* Mode Page Policy */
9822	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9823	/* SCSI Ports */
9824	pages->page_list[5] = SVPD_SCSI_PORTS;
9825	/* Third-party Copy */
9826	pages->page_list[6] = SVPD_SCSI_TPC;
9827	/* Block limits */
9828	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9829	/* Block Device Characteristics */
9830	pages->page_list[8] = SVPD_BDC;
9831	/* Logical Block Provisioning */
9832	pages->page_list[9] = SVPD_LBP;
9833
9834	ctsio->scsi_status = SCSI_STATUS_OK;
9835
9836	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9837	ctsio->be_move_done = ctl_config_move_done;
9838	ctl_datamove((union ctl_io *)ctsio);
9839
9840	return (CTL_RETVAL_COMPLETE);
9841}
9842
9843static int
9844ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9845{
9846	struct scsi_vpd_unit_serial_number *sn_ptr;
9847	struct ctl_lun *lun;
9848
9849	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9850
9851	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9852	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9853	ctsio->kern_sg_entries = 0;
9854
9855	if (sizeof(*sn_ptr) < alloc_len) {
9856		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9857		ctsio->kern_data_len = sizeof(*sn_ptr);
9858		ctsio->kern_total_len = sizeof(*sn_ptr);
9859	} else {
9860		ctsio->residual = 0;
9861		ctsio->kern_data_len = alloc_len;
9862		ctsio->kern_total_len = alloc_len;
9863	}
9864	ctsio->kern_data_resid = 0;
9865	ctsio->kern_rel_offset = 0;
9866	ctsio->kern_sg_entries = 0;
9867
9868	/*
9869	 * The control device is always connected.  The disk device, on the
9870	 * other hand, may not be online all the time.  Need to change this
9871	 * to figure out whether the disk device is actually online or not.
9872	 */
9873	if (lun != NULL)
9874		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9875				  lun->be_lun->lun_type;
9876	else
9877		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9878
9879	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9880	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9881	/*
9882	 * If we don't have a LUN, we just leave the serial number as
9883	 * all spaces.
9884	 */
9885	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9886	if (lun != NULL) {
9887		strncpy((char *)sn_ptr->serial_num,
9888			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9889	}
9890	ctsio->scsi_status = SCSI_STATUS_OK;
9891
9892	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9893	ctsio->be_move_done = ctl_config_move_done;
9894	ctl_datamove((union ctl_io *)ctsio);
9895
9896	return (CTL_RETVAL_COMPLETE);
9897}
9898
9899
9900static int
9901ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9902{
9903	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9904	struct ctl_lun *lun;
9905	int data_len;
9906
9907	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9908
9909	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9910	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9911
9912	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9913	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9914	ctsio->kern_sg_entries = 0;
9915
9916	if (data_len < alloc_len) {
9917		ctsio->residual = alloc_len - data_len;
9918		ctsio->kern_data_len = data_len;
9919		ctsio->kern_total_len = data_len;
9920	} else {
9921		ctsio->residual = 0;
9922		ctsio->kern_data_len = alloc_len;
9923		ctsio->kern_total_len = alloc_len;
9924	}
9925	ctsio->kern_data_resid = 0;
9926	ctsio->kern_rel_offset = 0;
9927	ctsio->kern_sg_entries = 0;
9928
9929	/*
9930	 * The control device is always connected.  The disk device, on the
9931	 * other hand, may not be online all the time.
9932	 */
9933	if (lun != NULL)
9934		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9935				     lun->be_lun->lun_type;
9936	else
9937		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9938	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9939	eid_ptr->page_length = data_len - 4;
9940	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9941	eid_ptr->flags3 = SVPD_EID_V_SUP;
9942
9943	ctsio->scsi_status = SCSI_STATUS_OK;
9944	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9945	ctsio->be_move_done = ctl_config_move_done;
9946	ctl_datamove((union ctl_io *)ctsio);
9947
9948	return (CTL_RETVAL_COMPLETE);
9949}
9950
9951static int
9952ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9953{
9954	struct scsi_vpd_mode_page_policy *mpp_ptr;
9955	struct ctl_lun *lun;
9956	int data_len;
9957
9958	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9959
9960	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9961	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9962
9963	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9964	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9965	ctsio->kern_sg_entries = 0;
9966
9967	if (data_len < alloc_len) {
9968		ctsio->residual = alloc_len - data_len;
9969		ctsio->kern_data_len = data_len;
9970		ctsio->kern_total_len = data_len;
9971	} else {
9972		ctsio->residual = 0;
9973		ctsio->kern_data_len = alloc_len;
9974		ctsio->kern_total_len = alloc_len;
9975	}
9976	ctsio->kern_data_resid = 0;
9977	ctsio->kern_rel_offset = 0;
9978	ctsio->kern_sg_entries = 0;
9979
9980	/*
9981	 * The control device is always connected.  The disk device, on the
9982	 * other hand, may not be online all the time.
9983	 */
9984	if (lun != NULL)
9985		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9986				     lun->be_lun->lun_type;
9987	else
9988		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9989	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9990	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9991	mpp_ptr->descr[0].page_code = 0x3f;
9992	mpp_ptr->descr[0].subpage_code = 0xff;
9993	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9994
9995	ctsio->scsi_status = SCSI_STATUS_OK;
9996	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9997	ctsio->be_move_done = ctl_config_move_done;
9998	ctl_datamove((union ctl_io *)ctsio);
9999
10000	return (CTL_RETVAL_COMPLETE);
10001}
10002
10003static int
10004ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
10005{
10006	struct scsi_vpd_device_id *devid_ptr;
10007	struct scsi_vpd_id_descriptor *desc;
10008	struct ctl_softc *ctl_softc;
10009	struct ctl_lun *lun;
10010	struct ctl_port *port;
10011	int data_len;
10012	uint8_t proto;
10013
10014	ctl_softc = control_softc;
10015
10016	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
10017	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10018
10019	data_len = sizeof(struct scsi_vpd_device_id) +
10020	    sizeof(struct scsi_vpd_id_descriptor) +
10021		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
10022	    sizeof(struct scsi_vpd_id_descriptor) +
10023		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
10024	if (lun && lun->lun_devid)
10025		data_len += lun->lun_devid->len;
10026	if (port->port_devid)
10027		data_len += port->port_devid->len;
10028	if (port->target_devid)
10029		data_len += port->target_devid->len;
10030
10031	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10032	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
10033	ctsio->kern_sg_entries = 0;
10034
10035	if (data_len < alloc_len) {
10036		ctsio->residual = alloc_len - data_len;
10037		ctsio->kern_data_len = data_len;
10038		ctsio->kern_total_len = data_len;
10039	} else {
10040		ctsio->residual = 0;
10041		ctsio->kern_data_len = alloc_len;
10042		ctsio->kern_total_len = alloc_len;
10043	}
10044	ctsio->kern_data_resid = 0;
10045	ctsio->kern_rel_offset = 0;
10046	ctsio->kern_sg_entries = 0;
10047
10048	/*
10049	 * The control device is always connected.  The disk device, on the
10050	 * other hand, may not be online all the time.
10051	 */
10052	if (lun != NULL)
10053		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10054				     lun->be_lun->lun_type;
10055	else
10056		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10057	devid_ptr->page_code = SVPD_DEVICE_ID;
10058	scsi_ulto2b(data_len - 4, devid_ptr->length);
10059
10060	if (port->port_type == CTL_PORT_FC)
10061		proto = SCSI_PROTO_FC << 4;
10062	else if (port->port_type == CTL_PORT_ISCSI)
10063		proto = SCSI_PROTO_ISCSI << 4;
10064	else
10065		proto = SCSI_PROTO_SPI << 4;
10066	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
10067
10068	/*
10069	 * We're using a LUN association here.  i.e., this device ID is a
10070	 * per-LUN identifier.
10071	 */
10072	if (lun && lun->lun_devid) {
10073		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
10074		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10075		    lun->lun_devid->len);
10076	}
10077
10078	/*
10079	 * This is for the WWPN which is a port association.
10080	 */
10081	if (port->port_devid) {
10082		memcpy(desc, port->port_devid->data, port->port_devid->len);
10083		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10084		    port->port_devid->len);
10085	}
10086
10087	/*
10088	 * This is for the Relative Target Port(type 4h) identifier
10089	 */
10090	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10091	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10092	    SVPD_ID_TYPE_RELTARG;
10093	desc->length = 4;
10094	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
10095	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10096	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
10097
10098	/*
10099	 * This is for the Target Port Group(type 5h) identifier
10100	 */
10101	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10102	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10103	    SVPD_ID_TYPE_TPORTGRP;
10104	desc->length = 4;
10105	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
10106	    &desc->identifier[2]);
10107	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10108	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10109
10110	/*
10111	 * This is for the Target identifier
10112	 */
10113	if (port->target_devid) {
10114		memcpy(desc, port->target_devid->data, port->target_devid->len);
10115	}
10116
10117	ctsio->scsi_status = SCSI_STATUS_OK;
10118	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10119	ctsio->be_move_done = ctl_config_move_done;
10120	ctl_datamove((union ctl_io *)ctsio);
10121
10122	return (CTL_RETVAL_COMPLETE);
10123}
10124
10125static int
10126ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10127{
10128	struct ctl_softc *softc = control_softc;
10129	struct scsi_vpd_scsi_ports *sp;
10130	struct scsi_vpd_port_designation *pd;
10131	struct scsi_vpd_port_designation_cont *pdc;
10132	struct ctl_lun *lun;
10133	struct ctl_port *port;
10134	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10135	int num_target_port_groups, single;
10136
10137	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10138
10139	single = ctl_is_single;
10140	if (single)
10141		num_target_port_groups = 1;
10142	else
10143		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10144	num_target_ports = 0;
10145	iid_len = 0;
10146	id_len = 0;
10147	mtx_lock(&softc->ctl_lock);
10148	STAILQ_FOREACH(port, &softc->port_list, links) {
10149		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10150			continue;
10151		if (lun != NULL &&
10152		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10153		    CTL_MAX_LUNS)
10154			continue;
10155		num_target_ports++;
10156		if (port->init_devid)
10157			iid_len += port->init_devid->len;
10158		if (port->port_devid)
10159			id_len += port->port_devid->len;
10160	}
10161	mtx_unlock(&softc->ctl_lock);
10162
10163	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10164	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10165	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10166	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10167	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10168	ctsio->kern_sg_entries = 0;
10169
10170	if (data_len < alloc_len) {
10171		ctsio->residual = alloc_len - data_len;
10172		ctsio->kern_data_len = data_len;
10173		ctsio->kern_total_len = data_len;
10174	} else {
10175		ctsio->residual = 0;
10176		ctsio->kern_data_len = alloc_len;
10177		ctsio->kern_total_len = alloc_len;
10178	}
10179	ctsio->kern_data_resid = 0;
10180	ctsio->kern_rel_offset = 0;
10181	ctsio->kern_sg_entries = 0;
10182
10183	/*
10184	 * The control device is always connected.  The disk device, on the
10185	 * other hand, may not be online all the time.  Need to change this
10186	 * to figure out whether the disk device is actually online or not.
10187	 */
10188	if (lun != NULL)
10189		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10190				  lun->be_lun->lun_type;
10191	else
10192		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10193
10194	sp->page_code = SVPD_SCSI_PORTS;
10195	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10196	    sp->page_length);
10197	pd = &sp->design[0];
10198
10199	mtx_lock(&softc->ctl_lock);
10200	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10201		pg = 0;
10202	else
10203		pg = 1;
10204	for (g = 0; g < num_target_port_groups; g++) {
10205		STAILQ_FOREACH(port, &softc->port_list, links) {
10206			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10207				continue;
10208			if (lun != NULL &&
10209			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10210			    CTL_MAX_LUNS)
10211				continue;
10212			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10213			scsi_ulto2b(p, pd->relative_port_id);
10214			if (port->init_devid && g == pg) {
10215				iid_len = port->init_devid->len;
10216				memcpy(pd->initiator_transportid,
10217				    port->init_devid->data, port->init_devid->len);
10218			} else
10219				iid_len = 0;
10220			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10221			pdc = (struct scsi_vpd_port_designation_cont *)
10222			    (&pd->initiator_transportid[iid_len]);
10223			if (port->port_devid && g == pg) {
10224				id_len = port->port_devid->len;
10225				memcpy(pdc->target_port_descriptors,
10226				    port->port_devid->data, port->port_devid->len);
10227			} else
10228				id_len = 0;
10229			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10230			pd = (struct scsi_vpd_port_designation *)
10231			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10232		}
10233	}
10234	mtx_unlock(&softc->ctl_lock);
10235
10236	ctsio->scsi_status = SCSI_STATUS_OK;
10237	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10238	ctsio->be_move_done = ctl_config_move_done;
10239	ctl_datamove((union ctl_io *)ctsio);
10240
10241	return (CTL_RETVAL_COMPLETE);
10242}
10243
10244static int
10245ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10246{
10247	struct scsi_vpd_block_limits *bl_ptr;
10248	struct ctl_lun *lun;
10249	int bs;
10250
10251	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10252
10253	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10254	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10255	ctsio->kern_sg_entries = 0;
10256
10257	if (sizeof(*bl_ptr) < alloc_len) {
10258		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10259		ctsio->kern_data_len = sizeof(*bl_ptr);
10260		ctsio->kern_total_len = sizeof(*bl_ptr);
10261	} else {
10262		ctsio->residual = 0;
10263		ctsio->kern_data_len = alloc_len;
10264		ctsio->kern_total_len = alloc_len;
10265	}
10266	ctsio->kern_data_resid = 0;
10267	ctsio->kern_rel_offset = 0;
10268	ctsio->kern_sg_entries = 0;
10269
10270	/*
10271	 * The control device is always connected.  The disk device, on the
10272	 * other hand, may not be online all the time.  Need to change this
10273	 * to figure out whether the disk device is actually online or not.
10274	 */
10275	if (lun != NULL)
10276		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10277				  lun->be_lun->lun_type;
10278	else
10279		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10280
10281	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10282	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
10283	bl_ptr->max_cmp_write_len = 0xff;
10284	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10285	if (lun != NULL) {
10286		bs = lun->be_lun->blocksize;
10287		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10288		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10289			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10290			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10291			if (lun->be_lun->pblockexp != 0) {
10292				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10293				    bl_ptr->opt_unmap_grain);
10294				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10295				    bl_ptr->unmap_grain_align);
10296			}
10297		}
10298	}
10299	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10300
10301	ctsio->scsi_status = SCSI_STATUS_OK;
10302	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10303	ctsio->be_move_done = ctl_config_move_done;
10304	ctl_datamove((union ctl_io *)ctsio);
10305
10306	return (CTL_RETVAL_COMPLETE);
10307}
10308
10309static int
10310ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10311{
10312	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10313	struct ctl_lun *lun;
10314
10315	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10316
10317	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10318	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10319	ctsio->kern_sg_entries = 0;
10320
10321	if (sizeof(*bdc_ptr) < alloc_len) {
10322		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10323		ctsio->kern_data_len = sizeof(*bdc_ptr);
10324		ctsio->kern_total_len = sizeof(*bdc_ptr);
10325	} else {
10326		ctsio->residual = 0;
10327		ctsio->kern_data_len = alloc_len;
10328		ctsio->kern_total_len = alloc_len;
10329	}
10330	ctsio->kern_data_resid = 0;
10331	ctsio->kern_rel_offset = 0;
10332	ctsio->kern_sg_entries = 0;
10333
10334	/*
10335	 * The control device is always connected.  The disk device, on the
10336	 * other hand, may not be online all the time.  Need to change this
10337	 * to figure out whether the disk device is actually online or not.
10338	 */
10339	if (lun != NULL)
10340		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10341				  lun->be_lun->lun_type;
10342	else
10343		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10344	bdc_ptr->page_code = SVPD_BDC;
10345	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10346	scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10347	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10348
10349	ctsio->scsi_status = SCSI_STATUS_OK;
10350	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10351	ctsio->be_move_done = ctl_config_move_done;
10352	ctl_datamove((union ctl_io *)ctsio);
10353
10354	return (CTL_RETVAL_COMPLETE);
10355}
10356
10357static int
10358ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10359{
10360	struct scsi_vpd_logical_block_prov *lbp_ptr;
10361	struct ctl_lun *lun;
10362
10363	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10364
10365	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10366	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10367	ctsio->kern_sg_entries = 0;
10368
10369	if (sizeof(*lbp_ptr) < alloc_len) {
10370		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10371		ctsio->kern_data_len = sizeof(*lbp_ptr);
10372		ctsio->kern_total_len = sizeof(*lbp_ptr);
10373	} else {
10374		ctsio->residual = 0;
10375		ctsio->kern_data_len = alloc_len;
10376		ctsio->kern_total_len = alloc_len;
10377	}
10378	ctsio->kern_data_resid = 0;
10379	ctsio->kern_rel_offset = 0;
10380	ctsio->kern_sg_entries = 0;
10381
10382	/*
10383	 * The control device is always connected.  The disk device, on the
10384	 * other hand, may not be online all the time.  Need to change this
10385	 * to figure out whether the disk device is actually online or not.
10386	 */
10387	if (lun != NULL)
10388		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10389				  lun->be_lun->lun_type;
10390	else
10391		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10392
10393	lbp_ptr->page_code = SVPD_LBP;
10394	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10395	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10396		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10397		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10398		lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10399	}
10400
10401	ctsio->scsi_status = SCSI_STATUS_OK;
10402	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10403	ctsio->be_move_done = ctl_config_move_done;
10404	ctl_datamove((union ctl_io *)ctsio);
10405
10406	return (CTL_RETVAL_COMPLETE);
10407}
10408
10409static int
10410ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10411{
10412	struct scsi_inquiry *cdb;
10413	struct ctl_lun *lun;
10414	int alloc_len, retval;
10415
10416	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10417	cdb = (struct scsi_inquiry *)ctsio->cdb;
10418
10419	retval = CTL_RETVAL_COMPLETE;
10420
10421	alloc_len = scsi_2btoul(cdb->length);
10422
10423	switch (cdb->page_code) {
10424	case SVPD_SUPPORTED_PAGES:
10425		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10426		break;
10427	case SVPD_UNIT_SERIAL_NUMBER:
10428		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10429		break;
10430	case SVPD_DEVICE_ID:
10431		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10432		break;
10433	case SVPD_EXTENDED_INQUIRY_DATA:
10434		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10435		break;
10436	case SVPD_MODE_PAGE_POLICY:
10437		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10438		break;
10439	case SVPD_SCSI_PORTS:
10440		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10441		break;
10442	case SVPD_SCSI_TPC:
10443		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10444		break;
10445	case SVPD_BLOCK_LIMITS:
10446		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10447		break;
10448	case SVPD_BDC:
10449		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10450		break;
10451	case SVPD_LBP:
10452		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10453		break;
10454	default:
10455		ctl_set_invalid_field(ctsio,
10456				      /*sks_valid*/ 1,
10457				      /*command*/ 1,
10458				      /*field*/ 2,
10459				      /*bit_valid*/ 0,
10460				      /*bit*/ 0);
10461		ctl_done((union ctl_io *)ctsio);
10462		retval = CTL_RETVAL_COMPLETE;
10463		break;
10464	}
10465
10466	return (retval);
10467}
10468
10469static int
10470ctl_inquiry_std(struct ctl_scsiio *ctsio)
10471{
10472	struct scsi_inquiry_data *inq_ptr;
10473	struct scsi_inquiry *cdb;
10474	struct ctl_softc *ctl_softc;
10475	struct ctl_lun *lun;
10476	char *val;
10477	uint32_t alloc_len;
10478	ctl_port_type port_type;
10479
10480	ctl_softc = control_softc;
10481
10482	/*
10483	 * Figure out whether we're talking to a Fibre Channel port or not.
10484	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10485	 * SCSI front ends.
10486	 */
10487	port_type = ctl_softc->ctl_ports[
10488	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10489	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10490		port_type = CTL_PORT_SCSI;
10491
10492	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10493	cdb = (struct scsi_inquiry *)ctsio->cdb;
10494	alloc_len = scsi_2btoul(cdb->length);
10495
10496	/*
10497	 * We malloc the full inquiry data size here and fill it
10498	 * in.  If the user only asks for less, we'll give him
10499	 * that much.
10500	 */
10501	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
10502	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10503	ctsio->kern_sg_entries = 0;
10504	ctsio->kern_data_resid = 0;
10505	ctsio->kern_rel_offset = 0;
10506
10507	if (sizeof(*inq_ptr) < alloc_len) {
10508		ctsio->residual = alloc_len - sizeof(*inq_ptr);
10509		ctsio->kern_data_len = sizeof(*inq_ptr);
10510		ctsio->kern_total_len = sizeof(*inq_ptr);
10511	} else {
10512		ctsio->residual = 0;
10513		ctsio->kern_data_len = alloc_len;
10514		ctsio->kern_total_len = alloc_len;
10515	}
10516
10517	/*
10518	 * If we have a LUN configured, report it as connected.  Otherwise,
10519	 * report that it is offline or no device is supported, depending
10520	 * on the value of inquiry_pq_no_lun.
10521	 *
10522	 * According to the spec (SPC-4 r34), the peripheral qualifier
10523	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10524	 *
10525	 * "A peripheral device having the specified peripheral device type
10526	 * is not connected to this logical unit. However, the device
10527	 * server is capable of supporting the specified peripheral device
10528	 * type on this logical unit."
10529	 *
10530	 * According to the same spec, the peripheral qualifier
10531	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10532	 *
10533	 * "The device server is not capable of supporting a peripheral
10534	 * device on this logical unit. For this peripheral qualifier the
10535	 * peripheral device type shall be set to 1Fh. All other peripheral
10536	 * device type values are reserved for this peripheral qualifier."
10537	 *
10538	 * Given the text, it would seem that we probably want to report that
10539	 * the LUN is offline here.  There is no LUN connected, but we can
10540	 * support a LUN at the given LUN number.
10541	 *
10542	 * In the real world, though, it sounds like things are a little
10543	 * different:
10544	 *
10545	 * - Linux, when presented with a LUN with the offline peripheral
10546	 *   qualifier, will create an sg driver instance for it.  So when
10547	 *   you attach it to CTL, you wind up with a ton of sg driver
10548	 *   instances.  (One for every LUN that Linux bothered to probe.)
10549	 *   Linux does this despite the fact that it issues a REPORT LUNs
10550	 *   to LUN 0 to get the inventory of supported LUNs.
10551	 *
10552	 * - There is other anecdotal evidence (from Emulex folks) about
10553	 *   arrays that use the offline peripheral qualifier for LUNs that
10554	 *   are on the "passive" path in an active/passive array.
10555	 *
10556	 * So the solution is provide a hopefully reasonable default
10557	 * (return bad/no LUN) and allow the user to change the behavior
10558	 * with a tunable/sysctl variable.
10559	 */
10560	if (lun != NULL)
10561		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10562				  lun->be_lun->lun_type;
10563	else if (ctl_softc->inquiry_pq_no_lun == 0)
10564		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10565	else
10566		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10567
10568	/* RMB in byte 2 is 0 */
10569	inq_ptr->version = SCSI_REV_SPC4;
10570
10571	/*
10572	 * According to SAM-3, even if a device only supports a single
10573	 * level of LUN addressing, it should still set the HISUP bit:
10574	 *
10575	 * 4.9.1 Logical unit numbers overview
10576	 *
10577	 * All logical unit number formats described in this standard are
10578	 * hierarchical in structure even when only a single level in that
10579	 * hierarchy is used. The HISUP bit shall be set to one in the
10580	 * standard INQUIRY data (see SPC-2) when any logical unit number
10581	 * format described in this standard is used.  Non-hierarchical
10582	 * formats are outside the scope of this standard.
10583	 *
10584	 * Therefore we set the HiSup bit here.
10585	 *
10586	 * The reponse format is 2, per SPC-3.
10587	 */
10588	inq_ptr->response_format = SID_HiSup | 2;
10589
10590	inq_ptr->additional_length =
10591	    offsetof(struct scsi_inquiry_data, vendor_specific1) -
10592	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10593	CTL_DEBUG_PRINT(("additional_length = %d\n",
10594			 inq_ptr->additional_length));
10595
10596	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10597	/* 16 bit addressing */
10598	if (port_type == CTL_PORT_SCSI)
10599		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10600	/* XXX set the SID_MultiP bit here if we're actually going to
10601	   respond on multiple ports */
10602	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10603
10604	/* 16 bit data bus, synchronous transfers */
10605	if (port_type == CTL_PORT_SCSI)
10606		inq_ptr->flags = SID_WBus16 | SID_Sync;
10607	/*
10608	 * XXX KDM do we want to support tagged queueing on the control
10609	 * device at all?
10610	 */
10611	if ((lun == NULL)
10612	 || (lun->be_lun->lun_type != T_PROCESSOR))
10613		inq_ptr->flags |= SID_CmdQue;
10614	/*
10615	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10616	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10617	 * name and 4 bytes for the revision.
10618	 */
10619	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10620	    "vendor")) == NULL) {
10621		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10622	} else {
10623		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10624		strncpy(inq_ptr->vendor, val,
10625		    min(sizeof(inq_ptr->vendor), strlen(val)));
10626	}
10627	if (lun == NULL) {
10628		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10629		    sizeof(inq_ptr->product));
10630	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10631		switch (lun->be_lun->lun_type) {
10632		case T_DIRECT:
10633			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10634			    sizeof(inq_ptr->product));
10635			break;
10636		case T_PROCESSOR:
10637			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10638			    sizeof(inq_ptr->product));
10639			break;
10640		default:
10641			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10642			    sizeof(inq_ptr->product));
10643			break;
10644		}
10645	} else {
10646		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10647		strncpy(inq_ptr->product, val,
10648		    min(sizeof(inq_ptr->product), strlen(val)));
10649	}
10650
10651	/*
10652	 * XXX make this a macro somewhere so it automatically gets
10653	 * incremented when we make changes.
10654	 */
10655	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10656	    "revision")) == NULL) {
10657		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10658	} else {
10659		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10660		strncpy(inq_ptr->revision, val,
10661		    min(sizeof(inq_ptr->revision), strlen(val)));
10662	}
10663
10664	/*
10665	 * For parallel SCSI, we support double transition and single
10666	 * transition clocking.  We also support QAS (Quick Arbitration
10667	 * and Selection) and Information Unit transfers on both the
10668	 * control and array devices.
10669	 */
10670	if (port_type == CTL_PORT_SCSI)
10671		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10672				    SID_SPI_IUS;
10673
10674	/* SAM-5 (no version claimed) */
10675	scsi_ulto2b(0x00A0, inq_ptr->version1);
10676	/* SPC-4 (no version claimed) */
10677	scsi_ulto2b(0x0460, inq_ptr->version2);
10678	if (port_type == CTL_PORT_FC) {
10679		/* FCP-2 ANSI INCITS.350:2003 */
10680		scsi_ulto2b(0x0917, inq_ptr->version3);
10681	} else if (port_type == CTL_PORT_SCSI) {
10682		/* SPI-4 ANSI INCITS.362:200x */
10683		scsi_ulto2b(0x0B56, inq_ptr->version3);
10684	} else if (port_type == CTL_PORT_ISCSI) {
10685		/* iSCSI (no version claimed) */
10686		scsi_ulto2b(0x0960, inq_ptr->version3);
10687	} else if (port_type == CTL_PORT_SAS) {
10688		/* SAS (no version claimed) */
10689		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10690	}
10691
10692	if (lun == NULL) {
10693		/* SBC-3 (no version claimed) */
10694		scsi_ulto2b(0x04C0, inq_ptr->version4);
10695	} else {
10696		switch (lun->be_lun->lun_type) {
10697		case T_DIRECT:
10698			/* SBC-3 (no version claimed) */
10699			scsi_ulto2b(0x04C0, inq_ptr->version4);
10700			break;
10701		case T_PROCESSOR:
10702		default:
10703			break;
10704		}
10705	}
10706
10707	ctsio->scsi_status = SCSI_STATUS_OK;
10708	if (ctsio->kern_data_len > 0) {
10709		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10710		ctsio->be_move_done = ctl_config_move_done;
10711		ctl_datamove((union ctl_io *)ctsio);
10712	} else {
10713		ctsio->io_hdr.status = CTL_SUCCESS;
10714		ctl_done((union ctl_io *)ctsio);
10715	}
10716
10717	return (CTL_RETVAL_COMPLETE);
10718}
10719
10720int
10721ctl_inquiry(struct ctl_scsiio *ctsio)
10722{
10723	struct scsi_inquiry *cdb;
10724	int retval;
10725
10726	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10727
10728	cdb = (struct scsi_inquiry *)ctsio->cdb;
10729	if (cdb->byte2 & SI_EVPD)
10730		retval = ctl_inquiry_evpd(ctsio);
10731	else if (cdb->page_code == 0)
10732		retval = ctl_inquiry_std(ctsio);
10733	else {
10734		ctl_set_invalid_field(ctsio,
10735				      /*sks_valid*/ 1,
10736				      /*command*/ 1,
10737				      /*field*/ 2,
10738				      /*bit_valid*/ 0,
10739				      /*bit*/ 0);
10740		ctl_done((union ctl_io *)ctsio);
10741		return (CTL_RETVAL_COMPLETE);
10742	}
10743
10744	return (retval);
10745}
10746
10747/*
10748 * For known CDB types, parse the LBA and length.
10749 */
10750static int
10751ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10752{
10753	if (io->io_hdr.io_type != CTL_IO_SCSI)
10754		return (1);
10755
10756	switch (io->scsiio.cdb[0]) {
10757	case COMPARE_AND_WRITE: {
10758		struct scsi_compare_and_write *cdb;
10759
10760		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10761
10762		*lba = scsi_8btou64(cdb->addr);
10763		*len = cdb->length;
10764		break;
10765	}
10766	case READ_6:
10767	case WRITE_6: {
10768		struct scsi_rw_6 *cdb;
10769
10770		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10771
10772		*lba = scsi_3btoul(cdb->addr);
10773		/* only 5 bits are valid in the most significant address byte */
10774		*lba &= 0x1fffff;
10775		*len = cdb->length;
10776		break;
10777	}
10778	case READ_10:
10779	case WRITE_10: {
10780		struct scsi_rw_10 *cdb;
10781
10782		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10783
10784		*lba = scsi_4btoul(cdb->addr);
10785		*len = scsi_2btoul(cdb->length);
10786		break;
10787	}
10788	case WRITE_VERIFY_10: {
10789		struct scsi_write_verify_10 *cdb;
10790
10791		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10792
10793		*lba = scsi_4btoul(cdb->addr);
10794		*len = scsi_2btoul(cdb->length);
10795		break;
10796	}
10797	case READ_12:
10798	case WRITE_12: {
10799		struct scsi_rw_12 *cdb;
10800
10801		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10802
10803		*lba = scsi_4btoul(cdb->addr);
10804		*len = scsi_4btoul(cdb->length);
10805		break;
10806	}
10807	case WRITE_VERIFY_12: {
10808		struct scsi_write_verify_12 *cdb;
10809
10810		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10811
10812		*lba = scsi_4btoul(cdb->addr);
10813		*len = scsi_4btoul(cdb->length);
10814		break;
10815	}
10816	case READ_16:
10817	case WRITE_16: {
10818		struct scsi_rw_16 *cdb;
10819
10820		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10821
10822		*lba = scsi_8btou64(cdb->addr);
10823		*len = scsi_4btoul(cdb->length);
10824		break;
10825	}
10826	case WRITE_VERIFY_16: {
10827		struct scsi_write_verify_16 *cdb;
10828
10829		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10830
10831
10832		*lba = scsi_8btou64(cdb->addr);
10833		*len = scsi_4btoul(cdb->length);
10834		break;
10835	}
10836	case WRITE_SAME_10: {
10837		struct scsi_write_same_10 *cdb;
10838
10839		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10840
10841		*lba = scsi_4btoul(cdb->addr);
10842		*len = scsi_2btoul(cdb->length);
10843		break;
10844	}
10845	case WRITE_SAME_16: {
10846		struct scsi_write_same_16 *cdb;
10847
10848		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10849
10850		*lba = scsi_8btou64(cdb->addr);
10851		*len = scsi_4btoul(cdb->length);
10852		break;
10853	}
10854	case VERIFY_10: {
10855		struct scsi_verify_10 *cdb;
10856
10857		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10858
10859		*lba = scsi_4btoul(cdb->addr);
10860		*len = scsi_2btoul(cdb->length);
10861		break;
10862	}
10863	case VERIFY_12: {
10864		struct scsi_verify_12 *cdb;
10865
10866		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10867
10868		*lba = scsi_4btoul(cdb->addr);
10869		*len = scsi_4btoul(cdb->length);
10870		break;
10871	}
10872	case VERIFY_16: {
10873		struct scsi_verify_16 *cdb;
10874
10875		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10876
10877		*lba = scsi_8btou64(cdb->addr);
10878		*len = scsi_4btoul(cdb->length);
10879		break;
10880	}
10881	case UNMAP: {
10882		*lba = 0;
10883		*len = UINT64_MAX;
10884		break;
10885	}
10886	default:
10887		return (1);
10888		break; /* NOTREACHED */
10889	}
10890
10891	return (0);
10892}
10893
10894static ctl_action
10895ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10896{
10897	uint64_t endlba1, endlba2;
10898
10899	endlba1 = lba1 + len1 - 1;
10900	endlba2 = lba2 + len2 - 1;
10901
10902	if ((endlba1 < lba2)
10903	 || (endlba2 < lba1))
10904		return (CTL_ACTION_PASS);
10905	else
10906		return (CTL_ACTION_BLOCK);
10907}
10908
10909static int
10910ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10911{
10912	struct ctl_ptr_len_flags *ptrlen;
10913	struct scsi_unmap_desc *buf, *end, *range;
10914	uint64_t lba;
10915	uint32_t len;
10916
10917	/* If not UNMAP -- go other way. */
10918	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10919	    io->scsiio.cdb[0] != UNMAP)
10920		return (CTL_ACTION_ERROR);
10921
10922	/* If UNMAP without data -- block and wait for data. */
10923	ptrlen = (struct ctl_ptr_len_flags *)
10924	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10925	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10926	    ptrlen->ptr == NULL)
10927		return (CTL_ACTION_BLOCK);
10928
10929	/* UNMAP with data -- check for collision. */
10930	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10931	end = buf + ptrlen->len / sizeof(*buf);
10932	for (range = buf; range < end; range++) {
10933		lba = scsi_8btou64(range->lba);
10934		len = scsi_4btoul(range->length);
10935		if ((lba < lba2 + len2) && (lba + len > lba2))
10936			return (CTL_ACTION_BLOCK);
10937	}
10938	return (CTL_ACTION_PASS);
10939}
10940
10941static ctl_action
10942ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10943{
10944	uint64_t lba1, lba2;
10945	uint64_t len1, len2;
10946	int retval;
10947
10948	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10949		return (CTL_ACTION_ERROR);
10950
10951	retval = ctl_extent_check_unmap(io2, lba1, len1);
10952	if (retval != CTL_ACTION_ERROR)
10953		return (retval);
10954
10955	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10956		return (CTL_ACTION_ERROR);
10957
10958	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10959}
10960
10961static ctl_action
10962ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10963    union ctl_io *ooa_io)
10964{
10965	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10966	ctl_serialize_action *serialize_row;
10967
10968	/*
10969	 * The initiator attempted multiple untagged commands at the same
10970	 * time.  Can't do that.
10971	 */
10972	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10973	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10974	 && ((pending_io->io_hdr.nexus.targ_port ==
10975	      ooa_io->io_hdr.nexus.targ_port)
10976	  && (pending_io->io_hdr.nexus.initid.id ==
10977	      ooa_io->io_hdr.nexus.initid.id))
10978	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10979		return (CTL_ACTION_OVERLAP);
10980
10981	/*
10982	 * The initiator attempted to send multiple tagged commands with
10983	 * the same ID.  (It's fine if different initiators have the same
10984	 * tag ID.)
10985	 *
10986	 * Even if all of those conditions are true, we don't kill the I/O
10987	 * if the command ahead of us has been aborted.  We won't end up
10988	 * sending it to the FETD, and it's perfectly legal to resend a
10989	 * command with the same tag number as long as the previous
10990	 * instance of this tag number has been aborted somehow.
10991	 */
10992	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10993	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10994	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10995	 && ((pending_io->io_hdr.nexus.targ_port ==
10996	      ooa_io->io_hdr.nexus.targ_port)
10997	  && (pending_io->io_hdr.nexus.initid.id ==
10998	      ooa_io->io_hdr.nexus.initid.id))
10999	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
11000		return (CTL_ACTION_OVERLAP_TAG);
11001
11002	/*
11003	 * If we get a head of queue tag, SAM-3 says that we should
11004	 * immediately execute it.
11005	 *
11006	 * What happens if this command would normally block for some other
11007	 * reason?  e.g. a request sense with a head of queue tag
11008	 * immediately after a write.  Normally that would block, but this
11009	 * will result in its getting executed immediately...
11010	 *
11011	 * We currently return "pass" instead of "skip", so we'll end up
11012	 * going through the rest of the queue to check for overlapped tags.
11013	 *
11014	 * XXX KDM check for other types of blockage first??
11015	 */
11016	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11017		return (CTL_ACTION_PASS);
11018
11019	/*
11020	 * Ordered tags have to block until all items ahead of them
11021	 * have completed.  If we get called with an ordered tag, we always
11022	 * block, if something else is ahead of us in the queue.
11023	 */
11024	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11025		return (CTL_ACTION_BLOCK);
11026
11027	/*
11028	 * Simple tags get blocked until all head of queue and ordered tags
11029	 * ahead of them have completed.  I'm lumping untagged commands in
11030	 * with simple tags here.  XXX KDM is that the right thing to do?
11031	 */
11032	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11033	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11034	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11035	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11036		return (CTL_ACTION_BLOCK);
11037
11038	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio);
11039	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio);
11040
11041	serialize_row = ctl_serialize_table[ooa_entry->seridx];
11042
11043	switch (serialize_row[pending_entry->seridx]) {
11044	case CTL_SER_BLOCK:
11045		return (CTL_ACTION_BLOCK);
11046	case CTL_SER_EXTENT:
11047		return (ctl_extent_check(pending_io, ooa_io));
11048	case CTL_SER_EXTENTOPT:
11049		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11050		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11051			return (ctl_extent_check(pending_io, ooa_io));
11052		/* FALLTHROUGH */
11053	case CTL_SER_PASS:
11054		return (CTL_ACTION_PASS);
11055	case CTL_SER_BLOCKOPT:
11056		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11057		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11058			return (CTL_ACTION_BLOCK);
11059		return (CTL_ACTION_PASS);
11060	case CTL_SER_SKIP:
11061		return (CTL_ACTION_SKIP);
11062	default:
11063		panic("invalid serialization value %d",
11064		      serialize_row[pending_entry->seridx]);
11065	}
11066
11067	return (CTL_ACTION_ERROR);
11068}
11069
11070/*
11071 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11072 * Assumptions:
11073 * - pending_io is generally either incoming, or on the blocked queue
11074 * - starting I/O is the I/O we want to start the check with.
11075 */
11076static ctl_action
11077ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11078	      union ctl_io *starting_io)
11079{
11080	union ctl_io *ooa_io;
11081	ctl_action action;
11082
11083	mtx_assert(&lun->lun_lock, MA_OWNED);
11084
11085	/*
11086	 * Run back along the OOA queue, starting with the current
11087	 * blocked I/O and going through every I/O before it on the
11088	 * queue.  If starting_io is NULL, we'll just end up returning
11089	 * CTL_ACTION_PASS.
11090	 */
11091	for (ooa_io = starting_io; ooa_io != NULL;
11092	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11093	     ooa_links)){
11094
11095		/*
11096		 * This routine just checks to see whether
11097		 * cur_blocked is blocked by ooa_io, which is ahead
11098		 * of it in the queue.  It doesn't queue/dequeue
11099		 * cur_blocked.
11100		 */
11101		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11102		switch (action) {
11103		case CTL_ACTION_BLOCK:
11104		case CTL_ACTION_OVERLAP:
11105		case CTL_ACTION_OVERLAP_TAG:
11106		case CTL_ACTION_SKIP:
11107		case CTL_ACTION_ERROR:
11108			return (action);
11109			break; /* NOTREACHED */
11110		case CTL_ACTION_PASS:
11111			break;
11112		default:
11113			panic("invalid action %d", action);
11114			break;  /* NOTREACHED */
11115		}
11116	}
11117
11118	return (CTL_ACTION_PASS);
11119}
11120
11121/*
11122 * Assumptions:
11123 * - An I/O has just completed, and has been removed from the per-LUN OOA
11124 *   queue, so some items on the blocked queue may now be unblocked.
11125 */
11126static int
11127ctl_check_blocked(struct ctl_lun *lun)
11128{
11129	union ctl_io *cur_blocked, *next_blocked;
11130
11131	mtx_assert(&lun->lun_lock, MA_OWNED);
11132
11133	/*
11134	 * Run forward from the head of the blocked queue, checking each
11135	 * entry against the I/Os prior to it on the OOA queue to see if
11136	 * there is still any blockage.
11137	 *
11138	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11139	 * with our removing a variable on it while it is traversing the
11140	 * list.
11141	 */
11142	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11143	     cur_blocked != NULL; cur_blocked = next_blocked) {
11144		union ctl_io *prev_ooa;
11145		ctl_action action;
11146
11147		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11148							  blocked_links);
11149
11150		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11151						      ctl_ooaq, ooa_links);
11152
11153		/*
11154		 * If cur_blocked happens to be the first item in the OOA
11155		 * queue now, prev_ooa will be NULL, and the action
11156		 * returned will just be CTL_ACTION_PASS.
11157		 */
11158		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11159
11160		switch (action) {
11161		case CTL_ACTION_BLOCK:
11162			/* Nothing to do here, still blocked */
11163			break;
11164		case CTL_ACTION_OVERLAP:
11165		case CTL_ACTION_OVERLAP_TAG:
11166			/*
11167			 * This shouldn't happen!  In theory we've already
11168			 * checked this command for overlap...
11169			 */
11170			break;
11171		case CTL_ACTION_PASS:
11172		case CTL_ACTION_SKIP: {
11173			struct ctl_softc *softc;
11174			const struct ctl_cmd_entry *entry;
11175			uint32_t initidx;
11176			int isc_retval;
11177
11178			/*
11179			 * The skip case shouldn't happen, this transaction
11180			 * should have never made it onto the blocked queue.
11181			 */
11182			/*
11183			 * This I/O is no longer blocked, we can remove it
11184			 * from the blocked queue.  Since this is a TAILQ
11185			 * (doubly linked list), we can do O(1) removals
11186			 * from any place on the list.
11187			 */
11188			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11189				     blocked_links);
11190			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11191
11192			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11193				/*
11194				 * Need to send IO back to original side to
11195				 * run
11196				 */
11197				union ctl_ha_msg msg_info;
11198
11199				msg_info.hdr.original_sc =
11200					cur_blocked->io_hdr.original_sc;
11201				msg_info.hdr.serializing_sc = cur_blocked;
11202				msg_info.hdr.msg_type = CTL_MSG_R2R;
11203				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11204				     &msg_info, sizeof(msg_info), 0)) >
11205				     CTL_HA_STATUS_SUCCESS) {
11206					printf("CTL:Check Blocked error from "
11207					       "ctl_ha_msg_send %d\n",
11208					       isc_retval);
11209				}
11210				break;
11211			}
11212			entry = ctl_get_cmd_entry(&cur_blocked->scsiio);
11213			softc = control_softc;
11214
11215			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11216
11217			/*
11218			 * Check this I/O for LUN state changes that may
11219			 * have happened while this command was blocked.
11220			 * The LUN state may have been changed by a command
11221			 * ahead of us in the queue, so we need to re-check
11222			 * for any states that can be caused by SCSI
11223			 * commands.
11224			 */
11225			if (ctl_scsiio_lun_check(softc, lun, entry,
11226						 &cur_blocked->scsiio) == 0) {
11227				cur_blocked->io_hdr.flags |=
11228				                      CTL_FLAG_IS_WAS_ON_RTR;
11229				ctl_enqueue_rtr(cur_blocked);
11230			} else
11231				ctl_done(cur_blocked);
11232			break;
11233		}
11234		default:
11235			/*
11236			 * This probably shouldn't happen -- we shouldn't
11237			 * get CTL_ACTION_ERROR, or anything else.
11238			 */
11239			break;
11240		}
11241	}
11242
11243	return (CTL_RETVAL_COMPLETE);
11244}
11245
11246/*
11247 * This routine (with one exception) checks LUN flags that can be set by
11248 * commands ahead of us in the OOA queue.  These flags have to be checked
11249 * when a command initially comes in, and when we pull a command off the
11250 * blocked queue and are preparing to execute it.  The reason we have to
11251 * check these flags for commands on the blocked queue is that the LUN
11252 * state may have been changed by a command ahead of us while we're on the
11253 * blocked queue.
11254 *
11255 * Ordering is somewhat important with these checks, so please pay
11256 * careful attention to the placement of any new checks.
11257 */
11258static int
11259ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11260    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11261{
11262	int retval;
11263
11264	retval = 0;
11265
11266	mtx_assert(&lun->lun_lock, MA_OWNED);
11267
11268	/*
11269	 * If this shelf is a secondary shelf controller, we have to reject
11270	 * any media access commands.
11271	 */
11272#if 0
11273	/* No longer needed for HA */
11274	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11275	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11276		ctl_set_lun_standby(ctsio);
11277		retval = 1;
11278		goto bailout;
11279	}
11280#endif
11281
11282	/*
11283	 * Check for a reservation conflict.  If this command isn't allowed
11284	 * even on reserved LUNs, and if this initiator isn't the one who
11285	 * reserved us, reject the command with a reservation conflict.
11286	 */
11287	if ((lun->flags & CTL_LUN_RESERVED)
11288	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11289		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
11290		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
11291		 || (ctsio->io_hdr.nexus.targ_target.id !=
11292		     lun->rsv_nexus.targ_target.id)) {
11293			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11294			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11295			retval = 1;
11296			goto bailout;
11297		}
11298	}
11299
11300	if ( (lun->flags & CTL_LUN_PR_RESERVED)
11301	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11302		uint32_t residx;
11303
11304		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11305		/*
11306		 * if we aren't registered or it's a res holder type
11307		 * reservation and this isn't the res holder then set a
11308		 * conflict.
11309		 * NOTE: Commands which might be allowed on write exclusive
11310		 * type reservations are checked in the particular command
11311		 * for a conflict. Read and SSU are the only ones.
11312		 */
11313		if (!lun->per_res[residx].registered
11314		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11315			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11316			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11317			retval = 1;
11318			goto bailout;
11319		}
11320
11321	}
11322
11323	if ((lun->flags & CTL_LUN_OFFLINE)
11324	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11325		ctl_set_lun_not_ready(ctsio);
11326		retval = 1;
11327		goto bailout;
11328	}
11329
11330	/*
11331	 * If the LUN is stopped, see if this particular command is allowed
11332	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11333	 */
11334	if ((lun->flags & CTL_LUN_STOPPED)
11335	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11336		/* "Logical unit not ready, initializing cmd. required" */
11337		ctl_set_lun_stopped(ctsio);
11338		retval = 1;
11339		goto bailout;
11340	}
11341
11342	if ((lun->flags & CTL_LUN_INOPERABLE)
11343	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11344		/* "Medium format corrupted" */
11345		ctl_set_medium_format_corrupted(ctsio);
11346		retval = 1;
11347		goto bailout;
11348	}
11349
11350bailout:
11351	return (retval);
11352
11353}
11354
11355static void
11356ctl_failover_io(union ctl_io *io, int have_lock)
11357{
11358	ctl_set_busy(&io->scsiio);
11359	ctl_done(io);
11360}
11361
11362static void
11363ctl_failover(void)
11364{
11365	struct ctl_lun *lun;
11366	struct ctl_softc *ctl_softc;
11367	union ctl_io *next_io, *pending_io;
11368	union ctl_io *io;
11369	int lun_idx;
11370	int i;
11371
11372	ctl_softc = control_softc;
11373
11374	mtx_lock(&ctl_softc->ctl_lock);
11375	/*
11376	 * Remove any cmds from the other SC from the rtr queue.  These
11377	 * will obviously only be for LUNs for which we're the primary.
11378	 * We can't send status or get/send data for these commands.
11379	 * Since they haven't been executed yet, we can just remove them.
11380	 * We'll either abort them or delete them below, depending on
11381	 * which HA mode we're in.
11382	 */
11383#ifdef notyet
11384	mtx_lock(&ctl_softc->queue_lock);
11385	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11386	     io != NULL; io = next_io) {
11387		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11388		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11389			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11390				      ctl_io_hdr, links);
11391	}
11392	mtx_unlock(&ctl_softc->queue_lock);
11393#endif
11394
11395	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11396		lun = ctl_softc->ctl_luns[lun_idx];
11397		if (lun==NULL)
11398			continue;
11399
11400		/*
11401		 * Processor LUNs are primary on both sides.
11402		 * XXX will this always be true?
11403		 */
11404		if (lun->be_lun->lun_type == T_PROCESSOR)
11405			continue;
11406
11407		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11408		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11409			printf("FAILOVER: primary lun %d\n", lun_idx);
11410		        /*
11411			 * Remove all commands from the other SC. First from the
11412			 * blocked queue then from the ooa queue. Once we have
11413			 * removed them. Call ctl_check_blocked to see if there
11414			 * is anything that can run.
11415			 */
11416			for (io = (union ctl_io *)TAILQ_FIRST(
11417			     &lun->blocked_queue); io != NULL; io = next_io) {
11418
11419		        	next_io = (union ctl_io *)TAILQ_NEXT(
11420				    &io->io_hdr, blocked_links);
11421
11422				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11423					TAILQ_REMOVE(&lun->blocked_queue,
11424						     &io->io_hdr,blocked_links);
11425					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11426					TAILQ_REMOVE(&lun->ooa_queue,
11427						     &io->io_hdr, ooa_links);
11428
11429					ctl_free_io(io);
11430				}
11431			}
11432
11433			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11434	     		     io != NULL; io = next_io) {
11435
11436		        	next_io = (union ctl_io *)TAILQ_NEXT(
11437				    &io->io_hdr, ooa_links);
11438
11439				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11440
11441					TAILQ_REMOVE(&lun->ooa_queue,
11442						&io->io_hdr,
11443					     	ooa_links);
11444
11445					ctl_free_io(io);
11446				}
11447			}
11448			ctl_check_blocked(lun);
11449		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11450			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11451
11452			printf("FAILOVER: primary lun %d\n", lun_idx);
11453			/*
11454			 * Abort all commands from the other SC.  We can't
11455			 * send status back for them now.  These should get
11456			 * cleaned up when they are completed or come out
11457			 * for a datamove operation.
11458			 */
11459			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11460	     		     io != NULL; io = next_io) {
11461		        	next_io = (union ctl_io *)TAILQ_NEXT(
11462					&io->io_hdr, ooa_links);
11463
11464				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11465					io->io_hdr.flags |= CTL_FLAG_ABORT;
11466			}
11467		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11468			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11469
11470			printf("FAILOVER: secondary lun %d\n", lun_idx);
11471
11472			lun->flags |= CTL_LUN_PRIMARY_SC;
11473
11474			/*
11475			 * We send all I/O that was sent to this controller
11476			 * and redirected to the other side back with
11477			 * busy status, and have the initiator retry it.
11478			 * Figuring out how much data has been transferred,
11479			 * etc. and picking up where we left off would be
11480			 * very tricky.
11481			 *
11482			 * XXX KDM need to remove I/O from the blocked
11483			 * queue as well!
11484			 */
11485			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11486			     &lun->ooa_queue); pending_io != NULL;
11487			     pending_io = next_io) {
11488
11489				next_io =  (union ctl_io *)TAILQ_NEXT(
11490					&pending_io->io_hdr, ooa_links);
11491
11492				pending_io->io_hdr.flags &=
11493					~CTL_FLAG_SENT_2OTHER_SC;
11494
11495				if (pending_io->io_hdr.flags &
11496				    CTL_FLAG_IO_ACTIVE) {
11497					pending_io->io_hdr.flags |=
11498						CTL_FLAG_FAILOVER;
11499				} else {
11500					ctl_set_busy(&pending_io->scsiio);
11501					ctl_done(pending_io);
11502				}
11503			}
11504
11505			/*
11506			 * Build Unit Attention
11507			 */
11508			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11509				lun->pending_ua[i] |=
11510				                     CTL_UA_ASYM_ACC_CHANGE;
11511			}
11512		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11513			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11514			printf("FAILOVER: secondary lun %d\n", lun_idx);
11515			/*
11516			 * if the first io on the OOA is not on the RtR queue
11517			 * add it.
11518			 */
11519			lun->flags |= CTL_LUN_PRIMARY_SC;
11520
11521			pending_io = (union ctl_io *)TAILQ_FIRST(
11522			    &lun->ooa_queue);
11523			if (pending_io==NULL) {
11524				printf("Nothing on OOA queue\n");
11525				continue;
11526			}
11527
11528			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11529			if ((pending_io->io_hdr.flags &
11530			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11531				pending_io->io_hdr.flags |=
11532				    CTL_FLAG_IS_WAS_ON_RTR;
11533				ctl_enqueue_rtr(pending_io);
11534			}
11535#if 0
11536			else
11537			{
11538				printf("Tag 0x%04x is running\n",
11539				      pending_io->scsiio.tag_num);
11540			}
11541#endif
11542
11543			next_io = (union ctl_io *)TAILQ_NEXT(
11544			    &pending_io->io_hdr, ooa_links);
11545			for (pending_io=next_io; pending_io != NULL;
11546			     pending_io = next_io) {
11547				pending_io->io_hdr.flags &=
11548				    ~CTL_FLAG_SENT_2OTHER_SC;
11549				next_io = (union ctl_io *)TAILQ_NEXT(
11550					&pending_io->io_hdr, ooa_links);
11551				if (pending_io->io_hdr.flags &
11552				    CTL_FLAG_IS_WAS_ON_RTR) {
11553#if 0
11554				        printf("Tag 0x%04x is running\n",
11555				      		pending_io->scsiio.tag_num);
11556#endif
11557					continue;
11558				}
11559
11560				switch (ctl_check_ooa(lun, pending_io,
11561			            (union ctl_io *)TAILQ_PREV(
11562				    &pending_io->io_hdr, ctl_ooaq,
11563				    ooa_links))) {
11564
11565				case CTL_ACTION_BLOCK:
11566					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11567							  &pending_io->io_hdr,
11568							  blocked_links);
11569					pending_io->io_hdr.flags |=
11570					    CTL_FLAG_BLOCKED;
11571					break;
11572				case CTL_ACTION_PASS:
11573				case CTL_ACTION_SKIP:
11574					pending_io->io_hdr.flags |=
11575					    CTL_FLAG_IS_WAS_ON_RTR;
11576					ctl_enqueue_rtr(pending_io);
11577					break;
11578				case CTL_ACTION_OVERLAP:
11579					ctl_set_overlapped_cmd(
11580					    (struct ctl_scsiio *)pending_io);
11581					ctl_done(pending_io);
11582					break;
11583				case CTL_ACTION_OVERLAP_TAG:
11584					ctl_set_overlapped_tag(
11585					    (struct ctl_scsiio *)pending_io,
11586					    pending_io->scsiio.tag_num & 0xff);
11587					ctl_done(pending_io);
11588					break;
11589				case CTL_ACTION_ERROR:
11590				default:
11591					ctl_set_internal_failure(
11592						(struct ctl_scsiio *)pending_io,
11593						0,  // sks_valid
11594						0); //retry count
11595					ctl_done(pending_io);
11596					break;
11597				}
11598			}
11599
11600			/*
11601			 * Build Unit Attention
11602			 */
11603			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11604				lun->pending_ua[i] |=
11605				                     CTL_UA_ASYM_ACC_CHANGE;
11606			}
11607		} else {
11608			panic("Unhandled HA mode failover, LUN flags = %#x, "
11609			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11610		}
11611	}
11612	ctl_pause_rtr = 0;
11613	mtx_unlock(&ctl_softc->ctl_lock);
11614}
11615
11616static int
11617ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11618{
11619	struct ctl_lun *lun;
11620	const struct ctl_cmd_entry *entry;
11621	uint32_t initidx, targ_lun;
11622	int retval;
11623
11624	retval = 0;
11625
11626	lun = NULL;
11627
11628	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11629	if ((targ_lun < CTL_MAX_LUNS)
11630	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11631		lun = ctl_softc->ctl_luns[targ_lun];
11632		/*
11633		 * If the LUN is invalid, pretend that it doesn't exist.
11634		 * It will go away as soon as all pending I/O has been
11635		 * completed.
11636		 */
11637		if (lun->flags & CTL_LUN_DISABLED) {
11638			lun = NULL;
11639		} else {
11640			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11641			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11642				lun->be_lun;
11643			if (lun->be_lun->lun_type == T_PROCESSOR) {
11644				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11645			}
11646
11647			/*
11648			 * Every I/O goes into the OOA queue for a
11649			 * particular LUN, and stays there until completion.
11650			 */
11651			mtx_lock(&lun->lun_lock);
11652			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11653			    ooa_links);
11654		}
11655	} else {
11656		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11657		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11658	}
11659
11660	/* Get command entry and return error if it is unsuppotyed. */
11661	entry = ctl_validate_command(ctsio);
11662	if (entry == NULL) {
11663		if (lun)
11664			mtx_unlock(&lun->lun_lock);
11665		return (retval);
11666	}
11667
11668	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11669	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11670
11671	/*
11672	 * Check to see whether we can send this command to LUNs that don't
11673	 * exist.  This should pretty much only be the case for inquiry
11674	 * and request sense.  Further checks, below, really require having
11675	 * a LUN, so we can't really check the command anymore.  Just put
11676	 * it on the rtr queue.
11677	 */
11678	if (lun == NULL) {
11679		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11680			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11681			ctl_enqueue_rtr((union ctl_io *)ctsio);
11682			return (retval);
11683		}
11684
11685		ctl_set_unsupported_lun(ctsio);
11686		ctl_done((union ctl_io *)ctsio);
11687		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11688		return (retval);
11689	} else {
11690		/*
11691		 * Make sure we support this particular command on this LUN.
11692		 * e.g., we don't support writes to the control LUN.
11693		 */
11694		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11695			mtx_unlock(&lun->lun_lock);
11696			ctl_set_invalid_opcode(ctsio);
11697			ctl_done((union ctl_io *)ctsio);
11698			return (retval);
11699		}
11700	}
11701
11702	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11703
11704#ifdef CTL_WITH_CA
11705	/*
11706	 * If we've got a request sense, it'll clear the contingent
11707	 * allegiance condition.  Otherwise, if we have a CA condition for
11708	 * this initiator, clear it, because it sent down a command other
11709	 * than request sense.
11710	 */
11711	if ((ctsio->cdb[0] != REQUEST_SENSE)
11712	 && (ctl_is_set(lun->have_ca, initidx)))
11713		ctl_clear_mask(lun->have_ca, initidx);
11714#endif
11715
11716	/*
11717	 * If the command has this flag set, it handles its own unit
11718	 * attention reporting, we shouldn't do anything.  Otherwise we
11719	 * check for any pending unit attentions, and send them back to the
11720	 * initiator.  We only do this when a command initially comes in,
11721	 * not when we pull it off the blocked queue.
11722	 *
11723	 * According to SAM-3, section 5.3.2, the order that things get
11724	 * presented back to the host is basically unit attentions caused
11725	 * by some sort of reset event, busy status, reservation conflicts
11726	 * or task set full, and finally any other status.
11727	 *
11728	 * One issue here is that some of the unit attentions we report
11729	 * don't fall into the "reset" category (e.g. "reported luns data
11730	 * has changed").  So reporting it here, before the reservation
11731	 * check, may be technically wrong.  I guess the only thing to do
11732	 * would be to check for and report the reset events here, and then
11733	 * check for the other unit attention types after we check for a
11734	 * reservation conflict.
11735	 *
11736	 * XXX KDM need to fix this
11737	 */
11738	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11739		ctl_ua_type ua_type;
11740
11741		ua_type = lun->pending_ua[initidx];
11742		if (ua_type != CTL_UA_NONE) {
11743			scsi_sense_data_type sense_format;
11744
11745			if (lun != NULL)
11746				sense_format = (lun->flags &
11747				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11748				    SSD_TYPE_FIXED;
11749			else
11750				sense_format = SSD_TYPE_FIXED;
11751
11752			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11753					       sense_format);
11754			if (ua_type != CTL_UA_NONE) {
11755				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11756				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11757						       CTL_AUTOSENSE;
11758				ctsio->sense_len = SSD_FULL_SIZE;
11759				lun->pending_ua[initidx] &= ~ua_type;
11760				mtx_unlock(&lun->lun_lock);
11761				ctl_done((union ctl_io *)ctsio);
11762				return (retval);
11763			}
11764		}
11765	}
11766
11767
11768	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11769		mtx_unlock(&lun->lun_lock);
11770		ctl_done((union ctl_io *)ctsio);
11771		return (retval);
11772	}
11773
11774	/*
11775	 * XXX CHD this is where we want to send IO to other side if
11776	 * this LUN is secondary on this SC. We will need to make a copy
11777	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11778	 * the copy we send as FROM_OTHER.
11779	 * We also need to stuff the address of the original IO so we can
11780	 * find it easily. Something similar will need be done on the other
11781	 * side so when we are done we can find the copy.
11782	 */
11783	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11784		union ctl_ha_msg msg_info;
11785		int isc_retval;
11786
11787		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11788
11789		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11790		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11791#if 0
11792		printf("1. ctsio %p\n", ctsio);
11793#endif
11794		msg_info.hdr.serializing_sc = NULL;
11795		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11796		msg_info.scsi.tag_num = ctsio->tag_num;
11797		msg_info.scsi.tag_type = ctsio->tag_type;
11798		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11799
11800		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11801
11802		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11803		    (void *)&msg_info, sizeof(msg_info), 0)) >
11804		    CTL_HA_STATUS_SUCCESS) {
11805			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11806			       isc_retval);
11807			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11808		} else {
11809#if 0
11810			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11811#endif
11812		}
11813
11814		/*
11815		 * XXX KDM this I/O is off the incoming queue, but hasn't
11816		 * been inserted on any other queue.  We may need to come
11817		 * up with a holding queue while we wait for serialization
11818		 * so that we have an idea of what we're waiting for from
11819		 * the other side.
11820		 */
11821		mtx_unlock(&lun->lun_lock);
11822		return (retval);
11823	}
11824
11825	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11826			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11827			      ctl_ooaq, ooa_links))) {
11828	case CTL_ACTION_BLOCK:
11829		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11830		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11831				  blocked_links);
11832		mtx_unlock(&lun->lun_lock);
11833		return (retval);
11834	case CTL_ACTION_PASS:
11835	case CTL_ACTION_SKIP:
11836		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11837		mtx_unlock(&lun->lun_lock);
11838		ctl_enqueue_rtr((union ctl_io *)ctsio);
11839		break;
11840	case CTL_ACTION_OVERLAP:
11841		mtx_unlock(&lun->lun_lock);
11842		ctl_set_overlapped_cmd(ctsio);
11843		ctl_done((union ctl_io *)ctsio);
11844		break;
11845	case CTL_ACTION_OVERLAP_TAG:
11846		mtx_unlock(&lun->lun_lock);
11847		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11848		ctl_done((union ctl_io *)ctsio);
11849		break;
11850	case CTL_ACTION_ERROR:
11851	default:
11852		mtx_unlock(&lun->lun_lock);
11853		ctl_set_internal_failure(ctsio,
11854					 /*sks_valid*/ 0,
11855					 /*retry_count*/ 0);
11856		ctl_done((union ctl_io *)ctsio);
11857		break;
11858	}
11859	return (retval);
11860}
11861
11862const struct ctl_cmd_entry *
11863ctl_get_cmd_entry(struct ctl_scsiio *ctsio)
11864{
11865	const struct ctl_cmd_entry *entry;
11866	int service_action;
11867
11868	entry = &ctl_cmd_table[ctsio->cdb[0]];
11869	if (entry->flags & CTL_CMD_FLAG_SA5) {
11870		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11871		entry = &((const struct ctl_cmd_entry *)
11872		    entry->execute)[service_action];
11873	}
11874	return (entry);
11875}
11876
11877const struct ctl_cmd_entry *
11878ctl_validate_command(struct ctl_scsiio *ctsio)
11879{
11880	const struct ctl_cmd_entry *entry;
11881	int i;
11882	uint8_t diff;
11883
11884	entry = ctl_get_cmd_entry(ctsio);
11885	if (entry->execute == NULL) {
11886		ctl_set_invalid_opcode(ctsio);
11887		ctl_done((union ctl_io *)ctsio);
11888		return (NULL);
11889	}
11890	KASSERT(entry->length > 0,
11891	    ("Not defined length for command 0x%02x/0x%02x",
11892	     ctsio->cdb[0], ctsio->cdb[1]));
11893	for (i = 1; i < entry->length; i++) {
11894		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11895		if (diff == 0)
11896			continue;
11897		ctl_set_invalid_field(ctsio,
11898				      /*sks_valid*/ 1,
11899				      /*command*/ 1,
11900				      /*field*/ i,
11901				      /*bit_valid*/ 1,
11902				      /*bit*/ fls(diff) - 1);
11903		ctl_done((union ctl_io *)ctsio);
11904		return (NULL);
11905	}
11906	return (entry);
11907}
11908
11909static int
11910ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11911{
11912
11913	switch (lun_type) {
11914	case T_PROCESSOR:
11915		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11916		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11917			return (0);
11918		break;
11919	case T_DIRECT:
11920		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11921		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11922			return (0);
11923		break;
11924	default:
11925		return (0);
11926	}
11927	return (1);
11928}
11929
11930static int
11931ctl_scsiio(struct ctl_scsiio *ctsio)
11932{
11933	int retval;
11934	const struct ctl_cmd_entry *entry;
11935
11936	retval = CTL_RETVAL_COMPLETE;
11937
11938	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11939
11940	entry = ctl_get_cmd_entry(ctsio);
11941
11942	/*
11943	 * If this I/O has been aborted, just send it straight to
11944	 * ctl_done() without executing it.
11945	 */
11946	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11947		ctl_done((union ctl_io *)ctsio);
11948		goto bailout;
11949	}
11950
11951	/*
11952	 * All the checks should have been handled by ctl_scsiio_precheck().
11953	 * We should be clear now to just execute the I/O.
11954	 */
11955	retval = entry->execute(ctsio);
11956
11957bailout:
11958	return (retval);
11959}
11960
11961/*
11962 * Since we only implement one target right now, a bus reset simply resets
11963 * our single target.
11964 */
11965static int
11966ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11967{
11968	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11969}
11970
11971static int
11972ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11973		 ctl_ua_type ua_type)
11974{
11975	struct ctl_lun *lun;
11976	int retval;
11977
11978	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11979		union ctl_ha_msg msg_info;
11980
11981		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11982		msg_info.hdr.nexus = io->io_hdr.nexus;
11983		if (ua_type==CTL_UA_TARG_RESET)
11984			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11985		else
11986			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11987		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11988		msg_info.hdr.original_sc = NULL;
11989		msg_info.hdr.serializing_sc = NULL;
11990		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11991		    (void *)&msg_info, sizeof(msg_info), 0)) {
11992		}
11993	}
11994	retval = 0;
11995
11996	mtx_lock(&ctl_softc->ctl_lock);
11997	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11998		retval += ctl_lun_reset(lun, io, ua_type);
11999	mtx_unlock(&ctl_softc->ctl_lock);
12000
12001	return (retval);
12002}
12003
12004/*
12005 * The LUN should always be set.  The I/O is optional, and is used to
12006 * distinguish between I/Os sent by this initiator, and by other
12007 * initiators.  We set unit attention for initiators other than this one.
12008 * SAM-3 is vague on this point.  It does say that a unit attention should
12009 * be established for other initiators when a LUN is reset (see section
12010 * 5.7.3), but it doesn't specifically say that the unit attention should
12011 * be established for this particular initiator when a LUN is reset.  Here
12012 * is the relevant text, from SAM-3 rev 8:
12013 *
12014 * 5.7.2 When a SCSI initiator port aborts its own tasks
12015 *
12016 * When a SCSI initiator port causes its own task(s) to be aborted, no
12017 * notification that the task(s) have been aborted shall be returned to
12018 * the SCSI initiator port other than the completion response for the
12019 * command or task management function action that caused the task(s) to
12020 * be aborted and notification(s) associated with related effects of the
12021 * action (e.g., a reset unit attention condition).
12022 *
12023 * XXX KDM for now, we're setting unit attention for all initiators.
12024 */
12025static int
12026ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
12027{
12028	union ctl_io *xio;
12029#if 0
12030	uint32_t initindex;
12031#endif
12032	int i;
12033
12034	mtx_lock(&lun->lun_lock);
12035	/*
12036	 * Run through the OOA queue and abort each I/O.
12037	 */
12038#if 0
12039	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12040#endif
12041	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12042	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12043		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
12044	}
12045
12046	/*
12047	 * This version sets unit attention for every
12048	 */
12049#if 0
12050	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12051	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12052		if (initindex == i)
12053			continue;
12054		lun->pending_ua[i] |= ua_type;
12055	}
12056#endif
12057
12058	/*
12059	 * A reset (any kind, really) clears reservations established with
12060	 * RESERVE/RELEASE.  It does not clear reservations established
12061	 * with PERSISTENT RESERVE OUT, but we don't support that at the
12062	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
12063	 * reservations made with the RESERVE/RELEASE commands, because
12064	 * those commands are obsolete in SPC-3.
12065	 */
12066	lun->flags &= ~CTL_LUN_RESERVED;
12067
12068	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12069#ifdef CTL_WITH_CA
12070		ctl_clear_mask(lun->have_ca, i);
12071#endif
12072		lun->pending_ua[i] |= ua_type;
12073	}
12074	mtx_unlock(&lun->lun_lock);
12075
12076	return (0);
12077}
12078
12079static void
12080ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
12081    int other_sc)
12082{
12083	union ctl_io *xio;
12084
12085	mtx_assert(&lun->lun_lock, MA_OWNED);
12086
12087	/*
12088	 * Run through the OOA queue and attempt to find the given I/O.
12089	 * The target port, initiator ID, tag type and tag number have to
12090	 * match the values that we got from the initiator.  If we have an
12091	 * untagged command to abort, simply abort the first untagged command
12092	 * we come to.  We only allow one untagged command at a time of course.
12093	 */
12094	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12095	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12096
12097		if ((targ_port == UINT32_MAX ||
12098		     targ_port == xio->io_hdr.nexus.targ_port) &&
12099		    (init_id == UINT32_MAX ||
12100		     init_id == xio->io_hdr.nexus.initid.id)) {
12101			if (targ_port != xio->io_hdr.nexus.targ_port ||
12102			    init_id != xio->io_hdr.nexus.initid.id)
12103				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12104			xio->io_hdr.flags |= CTL_FLAG_ABORT;
12105			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12106				union ctl_ha_msg msg_info;
12107
12108				msg_info.hdr.nexus = xio->io_hdr.nexus;
12109				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12110				msg_info.task.tag_num = xio->scsiio.tag_num;
12111				msg_info.task.tag_type = xio->scsiio.tag_type;
12112				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12113				msg_info.hdr.original_sc = NULL;
12114				msg_info.hdr.serializing_sc = NULL;
12115				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12116				    (void *)&msg_info, sizeof(msg_info), 0);
12117			}
12118		}
12119	}
12120}
12121
12122static int
12123ctl_abort_task_set(union ctl_io *io)
12124{
12125	struct ctl_softc *softc = control_softc;
12126	struct ctl_lun *lun;
12127	uint32_t targ_lun;
12128
12129	/*
12130	 * Look up the LUN.
12131	 */
12132	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12133	mtx_lock(&softc->ctl_lock);
12134	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12135		lun = softc->ctl_luns[targ_lun];
12136	else {
12137		mtx_unlock(&softc->ctl_lock);
12138		return (1);
12139	}
12140
12141	mtx_lock(&lun->lun_lock);
12142	mtx_unlock(&softc->ctl_lock);
12143	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12144		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12145		    io->io_hdr.nexus.initid.id,
12146		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12147	} else { /* CTL_TASK_CLEAR_TASK_SET */
12148		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12149		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12150	}
12151	mtx_unlock(&lun->lun_lock);
12152	return (0);
12153}
12154
12155static int
12156ctl_i_t_nexus_reset(union ctl_io *io)
12157{
12158	struct ctl_softc *softc = control_softc;
12159	struct ctl_lun *lun;
12160	uint32_t initindex;
12161
12162	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12163	mtx_lock(&softc->ctl_lock);
12164	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12165		mtx_lock(&lun->lun_lock);
12166		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12167		    io->io_hdr.nexus.initid.id,
12168		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12169#ifdef CTL_WITH_CA
12170		ctl_clear_mask(lun->have_ca, initindex);
12171#endif
12172		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12173		mtx_unlock(&lun->lun_lock);
12174	}
12175	mtx_unlock(&softc->ctl_lock);
12176	return (0);
12177}
12178
12179static int
12180ctl_abort_task(union ctl_io *io)
12181{
12182	union ctl_io *xio;
12183	struct ctl_lun *lun;
12184	struct ctl_softc *ctl_softc;
12185#if 0
12186	struct sbuf sb;
12187	char printbuf[128];
12188#endif
12189	int found;
12190	uint32_t targ_lun;
12191
12192	ctl_softc = control_softc;
12193	found = 0;
12194
12195	/*
12196	 * Look up the LUN.
12197	 */
12198	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12199	mtx_lock(&ctl_softc->ctl_lock);
12200	if ((targ_lun < CTL_MAX_LUNS)
12201	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12202		lun = ctl_softc->ctl_luns[targ_lun];
12203	else {
12204		mtx_unlock(&ctl_softc->ctl_lock);
12205		return (1);
12206	}
12207
12208#if 0
12209	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12210	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12211#endif
12212
12213	mtx_lock(&lun->lun_lock);
12214	mtx_unlock(&ctl_softc->ctl_lock);
12215	/*
12216	 * Run through the OOA queue and attempt to find the given I/O.
12217	 * The target port, initiator ID, tag type and tag number have to
12218	 * match the values that we got from the initiator.  If we have an
12219	 * untagged command to abort, simply abort the first untagged command
12220	 * we come to.  We only allow one untagged command at a time of course.
12221	 */
12222#if 0
12223	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12224#endif
12225	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12226	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12227#if 0
12228		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12229
12230		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12231			    lun->lun, xio->scsiio.tag_num,
12232			    xio->scsiio.tag_type,
12233			    (xio->io_hdr.blocked_links.tqe_prev
12234			    == NULL) ? "" : " BLOCKED",
12235			    (xio->io_hdr.flags &
12236			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12237			    (xio->io_hdr.flags &
12238			    CTL_FLAG_ABORT) ? " ABORT" : "",
12239			    (xio->io_hdr.flags &
12240			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12241		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12242		sbuf_finish(&sb);
12243		printf("%s\n", sbuf_data(&sb));
12244#endif
12245
12246		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12247		 && (xio->io_hdr.nexus.initid.id ==
12248		     io->io_hdr.nexus.initid.id)) {
12249			/*
12250			 * If the abort says that the task is untagged, the
12251			 * task in the queue must be untagged.  Otherwise,
12252			 * we just check to see whether the tag numbers
12253			 * match.  This is because the QLogic firmware
12254			 * doesn't pass back the tag type in an abort
12255			 * request.
12256			 */
12257#if 0
12258			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12259			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12260			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12261#endif
12262			/*
12263			 * XXX KDM we've got problems with FC, because it
12264			 * doesn't send down a tag type with aborts.  So we
12265			 * can only really go by the tag number...
12266			 * This may cause problems with parallel SCSI.
12267			 * Need to figure that out!!
12268			 */
12269			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12270				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12271				found = 1;
12272				if ((io->io_hdr.flags &
12273				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12274				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12275					union ctl_ha_msg msg_info;
12276
12277					io->io_hdr.flags |=
12278					                CTL_FLAG_SENT_2OTHER_SC;
12279					msg_info.hdr.nexus = io->io_hdr.nexus;
12280					msg_info.task.task_action =
12281						CTL_TASK_ABORT_TASK;
12282					msg_info.task.tag_num =
12283						io->taskio.tag_num;
12284					msg_info.task.tag_type =
12285						io->taskio.tag_type;
12286					msg_info.hdr.msg_type =
12287						CTL_MSG_MANAGE_TASKS;
12288					msg_info.hdr.original_sc = NULL;
12289					msg_info.hdr.serializing_sc = NULL;
12290#if 0
12291					printf("Sent Abort to other side\n");
12292#endif
12293					if (CTL_HA_STATUS_SUCCESS !=
12294					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12295		    				(void *)&msg_info,
12296						sizeof(msg_info), 0)) {
12297					}
12298				}
12299#if 0
12300				printf("ctl_abort_task: found I/O to abort\n");
12301#endif
12302				break;
12303			}
12304		}
12305	}
12306	mtx_unlock(&lun->lun_lock);
12307
12308	if (found == 0) {
12309		/*
12310		 * This isn't really an error.  It's entirely possible for
12311		 * the abort and command completion to cross on the wire.
12312		 * This is more of an informative/diagnostic error.
12313		 */
12314#if 0
12315		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12316		       "%d:%d:%d:%d tag %d type %d\n",
12317		       io->io_hdr.nexus.initid.id,
12318		       io->io_hdr.nexus.targ_port,
12319		       io->io_hdr.nexus.targ_target.id,
12320		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12321		       io->taskio.tag_type);
12322#endif
12323	}
12324	return (0);
12325}
12326
12327static void
12328ctl_run_task(union ctl_io *io)
12329{
12330	struct ctl_softc *ctl_softc = control_softc;
12331	int retval = 1;
12332	const char *task_desc;
12333
12334	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12335
12336	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12337	    ("ctl_run_task: Unextected io_type %d\n",
12338	     io->io_hdr.io_type));
12339
12340	task_desc = ctl_scsi_task_string(&io->taskio);
12341	if (task_desc != NULL) {
12342#ifdef NEEDTOPORT
12343		csevent_log(CSC_CTL | CSC_SHELF_SW |
12344			    CTL_TASK_REPORT,
12345			    csevent_LogType_Trace,
12346			    csevent_Severity_Information,
12347			    csevent_AlertLevel_Green,
12348			    csevent_FRU_Firmware,
12349			    csevent_FRU_Unknown,
12350			    "CTL: received task: %s",task_desc);
12351#endif
12352	} else {
12353#ifdef NEEDTOPORT
12354		csevent_log(CSC_CTL | CSC_SHELF_SW |
12355			    CTL_TASK_REPORT,
12356			    csevent_LogType_Trace,
12357			    csevent_Severity_Information,
12358			    csevent_AlertLevel_Green,
12359			    csevent_FRU_Firmware,
12360			    csevent_FRU_Unknown,
12361			    "CTL: received unknown task "
12362			    "type: %d (%#x)",
12363			    io->taskio.task_action,
12364			    io->taskio.task_action);
12365#endif
12366	}
12367	switch (io->taskio.task_action) {
12368	case CTL_TASK_ABORT_TASK:
12369		retval = ctl_abort_task(io);
12370		break;
12371	case CTL_TASK_ABORT_TASK_SET:
12372	case CTL_TASK_CLEAR_TASK_SET:
12373		retval = ctl_abort_task_set(io);
12374		break;
12375	case CTL_TASK_CLEAR_ACA:
12376		break;
12377	case CTL_TASK_I_T_NEXUS_RESET:
12378		retval = ctl_i_t_nexus_reset(io);
12379		break;
12380	case CTL_TASK_LUN_RESET: {
12381		struct ctl_lun *lun;
12382		uint32_t targ_lun;
12383
12384		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12385		mtx_lock(&ctl_softc->ctl_lock);
12386		if ((targ_lun < CTL_MAX_LUNS)
12387		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12388			lun = ctl_softc->ctl_luns[targ_lun];
12389		else {
12390			mtx_unlock(&ctl_softc->ctl_lock);
12391			retval = 1;
12392			break;
12393		}
12394
12395		if (!(io->io_hdr.flags &
12396		    CTL_FLAG_FROM_OTHER_SC)) {
12397			union ctl_ha_msg msg_info;
12398
12399			io->io_hdr.flags |=
12400				CTL_FLAG_SENT_2OTHER_SC;
12401			msg_info.hdr.msg_type =
12402				CTL_MSG_MANAGE_TASKS;
12403			msg_info.hdr.nexus = io->io_hdr.nexus;
12404			msg_info.task.task_action =
12405				CTL_TASK_LUN_RESET;
12406			msg_info.hdr.original_sc = NULL;
12407			msg_info.hdr.serializing_sc = NULL;
12408			if (CTL_HA_STATUS_SUCCESS !=
12409			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12410			    (void *)&msg_info,
12411			    sizeof(msg_info), 0)) {
12412			}
12413		}
12414
12415		retval = ctl_lun_reset(lun, io,
12416				       CTL_UA_LUN_RESET);
12417		mtx_unlock(&ctl_softc->ctl_lock);
12418		break;
12419	}
12420	case CTL_TASK_TARGET_RESET:
12421		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12422		break;
12423	case CTL_TASK_BUS_RESET:
12424		retval = ctl_bus_reset(ctl_softc, io);
12425		break;
12426	case CTL_TASK_PORT_LOGIN:
12427		break;
12428	case CTL_TASK_PORT_LOGOUT:
12429		break;
12430	default:
12431		printf("ctl_run_task: got unknown task management event %d\n",
12432		       io->taskio.task_action);
12433		break;
12434	}
12435	if (retval == 0)
12436		io->io_hdr.status = CTL_SUCCESS;
12437	else
12438		io->io_hdr.status = CTL_ERROR;
12439	ctl_done(io);
12440}
12441
12442/*
12443 * For HA operation.  Handle commands that come in from the other
12444 * controller.
12445 */
12446static void
12447ctl_handle_isc(union ctl_io *io)
12448{
12449	int free_io;
12450	struct ctl_lun *lun;
12451	struct ctl_softc *ctl_softc;
12452	uint32_t targ_lun;
12453
12454	ctl_softc = control_softc;
12455
12456	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12457	lun = ctl_softc->ctl_luns[targ_lun];
12458
12459	switch (io->io_hdr.msg_type) {
12460	case CTL_MSG_SERIALIZE:
12461		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12462		break;
12463	case CTL_MSG_R2R: {
12464		const struct ctl_cmd_entry *entry;
12465
12466		/*
12467		 * This is only used in SER_ONLY mode.
12468		 */
12469		free_io = 0;
12470		entry = ctl_get_cmd_entry(&io->scsiio);
12471		mtx_lock(&lun->lun_lock);
12472		if (ctl_scsiio_lun_check(ctl_softc, lun,
12473		    entry, (struct ctl_scsiio *)io) != 0) {
12474			mtx_unlock(&lun->lun_lock);
12475			ctl_done(io);
12476			break;
12477		}
12478		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12479		mtx_unlock(&lun->lun_lock);
12480		ctl_enqueue_rtr(io);
12481		break;
12482	}
12483	case CTL_MSG_FINISH_IO:
12484		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12485			free_io = 0;
12486			ctl_done(io);
12487		} else {
12488			free_io = 1;
12489			mtx_lock(&lun->lun_lock);
12490			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12491				     ooa_links);
12492			ctl_check_blocked(lun);
12493			mtx_unlock(&lun->lun_lock);
12494		}
12495		break;
12496	case CTL_MSG_PERS_ACTION:
12497		ctl_hndl_per_res_out_on_other_sc(
12498			(union ctl_ha_msg *)&io->presio.pr_msg);
12499		free_io = 1;
12500		break;
12501	case CTL_MSG_BAD_JUJU:
12502		free_io = 0;
12503		ctl_done(io);
12504		break;
12505	case CTL_MSG_DATAMOVE:
12506		/* Only used in XFER mode */
12507		free_io = 0;
12508		ctl_datamove_remote(io);
12509		break;
12510	case CTL_MSG_DATAMOVE_DONE:
12511		/* Only used in XFER mode */
12512		free_io = 0;
12513		io->scsiio.be_move_done(io);
12514		break;
12515	default:
12516		free_io = 1;
12517		printf("%s: Invalid message type %d\n",
12518		       __func__, io->io_hdr.msg_type);
12519		break;
12520	}
12521	if (free_io)
12522		ctl_free_io(io);
12523
12524}
12525
12526
12527/*
12528 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12529 * there is no match.
12530 */
12531static ctl_lun_error_pattern
12532ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12533{
12534	const struct ctl_cmd_entry *entry;
12535	ctl_lun_error_pattern filtered_pattern, pattern;
12536
12537	pattern = desc->error_pattern;
12538
12539	/*
12540	 * XXX KDM we need more data passed into this function to match a
12541	 * custom pattern, and we actually need to implement custom pattern
12542	 * matching.
12543	 */
12544	if (pattern & CTL_LUN_PAT_CMD)
12545		return (CTL_LUN_PAT_CMD);
12546
12547	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12548		return (CTL_LUN_PAT_ANY);
12549
12550	entry = ctl_get_cmd_entry(ctsio);
12551
12552	filtered_pattern = entry->pattern & pattern;
12553
12554	/*
12555	 * If the user requested specific flags in the pattern (e.g.
12556	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12557	 * flags.
12558	 *
12559	 * If the user did not specify any flags, it doesn't matter whether
12560	 * or not the command supports the flags.
12561	 */
12562	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12563	     (pattern & ~CTL_LUN_PAT_MASK))
12564		return (CTL_LUN_PAT_NONE);
12565
12566	/*
12567	 * If the user asked for a range check, see if the requested LBA
12568	 * range overlaps with this command's LBA range.
12569	 */
12570	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12571		uint64_t lba1;
12572		uint64_t len1;
12573		ctl_action action;
12574		int retval;
12575
12576		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12577		if (retval != 0)
12578			return (CTL_LUN_PAT_NONE);
12579
12580		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12581					      desc->lba_range.len);
12582		/*
12583		 * A "pass" means that the LBA ranges don't overlap, so
12584		 * this doesn't match the user's range criteria.
12585		 */
12586		if (action == CTL_ACTION_PASS)
12587			return (CTL_LUN_PAT_NONE);
12588	}
12589
12590	return (filtered_pattern);
12591}
12592
12593static void
12594ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12595{
12596	struct ctl_error_desc *desc, *desc2;
12597
12598	mtx_assert(&lun->lun_lock, MA_OWNED);
12599
12600	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12601		ctl_lun_error_pattern pattern;
12602		/*
12603		 * Check to see whether this particular command matches
12604		 * the pattern in the descriptor.
12605		 */
12606		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12607		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12608			continue;
12609
12610		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12611		case CTL_LUN_INJ_ABORTED:
12612			ctl_set_aborted(&io->scsiio);
12613			break;
12614		case CTL_LUN_INJ_MEDIUM_ERR:
12615			ctl_set_medium_error(&io->scsiio);
12616			break;
12617		case CTL_LUN_INJ_UA:
12618			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12619			 * OCCURRED */
12620			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12621			break;
12622		case CTL_LUN_INJ_CUSTOM:
12623			/*
12624			 * We're assuming the user knows what he is doing.
12625			 * Just copy the sense information without doing
12626			 * checks.
12627			 */
12628			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12629			      ctl_min(sizeof(desc->custom_sense),
12630				      sizeof(io->scsiio.sense_data)));
12631			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12632			io->scsiio.sense_len = SSD_FULL_SIZE;
12633			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12634			break;
12635		case CTL_LUN_INJ_NONE:
12636		default:
12637			/*
12638			 * If this is an error injection type we don't know
12639			 * about, clear the continuous flag (if it is set)
12640			 * so it will get deleted below.
12641			 */
12642			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12643			break;
12644		}
12645		/*
12646		 * By default, each error injection action is a one-shot
12647		 */
12648		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12649			continue;
12650
12651		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12652
12653		free(desc, M_CTL);
12654	}
12655}
12656
12657#ifdef CTL_IO_DELAY
12658static void
12659ctl_datamove_timer_wakeup(void *arg)
12660{
12661	union ctl_io *io;
12662
12663	io = (union ctl_io *)arg;
12664
12665	ctl_datamove(io);
12666}
12667#endif /* CTL_IO_DELAY */
12668
12669void
12670ctl_datamove(union ctl_io *io)
12671{
12672	void (*fe_datamove)(union ctl_io *io);
12673
12674	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12675
12676	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12677
12678#ifdef CTL_TIME_IO
12679	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12680		char str[256];
12681		char path_str[64];
12682		struct sbuf sb;
12683
12684		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12685		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12686
12687		sbuf_cat(&sb, path_str);
12688		switch (io->io_hdr.io_type) {
12689		case CTL_IO_SCSI:
12690			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12691			sbuf_printf(&sb, "\n");
12692			sbuf_cat(&sb, path_str);
12693			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12694				    io->scsiio.tag_num, io->scsiio.tag_type);
12695			break;
12696		case CTL_IO_TASK:
12697			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12698				    "Tag Type: %d\n", io->taskio.task_action,
12699				    io->taskio.tag_num, io->taskio.tag_type);
12700			break;
12701		default:
12702			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12703			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12704			break;
12705		}
12706		sbuf_cat(&sb, path_str);
12707		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12708			    (intmax_t)time_uptime - io->io_hdr.start_time);
12709		sbuf_finish(&sb);
12710		printf("%s", sbuf_data(&sb));
12711	}
12712#endif /* CTL_TIME_IO */
12713
12714#ifdef CTL_IO_DELAY
12715	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12716		struct ctl_lun *lun;
12717
12718		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12719
12720		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12721	} else {
12722		struct ctl_lun *lun;
12723
12724		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12725		if ((lun != NULL)
12726		 && (lun->delay_info.datamove_delay > 0)) {
12727			struct callout *callout;
12728
12729			callout = (struct callout *)&io->io_hdr.timer_bytes;
12730			callout_init(callout, /*mpsafe*/ 1);
12731			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12732			callout_reset(callout,
12733				      lun->delay_info.datamove_delay * hz,
12734				      ctl_datamove_timer_wakeup, io);
12735			if (lun->delay_info.datamove_type ==
12736			    CTL_DELAY_TYPE_ONESHOT)
12737				lun->delay_info.datamove_delay = 0;
12738			return;
12739		}
12740	}
12741#endif
12742
12743	/*
12744	 * This command has been aborted.  Set the port status, so we fail
12745	 * the data move.
12746	 */
12747	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12748		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12749		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12750		       io->io_hdr.nexus.targ_port,
12751		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12752		       io->io_hdr.nexus.targ_lun);
12753		io->io_hdr.port_status = 31337;
12754		/*
12755		 * Note that the backend, in this case, will get the
12756		 * callback in its context.  In other cases it may get
12757		 * called in the frontend's interrupt thread context.
12758		 */
12759		io->scsiio.be_move_done(io);
12760		return;
12761	}
12762
12763	/*
12764	 * If we're in XFER mode and this I/O is from the other shelf
12765	 * controller, we need to send the DMA to the other side to
12766	 * actually transfer the data to/from the host.  In serialize only
12767	 * mode the transfer happens below CTL and ctl_datamove() is only
12768	 * called on the machine that originally received the I/O.
12769	 */
12770	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12771	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12772		union ctl_ha_msg msg;
12773		uint32_t sg_entries_sent;
12774		int do_sg_copy;
12775		int i;
12776
12777		memset(&msg, 0, sizeof(msg));
12778		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12779		msg.hdr.original_sc = io->io_hdr.original_sc;
12780		msg.hdr.serializing_sc = io;
12781		msg.hdr.nexus = io->io_hdr.nexus;
12782		msg.dt.flags = io->io_hdr.flags;
12783		/*
12784		 * We convert everything into a S/G list here.  We can't
12785		 * pass by reference, only by value between controllers.
12786		 * So we can't pass a pointer to the S/G list, only as many
12787		 * S/G entries as we can fit in here.  If it's possible for
12788		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12789		 * then we need to break this up into multiple transfers.
12790		 */
12791		if (io->scsiio.kern_sg_entries == 0) {
12792			msg.dt.kern_sg_entries = 1;
12793			/*
12794			 * If this is in cached memory, flush the cache
12795			 * before we send the DMA request to the other
12796			 * controller.  We want to do this in either the
12797			 * read or the write case.  The read case is
12798			 * straightforward.  In the write case, we want to
12799			 * make sure nothing is in the local cache that
12800			 * could overwrite the DMAed data.
12801			 */
12802			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12803				/*
12804				 * XXX KDM use bus_dmamap_sync() here.
12805				 */
12806			}
12807
12808			/*
12809			 * Convert to a physical address if this is a
12810			 * virtual address.
12811			 */
12812			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12813				msg.dt.sg_list[0].addr =
12814					io->scsiio.kern_data_ptr;
12815			} else {
12816				/*
12817				 * XXX KDM use busdma here!
12818				 */
12819#if 0
12820				msg.dt.sg_list[0].addr = (void *)
12821					vtophys(io->scsiio.kern_data_ptr);
12822#endif
12823			}
12824
12825			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12826			do_sg_copy = 0;
12827		} else {
12828			struct ctl_sg_entry *sgl;
12829
12830			do_sg_copy = 1;
12831			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12832			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12833			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12834				/*
12835				 * XXX KDM use bus_dmamap_sync() here.
12836				 */
12837			}
12838		}
12839
12840		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12841		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12842		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12843		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12844		msg.dt.sg_sequence = 0;
12845
12846		/*
12847		 * Loop until we've sent all of the S/G entries.  On the
12848		 * other end, we'll recompose these S/G entries into one
12849		 * contiguous list before passing it to the
12850		 */
12851		for (sg_entries_sent = 0; sg_entries_sent <
12852		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12853			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12854				sizeof(msg.dt.sg_list[0])),
12855				msg.dt.kern_sg_entries - sg_entries_sent);
12856
12857			if (do_sg_copy != 0) {
12858				struct ctl_sg_entry *sgl;
12859				int j;
12860
12861				sgl = (struct ctl_sg_entry *)
12862					io->scsiio.kern_data_ptr;
12863				/*
12864				 * If this is in cached memory, flush the cache
12865				 * before we send the DMA request to the other
12866				 * controller.  We want to do this in either
12867				 * the * read or the write case.  The read
12868				 * case is straightforward.  In the write
12869				 * case, we want to make sure nothing is
12870				 * in the local cache that could overwrite
12871				 * the DMAed data.
12872				 */
12873
12874				for (i = sg_entries_sent, j = 0;
12875				     i < msg.dt.cur_sg_entries; i++, j++) {
12876					if ((io->io_hdr.flags &
12877					     CTL_FLAG_NO_DATASYNC) == 0) {
12878						/*
12879						 * XXX KDM use bus_dmamap_sync()
12880						 */
12881					}
12882					if ((io->io_hdr.flags &
12883					     CTL_FLAG_BUS_ADDR) == 0) {
12884						/*
12885						 * XXX KDM use busdma.
12886						 */
12887#if 0
12888						msg.dt.sg_list[j].addr =(void *)
12889						       vtophys(sgl[i].addr);
12890#endif
12891					} else {
12892						msg.dt.sg_list[j].addr =
12893							sgl[i].addr;
12894					}
12895					msg.dt.sg_list[j].len = sgl[i].len;
12896				}
12897			}
12898
12899			sg_entries_sent += msg.dt.cur_sg_entries;
12900			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12901				msg.dt.sg_last = 1;
12902			else
12903				msg.dt.sg_last = 0;
12904
12905			/*
12906			 * XXX KDM drop and reacquire the lock here?
12907			 */
12908			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12909			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12910				/*
12911				 * XXX do something here.
12912				 */
12913			}
12914
12915			msg.dt.sent_sg_entries = sg_entries_sent;
12916		}
12917		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12918		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12919			ctl_failover_io(io, /*have_lock*/ 0);
12920
12921	} else {
12922
12923		/*
12924		 * Lookup the fe_datamove() function for this particular
12925		 * front end.
12926		 */
12927		fe_datamove =
12928		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12929
12930		fe_datamove(io);
12931	}
12932}
12933
12934static void
12935ctl_send_datamove_done(union ctl_io *io, int have_lock)
12936{
12937	union ctl_ha_msg msg;
12938	int isc_status;
12939
12940	memset(&msg, 0, sizeof(msg));
12941
12942	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12943	msg.hdr.original_sc = io;
12944	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12945	msg.hdr.nexus = io->io_hdr.nexus;
12946	msg.hdr.status = io->io_hdr.status;
12947	msg.scsi.tag_num = io->scsiio.tag_num;
12948	msg.scsi.tag_type = io->scsiio.tag_type;
12949	msg.scsi.scsi_status = io->scsiio.scsi_status;
12950	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12951	       sizeof(io->scsiio.sense_data));
12952	msg.scsi.sense_len = io->scsiio.sense_len;
12953	msg.scsi.sense_residual = io->scsiio.sense_residual;
12954	msg.scsi.fetd_status = io->io_hdr.port_status;
12955	msg.scsi.residual = io->scsiio.residual;
12956	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12957
12958	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12959		ctl_failover_io(io, /*have_lock*/ have_lock);
12960		return;
12961	}
12962
12963	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12964	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12965		/* XXX do something if this fails */
12966	}
12967
12968}
12969
12970/*
12971 * The DMA to the remote side is done, now we need to tell the other side
12972 * we're done so it can continue with its data movement.
12973 */
12974static void
12975ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12976{
12977	union ctl_io *io;
12978
12979	io = rq->context;
12980
12981	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12982		printf("%s: ISC DMA write failed with error %d", __func__,
12983		       rq->ret);
12984		ctl_set_internal_failure(&io->scsiio,
12985					 /*sks_valid*/ 1,
12986					 /*retry_count*/ rq->ret);
12987	}
12988
12989	ctl_dt_req_free(rq);
12990
12991	/*
12992	 * In this case, we had to malloc the memory locally.  Free it.
12993	 */
12994	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12995		int i;
12996		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12997			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12998	}
12999	/*
13000	 * The data is in local and remote memory, so now we need to send
13001	 * status (good or back) back to the other side.
13002	 */
13003	ctl_send_datamove_done(io, /*have_lock*/ 0);
13004}
13005
13006/*
13007 * We've moved the data from the host/controller into local memory.  Now we
13008 * need to push it over to the remote controller's memory.
13009 */
13010static int
13011ctl_datamove_remote_dm_write_cb(union ctl_io *io)
13012{
13013	int retval;
13014
13015	retval = 0;
13016
13017	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
13018					  ctl_datamove_remote_write_cb);
13019
13020	return (retval);
13021}
13022
13023static void
13024ctl_datamove_remote_write(union ctl_io *io)
13025{
13026	int retval;
13027	void (*fe_datamove)(union ctl_io *io);
13028
13029	/*
13030	 * - Get the data from the host/HBA into local memory.
13031	 * - DMA memory from the local controller to the remote controller.
13032	 * - Send status back to the remote controller.
13033	 */
13034
13035	retval = ctl_datamove_remote_sgl_setup(io);
13036	if (retval != 0)
13037		return;
13038
13039	/* Switch the pointer over so the FETD knows what to do */
13040	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13041
13042	/*
13043	 * Use a custom move done callback, since we need to send completion
13044	 * back to the other controller, not to the backend on this side.
13045	 */
13046	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
13047
13048	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13049
13050	fe_datamove(io);
13051
13052	return;
13053
13054}
13055
13056static int
13057ctl_datamove_remote_dm_read_cb(union ctl_io *io)
13058{
13059#if 0
13060	char str[256];
13061	char path_str[64];
13062	struct sbuf sb;
13063#endif
13064
13065	/*
13066	 * In this case, we had to malloc the memory locally.  Free it.
13067	 */
13068	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13069		int i;
13070		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13071			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13072	}
13073
13074#if 0
13075	scsi_path_string(io, path_str, sizeof(path_str));
13076	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13077	sbuf_cat(&sb, path_str);
13078	scsi_command_string(&io->scsiio, NULL, &sb);
13079	sbuf_printf(&sb, "\n");
13080	sbuf_cat(&sb, path_str);
13081	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13082		    io->scsiio.tag_num, io->scsiio.tag_type);
13083	sbuf_cat(&sb, path_str);
13084	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
13085		    io->io_hdr.flags, io->io_hdr.status);
13086	sbuf_finish(&sb);
13087	printk("%s", sbuf_data(&sb));
13088#endif
13089
13090
13091	/*
13092	 * The read is done, now we need to send status (good or bad) back
13093	 * to the other side.
13094	 */
13095	ctl_send_datamove_done(io, /*have_lock*/ 0);
13096
13097	return (0);
13098}
13099
13100static void
13101ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
13102{
13103	union ctl_io *io;
13104	void (*fe_datamove)(union ctl_io *io);
13105
13106	io = rq->context;
13107
13108	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13109		printf("%s: ISC DMA read failed with error %d", __func__,
13110		       rq->ret);
13111		ctl_set_internal_failure(&io->scsiio,
13112					 /*sks_valid*/ 1,
13113					 /*retry_count*/ rq->ret);
13114	}
13115
13116	ctl_dt_req_free(rq);
13117
13118	/* Switch the pointer over so the FETD knows what to do */
13119	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13120
13121	/*
13122	 * Use a custom move done callback, since we need to send completion
13123	 * back to the other controller, not to the backend on this side.
13124	 */
13125	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13126
13127	/* XXX KDM add checks like the ones in ctl_datamove? */
13128
13129	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13130
13131	fe_datamove(io);
13132}
13133
13134static int
13135ctl_datamove_remote_sgl_setup(union ctl_io *io)
13136{
13137	struct ctl_sg_entry *local_sglist, *remote_sglist;
13138	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13139	struct ctl_softc *softc;
13140	int retval;
13141	int i;
13142
13143	retval = 0;
13144	softc = control_softc;
13145
13146	local_sglist = io->io_hdr.local_sglist;
13147	local_dma_sglist = io->io_hdr.local_dma_sglist;
13148	remote_sglist = io->io_hdr.remote_sglist;
13149	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13150
13151	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13152		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13153			local_sglist[i].len = remote_sglist[i].len;
13154
13155			/*
13156			 * XXX Detect the situation where the RS-level I/O
13157			 * redirector on the other side has already read the
13158			 * data off of the AOR RS on this side, and
13159			 * transferred it to remote (mirror) memory on the
13160			 * other side.  Since we already have the data in
13161			 * memory here, we just need to use it.
13162			 *
13163			 * XXX KDM this can probably be removed once we
13164			 * get the cache device code in and take the
13165			 * current AOR implementation out.
13166			 */
13167#ifdef NEEDTOPORT
13168			if ((remote_sglist[i].addr >=
13169			     (void *)vtophys(softc->mirr->addr))
13170			 && (remote_sglist[i].addr <
13171			     ((void *)vtophys(softc->mirr->addr) +
13172			     CacheMirrorOffset))) {
13173				local_sglist[i].addr = remote_sglist[i].addr -
13174					CacheMirrorOffset;
13175				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13176				     CTL_FLAG_DATA_IN)
13177					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13178			} else {
13179				local_sglist[i].addr = remote_sglist[i].addr +
13180					CacheMirrorOffset;
13181			}
13182#endif
13183#if 0
13184			printf("%s: local %p, remote %p, len %d\n",
13185			       __func__, local_sglist[i].addr,
13186			       remote_sglist[i].addr, local_sglist[i].len);
13187#endif
13188		}
13189	} else {
13190		uint32_t len_to_go;
13191
13192		/*
13193		 * In this case, we don't have automatically allocated
13194		 * memory for this I/O on this controller.  This typically
13195		 * happens with internal CTL I/O -- e.g. inquiry, mode
13196		 * sense, etc.  Anything coming from RAIDCore will have
13197		 * a mirror area available.
13198		 */
13199		len_to_go = io->scsiio.kern_data_len;
13200
13201		/*
13202		 * Clear the no datasync flag, we have to use malloced
13203		 * buffers.
13204		 */
13205		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13206
13207		/*
13208		 * The difficult thing here is that the size of the various
13209		 * S/G segments may be different than the size from the
13210		 * remote controller.  That'll make it harder when DMAing
13211		 * the data back to the other side.
13212		 */
13213		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13214		     sizeof(io->io_hdr.remote_sglist[0])) &&
13215		     (len_to_go > 0); i++) {
13216			local_sglist[i].len = ctl_min(len_to_go, 131072);
13217			CTL_SIZE_8B(local_dma_sglist[i].len,
13218				    local_sglist[i].len);
13219			local_sglist[i].addr =
13220				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13221
13222			local_dma_sglist[i].addr = local_sglist[i].addr;
13223
13224			if (local_sglist[i].addr == NULL) {
13225				int j;
13226
13227				printf("malloc failed for %zd bytes!",
13228				       local_dma_sglist[i].len);
13229				for (j = 0; j < i; j++) {
13230					free(local_sglist[j].addr, M_CTL);
13231				}
13232				ctl_set_internal_failure(&io->scsiio,
13233							 /*sks_valid*/ 1,
13234							 /*retry_count*/ 4857);
13235				retval = 1;
13236				goto bailout_error;
13237
13238			}
13239			/* XXX KDM do we need a sync here? */
13240
13241			len_to_go -= local_sglist[i].len;
13242		}
13243		/*
13244		 * Reset the number of S/G entries accordingly.  The
13245		 * original number of S/G entries is available in
13246		 * rem_sg_entries.
13247		 */
13248		io->scsiio.kern_sg_entries = i;
13249
13250#if 0
13251		printf("%s: kern_sg_entries = %d\n", __func__,
13252		       io->scsiio.kern_sg_entries);
13253		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13254			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13255			       local_sglist[i].addr, local_sglist[i].len,
13256			       local_dma_sglist[i].len);
13257#endif
13258	}
13259
13260
13261	return (retval);
13262
13263bailout_error:
13264
13265	ctl_send_datamove_done(io, /*have_lock*/ 0);
13266
13267	return (retval);
13268}
13269
13270static int
13271ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13272			 ctl_ha_dt_cb callback)
13273{
13274	struct ctl_ha_dt_req *rq;
13275	struct ctl_sg_entry *remote_sglist, *local_sglist;
13276	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13277	uint32_t local_used, remote_used, total_used;
13278	int retval;
13279	int i, j;
13280
13281	retval = 0;
13282
13283	rq = ctl_dt_req_alloc();
13284
13285	/*
13286	 * If we failed to allocate the request, and if the DMA didn't fail
13287	 * anyway, set busy status.  This is just a resource allocation
13288	 * failure.
13289	 */
13290	if ((rq == NULL)
13291	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13292		ctl_set_busy(&io->scsiio);
13293
13294	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13295
13296		if (rq != NULL)
13297			ctl_dt_req_free(rq);
13298
13299		/*
13300		 * The data move failed.  We need to return status back
13301		 * to the other controller.  No point in trying to DMA
13302		 * data to the remote controller.
13303		 */
13304
13305		ctl_send_datamove_done(io, /*have_lock*/ 0);
13306
13307		retval = 1;
13308
13309		goto bailout;
13310	}
13311
13312	local_sglist = io->io_hdr.local_sglist;
13313	local_dma_sglist = io->io_hdr.local_dma_sglist;
13314	remote_sglist = io->io_hdr.remote_sglist;
13315	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13316	local_used = 0;
13317	remote_used = 0;
13318	total_used = 0;
13319
13320	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13321		rq->ret = CTL_HA_STATUS_SUCCESS;
13322		rq->context = io;
13323		callback(rq);
13324		goto bailout;
13325	}
13326
13327	/*
13328	 * Pull/push the data over the wire from/to the other controller.
13329	 * This takes into account the possibility that the local and
13330	 * remote sglists may not be identical in terms of the size of
13331	 * the elements and the number of elements.
13332	 *
13333	 * One fundamental assumption here is that the length allocated for
13334	 * both the local and remote sglists is identical.  Otherwise, we've
13335	 * essentially got a coding error of some sort.
13336	 */
13337	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13338		int isc_ret;
13339		uint32_t cur_len, dma_length;
13340		uint8_t *tmp_ptr;
13341
13342		rq->id = CTL_HA_DATA_CTL;
13343		rq->command = command;
13344		rq->context = io;
13345
13346		/*
13347		 * Both pointers should be aligned.  But it is possible
13348		 * that the allocation length is not.  They should both
13349		 * also have enough slack left over at the end, though,
13350		 * to round up to the next 8 byte boundary.
13351		 */
13352		cur_len = ctl_min(local_sglist[i].len - local_used,
13353				  remote_sglist[j].len - remote_used);
13354
13355		/*
13356		 * In this case, we have a size issue and need to decrease
13357		 * the size, except in the case where we actually have less
13358		 * than 8 bytes left.  In that case, we need to increase
13359		 * the DMA length to get the last bit.
13360		 */
13361		if ((cur_len & 0x7) != 0) {
13362			if (cur_len > 0x7) {
13363				cur_len = cur_len - (cur_len & 0x7);
13364				dma_length = cur_len;
13365			} else {
13366				CTL_SIZE_8B(dma_length, cur_len);
13367			}
13368
13369		} else
13370			dma_length = cur_len;
13371
13372		/*
13373		 * If we had to allocate memory for this I/O, instead of using
13374		 * the non-cached mirror memory, we'll need to flush the cache
13375		 * before trying to DMA to the other controller.
13376		 *
13377		 * We could end up doing this multiple times for the same
13378		 * segment if we have a larger local segment than remote
13379		 * segment.  That shouldn't be an issue.
13380		 */
13381		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13382			/*
13383			 * XXX KDM use bus_dmamap_sync() here.
13384			 */
13385		}
13386
13387		rq->size = dma_length;
13388
13389		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13390		tmp_ptr += local_used;
13391
13392		/* Use physical addresses when talking to ISC hardware */
13393		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13394			/* XXX KDM use busdma */
13395#if 0
13396			rq->local = vtophys(tmp_ptr);
13397#endif
13398		} else
13399			rq->local = tmp_ptr;
13400
13401		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13402		tmp_ptr += remote_used;
13403		rq->remote = tmp_ptr;
13404
13405		rq->callback = NULL;
13406
13407		local_used += cur_len;
13408		if (local_used >= local_sglist[i].len) {
13409			i++;
13410			local_used = 0;
13411		}
13412
13413		remote_used += cur_len;
13414		if (remote_used >= remote_sglist[j].len) {
13415			j++;
13416			remote_used = 0;
13417		}
13418		total_used += cur_len;
13419
13420		if (total_used >= io->scsiio.kern_data_len)
13421			rq->callback = callback;
13422
13423		if ((rq->size & 0x7) != 0) {
13424			printf("%s: warning: size %d is not on 8b boundary\n",
13425			       __func__, rq->size);
13426		}
13427		if (((uintptr_t)rq->local & 0x7) != 0) {
13428			printf("%s: warning: local %p not on 8b boundary\n",
13429			       __func__, rq->local);
13430		}
13431		if (((uintptr_t)rq->remote & 0x7) != 0) {
13432			printf("%s: warning: remote %p not on 8b boundary\n",
13433			       __func__, rq->local);
13434		}
13435#if 0
13436		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13437		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13438		       rq->local, rq->remote, rq->size);
13439#endif
13440
13441		isc_ret = ctl_dt_single(rq);
13442		if (isc_ret == CTL_HA_STATUS_WAIT)
13443			continue;
13444
13445		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13446			rq->ret = CTL_HA_STATUS_SUCCESS;
13447		} else {
13448			rq->ret = isc_ret;
13449		}
13450		callback(rq);
13451		goto bailout;
13452	}
13453
13454bailout:
13455	return (retval);
13456
13457}
13458
13459static void
13460ctl_datamove_remote_read(union ctl_io *io)
13461{
13462	int retval;
13463	int i;
13464
13465	/*
13466	 * This will send an error to the other controller in the case of a
13467	 * failure.
13468	 */
13469	retval = ctl_datamove_remote_sgl_setup(io);
13470	if (retval != 0)
13471		return;
13472
13473	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13474					  ctl_datamove_remote_read_cb);
13475	if ((retval != 0)
13476	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13477		/*
13478		 * Make sure we free memory if there was an error..  The
13479		 * ctl_datamove_remote_xfer() function will send the
13480		 * datamove done message, or call the callback with an
13481		 * error if there is a problem.
13482		 */
13483		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13484			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13485	}
13486
13487	return;
13488}
13489
13490/*
13491 * Process a datamove request from the other controller.  This is used for
13492 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13493 * first.  Once that is complete, the data gets DMAed into the remote
13494 * controller's memory.  For reads, we DMA from the remote controller's
13495 * memory into our memory first, and then move it out to the FETD.
13496 */
13497static void
13498ctl_datamove_remote(union ctl_io *io)
13499{
13500	struct ctl_softc *softc;
13501
13502	softc = control_softc;
13503
13504	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13505
13506	/*
13507	 * Note that we look for an aborted I/O here, but don't do some of
13508	 * the other checks that ctl_datamove() normally does.
13509	 * We don't need to run the datamove delay code, since that should
13510	 * have been done if need be on the other controller.
13511	 */
13512	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13513		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13514		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13515		       io->io_hdr.nexus.targ_port,
13516		       io->io_hdr.nexus.targ_target.id,
13517		       io->io_hdr.nexus.targ_lun);
13518		io->io_hdr.port_status = 31338;
13519		ctl_send_datamove_done(io, /*have_lock*/ 0);
13520		return;
13521	}
13522
13523	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13524		ctl_datamove_remote_write(io);
13525	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13526		ctl_datamove_remote_read(io);
13527	} else {
13528		union ctl_ha_msg msg;
13529		struct scsi_sense_data *sense;
13530		uint8_t sks[3];
13531		int retry_count;
13532
13533		memset(&msg, 0, sizeof(msg));
13534
13535		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13536		msg.hdr.status = CTL_SCSI_ERROR;
13537		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13538
13539		retry_count = 4243;
13540
13541		sense = &msg.scsi.sense_data;
13542		sks[0] = SSD_SCS_VALID;
13543		sks[1] = (retry_count >> 8) & 0xff;
13544		sks[2] = retry_count & 0xff;
13545
13546		/* "Internal target failure" */
13547		scsi_set_sense_data(sense,
13548				    /*sense_format*/ SSD_TYPE_NONE,
13549				    /*current_error*/ 1,
13550				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13551				    /*asc*/ 0x44,
13552				    /*ascq*/ 0x00,
13553				    /*type*/ SSD_ELEM_SKS,
13554				    /*size*/ sizeof(sks),
13555				    /*data*/ sks,
13556				    SSD_ELEM_NONE);
13557
13558		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13559		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13560			ctl_failover_io(io, /*have_lock*/ 1);
13561			return;
13562		}
13563
13564		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13565		    CTL_HA_STATUS_SUCCESS) {
13566			/* XXX KDM what to do if this fails? */
13567		}
13568		return;
13569	}
13570
13571}
13572
13573static int
13574ctl_process_done(union ctl_io *io)
13575{
13576	struct ctl_lun *lun;
13577	struct ctl_softc *ctl_softc;
13578	void (*fe_done)(union ctl_io *io);
13579	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13580
13581	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13582
13583	fe_done =
13584	    control_softc->ctl_ports[targ_port]->fe_done;
13585
13586#ifdef CTL_TIME_IO
13587	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13588		char str[256];
13589		char path_str[64];
13590		struct sbuf sb;
13591
13592		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13593		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13594
13595		sbuf_cat(&sb, path_str);
13596		switch (io->io_hdr.io_type) {
13597		case CTL_IO_SCSI:
13598			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13599			sbuf_printf(&sb, "\n");
13600			sbuf_cat(&sb, path_str);
13601			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13602				    io->scsiio.tag_num, io->scsiio.tag_type);
13603			break;
13604		case CTL_IO_TASK:
13605			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13606				    "Tag Type: %d\n", io->taskio.task_action,
13607				    io->taskio.tag_num, io->taskio.tag_type);
13608			break;
13609		default:
13610			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13611			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13612			break;
13613		}
13614		sbuf_cat(&sb, path_str);
13615		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13616			    (intmax_t)time_uptime - io->io_hdr.start_time);
13617		sbuf_finish(&sb);
13618		printf("%s", sbuf_data(&sb));
13619	}
13620#endif /* CTL_TIME_IO */
13621
13622	switch (io->io_hdr.io_type) {
13623	case CTL_IO_SCSI:
13624		break;
13625	case CTL_IO_TASK:
13626		if (bootverbose || verbose > 0)
13627			ctl_io_error_print(io, NULL);
13628		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13629			ctl_free_io(io);
13630		else
13631			fe_done(io);
13632		return (CTL_RETVAL_COMPLETE);
13633		break;
13634	default:
13635		printf("ctl_process_done: invalid io type %d\n",
13636		       io->io_hdr.io_type);
13637		panic("ctl_process_done: invalid io type %d\n",
13638		      io->io_hdr.io_type);
13639		break; /* NOTREACHED */
13640	}
13641
13642	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13643	if (lun == NULL) {
13644		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13645				 io->io_hdr.nexus.targ_mapped_lun));
13646		fe_done(io);
13647		goto bailout;
13648	}
13649	ctl_softc = lun->ctl_softc;
13650
13651	mtx_lock(&lun->lun_lock);
13652
13653	/*
13654	 * Check to see if we have any errors to inject here.  We only
13655	 * inject errors for commands that don't already have errors set.
13656	 */
13657	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13658	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13659		ctl_inject_error(lun, io);
13660
13661	/*
13662	 * XXX KDM how do we treat commands that aren't completed
13663	 * successfully?
13664	 *
13665	 * XXX KDM should we also track I/O latency?
13666	 */
13667	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13668	    io->io_hdr.io_type == CTL_IO_SCSI) {
13669#ifdef CTL_TIME_IO
13670		struct bintime cur_bt;
13671#endif
13672		int type;
13673
13674		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13675		    CTL_FLAG_DATA_IN)
13676			type = CTL_STATS_READ;
13677		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13678		    CTL_FLAG_DATA_OUT)
13679			type = CTL_STATS_WRITE;
13680		else
13681			type = CTL_STATS_NO_IO;
13682
13683		lun->stats.ports[targ_port].bytes[type] +=
13684		    io->scsiio.kern_total_len;
13685		lun->stats.ports[targ_port].operations[type]++;
13686#ifdef CTL_TIME_IO
13687		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13688		   &io->io_hdr.dma_bt);
13689		lun->stats.ports[targ_port].num_dmas[type] +=
13690		    io->io_hdr.num_dmas;
13691		getbintime(&cur_bt);
13692		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13693		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13694#endif
13695	}
13696
13697	/*
13698	 * Remove this from the OOA queue.
13699	 */
13700	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13701
13702	/*
13703	 * Run through the blocked queue on this LUN and see if anything
13704	 * has become unblocked, now that this transaction is done.
13705	 */
13706	ctl_check_blocked(lun);
13707
13708	/*
13709	 * If the LUN has been invalidated, free it if there is nothing
13710	 * left on its OOA queue.
13711	 */
13712	if ((lun->flags & CTL_LUN_INVALID)
13713	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13714		mtx_unlock(&lun->lun_lock);
13715		mtx_lock(&ctl_softc->ctl_lock);
13716		ctl_free_lun(lun);
13717		mtx_unlock(&ctl_softc->ctl_lock);
13718	} else
13719		mtx_unlock(&lun->lun_lock);
13720
13721	/*
13722	 * If this command has been aborted, make sure we set the status
13723	 * properly.  The FETD is responsible for freeing the I/O and doing
13724	 * whatever it needs to do to clean up its state.
13725	 */
13726	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13727		ctl_set_task_aborted(&io->scsiio);
13728
13729	/*
13730	 * We print out status for every task management command.  For SCSI
13731	 * commands, we filter out any unit attention errors; they happen
13732	 * on every boot, and would clutter up the log.  Note:  task
13733	 * management commands aren't printed here, they are printed above,
13734	 * since they should never even make it down here.
13735	 */
13736	switch (io->io_hdr.io_type) {
13737	case CTL_IO_SCSI: {
13738		int error_code, sense_key, asc, ascq;
13739
13740		sense_key = 0;
13741
13742		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13743		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13744			/*
13745			 * Since this is just for printing, no need to
13746			 * show errors here.
13747			 */
13748			scsi_extract_sense_len(&io->scsiio.sense_data,
13749					       io->scsiio.sense_len,
13750					       &error_code,
13751					       &sense_key,
13752					       &asc,
13753					       &ascq,
13754					       /*show_errors*/ 0);
13755		}
13756
13757		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13758		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13759		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13760		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13761
13762			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13763				ctl_softc->skipped_prints++;
13764			} else {
13765				uint32_t skipped_prints;
13766
13767				skipped_prints = ctl_softc->skipped_prints;
13768
13769				ctl_softc->skipped_prints = 0;
13770				ctl_softc->last_print_jiffies = time_uptime;
13771
13772				if (skipped_prints > 0) {
13773#ifdef NEEDTOPORT
13774					csevent_log(CSC_CTL | CSC_SHELF_SW |
13775					    CTL_ERROR_REPORT,
13776					    csevent_LogType_Trace,
13777					    csevent_Severity_Information,
13778					    csevent_AlertLevel_Green,
13779					    csevent_FRU_Firmware,
13780					    csevent_FRU_Unknown,
13781					    "High CTL error volume, %d prints "
13782					    "skipped", skipped_prints);
13783#endif
13784				}
13785				if (bootverbose || verbose > 0)
13786					ctl_io_error_print(io, NULL);
13787			}
13788		}
13789		break;
13790	}
13791	case CTL_IO_TASK:
13792		if (bootverbose || verbose > 0)
13793			ctl_io_error_print(io, NULL);
13794		break;
13795	default:
13796		break;
13797	}
13798
13799	/*
13800	 * Tell the FETD or the other shelf controller we're done with this
13801	 * command.  Note that only SCSI commands get to this point.  Task
13802	 * management commands are completed above.
13803	 *
13804	 * We only send status to the other controller if we're in XFER
13805	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13806	 * received the I/O (from CTL's perspective), and so the status is
13807	 * generated there.
13808	 *
13809	 * XXX KDM if we hold the lock here, we could cause a deadlock
13810	 * if the frontend comes back in in this context to queue
13811	 * something.
13812	 */
13813	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13814	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13815		union ctl_ha_msg msg;
13816
13817		memset(&msg, 0, sizeof(msg));
13818		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13819		msg.hdr.original_sc = io->io_hdr.original_sc;
13820		msg.hdr.nexus = io->io_hdr.nexus;
13821		msg.hdr.status = io->io_hdr.status;
13822		msg.scsi.scsi_status = io->scsiio.scsi_status;
13823		msg.scsi.tag_num = io->scsiio.tag_num;
13824		msg.scsi.tag_type = io->scsiio.tag_type;
13825		msg.scsi.sense_len = io->scsiio.sense_len;
13826		msg.scsi.sense_residual = io->scsiio.sense_residual;
13827		msg.scsi.residual = io->scsiio.residual;
13828		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13829		       sizeof(io->scsiio.sense_data));
13830		/*
13831		 * We copy this whether or not this is an I/O-related
13832		 * command.  Otherwise, we'd have to go and check to see
13833		 * whether it's a read/write command, and it really isn't
13834		 * worth it.
13835		 */
13836		memcpy(&msg.scsi.lbalen,
13837		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13838		       sizeof(msg.scsi.lbalen));
13839
13840		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13841				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13842			/* XXX do something here */
13843		}
13844
13845		ctl_free_io(io);
13846	} else
13847		fe_done(io);
13848
13849bailout:
13850
13851	return (CTL_RETVAL_COMPLETE);
13852}
13853
13854#ifdef CTL_WITH_CA
13855/*
13856 * Front end should call this if it doesn't do autosense.  When the request
13857 * sense comes back in from the initiator, we'll dequeue this and send it.
13858 */
13859int
13860ctl_queue_sense(union ctl_io *io)
13861{
13862	struct ctl_lun *lun;
13863	struct ctl_softc *ctl_softc;
13864	uint32_t initidx, targ_lun;
13865
13866	ctl_softc = control_softc;
13867
13868	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13869
13870	/*
13871	 * LUN lookup will likely move to the ctl_work_thread() once we
13872	 * have our new queueing infrastructure (that doesn't put things on
13873	 * a per-LUN queue initially).  That is so that we can handle
13874	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13875	 * can't deal with that right now.
13876	 */
13877	mtx_lock(&ctl_softc->ctl_lock);
13878
13879	/*
13880	 * If we don't have a LUN for this, just toss the sense
13881	 * information.
13882	 */
13883	targ_lun = io->io_hdr.nexus.targ_lun;
13884	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13885	if ((targ_lun < CTL_MAX_LUNS)
13886	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13887		lun = ctl_softc->ctl_luns[targ_lun];
13888	else
13889		goto bailout;
13890
13891	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13892
13893	mtx_lock(&lun->lun_lock);
13894	/*
13895	 * Already have CA set for this LUN...toss the sense information.
13896	 */
13897	if (ctl_is_set(lun->have_ca, initidx)) {
13898		mtx_unlock(&lun->lun_lock);
13899		goto bailout;
13900	}
13901
13902	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13903	       ctl_min(sizeof(lun->pending_sense[initidx]),
13904	       sizeof(io->scsiio.sense_data)));
13905	ctl_set_mask(lun->have_ca, initidx);
13906	mtx_unlock(&lun->lun_lock);
13907
13908bailout:
13909	mtx_unlock(&ctl_softc->ctl_lock);
13910
13911	ctl_free_io(io);
13912
13913	return (CTL_RETVAL_COMPLETE);
13914}
13915#endif
13916
13917/*
13918 * Primary command inlet from frontend ports.  All SCSI and task I/O
13919 * requests must go through this function.
13920 */
13921int
13922ctl_queue(union ctl_io *io)
13923{
13924	struct ctl_softc *ctl_softc;
13925
13926	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13927
13928	ctl_softc = control_softc;
13929
13930#ifdef CTL_TIME_IO
13931	io->io_hdr.start_time = time_uptime;
13932	getbintime(&io->io_hdr.start_bt);
13933#endif /* CTL_TIME_IO */
13934
13935	/* Map FE-specific LUN ID into global one. */
13936	io->io_hdr.nexus.targ_mapped_lun =
13937	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13938
13939	switch (io->io_hdr.io_type) {
13940	case CTL_IO_SCSI:
13941	case CTL_IO_TASK:
13942		ctl_enqueue_incoming(io);
13943		break;
13944	default:
13945		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13946		return (EINVAL);
13947	}
13948
13949	return (CTL_RETVAL_COMPLETE);
13950}
13951
13952#ifdef CTL_IO_DELAY
13953static void
13954ctl_done_timer_wakeup(void *arg)
13955{
13956	union ctl_io *io;
13957
13958	io = (union ctl_io *)arg;
13959	ctl_done(io);
13960}
13961#endif /* CTL_IO_DELAY */
13962
13963void
13964ctl_done(union ctl_io *io)
13965{
13966	struct ctl_softc *ctl_softc;
13967
13968	ctl_softc = control_softc;
13969
13970	/*
13971	 * Enable this to catch duplicate completion issues.
13972	 */
13973#if 0
13974	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13975		printf("%s: type %d msg %d cdb %x iptl: "
13976		       "%d:%d:%d:%d tag 0x%04x "
13977		       "flag %#x status %x\n",
13978			__func__,
13979			io->io_hdr.io_type,
13980			io->io_hdr.msg_type,
13981			io->scsiio.cdb[0],
13982			io->io_hdr.nexus.initid.id,
13983			io->io_hdr.nexus.targ_port,
13984			io->io_hdr.nexus.targ_target.id,
13985			io->io_hdr.nexus.targ_lun,
13986			(io->io_hdr.io_type ==
13987			CTL_IO_TASK) ?
13988			io->taskio.tag_num :
13989			io->scsiio.tag_num,
13990		        io->io_hdr.flags,
13991			io->io_hdr.status);
13992	} else
13993		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13994#endif
13995
13996	/*
13997	 * This is an internal copy of an I/O, and should not go through
13998	 * the normal done processing logic.
13999	 */
14000	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
14001		return;
14002
14003	/*
14004	 * We need to send a msg to the serializing shelf to finish the IO
14005	 * as well.  We don't send a finish message to the other shelf if
14006	 * this is a task management command.  Task management commands
14007	 * aren't serialized in the OOA queue, but rather just executed on
14008	 * both shelf controllers for commands that originated on that
14009	 * controller.
14010	 */
14011	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
14012	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
14013		union ctl_ha_msg msg_io;
14014
14015		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
14016		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
14017		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
14018		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
14019		}
14020		/* continue on to finish IO */
14021	}
14022#ifdef CTL_IO_DELAY
14023	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
14024		struct ctl_lun *lun;
14025
14026		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14027
14028		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
14029	} else {
14030		struct ctl_lun *lun;
14031
14032		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14033
14034		if ((lun != NULL)
14035		 && (lun->delay_info.done_delay > 0)) {
14036			struct callout *callout;
14037
14038			callout = (struct callout *)&io->io_hdr.timer_bytes;
14039			callout_init(callout, /*mpsafe*/ 1);
14040			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
14041			callout_reset(callout,
14042				      lun->delay_info.done_delay * hz,
14043				      ctl_done_timer_wakeup, io);
14044			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
14045				lun->delay_info.done_delay = 0;
14046			return;
14047		}
14048	}
14049#endif /* CTL_IO_DELAY */
14050
14051	ctl_enqueue_done(io);
14052}
14053
14054int
14055ctl_isc(struct ctl_scsiio *ctsio)
14056{
14057	struct ctl_lun *lun;
14058	int retval;
14059
14060	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14061
14062	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
14063
14064	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
14065
14066	retval = lun->backend->data_submit((union ctl_io *)ctsio);
14067
14068	return (retval);
14069}
14070
14071
14072static void
14073ctl_work_thread(void *arg)
14074{
14075	struct ctl_thread *thr = (struct ctl_thread *)arg;
14076	struct ctl_softc *softc = thr->ctl_softc;
14077	union ctl_io *io;
14078	int retval;
14079
14080	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
14081
14082	for (;;) {
14083		retval = 0;
14084
14085		/*
14086		 * We handle the queues in this order:
14087		 * - ISC
14088		 * - done queue (to free up resources, unblock other commands)
14089		 * - RtR queue
14090		 * - incoming queue
14091		 *
14092		 * If those queues are empty, we break out of the loop and
14093		 * go to sleep.
14094		 */
14095		mtx_lock(&thr->queue_lock);
14096		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
14097		if (io != NULL) {
14098			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
14099			mtx_unlock(&thr->queue_lock);
14100			ctl_handle_isc(io);
14101			continue;
14102		}
14103		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
14104		if (io != NULL) {
14105			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
14106			/* clear any blocked commands, call fe_done */
14107			mtx_unlock(&thr->queue_lock);
14108			retval = ctl_process_done(io);
14109			continue;
14110		}
14111		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
14112		if (io != NULL) {
14113			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
14114			mtx_unlock(&thr->queue_lock);
14115			if (io->io_hdr.io_type == CTL_IO_TASK)
14116				ctl_run_task(io);
14117			else
14118				ctl_scsiio_precheck(softc, &io->scsiio);
14119			continue;
14120		}
14121		if (!ctl_pause_rtr) {
14122			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
14123			if (io != NULL) {
14124				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
14125				mtx_unlock(&thr->queue_lock);
14126				retval = ctl_scsiio(&io->scsiio);
14127				if (retval != CTL_RETVAL_COMPLETE)
14128					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
14129				continue;
14130			}
14131		}
14132
14133		/* Sleep until we have something to do. */
14134		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
14135	}
14136}
14137
14138static void
14139ctl_lun_thread(void *arg)
14140{
14141	struct ctl_softc *softc = (struct ctl_softc *)arg;
14142	struct ctl_be_lun *be_lun;
14143	int retval;
14144
14145	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14146
14147	for (;;) {
14148		retval = 0;
14149		mtx_lock(&softc->ctl_lock);
14150		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14151		if (be_lun != NULL) {
14152			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14153			mtx_unlock(&softc->ctl_lock);
14154			ctl_create_lun(be_lun);
14155			continue;
14156		}
14157
14158		/* Sleep until we have something to do. */
14159		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14160		    PDROP | PRIBIO, "-", 0);
14161	}
14162}
14163
14164static void
14165ctl_enqueue_incoming(union ctl_io *io)
14166{
14167	struct ctl_softc *softc = control_softc;
14168	struct ctl_thread *thr;
14169	u_int idx;
14170
14171	idx = (io->io_hdr.nexus.targ_port * 127 +
14172	       io->io_hdr.nexus.initid.id) % worker_threads;
14173	thr = &softc->threads[idx];
14174	mtx_lock(&thr->queue_lock);
14175	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14176	mtx_unlock(&thr->queue_lock);
14177	wakeup(thr);
14178}
14179
14180static void
14181ctl_enqueue_rtr(union ctl_io *io)
14182{
14183	struct ctl_softc *softc = control_softc;
14184	struct ctl_thread *thr;
14185
14186	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14187	mtx_lock(&thr->queue_lock);
14188	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14189	mtx_unlock(&thr->queue_lock);
14190	wakeup(thr);
14191}
14192
14193static void
14194ctl_enqueue_done(union ctl_io *io)
14195{
14196	struct ctl_softc *softc = control_softc;
14197	struct ctl_thread *thr;
14198
14199	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14200	mtx_lock(&thr->queue_lock);
14201	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14202	mtx_unlock(&thr->queue_lock);
14203	wakeup(thr);
14204}
14205
14206static void
14207ctl_enqueue_isc(union ctl_io *io)
14208{
14209	struct ctl_softc *softc = control_softc;
14210	struct ctl_thread *thr;
14211
14212	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14213	mtx_lock(&thr->queue_lock);
14214	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14215	mtx_unlock(&thr->queue_lock);
14216	wakeup(thr);
14217}
14218
14219/* Initialization and failover */
14220
14221void
14222ctl_init_isc_msg(void)
14223{
14224	printf("CTL: Still calling this thing\n");
14225}
14226
14227/*
14228 * Init component
14229 * 	Initializes component into configuration defined by bootMode
14230 *	(see hasc-sv.c)
14231 *  	returns hasc_Status:
14232 * 		OK
14233 *		ERROR - fatal error
14234 */
14235static ctl_ha_comp_status
14236ctl_isc_init(struct ctl_ha_component *c)
14237{
14238	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14239
14240	c->status = ret;
14241	return ret;
14242}
14243
14244/* Start component
14245 * 	Starts component in state requested. If component starts successfully,
14246 *	it must set its own state to the requestrd state
14247 *	When requested state is HASC_STATE_HA, the component may refine it
14248 * 	by adding _SLAVE or _MASTER flags.
14249 *	Currently allowed state transitions are:
14250 *	UNKNOWN->HA		- initial startup
14251 *	UNKNOWN->SINGLE - initial startup when no parter detected
14252 *	HA->SINGLE		- failover
14253 * returns ctl_ha_comp_status:
14254 * 		OK	- component successfully started in requested state
14255 *		FAILED  - could not start the requested state, failover may
14256 * 			  be possible
14257 *		ERROR	- fatal error detected, no future startup possible
14258 */
14259static ctl_ha_comp_status
14260ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14261{
14262	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14263
14264	printf("%s: go\n", __func__);
14265
14266	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14267	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14268		ctl_is_single = 0;
14269		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14270		    != CTL_HA_STATUS_SUCCESS) {
14271			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14272			ret = CTL_HA_COMP_STATUS_ERROR;
14273		}
14274	} else if (CTL_HA_STATE_IS_HA(c->state)
14275		&& CTL_HA_STATE_IS_SINGLE(state)){
14276		// HA->SINGLE transition
14277	        ctl_failover();
14278		ctl_is_single = 1;
14279	} else {
14280		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14281		       c->state, state);
14282		ret = CTL_HA_COMP_STATUS_ERROR;
14283	}
14284	if (CTL_HA_STATE_IS_SINGLE(state))
14285		ctl_is_single = 1;
14286
14287	c->state = state;
14288	c->status = ret;
14289	return ret;
14290}
14291
14292/*
14293 * Quiesce component
14294 * The component must clear any error conditions (set status to OK) and
14295 * prepare itself to another Start call
14296 * returns ctl_ha_comp_status:
14297 * 	OK
14298 *	ERROR
14299 */
14300static ctl_ha_comp_status
14301ctl_isc_quiesce(struct ctl_ha_component *c)
14302{
14303	int ret = CTL_HA_COMP_STATUS_OK;
14304
14305	ctl_pause_rtr = 1;
14306	c->status = ret;
14307	return ret;
14308}
14309
14310struct ctl_ha_component ctl_ha_component_ctlisc =
14311{
14312	.name = "CTL ISC",
14313	.state = CTL_HA_STATE_UNKNOWN,
14314	.init = ctl_isc_init,
14315	.start = ctl_isc_start,
14316	.quiesce = ctl_isc_quiesce
14317};
14318
14319/*
14320 *  vim: ts=8
14321 */
14322