ctl.c revision 272637
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: stable/10/sys/cam/ctl/ctl.c 272637 2014-10-06 13:25:36Z mav $");
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/kernel.h>
50#include <sys/types.h>
51#include <sys/kthread.h>
52#include <sys/bio.h>
53#include <sys/fcntl.h>
54#include <sys/lock.h>
55#include <sys/module.h>
56#include <sys/mutex.h>
57#include <sys/condvar.h>
58#include <sys/malloc.h>
59#include <sys/conf.h>
60#include <sys/ioccom.h>
61#include <sys/queue.h>
62#include <sys/sbuf.h>
63#include <sys/smp.h>
64#include <sys/endian.h>
65#include <sys/sysctl.h>
66
67#include <cam/cam.h>
68#include <cam/scsi/scsi_all.h>
69#include <cam/scsi/scsi_da.h>
70#include <cam/ctl/ctl_io.h>
71#include <cam/ctl/ctl.h>
72#include <cam/ctl/ctl_frontend.h>
73#include <cam/ctl/ctl_frontend_internal.h>
74#include <cam/ctl/ctl_util.h>
75#include <cam/ctl/ctl_backend.h>
76#include <cam/ctl/ctl_ioctl.h>
77#include <cam/ctl/ctl_ha.h>
78#include <cam/ctl/ctl_private.h>
79#include <cam/ctl/ctl_debug.h>
80#include <cam/ctl/ctl_scsi_all.h>
81#include <cam/ctl/ctl_error.h>
82
83struct ctl_softc *control_softc = NULL;
84
85/*
86 * Size and alignment macros needed for Copan-specific HA hardware.  These
87 * can go away when the HA code is re-written, and uses busdma for any
88 * hardware.
89 */
90#define	CTL_ALIGN_8B(target, source, type)				\
91	if (((uint32_t)source & 0x7) != 0)				\
92		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
93	else								\
94		target = (type)source;
95
96#define	CTL_SIZE_8B(target, size)					\
97	if ((size & 0x7) != 0)						\
98		target = size + (0x8 - (size & 0x7));			\
99	else								\
100		target = size;
101
102#define CTL_ALIGN_8B_MARGIN	16
103
104/*
105 * Template mode pages.
106 */
107
108/*
109 * Note that these are default values only.  The actual values will be
110 * filled in when the user does a mode sense.
111 */
112static struct copan_power_subpage power_page_default = {
113	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
114	/*subpage*/ PWR_SUBPAGE_CODE,
115	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
116			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
117	/*page_version*/ PWR_VERSION,
118	/* total_luns */ 26,
119	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
120	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
121		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
122		      0, 0, 0, 0, 0, 0}
123};
124
125static struct copan_power_subpage power_page_changeable = {
126	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
127	/*subpage*/ PWR_SUBPAGE_CODE,
128	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
129			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
130	/*page_version*/ 0,
131	/* total_luns */ 0,
132	/* max_active_luns*/ 0,
133	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
134		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
135		      0, 0, 0, 0, 0, 0}
136};
137
138static struct copan_aps_subpage aps_page_default = {
139	APS_PAGE_CODE | SMPH_SPF, //page_code
140	APS_SUBPAGE_CODE, //subpage
141	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
142	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
143	APS_VERSION, //page_version
144	0, //lock_active
145	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
146	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
147	0, 0, 0, 0, 0} //reserved
148};
149
150static struct copan_aps_subpage aps_page_changeable = {
151	APS_PAGE_CODE | SMPH_SPF, //page_code
152	APS_SUBPAGE_CODE, //subpage
153	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
154	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
155	0, //page_version
156	0, //lock_active
157	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
158	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
159	0, 0, 0, 0, 0} //reserved
160};
161
162static struct copan_debugconf_subpage debugconf_page_default = {
163	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
164	DBGCNF_SUBPAGE_CODE,		/* subpage */
165	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
166	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
167	DBGCNF_VERSION,			/* page_version */
168	{CTL_TIME_IO_DEFAULT_SECS>>8,
169	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
170};
171
172static struct copan_debugconf_subpage debugconf_page_changeable = {
173	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
174	DBGCNF_SUBPAGE_CODE,		/* subpage */
175	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
176	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
177	0,				/* page_version */
178	{0xff,0xff},			/* ctl_time_io_secs */
179};
180
181static struct scsi_format_page format_page_default = {
182	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
183	/*page_length*/sizeof(struct scsi_format_page) - 2,
184	/*tracks_per_zone*/ {0, 0},
185	/*alt_sectors_per_zone*/ {0, 0},
186	/*alt_tracks_per_zone*/ {0, 0},
187	/*alt_tracks_per_lun*/ {0, 0},
188	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
189			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
190	/*bytes_per_sector*/ {0, 0},
191	/*interleave*/ {0, 0},
192	/*track_skew*/ {0, 0},
193	/*cylinder_skew*/ {0, 0},
194	/*flags*/ SFP_HSEC,
195	/*reserved*/ {0, 0, 0}
196};
197
198static struct scsi_format_page format_page_changeable = {
199	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
200	/*page_length*/sizeof(struct scsi_format_page) - 2,
201	/*tracks_per_zone*/ {0, 0},
202	/*alt_sectors_per_zone*/ {0, 0},
203	/*alt_tracks_per_zone*/ {0, 0},
204	/*alt_tracks_per_lun*/ {0, 0},
205	/*sectors_per_track*/ {0, 0},
206	/*bytes_per_sector*/ {0, 0},
207	/*interleave*/ {0, 0},
208	/*track_skew*/ {0, 0},
209	/*cylinder_skew*/ {0, 0},
210	/*flags*/ 0,
211	/*reserved*/ {0, 0, 0}
212};
213
214static struct scsi_rigid_disk_page rigid_disk_page_default = {
215	/*page_code*/SMS_RIGID_DISK_PAGE,
216	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
217	/*cylinders*/ {0, 0, 0},
218	/*heads*/ CTL_DEFAULT_HEADS,
219	/*start_write_precomp*/ {0, 0, 0},
220	/*start_reduced_current*/ {0, 0, 0},
221	/*step_rate*/ {0, 0},
222	/*landing_zone_cylinder*/ {0, 0, 0},
223	/*rpl*/ SRDP_RPL_DISABLED,
224	/*rotational_offset*/ 0,
225	/*reserved1*/ 0,
226	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
227			   CTL_DEFAULT_ROTATION_RATE & 0xff},
228	/*reserved2*/ {0, 0}
229};
230
231static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
232	/*page_code*/SMS_RIGID_DISK_PAGE,
233	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
234	/*cylinders*/ {0, 0, 0},
235	/*heads*/ 0,
236	/*start_write_precomp*/ {0, 0, 0},
237	/*start_reduced_current*/ {0, 0, 0},
238	/*step_rate*/ {0, 0},
239	/*landing_zone_cylinder*/ {0, 0, 0},
240	/*rpl*/ 0,
241	/*rotational_offset*/ 0,
242	/*reserved1*/ 0,
243	/*rotation_rate*/ {0, 0},
244	/*reserved2*/ {0, 0}
245};
246
247static struct scsi_caching_page caching_page_default = {
248	/*page_code*/SMS_CACHING_PAGE,
249	/*page_length*/sizeof(struct scsi_caching_page) - 2,
250	/*flags1*/ SCP_DISC | SCP_WCE,
251	/*ret_priority*/ 0,
252	/*disable_pf_transfer_len*/ {0xff, 0xff},
253	/*min_prefetch*/ {0, 0},
254	/*max_prefetch*/ {0xff, 0xff},
255	/*max_pf_ceiling*/ {0xff, 0xff},
256	/*flags2*/ 0,
257	/*cache_segments*/ 0,
258	/*cache_seg_size*/ {0, 0},
259	/*reserved*/ 0,
260	/*non_cache_seg_size*/ {0, 0, 0}
261};
262
263static struct scsi_caching_page caching_page_changeable = {
264	/*page_code*/SMS_CACHING_PAGE,
265	/*page_length*/sizeof(struct scsi_caching_page) - 2,
266	/*flags1*/ SCP_WCE | SCP_RCD,
267	/*ret_priority*/ 0,
268	/*disable_pf_transfer_len*/ {0, 0},
269	/*min_prefetch*/ {0, 0},
270	/*max_prefetch*/ {0, 0},
271	/*max_pf_ceiling*/ {0, 0},
272	/*flags2*/ 0,
273	/*cache_segments*/ 0,
274	/*cache_seg_size*/ {0, 0},
275	/*reserved*/ 0,
276	/*non_cache_seg_size*/ {0, 0, 0}
277};
278
279static struct scsi_control_page control_page_default = {
280	/*page_code*/SMS_CONTROL_MODE_PAGE,
281	/*page_length*/sizeof(struct scsi_control_page) - 2,
282	/*rlec*/0,
283	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
284	/*eca_and_aen*/0,
285	/*flags4*/SCP_TAS,
286	/*aen_holdoff_period*/{0, 0},
287	/*busy_timeout_period*/{0, 0},
288	/*extended_selftest_completion_time*/{0, 0}
289};
290
291static struct scsi_control_page control_page_changeable = {
292	/*page_code*/SMS_CONTROL_MODE_PAGE,
293	/*page_length*/sizeof(struct scsi_control_page) - 2,
294	/*rlec*/SCP_DSENSE,
295	/*queue_flags*/SCP_QUEUE_ALG_MASK,
296	/*eca_and_aen*/0,
297	/*flags4*/0,
298	/*aen_holdoff_period*/{0, 0},
299	/*busy_timeout_period*/{0, 0},
300	/*extended_selftest_completion_time*/{0, 0}
301};
302
303
304/*
305 * XXX KDM move these into the softc.
306 */
307static int rcv_sync_msg;
308static int persis_offset;
309static uint8_t ctl_pause_rtr;
310static int     ctl_is_single = 1;
311static int     index_to_aps_page;
312
313SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
314static int worker_threads = -1;
315TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
316SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
317    &worker_threads, 1, "Number of worker threads");
318static int verbose = 0;
319TUNABLE_INT("kern.cam.ctl.verbose", &verbose);
320SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
321    &verbose, 0, "Show SCSI errors returned to initiator");
322
323/*
324 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
325 * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
326 * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
327 * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
328 */
329#define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
330
331static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
332				  int param);
333static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
334static int ctl_init(void);
335void ctl_shutdown(void);
336static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
337static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
338static void ctl_ioctl_online(void *arg);
339static void ctl_ioctl_offline(void *arg);
340static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
341static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
342static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
343static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
344static int ctl_ioctl_submit_wait(union ctl_io *io);
345static void ctl_ioctl_datamove(union ctl_io *io);
346static void ctl_ioctl_done(union ctl_io *io);
347static void ctl_ioctl_hard_startstop_callback(void *arg,
348					      struct cfi_metatask *metatask);
349static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
350static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
351			      struct ctl_ooa *ooa_hdr,
352			      struct ctl_ooa_entry *kern_entries);
353static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
354		     struct thread *td);
355static uint32_t ctl_map_lun(int port_num, uint32_t lun);
356static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
357#ifdef unused
358static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
359				   uint32_t targ_target, uint32_t targ_lun,
360				   int can_wait);
361static void ctl_kfree_io(union ctl_io *io);
362#endif /* unused */
363static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
364			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
365static int ctl_free_lun(struct ctl_lun *lun);
366static void ctl_create_lun(struct ctl_be_lun *be_lun);
367/**
368static void ctl_failover_change_pages(struct ctl_softc *softc,
369				      struct ctl_scsiio *ctsio, int master);
370**/
371
372static int ctl_do_mode_select(union ctl_io *io);
373static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
374			   uint64_t res_key, uint64_t sa_res_key,
375			   uint8_t type, uint32_t residx,
376			   struct ctl_scsiio *ctsio,
377			   struct scsi_per_res_out *cdb,
378			   struct scsi_per_res_out_parms* param);
379static void ctl_pro_preempt_other(struct ctl_lun *lun,
380				  union ctl_ha_msg *msg);
381static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
382static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
383static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
384static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
385static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
386static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
387static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
388					 int alloc_len);
389static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
390					 int alloc_len);
391static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
392static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
393static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
394static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
395static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
396static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
397static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
398    union ctl_io *pending_io, union ctl_io *ooa_io);
399static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
400				union ctl_io *starting_io);
401static int ctl_check_blocked(struct ctl_lun *lun);
402static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
403				struct ctl_lun *lun,
404				const struct ctl_cmd_entry *entry,
405				struct ctl_scsiio *ctsio);
406//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
407static void ctl_failover(void);
408static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
409			       struct ctl_scsiio *ctsio);
410static int ctl_scsiio(struct ctl_scsiio *ctsio);
411
412static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
413static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
414			    ctl_ua_type ua_type);
415static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
416			 ctl_ua_type ua_type);
417static int ctl_abort_task(union ctl_io *io);
418static int ctl_abort_task_set(union ctl_io *io);
419static int ctl_i_t_nexus_reset(union ctl_io *io);
420static void ctl_run_task(union ctl_io *io);
421#ifdef CTL_IO_DELAY
422static void ctl_datamove_timer_wakeup(void *arg);
423static void ctl_done_timer_wakeup(void *arg);
424#endif /* CTL_IO_DELAY */
425
426static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
427static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
428static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
429static void ctl_datamove_remote_write(union ctl_io *io);
430static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
431static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
432static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
433static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
434				    ctl_ha_dt_cb callback);
435static void ctl_datamove_remote_read(union ctl_io *io);
436static void ctl_datamove_remote(union ctl_io *io);
437static int ctl_process_done(union ctl_io *io);
438static void ctl_lun_thread(void *arg);
439static void ctl_work_thread(void *arg);
440static void ctl_enqueue_incoming(union ctl_io *io);
441static void ctl_enqueue_rtr(union ctl_io *io);
442static void ctl_enqueue_done(union ctl_io *io);
443static void ctl_enqueue_isc(union ctl_io *io);
444static const struct ctl_cmd_entry *
445    ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
446static const struct ctl_cmd_entry *
447    ctl_validate_command(struct ctl_scsiio *ctsio);
448static int ctl_cmd_applicable(uint8_t lun_type,
449    const struct ctl_cmd_entry *entry);
450
451/*
452 * Load the serialization table.  This isn't very pretty, but is probably
453 * the easiest way to do it.
454 */
455#include "ctl_ser_table.c"
456
457/*
458 * We only need to define open, close and ioctl routines for this driver.
459 */
460static struct cdevsw ctl_cdevsw = {
461	.d_version =	D_VERSION,
462	.d_flags =	0,
463	.d_open =	ctl_open,
464	.d_close =	ctl_close,
465	.d_ioctl =	ctl_ioctl,
466	.d_name =	"ctl",
467};
468
469
470MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
471MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
472
473static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
474
475static moduledata_t ctl_moduledata = {
476	"ctl",
477	ctl_module_event_handler,
478	NULL
479};
480
481DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
482MODULE_VERSION(ctl, 1);
483
484static struct ctl_frontend ioctl_frontend =
485{
486	.name = "ioctl",
487};
488
489static void
490ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
491			    union ctl_ha_msg *msg_info)
492{
493	struct ctl_scsiio *ctsio;
494
495	if (msg_info->hdr.original_sc == NULL) {
496		printf("%s: original_sc == NULL!\n", __func__);
497		/* XXX KDM now what? */
498		return;
499	}
500
501	ctsio = &msg_info->hdr.original_sc->scsiio;
502	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
503	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
504	ctsio->io_hdr.status = msg_info->hdr.status;
505	ctsio->scsi_status = msg_info->scsi.scsi_status;
506	ctsio->sense_len = msg_info->scsi.sense_len;
507	ctsio->sense_residual = msg_info->scsi.sense_residual;
508	ctsio->residual = msg_info->scsi.residual;
509	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
510	       sizeof(ctsio->sense_data));
511	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
512	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
513	ctl_enqueue_isc((union ctl_io *)ctsio);
514}
515
516static void
517ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
518				union ctl_ha_msg *msg_info)
519{
520	struct ctl_scsiio *ctsio;
521
522	if (msg_info->hdr.serializing_sc == NULL) {
523		printf("%s: serializing_sc == NULL!\n", __func__);
524		/* XXX KDM now what? */
525		return;
526	}
527
528	ctsio = &msg_info->hdr.serializing_sc->scsiio;
529#if 0
530	/*
531	 * Attempt to catch the situation where an I/O has
532	 * been freed, and we're using it again.
533	 */
534	if (ctsio->io_hdr.io_type == 0xff) {
535		union ctl_io *tmp_io;
536		tmp_io = (union ctl_io *)ctsio;
537		printf("%s: %p use after free!\n", __func__,
538		       ctsio);
539		printf("%s: type %d msg %d cdb %x iptl: "
540		       "%d:%d:%d:%d tag 0x%04x "
541		       "flag %#x status %x\n",
542			__func__,
543			tmp_io->io_hdr.io_type,
544			tmp_io->io_hdr.msg_type,
545			tmp_io->scsiio.cdb[0],
546			tmp_io->io_hdr.nexus.initid.id,
547			tmp_io->io_hdr.nexus.targ_port,
548			tmp_io->io_hdr.nexus.targ_target.id,
549			tmp_io->io_hdr.nexus.targ_lun,
550			(tmp_io->io_hdr.io_type ==
551			CTL_IO_TASK) ?
552			tmp_io->taskio.tag_num :
553			tmp_io->scsiio.tag_num,
554		        tmp_io->io_hdr.flags,
555			tmp_io->io_hdr.status);
556	}
557#endif
558	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
559	ctl_enqueue_isc((union ctl_io *)ctsio);
560}
561
562/*
563 * ISC (Inter Shelf Communication) event handler.  Events from the HA
564 * subsystem come in here.
565 */
566static void
567ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
568{
569	struct ctl_softc *ctl_softc;
570	union ctl_io *io;
571	struct ctl_prio *presio;
572	ctl_ha_status isc_status;
573
574	ctl_softc = control_softc;
575	io = NULL;
576
577
578#if 0
579	printf("CTL: Isc Msg event %d\n", event);
580#endif
581	if (event == CTL_HA_EVT_MSG_RECV) {
582		union ctl_ha_msg msg_info;
583
584		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
585					     sizeof(msg_info), /*wait*/ 0);
586#if 0
587		printf("CTL: msg_type %d\n", msg_info.msg_type);
588#endif
589		if (isc_status != 0) {
590			printf("Error receiving message, status = %d\n",
591			       isc_status);
592			return;
593		}
594
595		switch (msg_info.hdr.msg_type) {
596		case CTL_MSG_SERIALIZE:
597#if 0
598			printf("Serialize\n");
599#endif
600			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
601			if (io == NULL) {
602				printf("ctl_isc_event_handler: can't allocate "
603				       "ctl_io!\n");
604				/* Bad Juju */
605				/* Need to set busy and send msg back */
606				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
607				msg_info.hdr.status = CTL_SCSI_ERROR;
608				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
609				msg_info.scsi.sense_len = 0;
610			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
611				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
612				}
613				goto bailout;
614			}
615			ctl_zero_io(io);
616			// populate ctsio from msg_info
617			io->io_hdr.io_type = CTL_IO_SCSI;
618			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
619			io->io_hdr.original_sc = msg_info.hdr.original_sc;
620#if 0
621			printf("pOrig %x\n", (int)msg_info.original_sc);
622#endif
623			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
624					    CTL_FLAG_IO_ACTIVE;
625			/*
626			 * If we're in serialization-only mode, we don't
627			 * want to go through full done processing.  Thus
628			 * the COPY flag.
629			 *
630			 * XXX KDM add another flag that is more specific.
631			 */
632			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
633				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
634			io->io_hdr.nexus = msg_info.hdr.nexus;
635#if 0
636			printf("targ %d, port %d, iid %d, lun %d\n",
637			       io->io_hdr.nexus.targ_target.id,
638			       io->io_hdr.nexus.targ_port,
639			       io->io_hdr.nexus.initid.id,
640			       io->io_hdr.nexus.targ_lun);
641#endif
642			io->scsiio.tag_num = msg_info.scsi.tag_num;
643			io->scsiio.tag_type = msg_info.scsi.tag_type;
644			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
645			       CTL_MAX_CDBLEN);
646			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
647				const struct ctl_cmd_entry *entry;
648
649				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
650				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
651				io->io_hdr.flags |=
652					entry->flags & CTL_FLAG_DATA_MASK;
653			}
654			ctl_enqueue_isc(io);
655			break;
656
657		/* Performed on the Originating SC, XFER mode only */
658		case CTL_MSG_DATAMOVE: {
659			struct ctl_sg_entry *sgl;
660			int i, j;
661
662			io = msg_info.hdr.original_sc;
663			if (io == NULL) {
664				printf("%s: original_sc == NULL!\n", __func__);
665				/* XXX KDM do something here */
666				break;
667			}
668			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
669			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
670			/*
671			 * Keep track of this, we need to send it back over
672			 * when the datamove is complete.
673			 */
674			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
675
676			if (msg_info.dt.sg_sequence == 0) {
677				/*
678				 * XXX KDM we use the preallocated S/G list
679				 * here, but we'll need to change this to
680				 * dynamic allocation if we need larger S/G
681				 * lists.
682				 */
683				if (msg_info.dt.kern_sg_entries >
684				    sizeof(io->io_hdr.remote_sglist) /
685				    sizeof(io->io_hdr.remote_sglist[0])) {
686					printf("%s: number of S/G entries "
687					    "needed %u > allocated num %zd\n",
688					    __func__,
689					    msg_info.dt.kern_sg_entries,
690					    sizeof(io->io_hdr.remote_sglist)/
691					    sizeof(io->io_hdr.remote_sglist[0]));
692
693					/*
694					 * XXX KDM send a message back to
695					 * the other side to shut down the
696					 * DMA.  The error will come back
697					 * through via the normal channel.
698					 */
699					break;
700				}
701				sgl = io->io_hdr.remote_sglist;
702				memset(sgl, 0,
703				       sizeof(io->io_hdr.remote_sglist));
704
705				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
706
707				io->scsiio.kern_sg_entries =
708					msg_info.dt.kern_sg_entries;
709				io->scsiio.rem_sg_entries =
710					msg_info.dt.kern_sg_entries;
711				io->scsiio.kern_data_len =
712					msg_info.dt.kern_data_len;
713				io->scsiio.kern_total_len =
714					msg_info.dt.kern_total_len;
715				io->scsiio.kern_data_resid =
716					msg_info.dt.kern_data_resid;
717				io->scsiio.kern_rel_offset =
718					msg_info.dt.kern_rel_offset;
719				/*
720				 * Clear out per-DMA flags.
721				 */
722				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
723				/*
724				 * Add per-DMA flags that are set for this
725				 * particular DMA request.
726				 */
727				io->io_hdr.flags |= msg_info.dt.flags &
728						    CTL_FLAG_RDMA_MASK;
729			} else
730				sgl = (struct ctl_sg_entry *)
731					io->scsiio.kern_data_ptr;
732
733			for (i = msg_info.dt.sent_sg_entries, j = 0;
734			     i < (msg_info.dt.sent_sg_entries +
735			     msg_info.dt.cur_sg_entries); i++, j++) {
736				sgl[i].addr = msg_info.dt.sg_list[j].addr;
737				sgl[i].len = msg_info.dt.sg_list[j].len;
738
739#if 0
740				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
741				       __func__,
742				       msg_info.dt.sg_list[j].addr,
743				       msg_info.dt.sg_list[j].len,
744				       sgl[i].addr, sgl[i].len, j, i);
745#endif
746			}
747#if 0
748			memcpy(&sgl[msg_info.dt.sent_sg_entries],
749			       msg_info.dt.sg_list,
750			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
751#endif
752
753			/*
754			 * If this is the last piece of the I/O, we've got
755			 * the full S/G list.  Queue processing in the thread.
756			 * Otherwise wait for the next piece.
757			 */
758			if (msg_info.dt.sg_last != 0)
759				ctl_enqueue_isc(io);
760			break;
761		}
762		/* Performed on the Serializing (primary) SC, XFER mode only */
763		case CTL_MSG_DATAMOVE_DONE: {
764			if (msg_info.hdr.serializing_sc == NULL) {
765				printf("%s: serializing_sc == NULL!\n",
766				       __func__);
767				/* XXX KDM now what? */
768				break;
769			}
770			/*
771			 * We grab the sense information here in case
772			 * there was a failure, so we can return status
773			 * back to the initiator.
774			 */
775			io = msg_info.hdr.serializing_sc;
776			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
777			io->io_hdr.status = msg_info.hdr.status;
778			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
779			io->scsiio.sense_len = msg_info.scsi.sense_len;
780			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
781			io->io_hdr.port_status = msg_info.scsi.fetd_status;
782			io->scsiio.residual = msg_info.scsi.residual;
783			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
784			       sizeof(io->scsiio.sense_data));
785			ctl_enqueue_isc(io);
786			break;
787		}
788
789		/* Preformed on Originating SC, SER_ONLY mode */
790		case CTL_MSG_R2R:
791			io = msg_info.hdr.original_sc;
792			if (io == NULL) {
793				printf("%s: Major Bummer\n", __func__);
794				return;
795			} else {
796#if 0
797				printf("pOrig %x\n",(int) ctsio);
798#endif
799			}
800			io->io_hdr.msg_type = CTL_MSG_R2R;
801			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
802			ctl_enqueue_isc(io);
803			break;
804
805		/*
806		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
807		 * mode.
808		 * Performed on the Originating (i.e. secondary) SC in XFER
809		 * mode
810		 */
811		case CTL_MSG_FINISH_IO:
812			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
813				ctl_isc_handler_finish_xfer(ctl_softc,
814							    &msg_info);
815			else
816				ctl_isc_handler_finish_ser_only(ctl_softc,
817								&msg_info);
818			break;
819
820		/* Preformed on Originating SC */
821		case CTL_MSG_BAD_JUJU:
822			io = msg_info.hdr.original_sc;
823			if (io == NULL) {
824				printf("%s: Bad JUJU!, original_sc is NULL!\n",
825				       __func__);
826				break;
827			}
828			ctl_copy_sense_data(&msg_info, io);
829			/*
830			 * IO should have already been cleaned up on other
831			 * SC so clear this flag so we won't send a message
832			 * back to finish the IO there.
833			 */
834			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
835			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
836
837			/* io = msg_info.hdr.serializing_sc; */
838			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
839			ctl_enqueue_isc(io);
840			break;
841
842		/* Handle resets sent from the other side */
843		case CTL_MSG_MANAGE_TASKS: {
844			struct ctl_taskio *taskio;
845			taskio = (struct ctl_taskio *)ctl_alloc_io(
846				(void *)ctl_softc->othersc_pool);
847			if (taskio == NULL) {
848				printf("ctl_isc_event_handler: can't allocate "
849				       "ctl_io!\n");
850				/* Bad Juju */
851				/* should I just call the proper reset func
852				   here??? */
853				goto bailout;
854			}
855			ctl_zero_io((union ctl_io *)taskio);
856			taskio->io_hdr.io_type = CTL_IO_TASK;
857			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
858			taskio->io_hdr.nexus = msg_info.hdr.nexus;
859			taskio->task_action = msg_info.task.task_action;
860			taskio->tag_num = msg_info.task.tag_num;
861			taskio->tag_type = msg_info.task.tag_type;
862#ifdef CTL_TIME_IO
863			taskio->io_hdr.start_time = time_uptime;
864			getbintime(&taskio->io_hdr.start_bt);
865#if 0
866			cs_prof_gettime(&taskio->io_hdr.start_ticks);
867#endif
868#endif /* CTL_TIME_IO */
869			ctl_run_task((union ctl_io *)taskio);
870			break;
871		}
872		/* Persistent Reserve action which needs attention */
873		case CTL_MSG_PERS_ACTION:
874			presio = (struct ctl_prio *)ctl_alloc_io(
875				(void *)ctl_softc->othersc_pool);
876			if (presio == NULL) {
877				printf("ctl_isc_event_handler: can't allocate "
878				       "ctl_io!\n");
879				/* Bad Juju */
880				/* Need to set busy and send msg back */
881				goto bailout;
882			}
883			ctl_zero_io((union ctl_io *)presio);
884			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
885			presio->pr_msg = msg_info.pr;
886			ctl_enqueue_isc((union ctl_io *)presio);
887			break;
888		case CTL_MSG_SYNC_FE:
889			rcv_sync_msg = 1;
890			break;
891		case CTL_MSG_APS_LOCK: {
892			// It's quicker to execute this then to
893			// queue it.
894			struct ctl_lun *lun;
895			struct ctl_page_index *page_index;
896			struct copan_aps_subpage *current_sp;
897			uint32_t targ_lun;
898
899			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
900			lun = ctl_softc->ctl_luns[targ_lun];
901			mtx_lock(&lun->lun_lock);
902			page_index = &lun->mode_pages.index[index_to_aps_page];
903			current_sp = (struct copan_aps_subpage *)
904				     (page_index->page_data +
905				     (page_index->page_len * CTL_PAGE_CURRENT));
906
907			current_sp->lock_active = msg_info.aps.lock_flag;
908			mtx_unlock(&lun->lun_lock);
909		        break;
910		}
911		default:
912		        printf("How did I get here?\n");
913		}
914	} else if (event == CTL_HA_EVT_MSG_SENT) {
915		if (param != CTL_HA_STATUS_SUCCESS) {
916			printf("Bad status from ctl_ha_msg_send status %d\n",
917			       param);
918		}
919		return;
920	} else if (event == CTL_HA_EVT_DISCONNECT) {
921		printf("CTL: Got a disconnect from Isc\n");
922		return;
923	} else {
924		printf("ctl_isc_event_handler: Unknown event %d\n", event);
925		return;
926	}
927
928bailout:
929	return;
930}
931
932static void
933ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
934{
935	struct scsi_sense_data *sense;
936
937	sense = &dest->scsiio.sense_data;
938	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
939	dest->scsiio.scsi_status = src->scsi.scsi_status;
940	dest->scsiio.sense_len = src->scsi.sense_len;
941	dest->io_hdr.status = src->hdr.status;
942}
943
944static int
945ctl_init(void)
946{
947	struct ctl_softc *softc;
948	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
949	struct ctl_port *port;
950        uint8_t sc_id =0;
951	int i, error, retval;
952	//int isc_retval;
953
954	retval = 0;
955	ctl_pause_rtr = 0;
956        rcv_sync_msg = 0;
957
958	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
959			       M_WAITOK | M_ZERO);
960	softc = control_softc;
961
962	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
963			      "cam/ctl");
964
965	softc->dev->si_drv1 = softc;
966
967	/*
968	 * By default, return a "bad LUN" peripheral qualifier for unknown
969	 * LUNs.  The user can override this default using the tunable or
970	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
971	 */
972	softc->inquiry_pq_no_lun = 1;
973	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
974			  &softc->inquiry_pq_no_lun);
975	sysctl_ctx_init(&softc->sysctl_ctx);
976	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
977		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
978		CTLFLAG_RD, 0, "CAM Target Layer");
979
980	if (softc->sysctl_tree == NULL) {
981		printf("%s: unable to allocate sysctl tree\n", __func__);
982		destroy_dev(softc->dev);
983		free(control_softc, M_DEVBUF);
984		control_softc = NULL;
985		return (ENOMEM);
986	}
987
988	SYSCTL_ADD_INT(&softc->sysctl_ctx,
989		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
990		       "inquiry_pq_no_lun", CTLFLAG_RW,
991		       &softc->inquiry_pq_no_lun, 0,
992		       "Report no lun possible for invalid LUNs");
993
994	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
995	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
996	softc->open_count = 0;
997
998	/*
999	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1000	 * the drive.
1001	 */
1002	softc->flags = CTL_FLAG_REAL_SYNC;
1003
1004	/*
1005	 * In Copan's HA scheme, the "master" and "slave" roles are
1006	 * figured out through the slot the controller is in.  Although it
1007	 * is an active/active system, someone has to be in charge.
1008 	 */
1009#ifdef NEEDTOPORT
1010        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1011#endif
1012
1013        if (sc_id == 0) {
1014		softc->flags |= CTL_FLAG_MASTER_SHELF;
1015		persis_offset = 0;
1016	} else
1017		persis_offset = CTL_MAX_INITIATORS;
1018
1019	/*
1020	 * XXX KDM need to figure out where we want to get our target ID
1021	 * and WWID.  Is it different on each port?
1022	 */
1023	softc->target.id = 0;
1024	softc->target.wwid[0] = 0x12345678;
1025	softc->target.wwid[1] = 0x87654321;
1026	STAILQ_INIT(&softc->lun_list);
1027	STAILQ_INIT(&softc->pending_lun_queue);
1028	STAILQ_INIT(&softc->fe_list);
1029	STAILQ_INIT(&softc->port_list);
1030	STAILQ_INIT(&softc->be_list);
1031	STAILQ_INIT(&softc->io_pools);
1032	ctl_tpc_init(softc);
1033
1034	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1035			    &internal_pool)!= 0){
1036		printf("ctl: can't allocate %d entry internal pool, "
1037		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1038		return (ENOMEM);
1039	}
1040
1041	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1042			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1043		printf("ctl: can't allocate %d entry emergency pool, "
1044		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1045		ctl_pool_free(internal_pool);
1046		return (ENOMEM);
1047	}
1048
1049	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1050	                    &other_pool) != 0)
1051	{
1052		printf("ctl: can't allocate %d entry other SC pool, "
1053		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1054		ctl_pool_free(internal_pool);
1055		ctl_pool_free(emergency_pool);
1056		return (ENOMEM);
1057	}
1058
1059	softc->internal_pool = internal_pool;
1060	softc->emergency_pool = emergency_pool;
1061	softc->othersc_pool = other_pool;
1062
1063	if (worker_threads <= 0)
1064		worker_threads = max(1, mp_ncpus / 4);
1065	if (worker_threads > CTL_MAX_THREADS)
1066		worker_threads = CTL_MAX_THREADS;
1067
1068	for (i = 0; i < worker_threads; i++) {
1069		struct ctl_thread *thr = &softc->threads[i];
1070
1071		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1072		thr->ctl_softc = softc;
1073		STAILQ_INIT(&thr->incoming_queue);
1074		STAILQ_INIT(&thr->rtr_queue);
1075		STAILQ_INIT(&thr->done_queue);
1076		STAILQ_INIT(&thr->isc_queue);
1077
1078		error = kproc_kthread_add(ctl_work_thread, thr,
1079		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1080		if (error != 0) {
1081			printf("error creating CTL work thread!\n");
1082			ctl_pool_free(internal_pool);
1083			ctl_pool_free(emergency_pool);
1084			ctl_pool_free(other_pool);
1085			return (error);
1086		}
1087	}
1088	error = kproc_kthread_add(ctl_lun_thread, softc,
1089	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1090	if (error != 0) {
1091		printf("error creating CTL lun thread!\n");
1092		ctl_pool_free(internal_pool);
1093		ctl_pool_free(emergency_pool);
1094		ctl_pool_free(other_pool);
1095		return (error);
1096	}
1097	if (bootverbose)
1098		printf("ctl: CAM Target Layer loaded\n");
1099
1100	/*
1101	 * Initialize the ioctl front end.
1102	 */
1103	ctl_frontend_register(&ioctl_frontend);
1104	port = &softc->ioctl_info.port;
1105	port->frontend = &ioctl_frontend;
1106	sprintf(softc->ioctl_info.port_name, "ioctl");
1107	port->port_type = CTL_PORT_IOCTL;
1108	port->num_requested_ctl_io = 100;
1109	port->port_name = softc->ioctl_info.port_name;
1110	port->port_online = ctl_ioctl_online;
1111	port->port_offline = ctl_ioctl_offline;
1112	port->onoff_arg = &softc->ioctl_info;
1113	port->lun_enable = ctl_ioctl_lun_enable;
1114	port->lun_disable = ctl_ioctl_lun_disable;
1115	port->targ_lun_arg = &softc->ioctl_info;
1116	port->fe_datamove = ctl_ioctl_datamove;
1117	port->fe_done = ctl_ioctl_done;
1118	port->max_targets = 15;
1119	port->max_target_id = 15;
1120
1121	if (ctl_port_register(&softc->ioctl_info.port,
1122	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1123		printf("ctl: ioctl front end registration failed, will "
1124		       "continue anyway\n");
1125	}
1126
1127#ifdef CTL_IO_DELAY
1128	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1129		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1130		       sizeof(struct callout), CTL_TIMER_BYTES);
1131		return (EINVAL);
1132	}
1133#endif /* CTL_IO_DELAY */
1134
1135	return (0);
1136}
1137
1138void
1139ctl_shutdown(void)
1140{
1141	struct ctl_softc *softc;
1142	struct ctl_lun *lun, *next_lun;
1143	struct ctl_io_pool *pool;
1144
1145	softc = (struct ctl_softc *)control_softc;
1146
1147	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1148		printf("ctl: ioctl front end deregistration failed\n");
1149
1150	mtx_lock(&softc->ctl_lock);
1151
1152	/*
1153	 * Free up each LUN.
1154	 */
1155	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1156		next_lun = STAILQ_NEXT(lun, links);
1157		ctl_free_lun(lun);
1158	}
1159
1160	mtx_unlock(&softc->ctl_lock);
1161
1162	ctl_frontend_deregister(&ioctl_frontend);
1163
1164	/*
1165	 * This will rip the rug out from under any FETDs or anyone else
1166	 * that has a pool allocated.  Since we increment our module
1167	 * refcount any time someone outside the main CTL module allocates
1168	 * a pool, we shouldn't have any problems here.  The user won't be
1169	 * able to unload the CTL module until client modules have
1170	 * successfully unloaded.
1171	 */
1172	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1173		ctl_pool_free(pool);
1174
1175#if 0
1176	ctl_shutdown_thread(softc->work_thread);
1177	mtx_destroy(&softc->queue_lock);
1178#endif
1179
1180	ctl_tpc_shutdown(softc);
1181	mtx_destroy(&softc->pool_lock);
1182	mtx_destroy(&softc->ctl_lock);
1183
1184	destroy_dev(softc->dev);
1185
1186	sysctl_ctx_free(&softc->sysctl_ctx);
1187
1188	free(control_softc, M_DEVBUF);
1189	control_softc = NULL;
1190
1191	if (bootverbose)
1192		printf("ctl: CAM Target Layer unloaded\n");
1193}
1194
1195static int
1196ctl_module_event_handler(module_t mod, int what, void *arg)
1197{
1198
1199	switch (what) {
1200	case MOD_LOAD:
1201		return (ctl_init());
1202	case MOD_UNLOAD:
1203		return (EBUSY);
1204	default:
1205		return (EOPNOTSUPP);
1206	}
1207}
1208
1209/*
1210 * XXX KDM should we do some access checks here?  Bump a reference count to
1211 * prevent a CTL module from being unloaded while someone has it open?
1212 */
1213static int
1214ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1215{
1216	return (0);
1217}
1218
1219static int
1220ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1221{
1222	return (0);
1223}
1224
1225int
1226ctl_port_enable(ctl_port_type port_type)
1227{
1228	struct ctl_softc *softc;
1229	struct ctl_port *port;
1230
1231	if (ctl_is_single == 0) {
1232		union ctl_ha_msg msg_info;
1233		int isc_retval;
1234
1235#if 0
1236		printf("%s: HA mode, synchronizing frontend enable\n",
1237		        __func__);
1238#endif
1239		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1240	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1241		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1242			printf("Sync msg send error retval %d\n", isc_retval);
1243		}
1244		if (!rcv_sync_msg) {
1245			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1246			        sizeof(msg_info), 1);
1247		}
1248#if 0
1249        	printf("CTL:Frontend Enable\n");
1250	} else {
1251		printf("%s: single mode, skipping frontend synchronization\n",
1252		        __func__);
1253#endif
1254	}
1255
1256	softc = control_softc;
1257
1258	STAILQ_FOREACH(port, &softc->port_list, links) {
1259		if (port_type & port->port_type)
1260		{
1261#if 0
1262			printf("port %d\n", port->targ_port);
1263#endif
1264			ctl_port_online(port);
1265		}
1266	}
1267
1268	return (0);
1269}
1270
1271int
1272ctl_port_disable(ctl_port_type port_type)
1273{
1274	struct ctl_softc *softc;
1275	struct ctl_port *port;
1276
1277	softc = control_softc;
1278
1279	STAILQ_FOREACH(port, &softc->port_list, links) {
1280		if (port_type & port->port_type)
1281			ctl_port_offline(port);
1282	}
1283
1284	return (0);
1285}
1286
1287/*
1288 * Returns 0 for success, 1 for failure.
1289 * Currently the only failure mode is if there aren't enough entries
1290 * allocated.  So, in case of a failure, look at num_entries_dropped,
1291 * reallocate and try again.
1292 */
1293int
1294ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1295	      int *num_entries_filled, int *num_entries_dropped,
1296	      ctl_port_type port_type, int no_virtual)
1297{
1298	struct ctl_softc *softc;
1299	struct ctl_port *port;
1300	int entries_dropped, entries_filled;
1301	int retval;
1302	int i;
1303
1304	softc = control_softc;
1305
1306	retval = 0;
1307	entries_filled = 0;
1308	entries_dropped = 0;
1309
1310	i = 0;
1311	mtx_lock(&softc->ctl_lock);
1312	STAILQ_FOREACH(port, &softc->port_list, links) {
1313		struct ctl_port_entry *entry;
1314
1315		if ((port->port_type & port_type) == 0)
1316			continue;
1317
1318		if ((no_virtual != 0)
1319		 && (port->virtual_port != 0))
1320			continue;
1321
1322		if (entries_filled >= num_entries_alloced) {
1323			entries_dropped++;
1324			continue;
1325		}
1326		entry = &entries[i];
1327
1328		entry->port_type = port->port_type;
1329		strlcpy(entry->port_name, port->port_name,
1330			sizeof(entry->port_name));
1331		entry->physical_port = port->physical_port;
1332		entry->virtual_port = port->virtual_port;
1333		entry->wwnn = port->wwnn;
1334		entry->wwpn = port->wwpn;
1335
1336		i++;
1337		entries_filled++;
1338	}
1339
1340	mtx_unlock(&softc->ctl_lock);
1341
1342	if (entries_dropped > 0)
1343		retval = 1;
1344
1345	*num_entries_dropped = entries_dropped;
1346	*num_entries_filled = entries_filled;
1347
1348	return (retval);
1349}
1350
1351static void
1352ctl_ioctl_online(void *arg)
1353{
1354	struct ctl_ioctl_info *ioctl_info;
1355
1356	ioctl_info = (struct ctl_ioctl_info *)arg;
1357
1358	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1359}
1360
1361static void
1362ctl_ioctl_offline(void *arg)
1363{
1364	struct ctl_ioctl_info *ioctl_info;
1365
1366	ioctl_info = (struct ctl_ioctl_info *)arg;
1367
1368	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1369}
1370
1371/*
1372 * Remove an initiator by port number and initiator ID.
1373 * Returns 0 for success, -1 for failure.
1374 */
1375int
1376ctl_remove_initiator(struct ctl_port *port, int iid)
1377{
1378	struct ctl_softc *softc = control_softc;
1379
1380	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1381
1382	if (iid > CTL_MAX_INIT_PER_PORT) {
1383		printf("%s: initiator ID %u > maximun %u!\n",
1384		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1385		return (-1);
1386	}
1387
1388	mtx_lock(&softc->ctl_lock);
1389	port->wwpn_iid[iid].in_use--;
1390	port->wwpn_iid[iid].last_use = time_uptime;
1391	mtx_unlock(&softc->ctl_lock);
1392
1393	return (0);
1394}
1395
1396/*
1397 * Add an initiator to the initiator map.
1398 * Returns iid for success, < 0 for failure.
1399 */
1400int
1401ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1402{
1403	struct ctl_softc *softc = control_softc;
1404	time_t best_time;
1405	int i, best;
1406
1407	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1408
1409	if (iid >= CTL_MAX_INIT_PER_PORT) {
1410		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1411		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1412		free(name, M_CTL);
1413		return (-1);
1414	}
1415
1416	mtx_lock(&softc->ctl_lock);
1417
1418	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1419		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1420			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1421				iid = i;
1422				break;
1423			}
1424			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1425			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1426				iid = i;
1427				break;
1428			}
1429		}
1430	}
1431
1432	if (iid < 0) {
1433		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1434			if (port->wwpn_iid[i].in_use == 0 &&
1435			    port->wwpn_iid[i].wwpn == 0 &&
1436			    port->wwpn_iid[i].name == NULL) {
1437				iid = i;
1438				break;
1439			}
1440		}
1441	}
1442
1443	if (iid < 0) {
1444		best = -1;
1445		best_time = INT32_MAX;
1446		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1447			if (port->wwpn_iid[i].in_use == 0) {
1448				if (port->wwpn_iid[i].last_use < best_time) {
1449					best = i;
1450					best_time = port->wwpn_iid[i].last_use;
1451				}
1452			}
1453		}
1454		iid = best;
1455	}
1456
1457	if (iid < 0) {
1458		mtx_unlock(&softc->ctl_lock);
1459		free(name, M_CTL);
1460		return (-2);
1461	}
1462
1463	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1464		/*
1465		 * This is not an error yet.
1466		 */
1467		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1468#if 0
1469			printf("%s: port %d iid %u WWPN %#jx arrived"
1470			    " again\n", __func__, port->targ_port,
1471			    iid, (uintmax_t)wwpn);
1472#endif
1473			goto take;
1474		}
1475		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1476		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1477#if 0
1478			printf("%s: port %d iid %u name '%s' arrived"
1479			    " again\n", __func__, port->targ_port,
1480			    iid, name);
1481#endif
1482			goto take;
1483		}
1484
1485		/*
1486		 * This is an error, but what do we do about it?  The
1487		 * driver is telling us we have a new WWPN for this
1488		 * initiator ID, so we pretty much need to use it.
1489		 */
1490		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1491		    " but WWPN %#jx '%s' is still at that address\n",
1492		    __func__, port->targ_port, iid, wwpn, name,
1493		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1494		    port->wwpn_iid[iid].name);
1495
1496		/*
1497		 * XXX KDM clear have_ca and ua_pending on each LUN for
1498		 * this initiator.
1499		 */
1500	}
1501take:
1502	free(port->wwpn_iid[iid].name, M_CTL);
1503	port->wwpn_iid[iid].name = name;
1504	port->wwpn_iid[iid].wwpn = wwpn;
1505	port->wwpn_iid[iid].in_use++;
1506	mtx_unlock(&softc->ctl_lock);
1507
1508	return (iid);
1509}
1510
1511static int
1512ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1513{
1514	int len;
1515
1516	switch (port->port_type) {
1517	case CTL_PORT_FC:
1518	{
1519		struct scsi_transportid_fcp *id =
1520		    (struct scsi_transportid_fcp *)buf;
1521		if (port->wwpn_iid[iid].wwpn == 0)
1522			return (0);
1523		memset(id, 0, sizeof(*id));
1524		id->format_protocol = SCSI_PROTO_FC;
1525		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1526		return (sizeof(*id));
1527	}
1528	case CTL_PORT_ISCSI:
1529	{
1530		struct scsi_transportid_iscsi_port *id =
1531		    (struct scsi_transportid_iscsi_port *)buf;
1532		if (port->wwpn_iid[iid].name == NULL)
1533			return (0);
1534		memset(id, 0, 256);
1535		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1536		    SCSI_PROTO_ISCSI;
1537		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1538		len = roundup2(min(len, 252), 4);
1539		scsi_ulto2b(len, id->additional_length);
1540		return (sizeof(*id) + len);
1541	}
1542	case CTL_PORT_SAS:
1543	{
1544		struct scsi_transportid_sas *id =
1545		    (struct scsi_transportid_sas *)buf;
1546		if (port->wwpn_iid[iid].wwpn == 0)
1547			return (0);
1548		memset(id, 0, sizeof(*id));
1549		id->format_protocol = SCSI_PROTO_SAS;
1550		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1551		return (sizeof(*id));
1552	}
1553	default:
1554	{
1555		struct scsi_transportid_spi *id =
1556		    (struct scsi_transportid_spi *)buf;
1557		memset(id, 0, sizeof(*id));
1558		id->format_protocol = SCSI_PROTO_SPI;
1559		scsi_ulto2b(iid, id->scsi_addr);
1560		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1561		return (sizeof(*id));
1562	}
1563	}
1564}
1565
1566static int
1567ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1568{
1569	return (0);
1570}
1571
1572static int
1573ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1574{
1575	return (0);
1576}
1577
1578/*
1579 * Data movement routine for the CTL ioctl frontend port.
1580 */
1581static int
1582ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1583{
1584	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1585	struct ctl_sg_entry ext_entry, kern_entry;
1586	int ext_sglen, ext_sg_entries, kern_sg_entries;
1587	int ext_sg_start, ext_offset;
1588	int len_to_copy, len_copied;
1589	int kern_watermark, ext_watermark;
1590	int ext_sglist_malloced;
1591	int i, j;
1592
1593	ext_sglist_malloced = 0;
1594	ext_sg_start = 0;
1595	ext_offset = 0;
1596
1597	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1598
1599	/*
1600	 * If this flag is set, fake the data transfer.
1601	 */
1602	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1603		ctsio->ext_data_filled = ctsio->ext_data_len;
1604		goto bailout;
1605	}
1606
1607	/*
1608	 * To simplify things here, if we have a single buffer, stick it in
1609	 * a S/G entry and just make it a single entry S/G list.
1610	 */
1611	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1612		int len_seen;
1613
1614		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1615
1616		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1617							   M_WAITOK);
1618		ext_sglist_malloced = 1;
1619		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1620				   ext_sglen) != 0) {
1621			ctl_set_internal_failure(ctsio,
1622						 /*sks_valid*/ 0,
1623						 /*retry_count*/ 0);
1624			goto bailout;
1625		}
1626		ext_sg_entries = ctsio->ext_sg_entries;
1627		len_seen = 0;
1628		for (i = 0; i < ext_sg_entries; i++) {
1629			if ((len_seen + ext_sglist[i].len) >=
1630			     ctsio->ext_data_filled) {
1631				ext_sg_start = i;
1632				ext_offset = ctsio->ext_data_filled - len_seen;
1633				break;
1634			}
1635			len_seen += ext_sglist[i].len;
1636		}
1637	} else {
1638		ext_sglist = &ext_entry;
1639		ext_sglist->addr = ctsio->ext_data_ptr;
1640		ext_sglist->len = ctsio->ext_data_len;
1641		ext_sg_entries = 1;
1642		ext_sg_start = 0;
1643		ext_offset = ctsio->ext_data_filled;
1644	}
1645
1646	if (ctsio->kern_sg_entries > 0) {
1647		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1648		kern_sg_entries = ctsio->kern_sg_entries;
1649	} else {
1650		kern_sglist = &kern_entry;
1651		kern_sglist->addr = ctsio->kern_data_ptr;
1652		kern_sglist->len = ctsio->kern_data_len;
1653		kern_sg_entries = 1;
1654	}
1655
1656
1657	kern_watermark = 0;
1658	ext_watermark = ext_offset;
1659	len_copied = 0;
1660	for (i = ext_sg_start, j = 0;
1661	     i < ext_sg_entries && j < kern_sg_entries;) {
1662		uint8_t *ext_ptr, *kern_ptr;
1663
1664		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1665				      kern_sglist[j].len - kern_watermark);
1666
1667		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1668		ext_ptr = ext_ptr + ext_watermark;
1669		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1670			/*
1671			 * XXX KDM fix this!
1672			 */
1673			panic("need to implement bus address support");
1674#if 0
1675			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1676#endif
1677		} else
1678			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1679		kern_ptr = kern_ptr + kern_watermark;
1680
1681		kern_watermark += len_to_copy;
1682		ext_watermark += len_to_copy;
1683
1684		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1685		     CTL_FLAG_DATA_IN) {
1686			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1687					 "bytes to user\n", len_to_copy));
1688			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1689					 "to %p\n", kern_ptr, ext_ptr));
1690			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1691				ctl_set_internal_failure(ctsio,
1692							 /*sks_valid*/ 0,
1693							 /*retry_count*/ 0);
1694				goto bailout;
1695			}
1696		} else {
1697			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1698					 "bytes from user\n", len_to_copy));
1699			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1700					 "to %p\n", ext_ptr, kern_ptr));
1701			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1702				ctl_set_internal_failure(ctsio,
1703							 /*sks_valid*/ 0,
1704							 /*retry_count*/0);
1705				goto bailout;
1706			}
1707		}
1708
1709		len_copied += len_to_copy;
1710
1711		if (ext_sglist[i].len == ext_watermark) {
1712			i++;
1713			ext_watermark = 0;
1714		}
1715
1716		if (kern_sglist[j].len == kern_watermark) {
1717			j++;
1718			kern_watermark = 0;
1719		}
1720	}
1721
1722	ctsio->ext_data_filled += len_copied;
1723
1724	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1725			 "kern_sg_entries: %d\n", ext_sg_entries,
1726			 kern_sg_entries));
1727	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1728			 "kern_data_len = %d\n", ctsio->ext_data_len,
1729			 ctsio->kern_data_len));
1730
1731
1732	/* XXX KDM set residual?? */
1733bailout:
1734
1735	if (ext_sglist_malloced != 0)
1736		free(ext_sglist, M_CTL);
1737
1738	return (CTL_RETVAL_COMPLETE);
1739}
1740
1741/*
1742 * Serialize a command that went down the "wrong" side, and so was sent to
1743 * this controller for execution.  The logic is a little different than the
1744 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1745 * sent back to the other side, but in the success case, we execute the
1746 * command on this side (XFER mode) or tell the other side to execute it
1747 * (SER_ONLY mode).
1748 */
1749static int
1750ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1751{
1752	struct ctl_softc *ctl_softc;
1753	union ctl_ha_msg msg_info;
1754	struct ctl_lun *lun;
1755	int retval = 0;
1756	uint32_t targ_lun;
1757
1758	ctl_softc = control_softc;
1759
1760	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1761	lun = ctl_softc->ctl_luns[targ_lun];
1762	if (lun==NULL)
1763	{
1764		/*
1765		 * Why isn't LUN defined? The other side wouldn't
1766		 * send a cmd if the LUN is undefined.
1767		 */
1768		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1769
1770		/* "Logical unit not supported" */
1771		ctl_set_sense_data(&msg_info.scsi.sense_data,
1772				   lun,
1773				   /*sense_format*/SSD_TYPE_NONE,
1774				   /*current_error*/ 1,
1775				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1776				   /*asc*/ 0x25,
1777				   /*ascq*/ 0x00,
1778				   SSD_ELEM_NONE);
1779
1780		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1781		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1782		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1783		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1784		msg_info.hdr.serializing_sc = NULL;
1785		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1786	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1787				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1788		}
1789		return(1);
1790
1791	}
1792
1793	mtx_lock(&lun->lun_lock);
1794    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1795
1796	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1797		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1798		 ooa_links))) {
1799	case CTL_ACTION_BLOCK:
1800		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1801		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1802				  blocked_links);
1803		break;
1804	case CTL_ACTION_PASS:
1805	case CTL_ACTION_SKIP:
1806		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1807			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1808			ctl_enqueue_rtr((union ctl_io *)ctsio);
1809		} else {
1810
1811			/* send msg back to other side */
1812			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1813			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1814			msg_info.hdr.msg_type = CTL_MSG_R2R;
1815#if 0
1816			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1817#endif
1818		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1819			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1820			}
1821		}
1822		break;
1823	case CTL_ACTION_OVERLAP:
1824		/* OVERLAPPED COMMANDS ATTEMPTED */
1825		ctl_set_sense_data(&msg_info.scsi.sense_data,
1826				   lun,
1827				   /*sense_format*/SSD_TYPE_NONE,
1828				   /*current_error*/ 1,
1829				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1830				   /*asc*/ 0x4E,
1831				   /*ascq*/ 0x00,
1832				   SSD_ELEM_NONE);
1833
1834		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1835		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1836		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1837		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1838		msg_info.hdr.serializing_sc = NULL;
1839		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1840#if 0
1841		printf("BAD JUJU:Major Bummer Overlap\n");
1842#endif
1843		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1844		retval = 1;
1845		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1846		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1847		}
1848		break;
1849	case CTL_ACTION_OVERLAP_TAG:
1850		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1851		ctl_set_sense_data(&msg_info.scsi.sense_data,
1852				   lun,
1853				   /*sense_format*/SSD_TYPE_NONE,
1854				   /*current_error*/ 1,
1855				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1856				   /*asc*/ 0x4D,
1857				   /*ascq*/ ctsio->tag_num & 0xff,
1858				   SSD_ELEM_NONE);
1859
1860		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1861		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1862		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1863		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1864		msg_info.hdr.serializing_sc = NULL;
1865		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1866#if 0
1867		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1868#endif
1869		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1870		retval = 1;
1871		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1872		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1873		}
1874		break;
1875	case CTL_ACTION_ERROR:
1876	default:
1877		/* "Internal target failure" */
1878		ctl_set_sense_data(&msg_info.scsi.sense_data,
1879				   lun,
1880				   /*sense_format*/SSD_TYPE_NONE,
1881				   /*current_error*/ 1,
1882				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1883				   /*asc*/ 0x44,
1884				   /*ascq*/ 0x00,
1885				   SSD_ELEM_NONE);
1886
1887		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1888		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1889		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1890		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1891		msg_info.hdr.serializing_sc = NULL;
1892		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1893#if 0
1894		printf("BAD JUJU:Major Bummer HW Error\n");
1895#endif
1896		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1897		retval = 1;
1898		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1899		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1900		}
1901		break;
1902	}
1903	mtx_unlock(&lun->lun_lock);
1904	return (retval);
1905}
1906
1907static int
1908ctl_ioctl_submit_wait(union ctl_io *io)
1909{
1910	struct ctl_fe_ioctl_params params;
1911	ctl_fe_ioctl_state last_state;
1912	int done, retval;
1913
1914	retval = 0;
1915
1916	bzero(&params, sizeof(params));
1917
1918	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1919	cv_init(&params.sem, "ctlioccv");
1920	params.state = CTL_IOCTL_INPROG;
1921	last_state = params.state;
1922
1923	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1924
1925	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1926
1927	/* This shouldn't happen */
1928	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1929		return (retval);
1930
1931	done = 0;
1932
1933	do {
1934		mtx_lock(&params.ioctl_mtx);
1935		/*
1936		 * Check the state here, and don't sleep if the state has
1937		 * already changed (i.e. wakeup has already occured, but we
1938		 * weren't waiting yet).
1939		 */
1940		if (params.state == last_state) {
1941			/* XXX KDM cv_wait_sig instead? */
1942			cv_wait(&params.sem, &params.ioctl_mtx);
1943		}
1944		last_state = params.state;
1945
1946		switch (params.state) {
1947		case CTL_IOCTL_INPROG:
1948			/* Why did we wake up? */
1949			/* XXX KDM error here? */
1950			mtx_unlock(&params.ioctl_mtx);
1951			break;
1952		case CTL_IOCTL_DATAMOVE:
1953			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1954
1955			/*
1956			 * change last_state back to INPROG to avoid
1957			 * deadlock on subsequent data moves.
1958			 */
1959			params.state = last_state = CTL_IOCTL_INPROG;
1960
1961			mtx_unlock(&params.ioctl_mtx);
1962			ctl_ioctl_do_datamove(&io->scsiio);
1963			/*
1964			 * Note that in some cases, most notably writes,
1965			 * this will queue the I/O and call us back later.
1966			 * In other cases, generally reads, this routine
1967			 * will immediately call back and wake us up,
1968			 * probably using our own context.
1969			 */
1970			io->scsiio.be_move_done(io);
1971			break;
1972		case CTL_IOCTL_DONE:
1973			mtx_unlock(&params.ioctl_mtx);
1974			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1975			done = 1;
1976			break;
1977		default:
1978			mtx_unlock(&params.ioctl_mtx);
1979			/* XXX KDM error here? */
1980			break;
1981		}
1982	} while (done == 0);
1983
1984	mtx_destroy(&params.ioctl_mtx);
1985	cv_destroy(&params.sem);
1986
1987	return (CTL_RETVAL_COMPLETE);
1988}
1989
1990static void
1991ctl_ioctl_datamove(union ctl_io *io)
1992{
1993	struct ctl_fe_ioctl_params *params;
1994
1995	params = (struct ctl_fe_ioctl_params *)
1996		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1997
1998	mtx_lock(&params->ioctl_mtx);
1999	params->state = CTL_IOCTL_DATAMOVE;
2000	cv_broadcast(&params->sem);
2001	mtx_unlock(&params->ioctl_mtx);
2002}
2003
2004static void
2005ctl_ioctl_done(union ctl_io *io)
2006{
2007	struct ctl_fe_ioctl_params *params;
2008
2009	params = (struct ctl_fe_ioctl_params *)
2010		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2011
2012	mtx_lock(&params->ioctl_mtx);
2013	params->state = CTL_IOCTL_DONE;
2014	cv_broadcast(&params->sem);
2015	mtx_unlock(&params->ioctl_mtx);
2016}
2017
2018static void
2019ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2020{
2021	struct ctl_fe_ioctl_startstop_info *sd_info;
2022
2023	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2024
2025	sd_info->hs_info.status = metatask->status;
2026	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2027	sd_info->hs_info.luns_complete =
2028		metatask->taskinfo.startstop.luns_complete;
2029	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2030
2031	cv_broadcast(&sd_info->sem);
2032}
2033
2034static void
2035ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2036{
2037	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2038
2039	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2040
2041	mtx_lock(fe_bbr_info->lock);
2042	fe_bbr_info->bbr_info->status = metatask->status;
2043	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2044	fe_bbr_info->wakeup_done = 1;
2045	mtx_unlock(fe_bbr_info->lock);
2046
2047	cv_broadcast(&fe_bbr_info->sem);
2048}
2049
2050/*
2051 * Returns 0 for success, errno for failure.
2052 */
2053static int
2054ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2055		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2056{
2057	union ctl_io *io;
2058	int retval;
2059
2060	retval = 0;
2061
2062	mtx_lock(&lun->lun_lock);
2063	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2064	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2065	     ooa_links)) {
2066		struct ctl_ooa_entry *entry;
2067
2068		/*
2069		 * If we've got more than we can fit, just count the
2070		 * remaining entries.
2071		 */
2072		if (*cur_fill_num >= ooa_hdr->alloc_num)
2073			continue;
2074
2075		entry = &kern_entries[*cur_fill_num];
2076
2077		entry->tag_num = io->scsiio.tag_num;
2078		entry->lun_num = lun->lun;
2079#ifdef CTL_TIME_IO
2080		entry->start_bt = io->io_hdr.start_bt;
2081#endif
2082		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2083		entry->cdb_len = io->scsiio.cdb_len;
2084		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2085			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2086
2087		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2088			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2089
2090		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2091			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2092
2093		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2094			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2095
2096		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2097			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2098	}
2099	mtx_unlock(&lun->lun_lock);
2100
2101	return (retval);
2102}
2103
2104static void *
2105ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2106		 size_t error_str_len)
2107{
2108	void *kptr;
2109
2110	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2111
2112	if (copyin(user_addr, kptr, len) != 0) {
2113		snprintf(error_str, error_str_len, "Error copying %d bytes "
2114			 "from user address %p to kernel address %p", len,
2115			 user_addr, kptr);
2116		free(kptr, M_CTL);
2117		return (NULL);
2118	}
2119
2120	return (kptr);
2121}
2122
2123static void
2124ctl_free_args(int num_args, struct ctl_be_arg *args)
2125{
2126	int i;
2127
2128	if (args == NULL)
2129		return;
2130
2131	for (i = 0; i < num_args; i++) {
2132		free(args[i].kname, M_CTL);
2133		free(args[i].kvalue, M_CTL);
2134	}
2135
2136	free(args, M_CTL);
2137}
2138
2139static struct ctl_be_arg *
2140ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2141		char *error_str, size_t error_str_len)
2142{
2143	struct ctl_be_arg *args;
2144	int i;
2145
2146	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2147				error_str, error_str_len);
2148
2149	if (args == NULL)
2150		goto bailout;
2151
2152	for (i = 0; i < num_args; i++) {
2153		args[i].kname = NULL;
2154		args[i].kvalue = NULL;
2155	}
2156
2157	for (i = 0; i < num_args; i++) {
2158		uint8_t *tmpptr;
2159
2160		args[i].kname = ctl_copyin_alloc(args[i].name,
2161			args[i].namelen, error_str, error_str_len);
2162		if (args[i].kname == NULL)
2163			goto bailout;
2164
2165		if (args[i].kname[args[i].namelen - 1] != '\0') {
2166			snprintf(error_str, error_str_len, "Argument %d "
2167				 "name is not NUL-terminated", i);
2168			goto bailout;
2169		}
2170
2171		if (args[i].flags & CTL_BEARG_RD) {
2172			tmpptr = ctl_copyin_alloc(args[i].value,
2173				args[i].vallen, error_str, error_str_len);
2174			if (tmpptr == NULL)
2175				goto bailout;
2176			if ((args[i].flags & CTL_BEARG_ASCII)
2177			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2178				snprintf(error_str, error_str_len, "Argument "
2179				    "%d value is not NUL-terminated", i);
2180				goto bailout;
2181			}
2182			args[i].kvalue = tmpptr;
2183		} else {
2184			args[i].kvalue = malloc(args[i].vallen,
2185			    M_CTL, M_WAITOK | M_ZERO);
2186		}
2187	}
2188
2189	return (args);
2190bailout:
2191
2192	ctl_free_args(num_args, args);
2193
2194	return (NULL);
2195}
2196
2197static void
2198ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2199{
2200	int i;
2201
2202	for (i = 0; i < num_args; i++) {
2203		if (args[i].flags & CTL_BEARG_WR)
2204			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2205	}
2206}
2207
2208/*
2209 * Escape characters that are illegal or not recommended in XML.
2210 */
2211int
2212ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2213{
2214	int retval;
2215
2216	retval = 0;
2217
2218	for (; *str; str++) {
2219		switch (*str) {
2220		case '&':
2221			retval = sbuf_printf(sb, "&amp;");
2222			break;
2223		case '>':
2224			retval = sbuf_printf(sb, "&gt;");
2225			break;
2226		case '<':
2227			retval = sbuf_printf(sb, "&lt;");
2228			break;
2229		default:
2230			retval = sbuf_putc(sb, *str);
2231			break;
2232		}
2233
2234		if (retval != 0)
2235			break;
2236
2237	}
2238
2239	return (retval);
2240}
2241
2242static int
2243ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2244	  struct thread *td)
2245{
2246	struct ctl_softc *softc;
2247	int retval;
2248
2249	softc = control_softc;
2250
2251	retval = 0;
2252
2253	switch (cmd) {
2254	case CTL_IO: {
2255		union ctl_io *io;
2256		void *pool_tmp;
2257
2258		/*
2259		 * If we haven't been "enabled", don't allow any SCSI I/O
2260		 * to this FETD.
2261		 */
2262		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2263			retval = EPERM;
2264			break;
2265		}
2266
2267		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2268		if (io == NULL) {
2269			printf("ctl_ioctl: can't allocate ctl_io!\n");
2270			retval = ENOSPC;
2271			break;
2272		}
2273
2274		/*
2275		 * Need to save the pool reference so it doesn't get
2276		 * spammed by the user's ctl_io.
2277		 */
2278		pool_tmp = io->io_hdr.pool;
2279
2280		memcpy(io, (void *)addr, sizeof(*io));
2281
2282		io->io_hdr.pool = pool_tmp;
2283		/*
2284		 * No status yet, so make sure the status is set properly.
2285		 */
2286		io->io_hdr.status = CTL_STATUS_NONE;
2287
2288		/*
2289		 * The user sets the initiator ID, target and LUN IDs.
2290		 */
2291		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2292		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2293		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2294		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2295			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2296
2297		retval = ctl_ioctl_submit_wait(io);
2298
2299		if (retval != 0) {
2300			ctl_free_io(io);
2301			break;
2302		}
2303
2304		memcpy((void *)addr, io, sizeof(*io));
2305
2306		/* return this to our pool */
2307		ctl_free_io(io);
2308
2309		break;
2310	}
2311	case CTL_ENABLE_PORT:
2312	case CTL_DISABLE_PORT:
2313	case CTL_SET_PORT_WWNS: {
2314		struct ctl_port *port;
2315		struct ctl_port_entry *entry;
2316
2317		entry = (struct ctl_port_entry *)addr;
2318
2319		mtx_lock(&softc->ctl_lock);
2320		STAILQ_FOREACH(port, &softc->port_list, links) {
2321			int action, done;
2322
2323			action = 0;
2324			done = 0;
2325
2326			if ((entry->port_type == CTL_PORT_NONE)
2327			 && (entry->targ_port == port->targ_port)) {
2328				/*
2329				 * If the user only wants to enable or
2330				 * disable or set WWNs on a specific port,
2331				 * do the operation and we're done.
2332				 */
2333				action = 1;
2334				done = 1;
2335			} else if (entry->port_type & port->port_type) {
2336				/*
2337				 * Compare the user's type mask with the
2338				 * particular frontend type to see if we
2339				 * have a match.
2340				 */
2341				action = 1;
2342				done = 0;
2343
2344				/*
2345				 * Make sure the user isn't trying to set
2346				 * WWNs on multiple ports at the same time.
2347				 */
2348				if (cmd == CTL_SET_PORT_WWNS) {
2349					printf("%s: Can't set WWNs on "
2350					       "multiple ports\n", __func__);
2351					retval = EINVAL;
2352					break;
2353				}
2354			}
2355			if (action != 0) {
2356				/*
2357				 * XXX KDM we have to drop the lock here,
2358				 * because the online/offline operations
2359				 * can potentially block.  We need to
2360				 * reference count the frontends so they
2361				 * can't go away,
2362				 */
2363				mtx_unlock(&softc->ctl_lock);
2364
2365				if (cmd == CTL_ENABLE_PORT) {
2366					struct ctl_lun *lun;
2367
2368					STAILQ_FOREACH(lun, &softc->lun_list,
2369						       links) {
2370						port->lun_enable(port->targ_lun_arg,
2371						    lun->target,
2372						    lun->lun);
2373					}
2374
2375					ctl_port_online(port);
2376				} else if (cmd == CTL_DISABLE_PORT) {
2377					struct ctl_lun *lun;
2378
2379					ctl_port_offline(port);
2380
2381					STAILQ_FOREACH(lun, &softc->lun_list,
2382						       links) {
2383						port->lun_disable(
2384						    port->targ_lun_arg,
2385						    lun->target,
2386						    lun->lun);
2387					}
2388				}
2389
2390				mtx_lock(&softc->ctl_lock);
2391
2392				if (cmd == CTL_SET_PORT_WWNS)
2393					ctl_port_set_wwns(port,
2394					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2395					    1 : 0, entry->wwnn,
2396					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2397					    1 : 0, entry->wwpn);
2398			}
2399			if (done != 0)
2400				break;
2401		}
2402		mtx_unlock(&softc->ctl_lock);
2403		break;
2404	}
2405	case CTL_GET_PORT_LIST: {
2406		struct ctl_port *port;
2407		struct ctl_port_list *list;
2408		int i;
2409
2410		list = (struct ctl_port_list *)addr;
2411
2412		if (list->alloc_len != (list->alloc_num *
2413		    sizeof(struct ctl_port_entry))) {
2414			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2415			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2416			       "%zu\n", __func__, list->alloc_len,
2417			       list->alloc_num, sizeof(struct ctl_port_entry));
2418			retval = EINVAL;
2419			break;
2420		}
2421		list->fill_len = 0;
2422		list->fill_num = 0;
2423		list->dropped_num = 0;
2424		i = 0;
2425		mtx_lock(&softc->ctl_lock);
2426		STAILQ_FOREACH(port, &softc->port_list, links) {
2427			struct ctl_port_entry entry, *list_entry;
2428
2429			if (list->fill_num >= list->alloc_num) {
2430				list->dropped_num++;
2431				continue;
2432			}
2433
2434			entry.port_type = port->port_type;
2435			strlcpy(entry.port_name, port->port_name,
2436				sizeof(entry.port_name));
2437			entry.targ_port = port->targ_port;
2438			entry.physical_port = port->physical_port;
2439			entry.virtual_port = port->virtual_port;
2440			entry.wwnn = port->wwnn;
2441			entry.wwpn = port->wwpn;
2442			if (port->status & CTL_PORT_STATUS_ONLINE)
2443				entry.online = 1;
2444			else
2445				entry.online = 0;
2446
2447			list_entry = &list->entries[i];
2448
2449			retval = copyout(&entry, list_entry, sizeof(entry));
2450			if (retval != 0) {
2451				printf("%s: CTL_GET_PORT_LIST: copyout "
2452				       "returned %d\n", __func__, retval);
2453				break;
2454			}
2455			i++;
2456			list->fill_num++;
2457			list->fill_len += sizeof(entry);
2458		}
2459		mtx_unlock(&softc->ctl_lock);
2460
2461		/*
2462		 * If this is non-zero, we had a copyout fault, so there's
2463		 * probably no point in attempting to set the status inside
2464		 * the structure.
2465		 */
2466		if (retval != 0)
2467			break;
2468
2469		if (list->dropped_num > 0)
2470			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2471		else
2472			list->status = CTL_PORT_LIST_OK;
2473		break;
2474	}
2475	case CTL_DUMP_OOA: {
2476		struct ctl_lun *lun;
2477		union ctl_io *io;
2478		char printbuf[128];
2479		struct sbuf sb;
2480
2481		mtx_lock(&softc->ctl_lock);
2482		printf("Dumping OOA queues:\n");
2483		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2484			mtx_lock(&lun->lun_lock);
2485			for (io = (union ctl_io *)TAILQ_FIRST(
2486			     &lun->ooa_queue); io != NULL;
2487			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2488			     ooa_links)) {
2489				sbuf_new(&sb, printbuf, sizeof(printbuf),
2490					 SBUF_FIXEDLEN);
2491				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2492					    (intmax_t)lun->lun,
2493					    io->scsiio.tag_num,
2494					    (io->io_hdr.flags &
2495					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2496					    (io->io_hdr.flags &
2497					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2498					    (io->io_hdr.flags &
2499					    CTL_FLAG_ABORT) ? " ABORT" : "",
2500			                    (io->io_hdr.flags &
2501		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2502				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2503				sbuf_finish(&sb);
2504				printf("%s\n", sbuf_data(&sb));
2505			}
2506			mtx_unlock(&lun->lun_lock);
2507		}
2508		printf("OOA queues dump done\n");
2509		mtx_unlock(&softc->ctl_lock);
2510		break;
2511	}
2512	case CTL_GET_OOA: {
2513		struct ctl_lun *lun;
2514		struct ctl_ooa *ooa_hdr;
2515		struct ctl_ooa_entry *entries;
2516		uint32_t cur_fill_num;
2517
2518		ooa_hdr = (struct ctl_ooa *)addr;
2519
2520		if ((ooa_hdr->alloc_len == 0)
2521		 || (ooa_hdr->alloc_num == 0)) {
2522			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2523			       "must be non-zero\n", __func__,
2524			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2525			retval = EINVAL;
2526			break;
2527		}
2528
2529		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2530		    sizeof(struct ctl_ooa_entry))) {
2531			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2532			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2533			       __func__, ooa_hdr->alloc_len,
2534			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2535			retval = EINVAL;
2536			break;
2537		}
2538
2539		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2540		if (entries == NULL) {
2541			printf("%s: could not allocate %d bytes for OOA "
2542			       "dump\n", __func__, ooa_hdr->alloc_len);
2543			retval = ENOMEM;
2544			break;
2545		}
2546
2547		mtx_lock(&softc->ctl_lock);
2548		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2549		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2550		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2551			mtx_unlock(&softc->ctl_lock);
2552			free(entries, M_CTL);
2553			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2554			       __func__, (uintmax_t)ooa_hdr->lun_num);
2555			retval = EINVAL;
2556			break;
2557		}
2558
2559		cur_fill_num = 0;
2560
2561		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2562			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2563				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2564					ooa_hdr, entries);
2565				if (retval != 0)
2566					break;
2567			}
2568			if (retval != 0) {
2569				mtx_unlock(&softc->ctl_lock);
2570				free(entries, M_CTL);
2571				break;
2572			}
2573		} else {
2574			lun = softc->ctl_luns[ooa_hdr->lun_num];
2575
2576			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2577						    entries);
2578		}
2579		mtx_unlock(&softc->ctl_lock);
2580
2581		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2582		ooa_hdr->fill_len = ooa_hdr->fill_num *
2583			sizeof(struct ctl_ooa_entry);
2584		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2585		if (retval != 0) {
2586			printf("%s: error copying out %d bytes for OOA dump\n",
2587			       __func__, ooa_hdr->fill_len);
2588		}
2589
2590		getbintime(&ooa_hdr->cur_bt);
2591
2592		if (cur_fill_num > ooa_hdr->alloc_num) {
2593			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2594			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2595		} else {
2596			ooa_hdr->dropped_num = 0;
2597			ooa_hdr->status = CTL_OOA_OK;
2598		}
2599
2600		free(entries, M_CTL);
2601		break;
2602	}
2603	case CTL_CHECK_OOA: {
2604		union ctl_io *io;
2605		struct ctl_lun *lun;
2606		struct ctl_ooa_info *ooa_info;
2607
2608
2609		ooa_info = (struct ctl_ooa_info *)addr;
2610
2611		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2612			ooa_info->status = CTL_OOA_INVALID_LUN;
2613			break;
2614		}
2615		mtx_lock(&softc->ctl_lock);
2616		lun = softc->ctl_luns[ooa_info->lun_id];
2617		if (lun == NULL) {
2618			mtx_unlock(&softc->ctl_lock);
2619			ooa_info->status = CTL_OOA_INVALID_LUN;
2620			break;
2621		}
2622		mtx_lock(&lun->lun_lock);
2623		mtx_unlock(&softc->ctl_lock);
2624		ooa_info->num_entries = 0;
2625		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2626		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2627		     &io->io_hdr, ooa_links)) {
2628			ooa_info->num_entries++;
2629		}
2630		mtx_unlock(&lun->lun_lock);
2631
2632		ooa_info->status = CTL_OOA_SUCCESS;
2633
2634		break;
2635	}
2636	case CTL_HARD_START:
2637	case CTL_HARD_STOP: {
2638		struct ctl_fe_ioctl_startstop_info ss_info;
2639		struct cfi_metatask *metatask;
2640		struct mtx hs_mtx;
2641
2642		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2643
2644		cv_init(&ss_info.sem, "hard start/stop cv" );
2645
2646		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2647		if (metatask == NULL) {
2648			retval = ENOMEM;
2649			mtx_destroy(&hs_mtx);
2650			break;
2651		}
2652
2653		if (cmd == CTL_HARD_START)
2654			metatask->tasktype = CFI_TASK_STARTUP;
2655		else
2656			metatask->tasktype = CFI_TASK_SHUTDOWN;
2657
2658		metatask->callback = ctl_ioctl_hard_startstop_callback;
2659		metatask->callback_arg = &ss_info;
2660
2661		cfi_action(metatask);
2662
2663		/* Wait for the callback */
2664		mtx_lock(&hs_mtx);
2665		cv_wait_sig(&ss_info.sem, &hs_mtx);
2666		mtx_unlock(&hs_mtx);
2667
2668		/*
2669		 * All information has been copied from the metatask by the
2670		 * time cv_broadcast() is called, so we free the metatask here.
2671		 */
2672		cfi_free_metatask(metatask);
2673
2674		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2675
2676		mtx_destroy(&hs_mtx);
2677		break;
2678	}
2679	case CTL_BBRREAD: {
2680		struct ctl_bbrread_info *bbr_info;
2681		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2682		struct mtx bbr_mtx;
2683		struct cfi_metatask *metatask;
2684
2685		bbr_info = (struct ctl_bbrread_info *)addr;
2686
2687		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2688
2689		bzero(&bbr_mtx, sizeof(bbr_mtx));
2690		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2691
2692		fe_bbr_info.bbr_info = bbr_info;
2693		fe_bbr_info.lock = &bbr_mtx;
2694
2695		cv_init(&fe_bbr_info.sem, "BBR read cv");
2696		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2697
2698		if (metatask == NULL) {
2699			mtx_destroy(&bbr_mtx);
2700			cv_destroy(&fe_bbr_info.sem);
2701			retval = ENOMEM;
2702			break;
2703		}
2704		metatask->tasktype = CFI_TASK_BBRREAD;
2705		metatask->callback = ctl_ioctl_bbrread_callback;
2706		metatask->callback_arg = &fe_bbr_info;
2707		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2708		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2709		metatask->taskinfo.bbrread.len = bbr_info->len;
2710
2711		cfi_action(metatask);
2712
2713		mtx_lock(&bbr_mtx);
2714		while (fe_bbr_info.wakeup_done == 0)
2715			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2716		mtx_unlock(&bbr_mtx);
2717
2718		bbr_info->status = metatask->status;
2719		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2720		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2721		memcpy(&bbr_info->sense_data,
2722		       &metatask->taskinfo.bbrread.sense_data,
2723		       ctl_min(sizeof(bbr_info->sense_data),
2724			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2725
2726		cfi_free_metatask(metatask);
2727
2728		mtx_destroy(&bbr_mtx);
2729		cv_destroy(&fe_bbr_info.sem);
2730
2731		break;
2732	}
2733	case CTL_DELAY_IO: {
2734		struct ctl_io_delay_info *delay_info;
2735#ifdef CTL_IO_DELAY
2736		struct ctl_lun *lun;
2737#endif /* CTL_IO_DELAY */
2738
2739		delay_info = (struct ctl_io_delay_info *)addr;
2740
2741#ifdef CTL_IO_DELAY
2742		mtx_lock(&softc->ctl_lock);
2743
2744		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2745		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2746			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2747		} else {
2748			lun = softc->ctl_luns[delay_info->lun_id];
2749			mtx_lock(&lun->lun_lock);
2750
2751			delay_info->status = CTL_DELAY_STATUS_OK;
2752
2753			switch (delay_info->delay_type) {
2754			case CTL_DELAY_TYPE_CONT:
2755				break;
2756			case CTL_DELAY_TYPE_ONESHOT:
2757				break;
2758			default:
2759				delay_info->status =
2760					CTL_DELAY_STATUS_INVALID_TYPE;
2761				break;
2762			}
2763
2764			switch (delay_info->delay_loc) {
2765			case CTL_DELAY_LOC_DATAMOVE:
2766				lun->delay_info.datamove_type =
2767					delay_info->delay_type;
2768				lun->delay_info.datamove_delay =
2769					delay_info->delay_secs;
2770				break;
2771			case CTL_DELAY_LOC_DONE:
2772				lun->delay_info.done_type =
2773					delay_info->delay_type;
2774				lun->delay_info.done_delay =
2775					delay_info->delay_secs;
2776				break;
2777			default:
2778				delay_info->status =
2779					CTL_DELAY_STATUS_INVALID_LOC;
2780				break;
2781			}
2782			mtx_unlock(&lun->lun_lock);
2783		}
2784
2785		mtx_unlock(&softc->ctl_lock);
2786#else
2787		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2788#endif /* CTL_IO_DELAY */
2789		break;
2790	}
2791	case CTL_REALSYNC_SET: {
2792		int *syncstate;
2793
2794		syncstate = (int *)addr;
2795
2796		mtx_lock(&softc->ctl_lock);
2797		switch (*syncstate) {
2798		case 0:
2799			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2800			break;
2801		case 1:
2802			softc->flags |= CTL_FLAG_REAL_SYNC;
2803			break;
2804		default:
2805			retval = EINVAL;
2806			break;
2807		}
2808		mtx_unlock(&softc->ctl_lock);
2809		break;
2810	}
2811	case CTL_REALSYNC_GET: {
2812		int *syncstate;
2813
2814		syncstate = (int*)addr;
2815
2816		mtx_lock(&softc->ctl_lock);
2817		if (softc->flags & CTL_FLAG_REAL_SYNC)
2818			*syncstate = 1;
2819		else
2820			*syncstate = 0;
2821		mtx_unlock(&softc->ctl_lock);
2822
2823		break;
2824	}
2825	case CTL_SETSYNC:
2826	case CTL_GETSYNC: {
2827		struct ctl_sync_info *sync_info;
2828		struct ctl_lun *lun;
2829
2830		sync_info = (struct ctl_sync_info *)addr;
2831
2832		mtx_lock(&softc->ctl_lock);
2833		lun = softc->ctl_luns[sync_info->lun_id];
2834		if (lun == NULL) {
2835			mtx_unlock(&softc->ctl_lock);
2836			sync_info->status = CTL_GS_SYNC_NO_LUN;
2837		}
2838		/*
2839		 * Get or set the sync interval.  We're not bounds checking
2840		 * in the set case, hopefully the user won't do something
2841		 * silly.
2842		 */
2843		mtx_lock(&lun->lun_lock);
2844		mtx_unlock(&softc->ctl_lock);
2845		if (cmd == CTL_GETSYNC)
2846			sync_info->sync_interval = lun->sync_interval;
2847		else
2848			lun->sync_interval = sync_info->sync_interval;
2849		mtx_unlock(&lun->lun_lock);
2850
2851		sync_info->status = CTL_GS_SYNC_OK;
2852
2853		break;
2854	}
2855	case CTL_GETSTATS: {
2856		struct ctl_stats *stats;
2857		struct ctl_lun *lun;
2858		int i;
2859
2860		stats = (struct ctl_stats *)addr;
2861
2862		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2863		     stats->alloc_len) {
2864			stats->status = CTL_SS_NEED_MORE_SPACE;
2865			stats->num_luns = softc->num_luns;
2866			break;
2867		}
2868		/*
2869		 * XXX KDM no locking here.  If the LUN list changes,
2870		 * things can blow up.
2871		 */
2872		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2873		     i++, lun = STAILQ_NEXT(lun, links)) {
2874			retval = copyout(&lun->stats, &stats->lun_stats[i],
2875					 sizeof(lun->stats));
2876			if (retval != 0)
2877				break;
2878		}
2879		stats->num_luns = softc->num_luns;
2880		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2881				 softc->num_luns;
2882		stats->status = CTL_SS_OK;
2883#ifdef CTL_TIME_IO
2884		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2885#else
2886		stats->flags = CTL_STATS_FLAG_NONE;
2887#endif
2888		getnanouptime(&stats->timestamp);
2889		break;
2890	}
2891	case CTL_ERROR_INJECT: {
2892		struct ctl_error_desc *err_desc, *new_err_desc;
2893		struct ctl_lun *lun;
2894
2895		err_desc = (struct ctl_error_desc *)addr;
2896
2897		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2898				      M_WAITOK | M_ZERO);
2899		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2900
2901		mtx_lock(&softc->ctl_lock);
2902		lun = softc->ctl_luns[err_desc->lun_id];
2903		if (lun == NULL) {
2904			mtx_unlock(&softc->ctl_lock);
2905			free(new_err_desc, M_CTL);
2906			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2907			       __func__, (uintmax_t)err_desc->lun_id);
2908			retval = EINVAL;
2909			break;
2910		}
2911		mtx_lock(&lun->lun_lock);
2912		mtx_unlock(&softc->ctl_lock);
2913
2914		/*
2915		 * We could do some checking here to verify the validity
2916		 * of the request, but given the complexity of error
2917		 * injection requests, the checking logic would be fairly
2918		 * complex.
2919		 *
2920		 * For now, if the request is invalid, it just won't get
2921		 * executed and might get deleted.
2922		 */
2923		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2924
2925		/*
2926		 * XXX KDM check to make sure the serial number is unique,
2927		 * in case we somehow manage to wrap.  That shouldn't
2928		 * happen for a very long time, but it's the right thing to
2929		 * do.
2930		 */
2931		new_err_desc->serial = lun->error_serial;
2932		err_desc->serial = lun->error_serial;
2933		lun->error_serial++;
2934
2935		mtx_unlock(&lun->lun_lock);
2936		break;
2937	}
2938	case CTL_ERROR_INJECT_DELETE: {
2939		struct ctl_error_desc *delete_desc, *desc, *desc2;
2940		struct ctl_lun *lun;
2941		int delete_done;
2942
2943		delete_desc = (struct ctl_error_desc *)addr;
2944		delete_done = 0;
2945
2946		mtx_lock(&softc->ctl_lock);
2947		lun = softc->ctl_luns[delete_desc->lun_id];
2948		if (lun == NULL) {
2949			mtx_unlock(&softc->ctl_lock);
2950			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2951			       __func__, (uintmax_t)delete_desc->lun_id);
2952			retval = EINVAL;
2953			break;
2954		}
2955		mtx_lock(&lun->lun_lock);
2956		mtx_unlock(&softc->ctl_lock);
2957		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2958			if (desc->serial != delete_desc->serial)
2959				continue;
2960
2961			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2962				      links);
2963			free(desc, M_CTL);
2964			delete_done = 1;
2965		}
2966		mtx_unlock(&lun->lun_lock);
2967		if (delete_done == 0) {
2968			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2969			       "error serial %ju on LUN %u\n", __func__,
2970			       delete_desc->serial, delete_desc->lun_id);
2971			retval = EINVAL;
2972			break;
2973		}
2974		break;
2975	}
2976	case CTL_DUMP_STRUCTS: {
2977		int i, j, k, idx;
2978		struct ctl_port *port;
2979		struct ctl_frontend *fe;
2980
2981		mtx_lock(&softc->ctl_lock);
2982		printf("CTL Persistent Reservation information start:\n");
2983		for (i = 0; i < CTL_MAX_LUNS; i++) {
2984			struct ctl_lun *lun;
2985
2986			lun = softc->ctl_luns[i];
2987
2988			if ((lun == NULL)
2989			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2990				continue;
2991
2992			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2993				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2994					idx = j * CTL_MAX_INIT_PER_PORT + k;
2995					if (lun->per_res[idx].registered == 0)
2996						continue;
2997					printf("  LUN %d port %d iid %d key "
2998					       "%#jx\n", i, j, k,
2999					       (uintmax_t)scsi_8btou64(
3000					       lun->per_res[idx].res_key.key));
3001				}
3002			}
3003		}
3004		printf("CTL Persistent Reservation information end\n");
3005		printf("CTL Ports:\n");
3006		STAILQ_FOREACH(port, &softc->port_list, links) {
3007			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3008			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3009			       port->frontend->name, port->port_type,
3010			       port->physical_port, port->virtual_port,
3011			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3012			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3013				if (port->wwpn_iid[j].in_use == 0 &&
3014				    port->wwpn_iid[j].wwpn == 0 &&
3015				    port->wwpn_iid[j].name == NULL)
3016					continue;
3017
3018				printf("    iid %u use %d WWPN %#jx '%s'\n",
3019				    j, port->wwpn_iid[j].in_use,
3020				    (uintmax_t)port->wwpn_iid[j].wwpn,
3021				    port->wwpn_iid[j].name);
3022			}
3023		}
3024		printf("CTL Port information end\n");
3025		mtx_unlock(&softc->ctl_lock);
3026		/*
3027		 * XXX KDM calling this without a lock.  We'd likely want
3028		 * to drop the lock before calling the frontend's dump
3029		 * routine anyway.
3030		 */
3031		printf("CTL Frontends:\n");
3032		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3033			printf("  Frontend '%s'\n", fe->name);
3034			if (fe->fe_dump != NULL)
3035				fe->fe_dump();
3036		}
3037		printf("CTL Frontend information end\n");
3038		break;
3039	}
3040	case CTL_LUN_REQ: {
3041		struct ctl_lun_req *lun_req;
3042		struct ctl_backend_driver *backend;
3043
3044		lun_req = (struct ctl_lun_req *)addr;
3045
3046		backend = ctl_backend_find(lun_req->backend);
3047		if (backend == NULL) {
3048			lun_req->status = CTL_LUN_ERROR;
3049			snprintf(lun_req->error_str,
3050				 sizeof(lun_req->error_str),
3051				 "Backend \"%s\" not found.",
3052				 lun_req->backend);
3053			break;
3054		}
3055		if (lun_req->num_be_args > 0) {
3056			lun_req->kern_be_args = ctl_copyin_args(
3057				lun_req->num_be_args,
3058				lun_req->be_args,
3059				lun_req->error_str,
3060				sizeof(lun_req->error_str));
3061			if (lun_req->kern_be_args == NULL) {
3062				lun_req->status = CTL_LUN_ERROR;
3063				break;
3064			}
3065		}
3066
3067		retval = backend->ioctl(dev, cmd, addr, flag, td);
3068
3069		if (lun_req->num_be_args > 0) {
3070			ctl_copyout_args(lun_req->num_be_args,
3071				      lun_req->kern_be_args);
3072			ctl_free_args(lun_req->num_be_args,
3073				      lun_req->kern_be_args);
3074		}
3075		break;
3076	}
3077	case CTL_LUN_LIST: {
3078		struct sbuf *sb;
3079		struct ctl_lun *lun;
3080		struct ctl_lun_list *list;
3081		struct ctl_option *opt;
3082
3083		list = (struct ctl_lun_list *)addr;
3084
3085		/*
3086		 * Allocate a fixed length sbuf here, based on the length
3087		 * of the user's buffer.  We could allocate an auto-extending
3088		 * buffer, and then tell the user how much larger our
3089		 * amount of data is than his buffer, but that presents
3090		 * some problems:
3091		 *
3092		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3093		 *     we can't hold a lock while calling them with an
3094		 *     auto-extending buffer.
3095 		 *
3096		 * 2.  There is not currently a LUN reference counting
3097		 *     mechanism, outside of outstanding transactions on
3098		 *     the LUN's OOA queue.  So a LUN could go away on us
3099		 *     while we're getting the LUN number, backend-specific
3100		 *     information, etc.  Thus, given the way things
3101		 *     currently work, we need to hold the CTL lock while
3102		 *     grabbing LUN information.
3103		 *
3104		 * So, from the user's standpoint, the best thing to do is
3105		 * allocate what he thinks is a reasonable buffer length,
3106		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3107		 * double the buffer length and try again.  (And repeat
3108		 * that until he succeeds.)
3109		 */
3110		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3111		if (sb == NULL) {
3112			list->status = CTL_LUN_LIST_ERROR;
3113			snprintf(list->error_str, sizeof(list->error_str),
3114				 "Unable to allocate %d bytes for LUN list",
3115				 list->alloc_len);
3116			break;
3117		}
3118
3119		sbuf_printf(sb, "<ctllunlist>\n");
3120
3121		mtx_lock(&softc->ctl_lock);
3122		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3123			mtx_lock(&lun->lun_lock);
3124			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3125					     (uintmax_t)lun->lun);
3126
3127			/*
3128			 * Bail out as soon as we see that we've overfilled
3129			 * the buffer.
3130			 */
3131			if (retval != 0)
3132				break;
3133
3134			retval = sbuf_printf(sb, "\t<backend_type>%s"
3135					     "</backend_type>\n",
3136					     (lun->backend == NULL) ?  "none" :
3137					     lun->backend->name);
3138
3139			if (retval != 0)
3140				break;
3141
3142			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3143					     lun->be_lun->lun_type);
3144
3145			if (retval != 0)
3146				break;
3147
3148			if (lun->backend == NULL) {
3149				retval = sbuf_printf(sb, "</lun>\n");
3150				if (retval != 0)
3151					break;
3152				continue;
3153			}
3154
3155			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3156					     (lun->be_lun->maxlba > 0) ?
3157					     lun->be_lun->maxlba + 1 : 0);
3158
3159			if (retval != 0)
3160				break;
3161
3162			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3163					     lun->be_lun->blocksize);
3164
3165			if (retval != 0)
3166				break;
3167
3168			retval = sbuf_printf(sb, "\t<serial_number>");
3169
3170			if (retval != 0)
3171				break;
3172
3173			retval = ctl_sbuf_printf_esc(sb,
3174						     lun->be_lun->serial_num);
3175
3176			if (retval != 0)
3177				break;
3178
3179			retval = sbuf_printf(sb, "</serial_number>\n");
3180
3181			if (retval != 0)
3182				break;
3183
3184			retval = sbuf_printf(sb, "\t<device_id>");
3185
3186			if (retval != 0)
3187				break;
3188
3189			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3190
3191			if (retval != 0)
3192				break;
3193
3194			retval = sbuf_printf(sb, "</device_id>\n");
3195
3196			if (retval != 0)
3197				break;
3198
3199			if (lun->backend->lun_info != NULL) {
3200				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3201				if (retval != 0)
3202					break;
3203			}
3204			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3205				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3206				    opt->name, opt->value, opt->name);
3207				if (retval != 0)
3208					break;
3209			}
3210
3211			retval = sbuf_printf(sb, "</lun>\n");
3212
3213			if (retval != 0)
3214				break;
3215			mtx_unlock(&lun->lun_lock);
3216		}
3217		if (lun != NULL)
3218			mtx_unlock(&lun->lun_lock);
3219		mtx_unlock(&softc->ctl_lock);
3220
3221		if ((retval != 0)
3222		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3223			retval = 0;
3224			sbuf_delete(sb);
3225			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3226			snprintf(list->error_str, sizeof(list->error_str),
3227				 "Out of space, %d bytes is too small",
3228				 list->alloc_len);
3229			break;
3230		}
3231
3232		sbuf_finish(sb);
3233
3234		retval = copyout(sbuf_data(sb), list->lun_xml,
3235				 sbuf_len(sb) + 1);
3236
3237		list->fill_len = sbuf_len(sb) + 1;
3238		list->status = CTL_LUN_LIST_OK;
3239		sbuf_delete(sb);
3240		break;
3241	}
3242	case CTL_ISCSI: {
3243		struct ctl_iscsi *ci;
3244		struct ctl_frontend *fe;
3245
3246		ci = (struct ctl_iscsi *)addr;
3247
3248		fe = ctl_frontend_find("iscsi");
3249		if (fe == NULL) {
3250			ci->status = CTL_ISCSI_ERROR;
3251			snprintf(ci->error_str, sizeof(ci->error_str),
3252			    "Frontend \"iscsi\" not found.");
3253			break;
3254		}
3255
3256		retval = fe->ioctl(dev, cmd, addr, flag, td);
3257		break;
3258	}
3259	case CTL_PORT_REQ: {
3260		struct ctl_req *req;
3261		struct ctl_frontend *fe;
3262
3263		req = (struct ctl_req *)addr;
3264
3265		fe = ctl_frontend_find(req->driver);
3266		if (fe == NULL) {
3267			req->status = CTL_LUN_ERROR;
3268			snprintf(req->error_str, sizeof(req->error_str),
3269			    "Frontend \"%s\" not found.", req->driver);
3270			break;
3271		}
3272		if (req->num_args > 0) {
3273			req->kern_args = ctl_copyin_args(req->num_args,
3274			    req->args, req->error_str, sizeof(req->error_str));
3275			if (req->kern_args == NULL) {
3276				req->status = CTL_LUN_ERROR;
3277				break;
3278			}
3279		}
3280
3281		retval = fe->ioctl(dev, cmd, addr, flag, td);
3282
3283		if (req->num_args > 0) {
3284			ctl_copyout_args(req->num_args, req->kern_args);
3285			ctl_free_args(req->num_args, req->kern_args);
3286		}
3287		break;
3288	}
3289	case CTL_PORT_LIST: {
3290		struct sbuf *sb;
3291		struct ctl_port *port;
3292		struct ctl_lun_list *list;
3293		struct ctl_option *opt;
3294
3295		list = (struct ctl_lun_list *)addr;
3296
3297		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3298		if (sb == NULL) {
3299			list->status = CTL_LUN_LIST_ERROR;
3300			snprintf(list->error_str, sizeof(list->error_str),
3301				 "Unable to allocate %d bytes for LUN list",
3302				 list->alloc_len);
3303			break;
3304		}
3305
3306		sbuf_printf(sb, "<ctlportlist>\n");
3307
3308		mtx_lock(&softc->ctl_lock);
3309		STAILQ_FOREACH(port, &softc->port_list, links) {
3310			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3311					     (uintmax_t)port->targ_port);
3312
3313			/*
3314			 * Bail out as soon as we see that we've overfilled
3315			 * the buffer.
3316			 */
3317			if (retval != 0)
3318				break;
3319
3320			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3321			    "</frontend_type>\n", port->frontend->name);
3322			if (retval != 0)
3323				break;
3324
3325			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3326					     port->port_type);
3327			if (retval != 0)
3328				break;
3329
3330			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3331			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3332			if (retval != 0)
3333				break;
3334
3335			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3336			    port->port_name);
3337			if (retval != 0)
3338				break;
3339
3340			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3341			    port->physical_port);
3342			if (retval != 0)
3343				break;
3344
3345			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3346			    port->virtual_port);
3347			if (retval != 0)
3348				break;
3349
3350			retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3351			    (uintmax_t)port->wwnn);
3352			if (retval != 0)
3353				break;
3354
3355			retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3356			    (uintmax_t)port->wwpn);
3357			if (retval != 0)
3358				break;
3359
3360			if (port->port_info != NULL) {
3361				retval = port->port_info(port->onoff_arg, sb);
3362				if (retval != 0)
3363					break;
3364			}
3365			STAILQ_FOREACH(opt, &port->options, links) {
3366				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3367				    opt->name, opt->value, opt->name);
3368				if (retval != 0)
3369					break;
3370			}
3371
3372			retval = sbuf_printf(sb, "</targ_port>\n");
3373			if (retval != 0)
3374				break;
3375		}
3376		mtx_unlock(&softc->ctl_lock);
3377
3378		if ((retval != 0)
3379		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3380			retval = 0;
3381			sbuf_delete(sb);
3382			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3383			snprintf(list->error_str, sizeof(list->error_str),
3384				 "Out of space, %d bytes is too small",
3385				 list->alloc_len);
3386			break;
3387		}
3388
3389		sbuf_finish(sb);
3390
3391		retval = copyout(sbuf_data(sb), list->lun_xml,
3392				 sbuf_len(sb) + 1);
3393
3394		list->fill_len = sbuf_len(sb) + 1;
3395		list->status = CTL_LUN_LIST_OK;
3396		sbuf_delete(sb);
3397		break;
3398	}
3399	default: {
3400		/* XXX KDM should we fix this? */
3401#if 0
3402		struct ctl_backend_driver *backend;
3403		unsigned int type;
3404		int found;
3405
3406		found = 0;
3407
3408		/*
3409		 * We encode the backend type as the ioctl type for backend
3410		 * ioctls.  So parse it out here, and then search for a
3411		 * backend of this type.
3412		 */
3413		type = _IOC_TYPE(cmd);
3414
3415		STAILQ_FOREACH(backend, &softc->be_list, links) {
3416			if (backend->type == type) {
3417				found = 1;
3418				break;
3419			}
3420		}
3421		if (found == 0) {
3422			printf("ctl: unknown ioctl command %#lx or backend "
3423			       "%d\n", cmd, type);
3424			retval = EINVAL;
3425			break;
3426		}
3427		retval = backend->ioctl(dev, cmd, addr, flag, td);
3428#endif
3429		retval = ENOTTY;
3430		break;
3431	}
3432	}
3433	return (retval);
3434}
3435
3436uint32_t
3437ctl_get_initindex(struct ctl_nexus *nexus)
3438{
3439	if (nexus->targ_port < CTL_MAX_PORTS)
3440		return (nexus->initid.id +
3441			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3442	else
3443		return (nexus->initid.id +
3444		       ((nexus->targ_port - CTL_MAX_PORTS) *
3445			CTL_MAX_INIT_PER_PORT));
3446}
3447
3448uint32_t
3449ctl_get_resindex(struct ctl_nexus *nexus)
3450{
3451	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3452}
3453
3454uint32_t
3455ctl_port_idx(int port_num)
3456{
3457	if (port_num < CTL_MAX_PORTS)
3458		return(port_num);
3459	else
3460		return(port_num - CTL_MAX_PORTS);
3461}
3462
3463static uint32_t
3464ctl_map_lun(int port_num, uint32_t lun_id)
3465{
3466	struct ctl_port *port;
3467
3468	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3469	if (port == NULL)
3470		return (UINT32_MAX);
3471	if (port->lun_map == NULL)
3472		return (lun_id);
3473	return (port->lun_map(port->targ_lun_arg, lun_id));
3474}
3475
3476static uint32_t
3477ctl_map_lun_back(int port_num, uint32_t lun_id)
3478{
3479	struct ctl_port *port;
3480	uint32_t i;
3481
3482	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3483	if (port->lun_map == NULL)
3484		return (lun_id);
3485	for (i = 0; i < CTL_MAX_LUNS; i++) {
3486		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3487			return (i);
3488	}
3489	return (UINT32_MAX);
3490}
3491
3492/*
3493 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3494 * that are a power of 2.
3495 */
3496int
3497ctl_ffz(uint32_t *mask, uint32_t size)
3498{
3499	uint32_t num_chunks, num_pieces;
3500	int i, j;
3501
3502	num_chunks = (size >> 5);
3503	if (num_chunks == 0)
3504		num_chunks++;
3505	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3506
3507	for (i = 0; i < num_chunks; i++) {
3508		for (j = 0; j < num_pieces; j++) {
3509			if ((mask[i] & (1 << j)) == 0)
3510				return ((i << 5) + j);
3511		}
3512	}
3513
3514	return (-1);
3515}
3516
3517int
3518ctl_set_mask(uint32_t *mask, uint32_t bit)
3519{
3520	uint32_t chunk, piece;
3521
3522	chunk = bit >> 5;
3523	piece = bit % (sizeof(uint32_t) * 8);
3524
3525	if ((mask[chunk] & (1 << piece)) != 0)
3526		return (-1);
3527	else
3528		mask[chunk] |= (1 << piece);
3529
3530	return (0);
3531}
3532
3533int
3534ctl_clear_mask(uint32_t *mask, uint32_t bit)
3535{
3536	uint32_t chunk, piece;
3537
3538	chunk = bit >> 5;
3539	piece = bit % (sizeof(uint32_t) * 8);
3540
3541	if ((mask[chunk] & (1 << piece)) == 0)
3542		return (-1);
3543	else
3544		mask[chunk] &= ~(1 << piece);
3545
3546	return (0);
3547}
3548
3549int
3550ctl_is_set(uint32_t *mask, uint32_t bit)
3551{
3552	uint32_t chunk, piece;
3553
3554	chunk = bit >> 5;
3555	piece = bit % (sizeof(uint32_t) * 8);
3556
3557	if ((mask[chunk] & (1 << piece)) == 0)
3558		return (0);
3559	else
3560		return (1);
3561}
3562
3563#ifdef unused
3564/*
3565 * The bus, target and lun are optional, they can be filled in later.
3566 * can_wait is used to determine whether we can wait on the malloc or not.
3567 */
3568union ctl_io*
3569ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3570	      uint32_t targ_lun, int can_wait)
3571{
3572	union ctl_io *io;
3573
3574	if (can_wait)
3575		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3576	else
3577		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3578
3579	if (io != NULL) {
3580		io->io_hdr.io_type = io_type;
3581		io->io_hdr.targ_port = targ_port;
3582		/*
3583		 * XXX KDM this needs to change/go away.  We need to move
3584		 * to a preallocated pool of ctl_scsiio structures.
3585		 */
3586		io->io_hdr.nexus.targ_target.id = targ_target;
3587		io->io_hdr.nexus.targ_lun = targ_lun;
3588	}
3589
3590	return (io);
3591}
3592
3593void
3594ctl_kfree_io(union ctl_io *io)
3595{
3596	free(io, M_CTL);
3597}
3598#endif /* unused */
3599
3600/*
3601 * ctl_softc, pool_type, total_ctl_io are passed in.
3602 * npool is passed out.
3603 */
3604int
3605ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3606		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3607{
3608	uint32_t i;
3609	union ctl_io *cur_io, *next_io;
3610	struct ctl_io_pool *pool;
3611	int retval;
3612
3613	retval = 0;
3614
3615	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3616					    M_NOWAIT | M_ZERO);
3617	if (pool == NULL) {
3618		retval = ENOMEM;
3619		goto bailout;
3620	}
3621
3622	pool->type = pool_type;
3623	pool->ctl_softc = ctl_softc;
3624
3625	mtx_lock(&ctl_softc->pool_lock);
3626	pool->id = ctl_softc->cur_pool_id++;
3627	mtx_unlock(&ctl_softc->pool_lock);
3628
3629	pool->flags = CTL_POOL_FLAG_NONE;
3630	pool->refcount = 1;		/* Reference for validity. */
3631	STAILQ_INIT(&pool->free_queue);
3632
3633	/*
3634	 * XXX KDM other options here:
3635	 * - allocate a page at a time
3636	 * - allocate one big chunk of memory.
3637	 * Page allocation might work well, but would take a little more
3638	 * tracking.
3639	 */
3640	for (i = 0; i < total_ctl_io; i++) {
3641		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3642						M_NOWAIT);
3643		if (cur_io == NULL) {
3644			retval = ENOMEM;
3645			break;
3646		}
3647		cur_io->io_hdr.pool = pool;
3648		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3649		pool->total_ctl_io++;
3650		pool->free_ctl_io++;
3651	}
3652
3653	if (retval != 0) {
3654		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3655		     cur_io != NULL; cur_io = next_io) {
3656			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3657							      links);
3658			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3659				      ctl_io_hdr, links);
3660			free(cur_io, M_CTLIO);
3661		}
3662
3663		free(pool, M_CTL);
3664		goto bailout;
3665	}
3666	mtx_lock(&ctl_softc->pool_lock);
3667	ctl_softc->num_pools++;
3668	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3669	/*
3670	 * Increment our usage count if this is an external consumer, so we
3671	 * can't get unloaded until the external consumer (most likely a
3672	 * FETD) unloads and frees his pool.
3673	 *
3674	 * XXX KDM will this increment the caller's module use count, or
3675	 * mine?
3676	 */
3677#if 0
3678	if ((pool_type != CTL_POOL_EMERGENCY)
3679	 && (pool_type != CTL_POOL_INTERNAL)
3680	 && (pool_type != CTL_POOL_4OTHERSC))
3681		MOD_INC_USE_COUNT;
3682#endif
3683
3684	mtx_unlock(&ctl_softc->pool_lock);
3685
3686	*npool = pool;
3687
3688bailout:
3689
3690	return (retval);
3691}
3692
3693static int
3694ctl_pool_acquire(struct ctl_io_pool *pool)
3695{
3696
3697	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3698
3699	if (pool->flags & CTL_POOL_FLAG_INVALID)
3700		return (EINVAL);
3701
3702	pool->refcount++;
3703
3704	return (0);
3705}
3706
3707static void
3708ctl_pool_release(struct ctl_io_pool *pool)
3709{
3710	struct ctl_softc *ctl_softc = pool->ctl_softc;
3711	union ctl_io *io;
3712
3713	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3714
3715	if (--pool->refcount != 0)
3716		return;
3717
3718	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3719		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3720			      links);
3721		free(io, M_CTLIO);
3722	}
3723
3724	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3725	ctl_softc->num_pools--;
3726
3727	/*
3728	 * XXX KDM will this decrement the caller's usage count or mine?
3729	 */
3730#if 0
3731	if ((pool->type != CTL_POOL_EMERGENCY)
3732	 && (pool->type != CTL_POOL_INTERNAL)
3733	 && (pool->type != CTL_POOL_4OTHERSC))
3734		MOD_DEC_USE_COUNT;
3735#endif
3736
3737	free(pool, M_CTL);
3738}
3739
3740void
3741ctl_pool_free(struct ctl_io_pool *pool)
3742{
3743	struct ctl_softc *ctl_softc;
3744
3745	if (pool == NULL)
3746		return;
3747
3748	ctl_softc = pool->ctl_softc;
3749	mtx_lock(&ctl_softc->pool_lock);
3750	pool->flags |= CTL_POOL_FLAG_INVALID;
3751	ctl_pool_release(pool);
3752	mtx_unlock(&ctl_softc->pool_lock);
3753}
3754
3755/*
3756 * This routine does not block (except for spinlocks of course).
3757 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3758 * possible.
3759 */
3760union ctl_io *
3761ctl_alloc_io(void *pool_ref)
3762{
3763	union ctl_io *io;
3764	struct ctl_softc *ctl_softc;
3765	struct ctl_io_pool *pool, *npool;
3766	struct ctl_io_pool *emergency_pool;
3767
3768	pool = (struct ctl_io_pool *)pool_ref;
3769
3770	if (pool == NULL) {
3771		printf("%s: pool is NULL\n", __func__);
3772		return (NULL);
3773	}
3774
3775	emergency_pool = NULL;
3776
3777	ctl_softc = pool->ctl_softc;
3778
3779	mtx_lock(&ctl_softc->pool_lock);
3780	/*
3781	 * First, try to get the io structure from the user's pool.
3782	 */
3783	if (ctl_pool_acquire(pool) == 0) {
3784		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3785		if (io != NULL) {
3786			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3787			pool->total_allocated++;
3788			pool->free_ctl_io--;
3789			mtx_unlock(&ctl_softc->pool_lock);
3790			return (io);
3791		} else
3792			ctl_pool_release(pool);
3793	}
3794	/*
3795	 * If he doesn't have any io structures left, search for an
3796	 * emergency pool and grab one from there.
3797	 */
3798	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3799		if (npool->type != CTL_POOL_EMERGENCY)
3800			continue;
3801
3802		if (ctl_pool_acquire(npool) != 0)
3803			continue;
3804
3805		emergency_pool = npool;
3806
3807		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3808		if (io != NULL) {
3809			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3810			npool->total_allocated++;
3811			npool->free_ctl_io--;
3812			mtx_unlock(&ctl_softc->pool_lock);
3813			return (io);
3814		} else
3815			ctl_pool_release(npool);
3816	}
3817
3818	/* Drop the spinlock before we malloc */
3819	mtx_unlock(&ctl_softc->pool_lock);
3820
3821	/*
3822	 * The emergency pool (if it exists) didn't have one, so try an
3823	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3824	 */
3825	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3826	if (io != NULL) {
3827		/*
3828		 * If the emergency pool exists but is empty, add this
3829		 * ctl_io to its list when it gets freed.
3830		 */
3831		if (emergency_pool != NULL) {
3832			mtx_lock(&ctl_softc->pool_lock);
3833			if (ctl_pool_acquire(emergency_pool) == 0) {
3834				io->io_hdr.pool = emergency_pool;
3835				emergency_pool->total_ctl_io++;
3836				/*
3837				 * Need to bump this, otherwise
3838				 * total_allocated and total_freed won't
3839				 * match when we no longer have anything
3840				 * outstanding.
3841				 */
3842				emergency_pool->total_allocated++;
3843			}
3844			mtx_unlock(&ctl_softc->pool_lock);
3845		} else
3846			io->io_hdr.pool = NULL;
3847	}
3848
3849	return (io);
3850}
3851
3852void
3853ctl_free_io(union ctl_io *io)
3854{
3855	if (io == NULL)
3856		return;
3857
3858	/*
3859	 * If this ctl_io has a pool, return it to that pool.
3860	 */
3861	if (io->io_hdr.pool != NULL) {
3862		struct ctl_io_pool *pool;
3863
3864		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3865		mtx_lock(&pool->ctl_softc->pool_lock);
3866		io->io_hdr.io_type = 0xff;
3867		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3868		pool->total_freed++;
3869		pool->free_ctl_io++;
3870		ctl_pool_release(pool);
3871		mtx_unlock(&pool->ctl_softc->pool_lock);
3872	} else {
3873		/*
3874		 * Otherwise, just free it.  We probably malloced it and
3875		 * the emergency pool wasn't available.
3876		 */
3877		free(io, M_CTLIO);
3878	}
3879
3880}
3881
3882void
3883ctl_zero_io(union ctl_io *io)
3884{
3885	void *pool_ref;
3886
3887	if (io == NULL)
3888		return;
3889
3890	/*
3891	 * May need to preserve linked list pointers at some point too.
3892	 */
3893	pool_ref = io->io_hdr.pool;
3894
3895	memset(io, 0, sizeof(*io));
3896
3897	io->io_hdr.pool = pool_ref;
3898}
3899
3900/*
3901 * This routine is currently used for internal copies of ctl_ios that need
3902 * to persist for some reason after we've already returned status to the
3903 * FETD.  (Thus the flag set.)
3904 *
3905 * XXX XXX
3906 * Note that this makes a blind copy of all fields in the ctl_io, except
3907 * for the pool reference.  This includes any memory that has been
3908 * allocated!  That memory will no longer be valid after done has been
3909 * called, so this would be VERY DANGEROUS for command that actually does
3910 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3911 * start and stop commands, which don't transfer any data, so this is not a
3912 * problem.  If it is used for anything else, the caller would also need to
3913 * allocate data buffer space and this routine would need to be modified to
3914 * copy the data buffer(s) as well.
3915 */
3916void
3917ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3918{
3919	void *pool_ref;
3920
3921	if ((src == NULL)
3922	 || (dest == NULL))
3923		return;
3924
3925	/*
3926	 * May need to preserve linked list pointers at some point too.
3927	 */
3928	pool_ref = dest->io_hdr.pool;
3929
3930	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3931
3932	dest->io_hdr.pool = pool_ref;
3933	/*
3934	 * We need to know that this is an internal copy, and doesn't need
3935	 * to get passed back to the FETD that allocated it.
3936	 */
3937	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3938}
3939
3940#ifdef NEEDTOPORT
3941static void
3942ctl_update_power_subpage(struct copan_power_subpage *page)
3943{
3944	int num_luns, num_partitions, config_type;
3945	struct ctl_softc *softc;
3946	cs_BOOL_t aor_present, shelf_50pct_power;
3947	cs_raidset_personality_t rs_type;
3948	int max_active_luns;
3949
3950	softc = control_softc;
3951
3952	/* subtract out the processor LUN */
3953	num_luns = softc->num_luns - 1;
3954	/*
3955	 * Default to 7 LUNs active, which was the only number we allowed
3956	 * in the past.
3957	 */
3958	max_active_luns = 7;
3959
3960	num_partitions = config_GetRsPartitionInfo();
3961	config_type = config_GetConfigType();
3962	shelf_50pct_power = config_GetShelfPowerMode();
3963	aor_present = config_IsAorRsPresent();
3964
3965	rs_type = ddb_GetRsRaidType(1);
3966	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3967	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3968		EPRINT(0, "Unsupported RS type %d!", rs_type);
3969	}
3970
3971
3972	page->total_luns = num_luns;
3973
3974	switch (config_type) {
3975	case 40:
3976		/*
3977		 * In a 40 drive configuration, it doesn't matter what DC
3978		 * cards we have, whether we have AOR enabled or not,
3979		 * partitioning or not, or what type of RAIDset we have.
3980		 * In that scenario, we can power up every LUN we present
3981		 * to the user.
3982		 */
3983		max_active_luns = num_luns;
3984
3985		break;
3986	case 64:
3987		if (shelf_50pct_power == CS_FALSE) {
3988			/* 25% power */
3989			if (aor_present == CS_TRUE) {
3990				if (rs_type ==
3991				     CS_RAIDSET_PERSONALITY_RAID5) {
3992					max_active_luns = 7;
3993				} else if (rs_type ==
3994					 CS_RAIDSET_PERSONALITY_RAID1){
3995					max_active_luns = 14;
3996				} else {
3997					/* XXX KDM now what?? */
3998				}
3999			} else {
4000				if (rs_type ==
4001				     CS_RAIDSET_PERSONALITY_RAID5) {
4002					max_active_luns = 8;
4003				} else if (rs_type ==
4004					 CS_RAIDSET_PERSONALITY_RAID1){
4005					max_active_luns = 16;
4006				} else {
4007					/* XXX KDM now what?? */
4008				}
4009			}
4010		} else {
4011			/* 50% power */
4012			/*
4013			 * With 50% power in a 64 drive configuration, we
4014			 * can power all LUNs we present.
4015			 */
4016			max_active_luns = num_luns;
4017		}
4018		break;
4019	case 112:
4020		if (shelf_50pct_power == CS_FALSE) {
4021			/* 25% power */
4022			if (aor_present == CS_TRUE) {
4023				if (rs_type ==
4024				     CS_RAIDSET_PERSONALITY_RAID5) {
4025					max_active_luns = 7;
4026				} else if (rs_type ==
4027					 CS_RAIDSET_PERSONALITY_RAID1){
4028					max_active_luns = 14;
4029				} else {
4030					/* XXX KDM now what?? */
4031				}
4032			} else {
4033				if (rs_type ==
4034				     CS_RAIDSET_PERSONALITY_RAID5) {
4035					max_active_luns = 8;
4036				} else if (rs_type ==
4037					 CS_RAIDSET_PERSONALITY_RAID1){
4038					max_active_luns = 16;
4039				} else {
4040					/* XXX KDM now what?? */
4041				}
4042			}
4043		} else {
4044			/* 50% power */
4045			if (aor_present == CS_TRUE) {
4046				if (rs_type ==
4047				     CS_RAIDSET_PERSONALITY_RAID5) {
4048					max_active_luns = 14;
4049				} else if (rs_type ==
4050					 CS_RAIDSET_PERSONALITY_RAID1){
4051					/*
4052					 * We're assuming here that disk
4053					 * caching is enabled, and so we're
4054					 * able to power up half of each
4055					 * LUN, and cache all writes.
4056					 */
4057					max_active_luns = num_luns;
4058				} else {
4059					/* XXX KDM now what?? */
4060				}
4061			} else {
4062				if (rs_type ==
4063				     CS_RAIDSET_PERSONALITY_RAID5) {
4064					max_active_luns = 15;
4065				} else if (rs_type ==
4066					 CS_RAIDSET_PERSONALITY_RAID1){
4067					max_active_luns = 30;
4068				} else {
4069					/* XXX KDM now what?? */
4070				}
4071			}
4072		}
4073		break;
4074	default:
4075		/*
4076		 * In this case, we have an unknown configuration, so we
4077		 * just use the default from above.
4078		 */
4079		break;
4080	}
4081
4082	page->max_active_luns = max_active_luns;
4083#if 0
4084	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4085	       page->total_luns, page->max_active_luns);
4086#endif
4087}
4088#endif /* NEEDTOPORT */
4089
4090/*
4091 * This routine could be used in the future to load default and/or saved
4092 * mode page parameters for a particuar lun.
4093 */
4094static int
4095ctl_init_page_index(struct ctl_lun *lun)
4096{
4097	int i;
4098	struct ctl_page_index *page_index;
4099	struct ctl_softc *softc;
4100	const char *value;
4101
4102	memcpy(&lun->mode_pages.index, page_index_template,
4103	       sizeof(page_index_template));
4104
4105	softc = lun->ctl_softc;
4106
4107	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4108
4109		page_index = &lun->mode_pages.index[i];
4110		/*
4111		 * If this is a disk-only mode page, there's no point in
4112		 * setting it up.  For some pages, we have to have some
4113		 * basic information about the disk in order to calculate the
4114		 * mode page data.
4115		 */
4116		if ((lun->be_lun->lun_type != T_DIRECT)
4117		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4118			continue;
4119
4120		switch (page_index->page_code & SMPH_PC_MASK) {
4121		case SMS_FORMAT_DEVICE_PAGE: {
4122			struct scsi_format_page *format_page;
4123
4124			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4125				panic("subpage is incorrect!");
4126
4127			/*
4128			 * Sectors per track are set above.  Bytes per
4129			 * sector need to be set here on a per-LUN basis.
4130			 */
4131			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4132			       &format_page_default,
4133			       sizeof(format_page_default));
4134			memcpy(&lun->mode_pages.format_page[
4135			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4136			       sizeof(format_page_changeable));
4137			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4138			       &format_page_default,
4139			       sizeof(format_page_default));
4140			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4141			       &format_page_default,
4142			       sizeof(format_page_default));
4143
4144			format_page = &lun->mode_pages.format_page[
4145				CTL_PAGE_CURRENT];
4146			scsi_ulto2b(lun->be_lun->blocksize,
4147				    format_page->bytes_per_sector);
4148
4149			format_page = &lun->mode_pages.format_page[
4150				CTL_PAGE_DEFAULT];
4151			scsi_ulto2b(lun->be_lun->blocksize,
4152				    format_page->bytes_per_sector);
4153
4154			format_page = &lun->mode_pages.format_page[
4155				CTL_PAGE_SAVED];
4156			scsi_ulto2b(lun->be_lun->blocksize,
4157				    format_page->bytes_per_sector);
4158
4159			page_index->page_data =
4160				(uint8_t *)lun->mode_pages.format_page;
4161			break;
4162		}
4163		case SMS_RIGID_DISK_PAGE: {
4164			struct scsi_rigid_disk_page *rigid_disk_page;
4165			uint32_t sectors_per_cylinder;
4166			uint64_t cylinders;
4167#ifndef	__XSCALE__
4168			int shift;
4169#endif /* !__XSCALE__ */
4170
4171			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4172				panic("invalid subpage value %d",
4173				      page_index->subpage);
4174
4175			/*
4176			 * Rotation rate and sectors per track are set
4177			 * above.  We calculate the cylinders here based on
4178			 * capacity.  Due to the number of heads and
4179			 * sectors per track we're using, smaller arrays
4180			 * may turn out to have 0 cylinders.  Linux and
4181			 * FreeBSD don't pay attention to these mode pages
4182			 * to figure out capacity, but Solaris does.  It
4183			 * seems to deal with 0 cylinders just fine, and
4184			 * works out a fake geometry based on the capacity.
4185			 */
4186			memcpy(&lun->mode_pages.rigid_disk_page[
4187			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4188			       sizeof(rigid_disk_page_default));
4189			memcpy(&lun->mode_pages.rigid_disk_page[
4190			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4191			       sizeof(rigid_disk_page_changeable));
4192			memcpy(&lun->mode_pages.rigid_disk_page[
4193			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4194			       sizeof(rigid_disk_page_default));
4195			memcpy(&lun->mode_pages.rigid_disk_page[
4196			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4197			       sizeof(rigid_disk_page_default));
4198
4199			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4200				CTL_DEFAULT_HEADS;
4201
4202			/*
4203			 * The divide method here will be more accurate,
4204			 * probably, but results in floating point being
4205			 * used in the kernel on i386 (__udivdi3()).  On the
4206			 * XScale, though, __udivdi3() is implemented in
4207			 * software.
4208			 *
4209			 * The shift method for cylinder calculation is
4210			 * accurate if sectors_per_cylinder is a power of
4211			 * 2.  Otherwise it might be slightly off -- you
4212			 * might have a bit of a truncation problem.
4213			 */
4214#ifdef	__XSCALE__
4215			cylinders = (lun->be_lun->maxlba + 1) /
4216				sectors_per_cylinder;
4217#else
4218			for (shift = 31; shift > 0; shift--) {
4219				if (sectors_per_cylinder & (1 << shift))
4220					break;
4221			}
4222			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4223#endif
4224
4225			/*
4226			 * We've basically got 3 bytes, or 24 bits for the
4227			 * cylinder size in the mode page.  If we're over,
4228			 * just round down to 2^24.
4229			 */
4230			if (cylinders > 0xffffff)
4231				cylinders = 0xffffff;
4232
4233			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4234				CTL_PAGE_CURRENT];
4235			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4236
4237			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4238				CTL_PAGE_DEFAULT];
4239			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4240
4241			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4242				CTL_PAGE_SAVED];
4243			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4244
4245			page_index->page_data =
4246				(uint8_t *)lun->mode_pages.rigid_disk_page;
4247			break;
4248		}
4249		case SMS_CACHING_PAGE: {
4250			struct scsi_caching_page *caching_page;
4251
4252			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4253				panic("invalid subpage value %d",
4254				      page_index->subpage);
4255			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4256			       &caching_page_default,
4257			       sizeof(caching_page_default));
4258			memcpy(&lun->mode_pages.caching_page[
4259			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4260			       sizeof(caching_page_changeable));
4261			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4262			       &caching_page_default,
4263			       sizeof(caching_page_default));
4264			caching_page = &lun->mode_pages.caching_page[
4265			    CTL_PAGE_SAVED];
4266			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4267			if (value != NULL && strcmp(value, "off") == 0)
4268				caching_page->flags1 &= ~SCP_WCE;
4269			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4270			if (value != NULL && strcmp(value, "off") == 0)
4271				caching_page->flags1 |= SCP_RCD;
4272			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4273			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4274			       sizeof(caching_page_default));
4275			page_index->page_data =
4276				(uint8_t *)lun->mode_pages.caching_page;
4277			break;
4278		}
4279		case SMS_CONTROL_MODE_PAGE: {
4280			struct scsi_control_page *control_page;
4281
4282			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4283				panic("invalid subpage value %d",
4284				      page_index->subpage);
4285
4286			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4287			       &control_page_default,
4288			       sizeof(control_page_default));
4289			memcpy(&lun->mode_pages.control_page[
4290			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4291			       sizeof(control_page_changeable));
4292			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4293			       &control_page_default,
4294			       sizeof(control_page_default));
4295			control_page = &lun->mode_pages.control_page[
4296			    CTL_PAGE_SAVED];
4297			value = ctl_get_opt(&lun->be_lun->options, "reordering");
4298			if (value != NULL && strcmp(value, "unrestricted") == 0) {
4299				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4300				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4301			}
4302			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4303			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4304			       sizeof(control_page_default));
4305			page_index->page_data =
4306				(uint8_t *)lun->mode_pages.control_page;
4307			break;
4308
4309		}
4310		case SMS_VENDOR_SPECIFIC_PAGE:{
4311			switch (page_index->subpage) {
4312			case PWR_SUBPAGE_CODE: {
4313				struct copan_power_subpage *current_page,
4314							   *saved_page;
4315
4316				memcpy(&lun->mode_pages.power_subpage[
4317				       CTL_PAGE_CURRENT],
4318				       &power_page_default,
4319				       sizeof(power_page_default));
4320				memcpy(&lun->mode_pages.power_subpage[
4321				       CTL_PAGE_CHANGEABLE],
4322				       &power_page_changeable,
4323				       sizeof(power_page_changeable));
4324				memcpy(&lun->mode_pages.power_subpage[
4325				       CTL_PAGE_DEFAULT],
4326				       &power_page_default,
4327				       sizeof(power_page_default));
4328				memcpy(&lun->mode_pages.power_subpage[
4329				       CTL_PAGE_SAVED],
4330				       &power_page_default,
4331				       sizeof(power_page_default));
4332				page_index->page_data =
4333				    (uint8_t *)lun->mode_pages.power_subpage;
4334
4335				current_page = (struct copan_power_subpage *)
4336					(page_index->page_data +
4337					 (page_index->page_len *
4338					  CTL_PAGE_CURRENT));
4339			        saved_page = (struct copan_power_subpage *)
4340				        (page_index->page_data +
4341					 (page_index->page_len *
4342					  CTL_PAGE_SAVED));
4343				break;
4344			}
4345			case APS_SUBPAGE_CODE: {
4346				struct copan_aps_subpage *current_page,
4347							 *saved_page;
4348
4349				// This gets set multiple times but
4350				// it should always be the same. It's
4351				// only done during init so who cares.
4352				index_to_aps_page = i;
4353
4354				memcpy(&lun->mode_pages.aps_subpage[
4355				       CTL_PAGE_CURRENT],
4356				       &aps_page_default,
4357				       sizeof(aps_page_default));
4358				memcpy(&lun->mode_pages.aps_subpage[
4359				       CTL_PAGE_CHANGEABLE],
4360				       &aps_page_changeable,
4361				       sizeof(aps_page_changeable));
4362				memcpy(&lun->mode_pages.aps_subpage[
4363				       CTL_PAGE_DEFAULT],
4364				       &aps_page_default,
4365				       sizeof(aps_page_default));
4366				memcpy(&lun->mode_pages.aps_subpage[
4367				       CTL_PAGE_SAVED],
4368				       &aps_page_default,
4369				       sizeof(aps_page_default));
4370				page_index->page_data =
4371					(uint8_t *)lun->mode_pages.aps_subpage;
4372
4373				current_page = (struct copan_aps_subpage *)
4374					(page_index->page_data +
4375					 (page_index->page_len *
4376					  CTL_PAGE_CURRENT));
4377				saved_page = (struct copan_aps_subpage *)
4378					(page_index->page_data +
4379					 (page_index->page_len *
4380					  CTL_PAGE_SAVED));
4381				break;
4382			}
4383			case DBGCNF_SUBPAGE_CODE: {
4384				struct copan_debugconf_subpage *current_page,
4385							       *saved_page;
4386
4387				memcpy(&lun->mode_pages.debugconf_subpage[
4388				       CTL_PAGE_CURRENT],
4389				       &debugconf_page_default,
4390				       sizeof(debugconf_page_default));
4391				memcpy(&lun->mode_pages.debugconf_subpage[
4392				       CTL_PAGE_CHANGEABLE],
4393				       &debugconf_page_changeable,
4394				       sizeof(debugconf_page_changeable));
4395				memcpy(&lun->mode_pages.debugconf_subpage[
4396				       CTL_PAGE_DEFAULT],
4397				       &debugconf_page_default,
4398				       sizeof(debugconf_page_default));
4399				memcpy(&lun->mode_pages.debugconf_subpage[
4400				       CTL_PAGE_SAVED],
4401				       &debugconf_page_default,
4402				       sizeof(debugconf_page_default));
4403				page_index->page_data =
4404					(uint8_t *)lun->mode_pages.debugconf_subpage;
4405
4406				current_page = (struct copan_debugconf_subpage *)
4407					(page_index->page_data +
4408					 (page_index->page_len *
4409					  CTL_PAGE_CURRENT));
4410				saved_page = (struct copan_debugconf_subpage *)
4411					(page_index->page_data +
4412					 (page_index->page_len *
4413					  CTL_PAGE_SAVED));
4414				break;
4415			}
4416			default:
4417				panic("invalid subpage value %d",
4418				      page_index->subpage);
4419				break;
4420			}
4421   			break;
4422		}
4423		default:
4424			panic("invalid page value %d",
4425			      page_index->page_code & SMPH_PC_MASK);
4426			break;
4427    	}
4428	}
4429
4430	return (CTL_RETVAL_COMPLETE);
4431}
4432
4433/*
4434 * LUN allocation.
4435 *
4436 * Requirements:
4437 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4438 *   wants us to allocate the LUN and he can block.
4439 * - ctl_softc is always set
4440 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4441 *
4442 * Returns 0 for success, non-zero (errno) for failure.
4443 */
4444static int
4445ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4446	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4447{
4448	struct ctl_lun *nlun, *lun;
4449	struct ctl_port *port;
4450	struct scsi_vpd_id_descriptor *desc;
4451	struct scsi_vpd_id_t10 *t10id;
4452	const char *eui, *naa, *scsiname, *vendor;
4453	int lun_number, i, lun_malloced;
4454	int devidlen, idlen1, idlen2 = 0, len;
4455
4456	if (be_lun == NULL)
4457		return (EINVAL);
4458
4459	/*
4460	 * We currently only support Direct Access or Processor LUN types.
4461	 */
4462	switch (be_lun->lun_type) {
4463	case T_DIRECT:
4464		break;
4465	case T_PROCESSOR:
4466		break;
4467	case T_SEQUENTIAL:
4468	case T_CHANGER:
4469	default:
4470		be_lun->lun_config_status(be_lun->be_lun,
4471					  CTL_LUN_CONFIG_FAILURE);
4472		break;
4473	}
4474	if (ctl_lun == NULL) {
4475		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4476		lun_malloced = 1;
4477	} else {
4478		lun_malloced = 0;
4479		lun = ctl_lun;
4480	}
4481
4482	memset(lun, 0, sizeof(*lun));
4483	if (lun_malloced)
4484		lun->flags = CTL_LUN_MALLOCED;
4485
4486	/* Generate LUN ID. */
4487	devidlen = max(CTL_DEVID_MIN_LEN,
4488	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4489	idlen1 = sizeof(*t10id) + devidlen;
4490	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4491	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4492	if (scsiname != NULL) {
4493		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4494		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4495	}
4496	eui = ctl_get_opt(&be_lun->options, "eui");
4497	if (eui != NULL) {
4498		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4499	}
4500	naa = ctl_get_opt(&be_lun->options, "naa");
4501	if (naa != NULL) {
4502		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4503	}
4504	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4505	    M_CTL, M_WAITOK | M_ZERO);
4506	lun->lun_devid->len = len;
4507	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4508	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4509	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4510	desc->length = idlen1;
4511	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4512	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4513	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4514		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4515	} else {
4516		strncpy(t10id->vendor, vendor,
4517		    min(sizeof(t10id->vendor), strlen(vendor)));
4518	}
4519	strncpy((char *)t10id->vendor_spec_id,
4520	    (char *)be_lun->device_id, devidlen);
4521	if (scsiname != NULL) {
4522		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4523		    desc->length);
4524		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4525		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4526		    SVPD_ID_TYPE_SCSI_NAME;
4527		desc->length = idlen2;
4528		strlcpy(desc->identifier, scsiname, idlen2);
4529	}
4530	if (eui != NULL) {
4531		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4532		    desc->length);
4533		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4534		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4535		    SVPD_ID_TYPE_EUI64;
4536		desc->length = 8;
4537		scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4538	}
4539	if (naa != NULL) {
4540		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4541		    desc->length);
4542		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4543		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4544		    SVPD_ID_TYPE_NAA;
4545		desc->length = 8;
4546		scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4547	}
4548
4549	mtx_lock(&ctl_softc->ctl_lock);
4550	/*
4551	 * See if the caller requested a particular LUN number.  If so, see
4552	 * if it is available.  Otherwise, allocate the first available LUN.
4553	 */
4554	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4555		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4556		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4557			mtx_unlock(&ctl_softc->ctl_lock);
4558			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4559				printf("ctl: requested LUN ID %d is higher "
4560				       "than CTL_MAX_LUNS - 1 (%d)\n",
4561				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4562			} else {
4563				/*
4564				 * XXX KDM return an error, or just assign
4565				 * another LUN ID in this case??
4566				 */
4567				printf("ctl: requested LUN ID %d is already "
4568				       "in use\n", be_lun->req_lun_id);
4569			}
4570			if (lun->flags & CTL_LUN_MALLOCED)
4571				free(lun, M_CTL);
4572			be_lun->lun_config_status(be_lun->be_lun,
4573						  CTL_LUN_CONFIG_FAILURE);
4574			return (ENOSPC);
4575		}
4576		lun_number = be_lun->req_lun_id;
4577	} else {
4578		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4579		if (lun_number == -1) {
4580			mtx_unlock(&ctl_softc->ctl_lock);
4581			printf("ctl: can't allocate LUN on target %ju, out of "
4582			       "LUNs\n", (uintmax_t)target_id.id);
4583			if (lun->flags & CTL_LUN_MALLOCED)
4584				free(lun, M_CTL);
4585			be_lun->lun_config_status(be_lun->be_lun,
4586						  CTL_LUN_CONFIG_FAILURE);
4587			return (ENOSPC);
4588		}
4589	}
4590	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4591
4592	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4593	lun->target = target_id;
4594	lun->lun = lun_number;
4595	lun->be_lun = be_lun;
4596	/*
4597	 * The processor LUN is always enabled.  Disk LUNs come on line
4598	 * disabled, and must be enabled by the backend.
4599	 */
4600	lun->flags |= CTL_LUN_DISABLED;
4601	lun->backend = be_lun->be;
4602	be_lun->ctl_lun = lun;
4603	be_lun->lun_id = lun_number;
4604	atomic_add_int(&be_lun->be->num_luns, 1);
4605	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4606		lun->flags |= CTL_LUN_STOPPED;
4607
4608	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4609		lun->flags |= CTL_LUN_INOPERABLE;
4610
4611	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4612		lun->flags |= CTL_LUN_PRIMARY_SC;
4613
4614	lun->ctl_softc = ctl_softc;
4615	TAILQ_INIT(&lun->ooa_queue);
4616	TAILQ_INIT(&lun->blocked_queue);
4617	STAILQ_INIT(&lun->error_list);
4618	ctl_tpc_lun_init(lun);
4619
4620	/*
4621	 * Initialize the mode page index.
4622	 */
4623	ctl_init_page_index(lun);
4624
4625	/*
4626	 * Set the poweron UA for all initiators on this LUN only.
4627	 */
4628	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4629		lun->pending_ua[i] = CTL_UA_POWERON;
4630
4631	/*
4632	 * Now, before we insert this lun on the lun list, set the lun
4633	 * inventory changed UA for all other luns.
4634	 */
4635	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4636		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4637			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4638		}
4639	}
4640
4641	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4642
4643	ctl_softc->ctl_luns[lun_number] = lun;
4644
4645	ctl_softc->num_luns++;
4646
4647	/* Setup statistics gathering */
4648	lun->stats.device_type = be_lun->lun_type;
4649	lun->stats.lun_number = lun_number;
4650	if (lun->stats.device_type == T_DIRECT)
4651		lun->stats.blocksize = be_lun->blocksize;
4652	else
4653		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4654	for (i = 0;i < CTL_MAX_PORTS;i++)
4655		lun->stats.ports[i].targ_port = i;
4656
4657	mtx_unlock(&ctl_softc->ctl_lock);
4658
4659	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4660
4661	/*
4662	 * Run through each registered FETD and bring it online if it isn't
4663	 * already.  Enable the target ID if it hasn't been enabled, and
4664	 * enable this particular LUN.
4665	 */
4666	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4667		int retval;
4668
4669		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4670		if (retval != 0) {
4671			printf("ctl_alloc_lun: FETD %s port %d returned error "
4672			       "%d for lun_enable on target %ju lun %d\n",
4673			       port->port_name, port->targ_port, retval,
4674			       (uintmax_t)target_id.id, lun_number);
4675		} else
4676			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4677	}
4678	return (0);
4679}
4680
4681/*
4682 * Delete a LUN.
4683 * Assumptions:
4684 * - LUN has already been marked invalid and any pending I/O has been taken
4685 *   care of.
4686 */
4687static int
4688ctl_free_lun(struct ctl_lun *lun)
4689{
4690	struct ctl_softc *softc;
4691#if 0
4692	struct ctl_port *port;
4693#endif
4694	struct ctl_lun *nlun;
4695	int i;
4696
4697	softc = lun->ctl_softc;
4698
4699	mtx_assert(&softc->ctl_lock, MA_OWNED);
4700
4701	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4702
4703	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4704
4705	softc->ctl_luns[lun->lun] = NULL;
4706
4707	if (!TAILQ_EMPTY(&lun->ooa_queue))
4708		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4709
4710	softc->num_luns--;
4711
4712	/*
4713	 * XXX KDM this scheme only works for a single target/multiple LUN
4714	 * setup.  It needs to be revamped for a multiple target scheme.
4715	 *
4716	 * XXX KDM this results in port->lun_disable() getting called twice,
4717	 * once when ctl_disable_lun() is called, and a second time here.
4718	 * We really need to re-think the LUN disable semantics.  There
4719	 * should probably be several steps/levels to LUN removal:
4720	 *  - disable
4721	 *  - invalidate
4722	 *  - free
4723 	 *
4724	 * Right now we only have a disable method when communicating to
4725	 * the front end ports, at least for individual LUNs.
4726	 */
4727#if 0
4728	STAILQ_FOREACH(port, &softc->port_list, links) {
4729		int retval;
4730
4731		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4732					 lun->lun);
4733		if (retval != 0) {
4734			printf("ctl_free_lun: FETD %s port %d returned error "
4735			       "%d for lun_disable on target %ju lun %jd\n",
4736			       port->port_name, port->targ_port, retval,
4737			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4738		}
4739
4740		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4741			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4742
4743			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4744			if (retval != 0) {
4745				printf("ctl_free_lun: FETD %s port %d "
4746				       "returned error %d for targ_disable on "
4747				       "target %ju\n", port->port_name,
4748				       port->targ_port, retval,
4749				       (uintmax_t)lun->target.id);
4750			} else
4751				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4752
4753			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4754				continue;
4755
4756#if 0
4757			port->port_offline(port->onoff_arg);
4758			port->status &= ~CTL_PORT_STATUS_ONLINE;
4759#endif
4760		}
4761	}
4762#endif
4763
4764	/*
4765	 * Tell the backend to free resources, if this LUN has a backend.
4766	 */
4767	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4768	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4769
4770	ctl_tpc_lun_shutdown(lun);
4771	mtx_destroy(&lun->lun_lock);
4772	free(lun->lun_devid, M_CTL);
4773	if (lun->flags & CTL_LUN_MALLOCED)
4774		free(lun, M_CTL);
4775
4776	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4777		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4778			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4779		}
4780	}
4781
4782	return (0);
4783}
4784
4785static void
4786ctl_create_lun(struct ctl_be_lun *be_lun)
4787{
4788	struct ctl_softc *ctl_softc;
4789
4790	ctl_softc = control_softc;
4791
4792	/*
4793	 * ctl_alloc_lun() should handle all potential failure cases.
4794	 */
4795	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4796}
4797
4798int
4799ctl_add_lun(struct ctl_be_lun *be_lun)
4800{
4801	struct ctl_softc *ctl_softc = control_softc;
4802
4803	mtx_lock(&ctl_softc->ctl_lock);
4804	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4805	mtx_unlock(&ctl_softc->ctl_lock);
4806	wakeup(&ctl_softc->pending_lun_queue);
4807
4808	return (0);
4809}
4810
4811int
4812ctl_enable_lun(struct ctl_be_lun *be_lun)
4813{
4814	struct ctl_softc *ctl_softc;
4815	struct ctl_port *port, *nport;
4816	struct ctl_lun *lun;
4817	int retval;
4818
4819	ctl_softc = control_softc;
4820
4821	lun = (struct ctl_lun *)be_lun->ctl_lun;
4822
4823	mtx_lock(&ctl_softc->ctl_lock);
4824	mtx_lock(&lun->lun_lock);
4825	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4826		/*
4827		 * eh?  Why did we get called if the LUN is already
4828		 * enabled?
4829		 */
4830		mtx_unlock(&lun->lun_lock);
4831		mtx_unlock(&ctl_softc->ctl_lock);
4832		return (0);
4833	}
4834	lun->flags &= ~CTL_LUN_DISABLED;
4835	mtx_unlock(&lun->lun_lock);
4836
4837	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4838		nport = STAILQ_NEXT(port, links);
4839
4840		/*
4841		 * Drop the lock while we call the FETD's enable routine.
4842		 * This can lead to a callback into CTL (at least in the
4843		 * case of the internal initiator frontend.
4844		 */
4845		mtx_unlock(&ctl_softc->ctl_lock);
4846		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4847		mtx_lock(&ctl_softc->ctl_lock);
4848		if (retval != 0) {
4849			printf("%s: FETD %s port %d returned error "
4850			       "%d for lun_enable on target %ju lun %jd\n",
4851			       __func__, port->port_name, port->targ_port, retval,
4852			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4853		}
4854#if 0
4855		 else {
4856            /* NOTE:  TODO:  why does lun enable affect port status? */
4857			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4858		}
4859#endif
4860	}
4861
4862	mtx_unlock(&ctl_softc->ctl_lock);
4863
4864	return (0);
4865}
4866
4867int
4868ctl_disable_lun(struct ctl_be_lun *be_lun)
4869{
4870	struct ctl_softc *ctl_softc;
4871	struct ctl_port *port;
4872	struct ctl_lun *lun;
4873	int retval;
4874
4875	ctl_softc = control_softc;
4876
4877	lun = (struct ctl_lun *)be_lun->ctl_lun;
4878
4879	mtx_lock(&ctl_softc->ctl_lock);
4880	mtx_lock(&lun->lun_lock);
4881	if (lun->flags & CTL_LUN_DISABLED) {
4882		mtx_unlock(&lun->lun_lock);
4883		mtx_unlock(&ctl_softc->ctl_lock);
4884		return (0);
4885	}
4886	lun->flags |= CTL_LUN_DISABLED;
4887	mtx_unlock(&lun->lun_lock);
4888
4889	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4890		mtx_unlock(&ctl_softc->ctl_lock);
4891		/*
4892		 * Drop the lock before we call the frontend's disable
4893		 * routine, to avoid lock order reversals.
4894		 *
4895		 * XXX KDM what happens if the frontend list changes while
4896		 * we're traversing it?  It's unlikely, but should be handled.
4897		 */
4898		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4899					 lun->lun);
4900		mtx_lock(&ctl_softc->ctl_lock);
4901		if (retval != 0) {
4902			printf("ctl_alloc_lun: FETD %s port %d returned error "
4903			       "%d for lun_disable on target %ju lun %jd\n",
4904			       port->port_name, port->targ_port, retval,
4905			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4906		}
4907	}
4908
4909	mtx_unlock(&ctl_softc->ctl_lock);
4910
4911	return (0);
4912}
4913
4914int
4915ctl_start_lun(struct ctl_be_lun *be_lun)
4916{
4917	struct ctl_softc *ctl_softc;
4918	struct ctl_lun *lun;
4919
4920	ctl_softc = control_softc;
4921
4922	lun = (struct ctl_lun *)be_lun->ctl_lun;
4923
4924	mtx_lock(&lun->lun_lock);
4925	lun->flags &= ~CTL_LUN_STOPPED;
4926	mtx_unlock(&lun->lun_lock);
4927
4928	return (0);
4929}
4930
4931int
4932ctl_stop_lun(struct ctl_be_lun *be_lun)
4933{
4934	struct ctl_softc *ctl_softc;
4935	struct ctl_lun *lun;
4936
4937	ctl_softc = control_softc;
4938
4939	lun = (struct ctl_lun *)be_lun->ctl_lun;
4940
4941	mtx_lock(&lun->lun_lock);
4942	lun->flags |= CTL_LUN_STOPPED;
4943	mtx_unlock(&lun->lun_lock);
4944
4945	return (0);
4946}
4947
4948int
4949ctl_lun_offline(struct ctl_be_lun *be_lun)
4950{
4951	struct ctl_softc *ctl_softc;
4952	struct ctl_lun *lun;
4953
4954	ctl_softc = control_softc;
4955
4956	lun = (struct ctl_lun *)be_lun->ctl_lun;
4957
4958	mtx_lock(&lun->lun_lock);
4959	lun->flags |= CTL_LUN_OFFLINE;
4960	mtx_unlock(&lun->lun_lock);
4961
4962	return (0);
4963}
4964
4965int
4966ctl_lun_online(struct ctl_be_lun *be_lun)
4967{
4968	struct ctl_softc *ctl_softc;
4969	struct ctl_lun *lun;
4970
4971	ctl_softc = control_softc;
4972
4973	lun = (struct ctl_lun *)be_lun->ctl_lun;
4974
4975	mtx_lock(&lun->lun_lock);
4976	lun->flags &= ~CTL_LUN_OFFLINE;
4977	mtx_unlock(&lun->lun_lock);
4978
4979	return (0);
4980}
4981
4982int
4983ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4984{
4985	struct ctl_softc *ctl_softc;
4986	struct ctl_lun *lun;
4987
4988	ctl_softc = control_softc;
4989
4990	lun = (struct ctl_lun *)be_lun->ctl_lun;
4991
4992	mtx_lock(&lun->lun_lock);
4993
4994	/*
4995	 * The LUN needs to be disabled before it can be marked invalid.
4996	 */
4997	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4998		mtx_unlock(&lun->lun_lock);
4999		return (-1);
5000	}
5001	/*
5002	 * Mark the LUN invalid.
5003	 */
5004	lun->flags |= CTL_LUN_INVALID;
5005
5006	/*
5007	 * If there is nothing in the OOA queue, go ahead and free the LUN.
5008	 * If we have something in the OOA queue, we'll free it when the
5009	 * last I/O completes.
5010	 */
5011	if (TAILQ_EMPTY(&lun->ooa_queue)) {
5012		mtx_unlock(&lun->lun_lock);
5013		mtx_lock(&ctl_softc->ctl_lock);
5014		ctl_free_lun(lun);
5015		mtx_unlock(&ctl_softc->ctl_lock);
5016	} else
5017		mtx_unlock(&lun->lun_lock);
5018
5019	return (0);
5020}
5021
5022int
5023ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5024{
5025	struct ctl_softc *ctl_softc;
5026	struct ctl_lun *lun;
5027
5028	ctl_softc = control_softc;
5029	lun = (struct ctl_lun *)be_lun->ctl_lun;
5030
5031	mtx_lock(&lun->lun_lock);
5032	lun->flags |= CTL_LUN_INOPERABLE;
5033	mtx_unlock(&lun->lun_lock);
5034
5035	return (0);
5036}
5037
5038int
5039ctl_lun_operable(struct ctl_be_lun *be_lun)
5040{
5041	struct ctl_softc *ctl_softc;
5042	struct ctl_lun *lun;
5043
5044	ctl_softc = control_softc;
5045	lun = (struct ctl_lun *)be_lun->ctl_lun;
5046
5047	mtx_lock(&lun->lun_lock);
5048	lun->flags &= ~CTL_LUN_INOPERABLE;
5049	mtx_unlock(&lun->lun_lock);
5050
5051	return (0);
5052}
5053
5054int
5055ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5056		   int lock)
5057{
5058	struct ctl_softc *softc;
5059	struct ctl_lun *lun;
5060	struct copan_aps_subpage *current_sp;
5061	struct ctl_page_index *page_index;
5062	int i;
5063
5064	softc = control_softc;
5065
5066	mtx_lock(&softc->ctl_lock);
5067
5068	lun = (struct ctl_lun *)be_lun->ctl_lun;
5069	mtx_lock(&lun->lun_lock);
5070
5071	page_index = NULL;
5072	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5073		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5074		     APS_PAGE_CODE)
5075			continue;
5076
5077		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5078			continue;
5079		page_index = &lun->mode_pages.index[i];
5080	}
5081
5082	if (page_index == NULL) {
5083		mtx_unlock(&lun->lun_lock);
5084		mtx_unlock(&softc->ctl_lock);
5085		printf("%s: APS subpage not found for lun %ju!\n", __func__,
5086		       (uintmax_t)lun->lun);
5087		return (1);
5088	}
5089#if 0
5090	if ((softc->aps_locked_lun != 0)
5091	 && (softc->aps_locked_lun != lun->lun)) {
5092		printf("%s: attempt to lock LUN %llu when %llu is already "
5093		       "locked\n");
5094		mtx_unlock(&lun->lun_lock);
5095		mtx_unlock(&softc->ctl_lock);
5096		return (1);
5097	}
5098#endif
5099
5100	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5101		(page_index->page_len * CTL_PAGE_CURRENT));
5102
5103	if (lock != 0) {
5104		current_sp->lock_active = APS_LOCK_ACTIVE;
5105		softc->aps_locked_lun = lun->lun;
5106	} else {
5107		current_sp->lock_active = 0;
5108		softc->aps_locked_lun = 0;
5109	}
5110
5111
5112	/*
5113	 * If we're in HA mode, try to send the lock message to the other
5114	 * side.
5115	 */
5116	if (ctl_is_single == 0) {
5117		int isc_retval;
5118		union ctl_ha_msg lock_msg;
5119
5120		lock_msg.hdr.nexus = *nexus;
5121		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5122		if (lock != 0)
5123			lock_msg.aps.lock_flag = 1;
5124		else
5125			lock_msg.aps.lock_flag = 0;
5126		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5127					 sizeof(lock_msg), 0);
5128		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5129			printf("%s: APS (lock=%d) error returned from "
5130			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5131			mtx_unlock(&lun->lun_lock);
5132			mtx_unlock(&softc->ctl_lock);
5133			return (1);
5134		}
5135	}
5136
5137	mtx_unlock(&lun->lun_lock);
5138	mtx_unlock(&softc->ctl_lock);
5139
5140	return (0);
5141}
5142
5143void
5144ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5145{
5146	struct ctl_lun *lun;
5147	struct ctl_softc *softc;
5148	int i;
5149
5150	softc = control_softc;
5151
5152	lun = (struct ctl_lun *)be_lun->ctl_lun;
5153
5154	mtx_lock(&lun->lun_lock);
5155
5156	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5157		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5158
5159	mtx_unlock(&lun->lun_lock);
5160}
5161
5162/*
5163 * Backend "memory move is complete" callback for requests that never
5164 * make it down to say RAIDCore's configuration code.
5165 */
5166int
5167ctl_config_move_done(union ctl_io *io)
5168{
5169	int retval;
5170
5171	retval = CTL_RETVAL_COMPLETE;
5172
5173
5174	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5175	/*
5176	 * XXX KDM this shouldn't happen, but what if it does?
5177	 */
5178	if (io->io_hdr.io_type != CTL_IO_SCSI)
5179		panic("I/O type isn't CTL_IO_SCSI!");
5180
5181	if ((io->io_hdr.port_status == 0)
5182	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5183	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5184		io->io_hdr.status = CTL_SUCCESS;
5185	else if ((io->io_hdr.port_status != 0)
5186	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5187	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5188		/*
5189		 * For hardware error sense keys, the sense key
5190		 * specific value is defined to be a retry count,
5191		 * but we use it to pass back an internal FETD
5192		 * error code.  XXX KDM  Hopefully the FETD is only
5193		 * using 16 bits for an error code, since that's
5194		 * all the space we have in the sks field.
5195		 */
5196		ctl_set_internal_failure(&io->scsiio,
5197					 /*sks_valid*/ 1,
5198					 /*retry_count*/
5199					 io->io_hdr.port_status);
5200		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5201			free(io->scsiio.kern_data_ptr, M_CTL);
5202		ctl_done(io);
5203		goto bailout;
5204	}
5205
5206	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5207	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5208	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5209		/*
5210		 * XXX KDM just assuming a single pointer here, and not a
5211		 * S/G list.  If we start using S/G lists for config data,
5212		 * we'll need to know how to clean them up here as well.
5213		 */
5214		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5215			free(io->scsiio.kern_data_ptr, M_CTL);
5216		/* Hopefully the user has already set the status... */
5217		ctl_done(io);
5218	} else {
5219		/*
5220		 * XXX KDM now we need to continue data movement.  Some
5221		 * options:
5222		 * - call ctl_scsiio() again?  We don't do this for data
5223		 *   writes, because for those at least we know ahead of
5224		 *   time where the write will go and how long it is.  For
5225		 *   config writes, though, that information is largely
5226		 *   contained within the write itself, thus we need to
5227		 *   parse out the data again.
5228		 *
5229		 * - Call some other function once the data is in?
5230		 */
5231
5232		/*
5233		 * XXX KDM call ctl_scsiio() again for now, and check flag
5234		 * bits to see whether we're allocated or not.
5235		 */
5236		retval = ctl_scsiio(&io->scsiio);
5237	}
5238bailout:
5239	return (retval);
5240}
5241
5242/*
5243 * This gets called by a backend driver when it is done with a
5244 * data_submit method.
5245 */
5246void
5247ctl_data_submit_done(union ctl_io *io)
5248{
5249	/*
5250	 * If the IO_CONT flag is set, we need to call the supplied
5251	 * function to continue processing the I/O, instead of completing
5252	 * the I/O just yet.
5253	 *
5254	 * If there is an error, though, we don't want to keep processing.
5255	 * Instead, just send status back to the initiator.
5256	 */
5257	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5258	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5259	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5260	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5261		io->scsiio.io_cont(io);
5262		return;
5263	}
5264	ctl_done(io);
5265}
5266
5267/*
5268 * This gets called by a backend driver when it is done with a
5269 * configuration write.
5270 */
5271void
5272ctl_config_write_done(union ctl_io *io)
5273{
5274	uint8_t *buf;
5275
5276	/*
5277	 * If the IO_CONT flag is set, we need to call the supplied
5278	 * function to continue processing the I/O, instead of completing
5279	 * the I/O just yet.
5280	 *
5281	 * If there is an error, though, we don't want to keep processing.
5282	 * Instead, just send status back to the initiator.
5283	 */
5284	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5285	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5286	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5287		io->scsiio.io_cont(io);
5288		return;
5289	}
5290	/*
5291	 * Since a configuration write can be done for commands that actually
5292	 * have data allocated, like write buffer, and commands that have
5293	 * no data, like start/stop unit, we need to check here.
5294	 */
5295	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5296		buf = io->scsiio.kern_data_ptr;
5297	else
5298		buf = NULL;
5299	ctl_done(io);
5300	if (buf)
5301		free(buf, M_CTL);
5302}
5303
5304/*
5305 * SCSI release command.
5306 */
5307int
5308ctl_scsi_release(struct ctl_scsiio *ctsio)
5309{
5310	int length, longid, thirdparty_id, resv_id;
5311	struct ctl_softc *ctl_softc;
5312	struct ctl_lun *lun;
5313
5314	length = 0;
5315	resv_id = 0;
5316
5317	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5318
5319	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5320	ctl_softc = control_softc;
5321
5322	switch (ctsio->cdb[0]) {
5323	case RELEASE_10: {
5324		struct scsi_release_10 *cdb;
5325
5326		cdb = (struct scsi_release_10 *)ctsio->cdb;
5327
5328		if (cdb->byte2 & SR10_LONGID)
5329			longid = 1;
5330		else
5331			thirdparty_id = cdb->thirdparty_id;
5332
5333		resv_id = cdb->resv_id;
5334		length = scsi_2btoul(cdb->length);
5335		break;
5336	}
5337	}
5338
5339
5340	/*
5341	 * XXX KDM right now, we only support LUN reservation.  We don't
5342	 * support 3rd party reservations, or extent reservations, which
5343	 * might actually need the parameter list.  If we've gotten this
5344	 * far, we've got a LUN reservation.  Anything else got kicked out
5345	 * above.  So, according to SPC, ignore the length.
5346	 */
5347	length = 0;
5348
5349	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5350	 && (length > 0)) {
5351		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5352		ctsio->kern_data_len = length;
5353		ctsio->kern_total_len = length;
5354		ctsio->kern_data_resid = 0;
5355		ctsio->kern_rel_offset = 0;
5356		ctsio->kern_sg_entries = 0;
5357		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5358		ctsio->be_move_done = ctl_config_move_done;
5359		ctl_datamove((union ctl_io *)ctsio);
5360
5361		return (CTL_RETVAL_COMPLETE);
5362	}
5363
5364	if (length > 0)
5365		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5366
5367	mtx_lock(&lun->lun_lock);
5368
5369	/*
5370	 * According to SPC, it is not an error for an intiator to attempt
5371	 * to release a reservation on a LUN that isn't reserved, or that
5372	 * is reserved by another initiator.  The reservation can only be
5373	 * released, though, by the initiator who made it or by one of
5374	 * several reset type events.
5375	 */
5376	if (lun->flags & CTL_LUN_RESERVED) {
5377		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5378		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5379		 && (ctsio->io_hdr.nexus.targ_target.id ==
5380		     lun->rsv_nexus.targ_target.id)) {
5381			lun->flags &= ~CTL_LUN_RESERVED;
5382		}
5383	}
5384
5385	mtx_unlock(&lun->lun_lock);
5386
5387	ctsio->scsi_status = SCSI_STATUS_OK;
5388	ctsio->io_hdr.status = CTL_SUCCESS;
5389
5390	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5391		free(ctsio->kern_data_ptr, M_CTL);
5392		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5393	}
5394
5395	ctl_done((union ctl_io *)ctsio);
5396	return (CTL_RETVAL_COMPLETE);
5397}
5398
5399int
5400ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5401{
5402	int extent, thirdparty, longid;
5403	int resv_id, length;
5404	uint64_t thirdparty_id;
5405	struct ctl_softc *ctl_softc;
5406	struct ctl_lun *lun;
5407
5408	extent = 0;
5409	thirdparty = 0;
5410	longid = 0;
5411	resv_id = 0;
5412	length = 0;
5413	thirdparty_id = 0;
5414
5415	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5416
5417	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5418	ctl_softc = control_softc;
5419
5420	switch (ctsio->cdb[0]) {
5421	case RESERVE_10: {
5422		struct scsi_reserve_10 *cdb;
5423
5424		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5425
5426		if (cdb->byte2 & SR10_LONGID)
5427			longid = 1;
5428		else
5429			thirdparty_id = cdb->thirdparty_id;
5430
5431		resv_id = cdb->resv_id;
5432		length = scsi_2btoul(cdb->length);
5433		break;
5434	}
5435	}
5436
5437	/*
5438	 * XXX KDM right now, we only support LUN reservation.  We don't
5439	 * support 3rd party reservations, or extent reservations, which
5440	 * might actually need the parameter list.  If we've gotten this
5441	 * far, we've got a LUN reservation.  Anything else got kicked out
5442	 * above.  So, according to SPC, ignore the length.
5443	 */
5444	length = 0;
5445
5446	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5447	 && (length > 0)) {
5448		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5449		ctsio->kern_data_len = length;
5450		ctsio->kern_total_len = length;
5451		ctsio->kern_data_resid = 0;
5452		ctsio->kern_rel_offset = 0;
5453		ctsio->kern_sg_entries = 0;
5454		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5455		ctsio->be_move_done = ctl_config_move_done;
5456		ctl_datamove((union ctl_io *)ctsio);
5457
5458		return (CTL_RETVAL_COMPLETE);
5459	}
5460
5461	if (length > 0)
5462		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5463
5464	mtx_lock(&lun->lun_lock);
5465	if (lun->flags & CTL_LUN_RESERVED) {
5466		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5467		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5468		 || (ctsio->io_hdr.nexus.targ_target.id !=
5469		     lun->rsv_nexus.targ_target.id)) {
5470			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5471			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5472			goto bailout;
5473		}
5474	}
5475
5476	lun->flags |= CTL_LUN_RESERVED;
5477	lun->rsv_nexus = ctsio->io_hdr.nexus;
5478
5479	ctsio->scsi_status = SCSI_STATUS_OK;
5480	ctsio->io_hdr.status = CTL_SUCCESS;
5481
5482bailout:
5483	mtx_unlock(&lun->lun_lock);
5484
5485	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5486		free(ctsio->kern_data_ptr, M_CTL);
5487		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5488	}
5489
5490	ctl_done((union ctl_io *)ctsio);
5491	return (CTL_RETVAL_COMPLETE);
5492}
5493
5494int
5495ctl_start_stop(struct ctl_scsiio *ctsio)
5496{
5497	struct scsi_start_stop_unit *cdb;
5498	struct ctl_lun *lun;
5499	struct ctl_softc *ctl_softc;
5500	int retval;
5501
5502	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5503
5504	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5505	ctl_softc = control_softc;
5506	retval = 0;
5507
5508	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5509
5510	/*
5511	 * XXX KDM
5512	 * We don't support the immediate bit on a stop unit.  In order to
5513	 * do that, we would need to code up a way to know that a stop is
5514	 * pending, and hold off any new commands until it completes, one
5515	 * way or another.  Then we could accept or reject those commands
5516	 * depending on its status.  We would almost need to do the reverse
5517	 * of what we do below for an immediate start -- return the copy of
5518	 * the ctl_io to the FETD with status to send to the host (and to
5519	 * free the copy!) and then free the original I/O once the stop
5520	 * actually completes.  That way, the OOA queue mechanism can work
5521	 * to block commands that shouldn't proceed.  Another alternative
5522	 * would be to put the copy in the queue in place of the original,
5523	 * and return the original back to the caller.  That could be
5524	 * slightly safer..
5525	 */
5526	if ((cdb->byte2 & SSS_IMMED)
5527	 && ((cdb->how & SSS_START) == 0)) {
5528		ctl_set_invalid_field(ctsio,
5529				      /*sks_valid*/ 1,
5530				      /*command*/ 1,
5531				      /*field*/ 1,
5532				      /*bit_valid*/ 1,
5533				      /*bit*/ 0);
5534		ctl_done((union ctl_io *)ctsio);
5535		return (CTL_RETVAL_COMPLETE);
5536	}
5537
5538	if ((lun->flags & CTL_LUN_PR_RESERVED)
5539	 && ((cdb->how & SSS_START)==0)) {
5540		uint32_t residx;
5541
5542		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5543		if (!lun->per_res[residx].registered
5544		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5545
5546			ctl_set_reservation_conflict(ctsio);
5547			ctl_done((union ctl_io *)ctsio);
5548			return (CTL_RETVAL_COMPLETE);
5549		}
5550	}
5551
5552	/*
5553	 * If there is no backend on this device, we can't start or stop
5554	 * it.  In theory we shouldn't get any start/stop commands in the
5555	 * first place at this level if the LUN doesn't have a backend.
5556	 * That should get stopped by the command decode code.
5557	 */
5558	if (lun->backend == NULL) {
5559		ctl_set_invalid_opcode(ctsio);
5560		ctl_done((union ctl_io *)ctsio);
5561		return (CTL_RETVAL_COMPLETE);
5562	}
5563
5564	/*
5565	 * XXX KDM Copan-specific offline behavior.
5566	 * Figure out a reasonable way to port this?
5567	 */
5568#ifdef NEEDTOPORT
5569	mtx_lock(&lun->lun_lock);
5570
5571	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5572	 && (lun->flags & CTL_LUN_OFFLINE)) {
5573		/*
5574		 * If the LUN is offline, and the on/offline bit isn't set,
5575		 * reject the start or stop.  Otherwise, let it through.
5576		 */
5577		mtx_unlock(&lun->lun_lock);
5578		ctl_set_lun_not_ready(ctsio);
5579		ctl_done((union ctl_io *)ctsio);
5580	} else {
5581		mtx_unlock(&lun->lun_lock);
5582#endif /* NEEDTOPORT */
5583		/*
5584		 * This could be a start or a stop when we're online,
5585		 * or a stop/offline or start/online.  A start or stop when
5586		 * we're offline is covered in the case above.
5587		 */
5588		/*
5589		 * In the non-immediate case, we send the request to
5590		 * the backend and return status to the user when
5591		 * it is done.
5592		 *
5593		 * In the immediate case, we allocate a new ctl_io
5594		 * to hold a copy of the request, and send that to
5595		 * the backend.  We then set good status on the
5596		 * user's request and return it immediately.
5597		 */
5598		if (cdb->byte2 & SSS_IMMED) {
5599			union ctl_io *new_io;
5600
5601			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5602			if (new_io == NULL) {
5603				ctl_set_busy(ctsio);
5604				ctl_done((union ctl_io *)ctsio);
5605			} else {
5606				ctl_copy_io((union ctl_io *)ctsio,
5607					    new_io);
5608				retval = lun->backend->config_write(new_io);
5609				ctl_set_success(ctsio);
5610				ctl_done((union ctl_io *)ctsio);
5611			}
5612		} else {
5613			retval = lun->backend->config_write(
5614				(union ctl_io *)ctsio);
5615		}
5616#ifdef NEEDTOPORT
5617	}
5618#endif
5619	return (retval);
5620}
5621
5622/*
5623 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5624 * we don't really do anything with the LBA and length fields if the user
5625 * passes them in.  Instead we'll just flush out the cache for the entire
5626 * LUN.
5627 */
5628int
5629ctl_sync_cache(struct ctl_scsiio *ctsio)
5630{
5631	struct ctl_lun *lun;
5632	struct ctl_softc *ctl_softc;
5633	uint64_t starting_lba;
5634	uint32_t block_count;
5635	int retval;
5636
5637	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5638
5639	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5640	ctl_softc = control_softc;
5641	retval = 0;
5642
5643	switch (ctsio->cdb[0]) {
5644	case SYNCHRONIZE_CACHE: {
5645		struct scsi_sync_cache *cdb;
5646		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5647
5648		starting_lba = scsi_4btoul(cdb->begin_lba);
5649		block_count = scsi_2btoul(cdb->lb_count);
5650		break;
5651	}
5652	case SYNCHRONIZE_CACHE_16: {
5653		struct scsi_sync_cache_16 *cdb;
5654		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5655
5656		starting_lba = scsi_8btou64(cdb->begin_lba);
5657		block_count = scsi_4btoul(cdb->lb_count);
5658		break;
5659	}
5660	default:
5661		ctl_set_invalid_opcode(ctsio);
5662		ctl_done((union ctl_io *)ctsio);
5663		goto bailout;
5664		break; /* NOTREACHED */
5665	}
5666
5667	/*
5668	 * We check the LBA and length, but don't do anything with them.
5669	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5670	 * get flushed.  This check will just help satisfy anyone who wants
5671	 * to see an error for an out of range LBA.
5672	 */
5673	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5674		ctl_set_lba_out_of_range(ctsio);
5675		ctl_done((union ctl_io *)ctsio);
5676		goto bailout;
5677	}
5678
5679	/*
5680	 * If this LUN has no backend, we can't flush the cache anyway.
5681	 */
5682	if (lun->backend == NULL) {
5683		ctl_set_invalid_opcode(ctsio);
5684		ctl_done((union ctl_io *)ctsio);
5685		goto bailout;
5686	}
5687
5688	/*
5689	 * Check to see whether we're configured to send the SYNCHRONIZE
5690	 * CACHE command directly to the back end.
5691	 */
5692	mtx_lock(&lun->lun_lock);
5693	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5694	 && (++(lun->sync_count) >= lun->sync_interval)) {
5695		lun->sync_count = 0;
5696		mtx_unlock(&lun->lun_lock);
5697		retval = lun->backend->config_write((union ctl_io *)ctsio);
5698	} else {
5699		mtx_unlock(&lun->lun_lock);
5700		ctl_set_success(ctsio);
5701		ctl_done((union ctl_io *)ctsio);
5702	}
5703
5704bailout:
5705
5706	return (retval);
5707}
5708
5709int
5710ctl_format(struct ctl_scsiio *ctsio)
5711{
5712	struct scsi_format *cdb;
5713	struct ctl_lun *lun;
5714	struct ctl_softc *ctl_softc;
5715	int length, defect_list_len;
5716
5717	CTL_DEBUG_PRINT(("ctl_format\n"));
5718
5719	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5720	ctl_softc = control_softc;
5721
5722	cdb = (struct scsi_format *)ctsio->cdb;
5723
5724	length = 0;
5725	if (cdb->byte2 & SF_FMTDATA) {
5726		if (cdb->byte2 & SF_LONGLIST)
5727			length = sizeof(struct scsi_format_header_long);
5728		else
5729			length = sizeof(struct scsi_format_header_short);
5730	}
5731
5732	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5733	 && (length > 0)) {
5734		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5735		ctsio->kern_data_len = length;
5736		ctsio->kern_total_len = length;
5737		ctsio->kern_data_resid = 0;
5738		ctsio->kern_rel_offset = 0;
5739		ctsio->kern_sg_entries = 0;
5740		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5741		ctsio->be_move_done = ctl_config_move_done;
5742		ctl_datamove((union ctl_io *)ctsio);
5743
5744		return (CTL_RETVAL_COMPLETE);
5745	}
5746
5747	defect_list_len = 0;
5748
5749	if (cdb->byte2 & SF_FMTDATA) {
5750		if (cdb->byte2 & SF_LONGLIST) {
5751			struct scsi_format_header_long *header;
5752
5753			header = (struct scsi_format_header_long *)
5754				ctsio->kern_data_ptr;
5755
5756			defect_list_len = scsi_4btoul(header->defect_list_len);
5757			if (defect_list_len != 0) {
5758				ctl_set_invalid_field(ctsio,
5759						      /*sks_valid*/ 1,
5760						      /*command*/ 0,
5761						      /*field*/ 2,
5762						      /*bit_valid*/ 0,
5763						      /*bit*/ 0);
5764				goto bailout;
5765			}
5766		} else {
5767			struct scsi_format_header_short *header;
5768
5769			header = (struct scsi_format_header_short *)
5770				ctsio->kern_data_ptr;
5771
5772			defect_list_len = scsi_2btoul(header->defect_list_len);
5773			if (defect_list_len != 0) {
5774				ctl_set_invalid_field(ctsio,
5775						      /*sks_valid*/ 1,
5776						      /*command*/ 0,
5777						      /*field*/ 2,
5778						      /*bit_valid*/ 0,
5779						      /*bit*/ 0);
5780				goto bailout;
5781			}
5782		}
5783	}
5784
5785	/*
5786	 * The format command will clear out the "Medium format corrupted"
5787	 * status if set by the configuration code.  That status is really
5788	 * just a way to notify the host that we have lost the media, and
5789	 * get them to issue a command that will basically make them think
5790	 * they're blowing away the media.
5791	 */
5792	mtx_lock(&lun->lun_lock);
5793	lun->flags &= ~CTL_LUN_INOPERABLE;
5794	mtx_unlock(&lun->lun_lock);
5795
5796	ctsio->scsi_status = SCSI_STATUS_OK;
5797	ctsio->io_hdr.status = CTL_SUCCESS;
5798bailout:
5799
5800	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5801		free(ctsio->kern_data_ptr, M_CTL);
5802		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5803	}
5804
5805	ctl_done((union ctl_io *)ctsio);
5806	return (CTL_RETVAL_COMPLETE);
5807}
5808
5809int
5810ctl_read_buffer(struct ctl_scsiio *ctsio)
5811{
5812	struct scsi_read_buffer *cdb;
5813	struct ctl_lun *lun;
5814	int buffer_offset, len;
5815	static uint8_t descr[4];
5816	static uint8_t echo_descr[4] = { 0 };
5817
5818	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5819
5820	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5821	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5822
5823	if (lun->flags & CTL_LUN_PR_RESERVED) {
5824		uint32_t residx;
5825
5826		/*
5827		 * XXX KDM need a lock here.
5828		 */
5829		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5830		if ((lun->res_type == SPR_TYPE_EX_AC
5831		  && residx != lun->pr_res_idx)
5832		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5833		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5834		  && !lun->per_res[residx].registered)) {
5835			ctl_set_reservation_conflict(ctsio);
5836			ctl_done((union ctl_io *)ctsio);
5837			return (CTL_RETVAL_COMPLETE);
5838	        }
5839	}
5840
5841	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5842	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5843	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5844		ctl_set_invalid_field(ctsio,
5845				      /*sks_valid*/ 1,
5846				      /*command*/ 1,
5847				      /*field*/ 1,
5848				      /*bit_valid*/ 1,
5849				      /*bit*/ 4);
5850		ctl_done((union ctl_io *)ctsio);
5851		return (CTL_RETVAL_COMPLETE);
5852	}
5853
5854	len = scsi_3btoul(cdb->length);
5855	buffer_offset = scsi_3btoul(cdb->offset);
5856
5857	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5858		ctl_set_invalid_field(ctsio,
5859				      /*sks_valid*/ 1,
5860				      /*command*/ 1,
5861				      /*field*/ 6,
5862				      /*bit_valid*/ 0,
5863				      /*bit*/ 0);
5864		ctl_done((union ctl_io *)ctsio);
5865		return (CTL_RETVAL_COMPLETE);
5866	}
5867
5868	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5869		descr[0] = 0;
5870		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5871		ctsio->kern_data_ptr = descr;
5872		len = min(len, sizeof(descr));
5873	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5874		ctsio->kern_data_ptr = echo_descr;
5875		len = min(len, sizeof(echo_descr));
5876	} else
5877		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5878	ctsio->kern_data_len = len;
5879	ctsio->kern_total_len = len;
5880	ctsio->kern_data_resid = 0;
5881	ctsio->kern_rel_offset = 0;
5882	ctsio->kern_sg_entries = 0;
5883	ctsio->be_move_done = ctl_config_move_done;
5884	ctl_datamove((union ctl_io *)ctsio);
5885
5886	return (CTL_RETVAL_COMPLETE);
5887}
5888
5889int
5890ctl_write_buffer(struct ctl_scsiio *ctsio)
5891{
5892	struct scsi_write_buffer *cdb;
5893	struct ctl_lun *lun;
5894	int buffer_offset, len;
5895
5896	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5897
5898	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5899	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5900
5901	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5902		ctl_set_invalid_field(ctsio,
5903				      /*sks_valid*/ 1,
5904				      /*command*/ 1,
5905				      /*field*/ 1,
5906				      /*bit_valid*/ 1,
5907				      /*bit*/ 4);
5908		ctl_done((union ctl_io *)ctsio);
5909		return (CTL_RETVAL_COMPLETE);
5910	}
5911
5912	len = scsi_3btoul(cdb->length);
5913	buffer_offset = scsi_3btoul(cdb->offset);
5914
5915	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5916		ctl_set_invalid_field(ctsio,
5917				      /*sks_valid*/ 1,
5918				      /*command*/ 1,
5919				      /*field*/ 6,
5920				      /*bit_valid*/ 0,
5921				      /*bit*/ 0);
5922		ctl_done((union ctl_io *)ctsio);
5923		return (CTL_RETVAL_COMPLETE);
5924	}
5925
5926	/*
5927	 * If we've got a kernel request that hasn't been malloced yet,
5928	 * malloc it and tell the caller the data buffer is here.
5929	 */
5930	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5931		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5932		ctsio->kern_data_len = len;
5933		ctsio->kern_total_len = len;
5934		ctsio->kern_data_resid = 0;
5935		ctsio->kern_rel_offset = 0;
5936		ctsio->kern_sg_entries = 0;
5937		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5938		ctsio->be_move_done = ctl_config_move_done;
5939		ctl_datamove((union ctl_io *)ctsio);
5940
5941		return (CTL_RETVAL_COMPLETE);
5942	}
5943
5944	ctl_done((union ctl_io *)ctsio);
5945
5946	return (CTL_RETVAL_COMPLETE);
5947}
5948
5949int
5950ctl_write_same(struct ctl_scsiio *ctsio)
5951{
5952	struct ctl_lun *lun;
5953	struct ctl_lba_len_flags *lbalen;
5954	uint64_t lba;
5955	uint32_t num_blocks;
5956	int len, retval;
5957	uint8_t byte2;
5958
5959	retval = CTL_RETVAL_COMPLETE;
5960
5961	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5962
5963	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5964
5965	switch (ctsio->cdb[0]) {
5966	case WRITE_SAME_10: {
5967		struct scsi_write_same_10 *cdb;
5968
5969		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5970
5971		lba = scsi_4btoul(cdb->addr);
5972		num_blocks = scsi_2btoul(cdb->length);
5973		byte2 = cdb->byte2;
5974		break;
5975	}
5976	case WRITE_SAME_16: {
5977		struct scsi_write_same_16 *cdb;
5978
5979		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5980
5981		lba = scsi_8btou64(cdb->addr);
5982		num_blocks = scsi_4btoul(cdb->length);
5983		byte2 = cdb->byte2;
5984		break;
5985	}
5986	default:
5987		/*
5988		 * We got a command we don't support.  This shouldn't
5989		 * happen, commands should be filtered out above us.
5990		 */
5991		ctl_set_invalid_opcode(ctsio);
5992		ctl_done((union ctl_io *)ctsio);
5993
5994		return (CTL_RETVAL_COMPLETE);
5995		break; /* NOTREACHED */
5996	}
5997
5998	/* NDOB flag can be used only together with UNMAP */
5999	if ((byte2 & (SWS_NDOB | SWS_UNMAP)) == SWS_NDOB) {
6000		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
6001		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
6002		ctl_done((union ctl_io *)ctsio);
6003		return (CTL_RETVAL_COMPLETE);
6004	}
6005
6006	/*
6007	 * The first check is to make sure we're in bounds, the second
6008	 * check is to catch wrap-around problems.  If the lba + num blocks
6009	 * is less than the lba, then we've wrapped around and the block
6010	 * range is invalid anyway.
6011	 */
6012	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6013	 || ((lba + num_blocks) < lba)) {
6014		ctl_set_lba_out_of_range(ctsio);
6015		ctl_done((union ctl_io *)ctsio);
6016		return (CTL_RETVAL_COMPLETE);
6017	}
6018
6019	/* Zero number of blocks means "to the last logical block" */
6020	if (num_blocks == 0) {
6021		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
6022			ctl_set_invalid_field(ctsio,
6023					      /*sks_valid*/ 0,
6024					      /*command*/ 1,
6025					      /*field*/ 0,
6026					      /*bit_valid*/ 0,
6027					      /*bit*/ 0);
6028			ctl_done((union ctl_io *)ctsio);
6029			return (CTL_RETVAL_COMPLETE);
6030		}
6031		num_blocks = (lun->be_lun->maxlba + 1) - lba;
6032	}
6033
6034	len = lun->be_lun->blocksize;
6035
6036	/*
6037	 * If we've got a kernel request that hasn't been malloced yet,
6038	 * malloc it and tell the caller the data buffer is here.
6039	 */
6040	if ((byte2 & SWS_NDOB) == 0 &&
6041	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6042		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6043		ctsio->kern_data_len = len;
6044		ctsio->kern_total_len = len;
6045		ctsio->kern_data_resid = 0;
6046		ctsio->kern_rel_offset = 0;
6047		ctsio->kern_sg_entries = 0;
6048		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6049		ctsio->be_move_done = ctl_config_move_done;
6050		ctl_datamove((union ctl_io *)ctsio);
6051
6052		return (CTL_RETVAL_COMPLETE);
6053	}
6054
6055	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6056	lbalen->lba = lba;
6057	lbalen->len = num_blocks;
6058	lbalen->flags = byte2;
6059	retval = lun->backend->config_write((union ctl_io *)ctsio);
6060
6061	return (retval);
6062}
6063
6064int
6065ctl_unmap(struct ctl_scsiio *ctsio)
6066{
6067	struct ctl_lun *lun;
6068	struct scsi_unmap *cdb;
6069	struct ctl_ptr_len_flags *ptrlen;
6070	struct scsi_unmap_header *hdr;
6071	struct scsi_unmap_desc *buf, *end, *endnz, *range;
6072	uint64_t lba;
6073	uint32_t num_blocks;
6074	int len, retval;
6075	uint8_t byte2;
6076
6077	retval = CTL_RETVAL_COMPLETE;
6078
6079	CTL_DEBUG_PRINT(("ctl_unmap\n"));
6080
6081	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6082	cdb = (struct scsi_unmap *)ctsio->cdb;
6083
6084	len = scsi_2btoul(cdb->length);
6085	byte2 = cdb->byte2;
6086
6087	/*
6088	 * If we've got a kernel request that hasn't been malloced yet,
6089	 * malloc it and tell the caller the data buffer is here.
6090	 */
6091	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6092		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6093		ctsio->kern_data_len = len;
6094		ctsio->kern_total_len = len;
6095		ctsio->kern_data_resid = 0;
6096		ctsio->kern_rel_offset = 0;
6097		ctsio->kern_sg_entries = 0;
6098		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6099		ctsio->be_move_done = ctl_config_move_done;
6100		ctl_datamove((union ctl_io *)ctsio);
6101
6102		return (CTL_RETVAL_COMPLETE);
6103	}
6104
6105	len = ctsio->kern_total_len - ctsio->kern_data_resid;
6106	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6107	if (len < sizeof (*hdr) ||
6108	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6109	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6110	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6111		ctl_set_invalid_field(ctsio,
6112				      /*sks_valid*/ 0,
6113				      /*command*/ 0,
6114				      /*field*/ 0,
6115				      /*bit_valid*/ 0,
6116				      /*bit*/ 0);
6117		ctl_done((union ctl_io *)ctsio);
6118		return (CTL_RETVAL_COMPLETE);
6119	}
6120	len = scsi_2btoul(hdr->desc_length);
6121	buf = (struct scsi_unmap_desc *)(hdr + 1);
6122	end = buf + len / sizeof(*buf);
6123
6124	endnz = buf;
6125	for (range = buf; range < end; range++) {
6126		lba = scsi_8btou64(range->lba);
6127		num_blocks = scsi_4btoul(range->length);
6128		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6129		 || ((lba + num_blocks) < lba)) {
6130			ctl_set_lba_out_of_range(ctsio);
6131			ctl_done((union ctl_io *)ctsio);
6132			return (CTL_RETVAL_COMPLETE);
6133		}
6134		if (num_blocks != 0)
6135			endnz = range + 1;
6136	}
6137
6138	/*
6139	 * Block backend can not handle zero last range.
6140	 * Filter it out and return if there is nothing left.
6141	 */
6142	len = (uint8_t *)endnz - (uint8_t *)buf;
6143	if (len == 0) {
6144		ctl_set_success(ctsio);
6145		ctl_done((union ctl_io *)ctsio);
6146		return (CTL_RETVAL_COMPLETE);
6147	}
6148
6149	mtx_lock(&lun->lun_lock);
6150	ptrlen = (struct ctl_ptr_len_flags *)
6151	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6152	ptrlen->ptr = (void *)buf;
6153	ptrlen->len = len;
6154	ptrlen->flags = byte2;
6155	ctl_check_blocked(lun);
6156	mtx_unlock(&lun->lun_lock);
6157
6158	retval = lun->backend->config_write((union ctl_io *)ctsio);
6159	return (retval);
6160}
6161
6162/*
6163 * Note that this function currently doesn't actually do anything inside
6164 * CTL to enforce things if the DQue bit is turned on.
6165 *
6166 * Also note that this function can't be used in the default case, because
6167 * the DQue bit isn't set in the changeable mask for the control mode page
6168 * anyway.  This is just here as an example for how to implement a page
6169 * handler, and a placeholder in case we want to allow the user to turn
6170 * tagged queueing on and off.
6171 *
6172 * The D_SENSE bit handling is functional, however, and will turn
6173 * descriptor sense on and off for a given LUN.
6174 */
6175int
6176ctl_control_page_handler(struct ctl_scsiio *ctsio,
6177			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6178{
6179	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6180	struct ctl_lun *lun;
6181	struct ctl_softc *softc;
6182	int set_ua;
6183	uint32_t initidx;
6184
6185	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6186	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6187	set_ua = 0;
6188
6189	user_cp = (struct scsi_control_page *)page_ptr;
6190	current_cp = (struct scsi_control_page *)
6191		(page_index->page_data + (page_index->page_len *
6192		CTL_PAGE_CURRENT));
6193	saved_cp = (struct scsi_control_page *)
6194		(page_index->page_data + (page_index->page_len *
6195		CTL_PAGE_SAVED));
6196
6197	softc = control_softc;
6198
6199	mtx_lock(&lun->lun_lock);
6200	if (((current_cp->rlec & SCP_DSENSE) == 0)
6201	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6202		/*
6203		 * Descriptor sense is currently turned off and the user
6204		 * wants to turn it on.
6205		 */
6206		current_cp->rlec |= SCP_DSENSE;
6207		saved_cp->rlec |= SCP_DSENSE;
6208		lun->flags |= CTL_LUN_SENSE_DESC;
6209		set_ua = 1;
6210	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6211		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6212		/*
6213		 * Descriptor sense is currently turned on, and the user
6214		 * wants to turn it off.
6215		 */
6216		current_cp->rlec &= ~SCP_DSENSE;
6217		saved_cp->rlec &= ~SCP_DSENSE;
6218		lun->flags &= ~CTL_LUN_SENSE_DESC;
6219		set_ua = 1;
6220	}
6221	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6222	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6223		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6224		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6225		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6226		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6227		set_ua = 1;
6228	}
6229	if (set_ua != 0) {
6230		int i;
6231		/*
6232		 * Let other initiators know that the mode
6233		 * parameters for this LUN have changed.
6234		 */
6235		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6236			if (i == initidx)
6237				continue;
6238
6239			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6240		}
6241	}
6242	mtx_unlock(&lun->lun_lock);
6243
6244	return (0);
6245}
6246
6247int
6248ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6249		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6250{
6251	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6252	struct ctl_lun *lun;
6253	int set_ua;
6254	uint32_t initidx;
6255
6256	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6257	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6258	set_ua = 0;
6259
6260	user_cp = (struct scsi_caching_page *)page_ptr;
6261	current_cp = (struct scsi_caching_page *)
6262		(page_index->page_data + (page_index->page_len *
6263		CTL_PAGE_CURRENT));
6264	saved_cp = (struct scsi_caching_page *)
6265		(page_index->page_data + (page_index->page_len *
6266		CTL_PAGE_SAVED));
6267
6268	mtx_lock(&lun->lun_lock);
6269	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6270	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6271		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6272		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6273		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6274		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6275		set_ua = 1;
6276	}
6277	if (set_ua != 0) {
6278		int i;
6279		/*
6280		 * Let other initiators know that the mode
6281		 * parameters for this LUN have changed.
6282		 */
6283		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6284			if (i == initidx)
6285				continue;
6286
6287			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6288		}
6289	}
6290	mtx_unlock(&lun->lun_lock);
6291
6292	return (0);
6293}
6294
6295int
6296ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6297		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6298{
6299	return (0);
6300}
6301
6302int
6303ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6304			   struct ctl_page_index *page_index, int pc)
6305{
6306	struct copan_power_subpage *page;
6307
6308	page = (struct copan_power_subpage *)page_index->page_data +
6309		(page_index->page_len * pc);
6310
6311	switch (pc) {
6312	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6313		/*
6314		 * We don't update the changable bits for this page.
6315		 */
6316		break;
6317	case SMS_PAGE_CTRL_CURRENT >> 6:
6318	case SMS_PAGE_CTRL_DEFAULT >> 6:
6319	case SMS_PAGE_CTRL_SAVED >> 6:
6320#ifdef NEEDTOPORT
6321		ctl_update_power_subpage(page);
6322#endif
6323		break;
6324	default:
6325#ifdef NEEDTOPORT
6326		EPRINT(0, "Invalid PC %d!!", pc);
6327#endif
6328		break;
6329	}
6330	return (0);
6331}
6332
6333
6334int
6335ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6336		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6337{
6338	struct copan_aps_subpage *user_sp;
6339	struct copan_aps_subpage *current_sp;
6340	union ctl_modepage_info *modepage_info;
6341	struct ctl_softc *softc;
6342	struct ctl_lun *lun;
6343	int retval;
6344
6345	retval = CTL_RETVAL_COMPLETE;
6346	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6347		     (page_index->page_len * CTL_PAGE_CURRENT));
6348	softc = control_softc;
6349	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6350
6351	user_sp = (struct copan_aps_subpage *)page_ptr;
6352
6353	modepage_info = (union ctl_modepage_info *)
6354		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6355
6356	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6357	modepage_info->header.subpage = page_index->subpage;
6358	modepage_info->aps.lock_active = user_sp->lock_active;
6359
6360	mtx_lock(&softc->ctl_lock);
6361
6362	/*
6363	 * If there is a request to lock the LUN and another LUN is locked
6364	 * this is an error. If the requested LUN is already locked ignore
6365	 * the request. If no LUN is locked attempt to lock it.
6366	 * if there is a request to unlock the LUN and the LUN is currently
6367	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6368	 * if another LUN is locked or no LUN is locked.
6369	 */
6370	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6371		if (softc->aps_locked_lun == lun->lun) {
6372			/*
6373			 * This LUN is already locked, so we're done.
6374			 */
6375			retval = CTL_RETVAL_COMPLETE;
6376		} else if (softc->aps_locked_lun == 0) {
6377			/*
6378			 * No one has the lock, pass the request to the
6379			 * backend.
6380			 */
6381			retval = lun->backend->config_write(
6382				(union ctl_io *)ctsio);
6383		} else {
6384			/*
6385			 * Someone else has the lock, throw out the request.
6386			 */
6387			ctl_set_already_locked(ctsio);
6388			free(ctsio->kern_data_ptr, M_CTL);
6389			ctl_done((union ctl_io *)ctsio);
6390
6391			/*
6392			 * Set the return value so that ctl_do_mode_select()
6393			 * won't try to complete the command.  We already
6394			 * completed it here.
6395			 */
6396			retval = CTL_RETVAL_ERROR;
6397		}
6398	} else if (softc->aps_locked_lun == lun->lun) {
6399		/*
6400		 * This LUN is locked, so pass the unlock request to the
6401		 * backend.
6402		 */
6403		retval = lun->backend->config_write((union ctl_io *)ctsio);
6404	}
6405	mtx_unlock(&softc->ctl_lock);
6406
6407	return (retval);
6408}
6409
6410int
6411ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6412				struct ctl_page_index *page_index,
6413				uint8_t *page_ptr)
6414{
6415	uint8_t *c;
6416	int i;
6417
6418	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6419	ctl_time_io_secs =
6420		(c[0] << 8) |
6421		(c[1] << 0) |
6422		0;
6423	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6424	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6425	printf("page data:");
6426	for (i=0; i<8; i++)
6427		printf(" %.2x",page_ptr[i]);
6428	printf("\n");
6429	return (0);
6430}
6431
6432int
6433ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6434			       struct ctl_page_index *page_index,
6435			       int pc)
6436{
6437	struct copan_debugconf_subpage *page;
6438
6439	page = (struct copan_debugconf_subpage *)page_index->page_data +
6440		(page_index->page_len * pc);
6441
6442	switch (pc) {
6443	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6444	case SMS_PAGE_CTRL_DEFAULT >> 6:
6445	case SMS_PAGE_CTRL_SAVED >> 6:
6446		/*
6447		 * We don't update the changable or default bits for this page.
6448		 */
6449		break;
6450	case SMS_PAGE_CTRL_CURRENT >> 6:
6451		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6452		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6453		break;
6454	default:
6455#ifdef NEEDTOPORT
6456		EPRINT(0, "Invalid PC %d!!", pc);
6457#endif /* NEEDTOPORT */
6458		break;
6459	}
6460	return (0);
6461}
6462
6463
6464static int
6465ctl_do_mode_select(union ctl_io *io)
6466{
6467	struct scsi_mode_page_header *page_header;
6468	struct ctl_page_index *page_index;
6469	struct ctl_scsiio *ctsio;
6470	int control_dev, page_len;
6471	int page_len_offset, page_len_size;
6472	union ctl_modepage_info *modepage_info;
6473	struct ctl_lun *lun;
6474	int *len_left, *len_used;
6475	int retval, i;
6476
6477	ctsio = &io->scsiio;
6478	page_index = NULL;
6479	page_len = 0;
6480	retval = CTL_RETVAL_COMPLETE;
6481
6482	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6483
6484	if (lun->be_lun->lun_type != T_DIRECT)
6485		control_dev = 1;
6486	else
6487		control_dev = 0;
6488
6489	modepage_info = (union ctl_modepage_info *)
6490		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6491	len_left = &modepage_info->header.len_left;
6492	len_used = &modepage_info->header.len_used;
6493
6494do_next_page:
6495
6496	page_header = (struct scsi_mode_page_header *)
6497		(ctsio->kern_data_ptr + *len_used);
6498
6499	if (*len_left == 0) {
6500		free(ctsio->kern_data_ptr, M_CTL);
6501		ctl_set_success(ctsio);
6502		ctl_done((union ctl_io *)ctsio);
6503		return (CTL_RETVAL_COMPLETE);
6504	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6505
6506		free(ctsio->kern_data_ptr, M_CTL);
6507		ctl_set_param_len_error(ctsio);
6508		ctl_done((union ctl_io *)ctsio);
6509		return (CTL_RETVAL_COMPLETE);
6510
6511	} else if ((page_header->page_code & SMPH_SPF)
6512		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6513
6514		free(ctsio->kern_data_ptr, M_CTL);
6515		ctl_set_param_len_error(ctsio);
6516		ctl_done((union ctl_io *)ctsio);
6517		return (CTL_RETVAL_COMPLETE);
6518	}
6519
6520
6521	/*
6522	 * XXX KDM should we do something with the block descriptor?
6523	 */
6524	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6525
6526		if ((control_dev != 0)
6527		 && (lun->mode_pages.index[i].page_flags &
6528		     CTL_PAGE_FLAG_DISK_ONLY))
6529			continue;
6530
6531		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6532		    (page_header->page_code & SMPH_PC_MASK))
6533			continue;
6534
6535		/*
6536		 * If neither page has a subpage code, then we've got a
6537		 * match.
6538		 */
6539		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6540		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6541			page_index = &lun->mode_pages.index[i];
6542			page_len = page_header->page_length;
6543			break;
6544		}
6545
6546		/*
6547		 * If both pages have subpages, then the subpage numbers
6548		 * have to match.
6549		 */
6550		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6551		  && (page_header->page_code & SMPH_SPF)) {
6552			struct scsi_mode_page_header_sp *sph;
6553
6554			sph = (struct scsi_mode_page_header_sp *)page_header;
6555
6556			if (lun->mode_pages.index[i].subpage ==
6557			    sph->subpage) {
6558				page_index = &lun->mode_pages.index[i];
6559				page_len = scsi_2btoul(sph->page_length);
6560				break;
6561			}
6562		}
6563	}
6564
6565	/*
6566	 * If we couldn't find the page, or if we don't have a mode select
6567	 * handler for it, send back an error to the user.
6568	 */
6569	if ((page_index == NULL)
6570	 || (page_index->select_handler == NULL)) {
6571		ctl_set_invalid_field(ctsio,
6572				      /*sks_valid*/ 1,
6573				      /*command*/ 0,
6574				      /*field*/ *len_used,
6575				      /*bit_valid*/ 0,
6576				      /*bit*/ 0);
6577		free(ctsio->kern_data_ptr, M_CTL);
6578		ctl_done((union ctl_io *)ctsio);
6579		return (CTL_RETVAL_COMPLETE);
6580	}
6581
6582	if (page_index->page_code & SMPH_SPF) {
6583		page_len_offset = 2;
6584		page_len_size = 2;
6585	} else {
6586		page_len_size = 1;
6587		page_len_offset = 1;
6588	}
6589
6590	/*
6591	 * If the length the initiator gives us isn't the one we specify in
6592	 * the mode page header, or if they didn't specify enough data in
6593	 * the CDB to avoid truncating this page, kick out the request.
6594	 */
6595	if ((page_len != (page_index->page_len - page_len_offset -
6596			  page_len_size))
6597	 || (*len_left < page_index->page_len)) {
6598
6599
6600		ctl_set_invalid_field(ctsio,
6601				      /*sks_valid*/ 1,
6602				      /*command*/ 0,
6603				      /*field*/ *len_used + page_len_offset,
6604				      /*bit_valid*/ 0,
6605				      /*bit*/ 0);
6606		free(ctsio->kern_data_ptr, M_CTL);
6607		ctl_done((union ctl_io *)ctsio);
6608		return (CTL_RETVAL_COMPLETE);
6609	}
6610
6611	/*
6612	 * Run through the mode page, checking to make sure that the bits
6613	 * the user changed are actually legal for him to change.
6614	 */
6615	for (i = 0; i < page_index->page_len; i++) {
6616		uint8_t *user_byte, *change_mask, *current_byte;
6617		int bad_bit;
6618		int j;
6619
6620		user_byte = (uint8_t *)page_header + i;
6621		change_mask = page_index->page_data +
6622			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6623		current_byte = page_index->page_data +
6624			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6625
6626		/*
6627		 * Check to see whether the user set any bits in this byte
6628		 * that he is not allowed to set.
6629		 */
6630		if ((*user_byte & ~(*change_mask)) ==
6631		    (*current_byte & ~(*change_mask)))
6632			continue;
6633
6634		/*
6635		 * Go through bit by bit to determine which one is illegal.
6636		 */
6637		bad_bit = 0;
6638		for (j = 7; j >= 0; j--) {
6639			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6640			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6641				bad_bit = i;
6642				break;
6643			}
6644		}
6645		ctl_set_invalid_field(ctsio,
6646				      /*sks_valid*/ 1,
6647				      /*command*/ 0,
6648				      /*field*/ *len_used + i,
6649				      /*bit_valid*/ 1,
6650				      /*bit*/ bad_bit);
6651		free(ctsio->kern_data_ptr, M_CTL);
6652		ctl_done((union ctl_io *)ctsio);
6653		return (CTL_RETVAL_COMPLETE);
6654	}
6655
6656	/*
6657	 * Decrement these before we call the page handler, since we may
6658	 * end up getting called back one way or another before the handler
6659	 * returns to this context.
6660	 */
6661	*len_left -= page_index->page_len;
6662	*len_used += page_index->page_len;
6663
6664	retval = page_index->select_handler(ctsio, page_index,
6665					    (uint8_t *)page_header);
6666
6667	/*
6668	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6669	 * wait until this queued command completes to finish processing
6670	 * the mode page.  If it returns anything other than
6671	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6672	 * already set the sense information, freed the data pointer, and
6673	 * completed the io for us.
6674	 */
6675	if (retval != CTL_RETVAL_COMPLETE)
6676		goto bailout_no_done;
6677
6678	/*
6679	 * If the initiator sent us more than one page, parse the next one.
6680	 */
6681	if (*len_left > 0)
6682		goto do_next_page;
6683
6684	ctl_set_success(ctsio);
6685	free(ctsio->kern_data_ptr, M_CTL);
6686	ctl_done((union ctl_io *)ctsio);
6687
6688bailout_no_done:
6689
6690	return (CTL_RETVAL_COMPLETE);
6691
6692}
6693
6694int
6695ctl_mode_select(struct ctl_scsiio *ctsio)
6696{
6697	int param_len, pf, sp;
6698	int header_size, bd_len;
6699	int len_left, len_used;
6700	struct ctl_page_index *page_index;
6701	struct ctl_lun *lun;
6702	int control_dev, page_len;
6703	union ctl_modepage_info *modepage_info;
6704	int retval;
6705
6706	pf = 0;
6707	sp = 0;
6708	page_len = 0;
6709	len_used = 0;
6710	len_left = 0;
6711	retval = 0;
6712	bd_len = 0;
6713	page_index = NULL;
6714
6715	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6716
6717	if (lun->be_lun->lun_type != T_DIRECT)
6718		control_dev = 1;
6719	else
6720		control_dev = 0;
6721
6722	switch (ctsio->cdb[0]) {
6723	case MODE_SELECT_6: {
6724		struct scsi_mode_select_6 *cdb;
6725
6726		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6727
6728		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6729		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6730
6731		param_len = cdb->length;
6732		header_size = sizeof(struct scsi_mode_header_6);
6733		break;
6734	}
6735	case MODE_SELECT_10: {
6736		struct scsi_mode_select_10 *cdb;
6737
6738		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6739
6740		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6741		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6742
6743		param_len = scsi_2btoul(cdb->length);
6744		header_size = sizeof(struct scsi_mode_header_10);
6745		break;
6746	}
6747	default:
6748		ctl_set_invalid_opcode(ctsio);
6749		ctl_done((union ctl_io *)ctsio);
6750		return (CTL_RETVAL_COMPLETE);
6751		break; /* NOTREACHED */
6752	}
6753
6754	/*
6755	 * From SPC-3:
6756	 * "A parameter list length of zero indicates that the Data-Out Buffer
6757	 * shall be empty. This condition shall not be considered as an error."
6758	 */
6759	if (param_len == 0) {
6760		ctl_set_success(ctsio);
6761		ctl_done((union ctl_io *)ctsio);
6762		return (CTL_RETVAL_COMPLETE);
6763	}
6764
6765	/*
6766	 * Since we'll hit this the first time through, prior to
6767	 * allocation, we don't need to free a data buffer here.
6768	 */
6769	if (param_len < header_size) {
6770		ctl_set_param_len_error(ctsio);
6771		ctl_done((union ctl_io *)ctsio);
6772		return (CTL_RETVAL_COMPLETE);
6773	}
6774
6775	/*
6776	 * Allocate the data buffer and grab the user's data.  In theory,
6777	 * we shouldn't have to sanity check the parameter list length here
6778	 * because the maximum size is 64K.  We should be able to malloc
6779	 * that much without too many problems.
6780	 */
6781	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6782		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6783		ctsio->kern_data_len = param_len;
6784		ctsio->kern_total_len = param_len;
6785		ctsio->kern_data_resid = 0;
6786		ctsio->kern_rel_offset = 0;
6787		ctsio->kern_sg_entries = 0;
6788		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6789		ctsio->be_move_done = ctl_config_move_done;
6790		ctl_datamove((union ctl_io *)ctsio);
6791
6792		return (CTL_RETVAL_COMPLETE);
6793	}
6794
6795	switch (ctsio->cdb[0]) {
6796	case MODE_SELECT_6: {
6797		struct scsi_mode_header_6 *mh6;
6798
6799		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6800		bd_len = mh6->blk_desc_len;
6801		break;
6802	}
6803	case MODE_SELECT_10: {
6804		struct scsi_mode_header_10 *mh10;
6805
6806		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6807		bd_len = scsi_2btoul(mh10->blk_desc_len);
6808		break;
6809	}
6810	default:
6811		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6812		break;
6813	}
6814
6815	if (param_len < (header_size + bd_len)) {
6816		free(ctsio->kern_data_ptr, M_CTL);
6817		ctl_set_param_len_error(ctsio);
6818		ctl_done((union ctl_io *)ctsio);
6819		return (CTL_RETVAL_COMPLETE);
6820	}
6821
6822	/*
6823	 * Set the IO_CONT flag, so that if this I/O gets passed to
6824	 * ctl_config_write_done(), it'll get passed back to
6825	 * ctl_do_mode_select() for further processing, or completion if
6826	 * we're all done.
6827	 */
6828	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6829	ctsio->io_cont = ctl_do_mode_select;
6830
6831	modepage_info = (union ctl_modepage_info *)
6832		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6833
6834	memset(modepage_info, 0, sizeof(*modepage_info));
6835
6836	len_left = param_len - header_size - bd_len;
6837	len_used = header_size + bd_len;
6838
6839	modepage_info->header.len_left = len_left;
6840	modepage_info->header.len_used = len_used;
6841
6842	return (ctl_do_mode_select((union ctl_io *)ctsio));
6843}
6844
6845int
6846ctl_mode_sense(struct ctl_scsiio *ctsio)
6847{
6848	struct ctl_lun *lun;
6849	int pc, page_code, dbd, llba, subpage;
6850	int alloc_len, page_len, header_len, total_len;
6851	struct scsi_mode_block_descr *block_desc;
6852	struct ctl_page_index *page_index;
6853	int control_dev;
6854
6855	dbd = 0;
6856	llba = 0;
6857	block_desc = NULL;
6858	page_index = NULL;
6859
6860	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6861
6862	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6863
6864	if (lun->be_lun->lun_type != T_DIRECT)
6865		control_dev = 1;
6866	else
6867		control_dev = 0;
6868
6869	if (lun->flags & CTL_LUN_PR_RESERVED) {
6870		uint32_t residx;
6871
6872		/*
6873		 * XXX KDM need a lock here.
6874		 */
6875		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6876		if ((lun->res_type == SPR_TYPE_EX_AC
6877		  && residx != lun->pr_res_idx)
6878		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6879		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6880		  && !lun->per_res[residx].registered)) {
6881			ctl_set_reservation_conflict(ctsio);
6882			ctl_done((union ctl_io *)ctsio);
6883			return (CTL_RETVAL_COMPLETE);
6884		}
6885	}
6886
6887	switch (ctsio->cdb[0]) {
6888	case MODE_SENSE_6: {
6889		struct scsi_mode_sense_6 *cdb;
6890
6891		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6892
6893		header_len = sizeof(struct scsi_mode_hdr_6);
6894		if (cdb->byte2 & SMS_DBD)
6895			dbd = 1;
6896		else
6897			header_len += sizeof(struct scsi_mode_block_descr);
6898
6899		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6900		page_code = cdb->page & SMS_PAGE_CODE;
6901		subpage = cdb->subpage;
6902		alloc_len = cdb->length;
6903		break;
6904	}
6905	case MODE_SENSE_10: {
6906		struct scsi_mode_sense_10 *cdb;
6907
6908		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6909
6910		header_len = sizeof(struct scsi_mode_hdr_10);
6911
6912		if (cdb->byte2 & SMS_DBD)
6913			dbd = 1;
6914		else
6915			header_len += sizeof(struct scsi_mode_block_descr);
6916		if (cdb->byte2 & SMS10_LLBAA)
6917			llba = 1;
6918		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6919		page_code = cdb->page & SMS_PAGE_CODE;
6920		subpage = cdb->subpage;
6921		alloc_len = scsi_2btoul(cdb->length);
6922		break;
6923	}
6924	default:
6925		ctl_set_invalid_opcode(ctsio);
6926		ctl_done((union ctl_io *)ctsio);
6927		return (CTL_RETVAL_COMPLETE);
6928		break; /* NOTREACHED */
6929	}
6930
6931	/*
6932	 * We have to make a first pass through to calculate the size of
6933	 * the pages that match the user's query.  Then we allocate enough
6934	 * memory to hold it, and actually copy the data into the buffer.
6935	 */
6936	switch (page_code) {
6937	case SMS_ALL_PAGES_PAGE: {
6938		int i;
6939
6940		page_len = 0;
6941
6942		/*
6943		 * At the moment, values other than 0 and 0xff here are
6944		 * reserved according to SPC-3.
6945		 */
6946		if ((subpage != SMS_SUBPAGE_PAGE_0)
6947		 && (subpage != SMS_SUBPAGE_ALL)) {
6948			ctl_set_invalid_field(ctsio,
6949					      /*sks_valid*/ 1,
6950					      /*command*/ 1,
6951					      /*field*/ 3,
6952					      /*bit_valid*/ 0,
6953					      /*bit*/ 0);
6954			ctl_done((union ctl_io *)ctsio);
6955			return (CTL_RETVAL_COMPLETE);
6956		}
6957
6958		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6959			if ((control_dev != 0)
6960			 && (lun->mode_pages.index[i].page_flags &
6961			     CTL_PAGE_FLAG_DISK_ONLY))
6962				continue;
6963
6964			/*
6965			 * We don't use this subpage if the user didn't
6966			 * request all subpages.
6967			 */
6968			if ((lun->mode_pages.index[i].subpage != 0)
6969			 && (subpage == SMS_SUBPAGE_PAGE_0))
6970				continue;
6971
6972#if 0
6973			printf("found page %#x len %d\n",
6974			       lun->mode_pages.index[i].page_code &
6975			       SMPH_PC_MASK,
6976			       lun->mode_pages.index[i].page_len);
6977#endif
6978			page_len += lun->mode_pages.index[i].page_len;
6979		}
6980		break;
6981	}
6982	default: {
6983		int i;
6984
6985		page_len = 0;
6986
6987		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6988			/* Look for the right page code */
6989			if ((lun->mode_pages.index[i].page_code &
6990			     SMPH_PC_MASK) != page_code)
6991				continue;
6992
6993			/* Look for the right subpage or the subpage wildcard*/
6994			if ((lun->mode_pages.index[i].subpage != subpage)
6995			 && (subpage != SMS_SUBPAGE_ALL))
6996				continue;
6997
6998			/* Make sure the page is supported for this dev type */
6999			if ((control_dev != 0)
7000			 && (lun->mode_pages.index[i].page_flags &
7001			     CTL_PAGE_FLAG_DISK_ONLY))
7002				continue;
7003
7004#if 0
7005			printf("found page %#x len %d\n",
7006			       lun->mode_pages.index[i].page_code &
7007			       SMPH_PC_MASK,
7008			       lun->mode_pages.index[i].page_len);
7009#endif
7010
7011			page_len += lun->mode_pages.index[i].page_len;
7012		}
7013
7014		if (page_len == 0) {
7015			ctl_set_invalid_field(ctsio,
7016					      /*sks_valid*/ 1,
7017					      /*command*/ 1,
7018					      /*field*/ 2,
7019					      /*bit_valid*/ 1,
7020					      /*bit*/ 5);
7021			ctl_done((union ctl_io *)ctsio);
7022			return (CTL_RETVAL_COMPLETE);
7023		}
7024		break;
7025	}
7026	}
7027
7028	total_len = header_len + page_len;
7029#if 0
7030	printf("header_len = %d, page_len = %d, total_len = %d\n",
7031	       header_len, page_len, total_len);
7032#endif
7033
7034	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7035	ctsio->kern_sg_entries = 0;
7036	ctsio->kern_data_resid = 0;
7037	ctsio->kern_rel_offset = 0;
7038	if (total_len < alloc_len) {
7039		ctsio->residual = alloc_len - total_len;
7040		ctsio->kern_data_len = total_len;
7041		ctsio->kern_total_len = total_len;
7042	} else {
7043		ctsio->residual = 0;
7044		ctsio->kern_data_len = alloc_len;
7045		ctsio->kern_total_len = alloc_len;
7046	}
7047
7048	switch (ctsio->cdb[0]) {
7049	case MODE_SENSE_6: {
7050		struct scsi_mode_hdr_6 *header;
7051
7052		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7053
7054		header->datalen = ctl_min(total_len - 1, 254);
7055		if (control_dev == 0)
7056			header->dev_specific = 0x10; /* DPOFUA */
7057		if (dbd)
7058			header->block_descr_len = 0;
7059		else
7060			header->block_descr_len =
7061				sizeof(struct scsi_mode_block_descr);
7062		block_desc = (struct scsi_mode_block_descr *)&header[1];
7063		break;
7064	}
7065	case MODE_SENSE_10: {
7066		struct scsi_mode_hdr_10 *header;
7067		int datalen;
7068
7069		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7070
7071		datalen = ctl_min(total_len - 2, 65533);
7072		scsi_ulto2b(datalen, header->datalen);
7073		if (control_dev == 0)
7074			header->dev_specific = 0x10; /* DPOFUA */
7075		if (dbd)
7076			scsi_ulto2b(0, header->block_descr_len);
7077		else
7078			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7079				    header->block_descr_len);
7080		block_desc = (struct scsi_mode_block_descr *)&header[1];
7081		break;
7082	}
7083	default:
7084		panic("invalid CDB type %#x", ctsio->cdb[0]);
7085		break; /* NOTREACHED */
7086	}
7087
7088	/*
7089	 * If we've got a disk, use its blocksize in the block
7090	 * descriptor.  Otherwise, just set it to 0.
7091	 */
7092	if (dbd == 0) {
7093		if (control_dev == 0)
7094			scsi_ulto3b(lun->be_lun->blocksize,
7095				    block_desc->block_len);
7096		else
7097			scsi_ulto3b(0, block_desc->block_len);
7098	}
7099
7100	switch (page_code) {
7101	case SMS_ALL_PAGES_PAGE: {
7102		int i, data_used;
7103
7104		data_used = header_len;
7105		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7106			struct ctl_page_index *page_index;
7107
7108			page_index = &lun->mode_pages.index[i];
7109
7110			if ((control_dev != 0)
7111			 && (page_index->page_flags &
7112			    CTL_PAGE_FLAG_DISK_ONLY))
7113				continue;
7114
7115			/*
7116			 * We don't use this subpage if the user didn't
7117			 * request all subpages.  We already checked (above)
7118			 * to make sure the user only specified a subpage
7119			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7120			 */
7121			if ((page_index->subpage != 0)
7122			 && (subpage == SMS_SUBPAGE_PAGE_0))
7123				continue;
7124
7125			/*
7126			 * Call the handler, if it exists, to update the
7127			 * page to the latest values.
7128			 */
7129			if (page_index->sense_handler != NULL)
7130				page_index->sense_handler(ctsio, page_index,pc);
7131
7132			memcpy(ctsio->kern_data_ptr + data_used,
7133			       page_index->page_data +
7134			       (page_index->page_len * pc),
7135			       page_index->page_len);
7136			data_used += page_index->page_len;
7137		}
7138		break;
7139	}
7140	default: {
7141		int i, data_used;
7142
7143		data_used = header_len;
7144
7145		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7146			struct ctl_page_index *page_index;
7147
7148			page_index = &lun->mode_pages.index[i];
7149
7150			/* Look for the right page code */
7151			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7152				continue;
7153
7154			/* Look for the right subpage or the subpage wildcard*/
7155			if ((page_index->subpage != subpage)
7156			 && (subpage != SMS_SUBPAGE_ALL))
7157				continue;
7158
7159			/* Make sure the page is supported for this dev type */
7160			if ((control_dev != 0)
7161			 && (page_index->page_flags &
7162			     CTL_PAGE_FLAG_DISK_ONLY))
7163				continue;
7164
7165			/*
7166			 * Call the handler, if it exists, to update the
7167			 * page to the latest values.
7168			 */
7169			if (page_index->sense_handler != NULL)
7170				page_index->sense_handler(ctsio, page_index,pc);
7171
7172			memcpy(ctsio->kern_data_ptr + data_used,
7173			       page_index->page_data +
7174			       (page_index->page_len * pc),
7175			       page_index->page_len);
7176			data_used += page_index->page_len;
7177		}
7178		break;
7179	}
7180	}
7181
7182	ctsio->scsi_status = SCSI_STATUS_OK;
7183
7184	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7185	ctsio->be_move_done = ctl_config_move_done;
7186	ctl_datamove((union ctl_io *)ctsio);
7187
7188	return (CTL_RETVAL_COMPLETE);
7189}
7190
7191int
7192ctl_read_capacity(struct ctl_scsiio *ctsio)
7193{
7194	struct scsi_read_capacity *cdb;
7195	struct scsi_read_capacity_data *data;
7196	struct ctl_lun *lun;
7197	uint32_t lba;
7198
7199	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7200
7201	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7202
7203	lba = scsi_4btoul(cdb->addr);
7204	if (((cdb->pmi & SRC_PMI) == 0)
7205	 && (lba != 0)) {
7206		ctl_set_invalid_field(/*ctsio*/ ctsio,
7207				      /*sks_valid*/ 1,
7208				      /*command*/ 1,
7209				      /*field*/ 2,
7210				      /*bit_valid*/ 0,
7211				      /*bit*/ 0);
7212		ctl_done((union ctl_io *)ctsio);
7213		return (CTL_RETVAL_COMPLETE);
7214	}
7215
7216	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7217
7218	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7219	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7220	ctsio->residual = 0;
7221	ctsio->kern_data_len = sizeof(*data);
7222	ctsio->kern_total_len = sizeof(*data);
7223	ctsio->kern_data_resid = 0;
7224	ctsio->kern_rel_offset = 0;
7225	ctsio->kern_sg_entries = 0;
7226
7227	/*
7228	 * If the maximum LBA is greater than 0xfffffffe, the user must
7229	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7230	 * serivce action set.
7231	 */
7232	if (lun->be_lun->maxlba > 0xfffffffe)
7233		scsi_ulto4b(0xffffffff, data->addr);
7234	else
7235		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7236
7237	/*
7238	 * XXX KDM this may not be 512 bytes...
7239	 */
7240	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7241
7242	ctsio->scsi_status = SCSI_STATUS_OK;
7243
7244	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7245	ctsio->be_move_done = ctl_config_move_done;
7246	ctl_datamove((union ctl_io *)ctsio);
7247
7248	return (CTL_RETVAL_COMPLETE);
7249}
7250
7251int
7252ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7253{
7254	struct scsi_read_capacity_16 *cdb;
7255	struct scsi_read_capacity_data_long *data;
7256	struct ctl_lun *lun;
7257	uint64_t lba;
7258	uint32_t alloc_len;
7259
7260	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7261
7262	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7263
7264	alloc_len = scsi_4btoul(cdb->alloc_len);
7265	lba = scsi_8btou64(cdb->addr);
7266
7267	if ((cdb->reladr & SRC16_PMI)
7268	 && (lba != 0)) {
7269		ctl_set_invalid_field(/*ctsio*/ ctsio,
7270				      /*sks_valid*/ 1,
7271				      /*command*/ 1,
7272				      /*field*/ 2,
7273				      /*bit_valid*/ 0,
7274				      /*bit*/ 0);
7275		ctl_done((union ctl_io *)ctsio);
7276		return (CTL_RETVAL_COMPLETE);
7277	}
7278
7279	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7280
7281	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7282	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7283
7284	if (sizeof(*data) < alloc_len) {
7285		ctsio->residual = alloc_len - sizeof(*data);
7286		ctsio->kern_data_len = sizeof(*data);
7287		ctsio->kern_total_len = sizeof(*data);
7288	} else {
7289		ctsio->residual = 0;
7290		ctsio->kern_data_len = alloc_len;
7291		ctsio->kern_total_len = alloc_len;
7292	}
7293	ctsio->kern_data_resid = 0;
7294	ctsio->kern_rel_offset = 0;
7295	ctsio->kern_sg_entries = 0;
7296
7297	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7298	/* XXX KDM this may not be 512 bytes... */
7299	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7300	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7301	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7302	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7303		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7304
7305	ctsio->scsi_status = SCSI_STATUS_OK;
7306
7307	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7308	ctsio->be_move_done = ctl_config_move_done;
7309	ctl_datamove((union ctl_io *)ctsio);
7310
7311	return (CTL_RETVAL_COMPLETE);
7312}
7313
7314int
7315ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7316{
7317	struct scsi_maintenance_in *cdb;
7318	int retval;
7319	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7320	int num_target_port_groups, num_target_ports, single;
7321	struct ctl_lun *lun;
7322	struct ctl_softc *softc;
7323	struct ctl_port *port;
7324	struct scsi_target_group_data *rtg_ptr;
7325	struct scsi_target_group_data_extended *rtg_ext_ptr;
7326	struct scsi_target_port_group_descriptor *tpg_desc;
7327
7328	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7329
7330	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7331	softc = control_softc;
7332	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7333
7334	retval = CTL_RETVAL_COMPLETE;
7335
7336	switch (cdb->byte2 & STG_PDF_MASK) {
7337	case STG_PDF_LENGTH:
7338		ext = 0;
7339		break;
7340	case STG_PDF_EXTENDED:
7341		ext = 1;
7342		break;
7343	default:
7344		ctl_set_invalid_field(/*ctsio*/ ctsio,
7345				      /*sks_valid*/ 1,
7346				      /*command*/ 1,
7347				      /*field*/ 2,
7348				      /*bit_valid*/ 1,
7349				      /*bit*/ 5);
7350		ctl_done((union ctl_io *)ctsio);
7351		return(retval);
7352	}
7353
7354	single = ctl_is_single;
7355	if (single)
7356		num_target_port_groups = 1;
7357	else
7358		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7359	num_target_ports = 0;
7360	mtx_lock(&softc->ctl_lock);
7361	STAILQ_FOREACH(port, &softc->port_list, links) {
7362		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7363			continue;
7364		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7365			continue;
7366		num_target_ports++;
7367	}
7368	mtx_unlock(&softc->ctl_lock);
7369
7370	if (ext)
7371		total_len = sizeof(struct scsi_target_group_data_extended);
7372	else
7373		total_len = sizeof(struct scsi_target_group_data);
7374	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7375		num_target_port_groups +
7376	    sizeof(struct scsi_target_port_descriptor) *
7377		num_target_ports * num_target_port_groups;
7378
7379	alloc_len = scsi_4btoul(cdb->length);
7380
7381	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7382
7383	ctsio->kern_sg_entries = 0;
7384
7385	if (total_len < alloc_len) {
7386		ctsio->residual = alloc_len - total_len;
7387		ctsio->kern_data_len = total_len;
7388		ctsio->kern_total_len = total_len;
7389	} else {
7390		ctsio->residual = 0;
7391		ctsio->kern_data_len = alloc_len;
7392		ctsio->kern_total_len = alloc_len;
7393	}
7394	ctsio->kern_data_resid = 0;
7395	ctsio->kern_rel_offset = 0;
7396
7397	if (ext) {
7398		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7399		    ctsio->kern_data_ptr;
7400		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7401		rtg_ext_ptr->format_type = 0x10;
7402		rtg_ext_ptr->implicit_transition_time = 0;
7403		tpg_desc = &rtg_ext_ptr->groups[0];
7404	} else {
7405		rtg_ptr = (struct scsi_target_group_data *)
7406		    ctsio->kern_data_ptr;
7407		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7408		tpg_desc = &rtg_ptr->groups[0];
7409	}
7410
7411	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7412	mtx_lock(&softc->ctl_lock);
7413	for (g = 0; g < num_target_port_groups; g++) {
7414		if (g == pg)
7415			tpg_desc->pref_state = TPG_PRIMARY |
7416			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7417		else
7418			tpg_desc->pref_state =
7419			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7420		tpg_desc->support = TPG_AO_SUP;
7421		if (!single)
7422			tpg_desc->support |= TPG_AN_SUP;
7423		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7424		tpg_desc->status = TPG_IMPLICIT;
7425		pc = 0;
7426		STAILQ_FOREACH(port, &softc->port_list, links) {
7427			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7428				continue;
7429			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7430			    CTL_MAX_LUNS)
7431				continue;
7432			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7433			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7434			    relative_target_port_identifier);
7435			pc++;
7436		}
7437		tpg_desc->target_port_count = pc;
7438		tpg_desc = (struct scsi_target_port_group_descriptor *)
7439		    &tpg_desc->descriptors[pc];
7440	}
7441	mtx_unlock(&softc->ctl_lock);
7442
7443	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7444	ctsio->be_move_done = ctl_config_move_done;
7445
7446	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7447			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7448			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7449			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7450			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7451
7452	ctl_datamove((union ctl_io *)ctsio);
7453	return(retval);
7454}
7455
7456int
7457ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7458{
7459	struct ctl_lun *lun;
7460	struct scsi_report_supported_opcodes *cdb;
7461	const struct ctl_cmd_entry *entry, *sentry;
7462	struct scsi_report_supported_opcodes_all *all;
7463	struct scsi_report_supported_opcodes_descr *descr;
7464	struct scsi_report_supported_opcodes_one *one;
7465	int retval;
7466	int alloc_len, total_len;
7467	int opcode, service_action, i, j, num;
7468
7469	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7470
7471	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7472	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7473
7474	retval = CTL_RETVAL_COMPLETE;
7475
7476	opcode = cdb->requested_opcode;
7477	service_action = scsi_2btoul(cdb->requested_service_action);
7478	switch (cdb->options & RSO_OPTIONS_MASK) {
7479	case RSO_OPTIONS_ALL:
7480		num = 0;
7481		for (i = 0; i < 256; i++) {
7482			entry = &ctl_cmd_table[i];
7483			if (entry->flags & CTL_CMD_FLAG_SA5) {
7484				for (j = 0; j < 32; j++) {
7485					sentry = &((const struct ctl_cmd_entry *)
7486					    entry->execute)[j];
7487					if (ctl_cmd_applicable(
7488					    lun->be_lun->lun_type, sentry))
7489						num++;
7490				}
7491			} else {
7492				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7493				    entry))
7494					num++;
7495			}
7496		}
7497		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7498		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7499		break;
7500	case RSO_OPTIONS_OC:
7501		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7502			ctl_set_invalid_field(/*ctsio*/ ctsio,
7503					      /*sks_valid*/ 1,
7504					      /*command*/ 1,
7505					      /*field*/ 2,
7506					      /*bit_valid*/ 1,
7507					      /*bit*/ 2);
7508			ctl_done((union ctl_io *)ctsio);
7509			return (CTL_RETVAL_COMPLETE);
7510		}
7511		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7512		break;
7513	case RSO_OPTIONS_OC_SA:
7514		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7515		    service_action >= 32) {
7516			ctl_set_invalid_field(/*ctsio*/ ctsio,
7517					      /*sks_valid*/ 1,
7518					      /*command*/ 1,
7519					      /*field*/ 2,
7520					      /*bit_valid*/ 1,
7521					      /*bit*/ 2);
7522			ctl_done((union ctl_io *)ctsio);
7523			return (CTL_RETVAL_COMPLETE);
7524		}
7525		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7526		break;
7527	default:
7528		ctl_set_invalid_field(/*ctsio*/ ctsio,
7529				      /*sks_valid*/ 1,
7530				      /*command*/ 1,
7531				      /*field*/ 2,
7532				      /*bit_valid*/ 1,
7533				      /*bit*/ 2);
7534		ctl_done((union ctl_io *)ctsio);
7535		return (CTL_RETVAL_COMPLETE);
7536	}
7537
7538	alloc_len = scsi_4btoul(cdb->length);
7539
7540	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7541
7542	ctsio->kern_sg_entries = 0;
7543
7544	if (total_len < alloc_len) {
7545		ctsio->residual = alloc_len - total_len;
7546		ctsio->kern_data_len = total_len;
7547		ctsio->kern_total_len = total_len;
7548	} else {
7549		ctsio->residual = 0;
7550		ctsio->kern_data_len = alloc_len;
7551		ctsio->kern_total_len = alloc_len;
7552	}
7553	ctsio->kern_data_resid = 0;
7554	ctsio->kern_rel_offset = 0;
7555
7556	switch (cdb->options & RSO_OPTIONS_MASK) {
7557	case RSO_OPTIONS_ALL:
7558		all = (struct scsi_report_supported_opcodes_all *)
7559		    ctsio->kern_data_ptr;
7560		num = 0;
7561		for (i = 0; i < 256; i++) {
7562			entry = &ctl_cmd_table[i];
7563			if (entry->flags & CTL_CMD_FLAG_SA5) {
7564				for (j = 0; j < 32; j++) {
7565					sentry = &((const struct ctl_cmd_entry *)
7566					    entry->execute)[j];
7567					if (!ctl_cmd_applicable(
7568					    lun->be_lun->lun_type, sentry))
7569						continue;
7570					descr = &all->descr[num++];
7571					descr->opcode = i;
7572					scsi_ulto2b(j, descr->service_action);
7573					descr->flags = RSO_SERVACTV;
7574					scsi_ulto2b(sentry->length,
7575					    descr->cdb_length);
7576				}
7577			} else {
7578				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7579				    entry))
7580					continue;
7581				descr = &all->descr[num++];
7582				descr->opcode = i;
7583				scsi_ulto2b(0, descr->service_action);
7584				descr->flags = 0;
7585				scsi_ulto2b(entry->length, descr->cdb_length);
7586			}
7587		}
7588		scsi_ulto4b(
7589		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7590		    all->length);
7591		break;
7592	case RSO_OPTIONS_OC:
7593		one = (struct scsi_report_supported_opcodes_one *)
7594		    ctsio->kern_data_ptr;
7595		entry = &ctl_cmd_table[opcode];
7596		goto fill_one;
7597	case RSO_OPTIONS_OC_SA:
7598		one = (struct scsi_report_supported_opcodes_one *)
7599		    ctsio->kern_data_ptr;
7600		entry = &ctl_cmd_table[opcode];
7601		entry = &((const struct ctl_cmd_entry *)
7602		    entry->execute)[service_action];
7603fill_one:
7604		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7605			one->support = 3;
7606			scsi_ulto2b(entry->length, one->cdb_length);
7607			one->cdb_usage[0] = opcode;
7608			memcpy(&one->cdb_usage[1], entry->usage,
7609			    entry->length - 1);
7610		} else
7611			one->support = 1;
7612		break;
7613	}
7614
7615	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7616	ctsio->be_move_done = ctl_config_move_done;
7617
7618	ctl_datamove((union ctl_io *)ctsio);
7619	return(retval);
7620}
7621
7622int
7623ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7624{
7625	struct ctl_lun *lun;
7626	struct scsi_report_supported_tmf *cdb;
7627	struct scsi_report_supported_tmf_data *data;
7628	int retval;
7629	int alloc_len, total_len;
7630
7631	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7632
7633	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7634	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7635
7636	retval = CTL_RETVAL_COMPLETE;
7637
7638	total_len = sizeof(struct scsi_report_supported_tmf_data);
7639	alloc_len = scsi_4btoul(cdb->length);
7640
7641	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7642
7643	ctsio->kern_sg_entries = 0;
7644
7645	if (total_len < alloc_len) {
7646		ctsio->residual = alloc_len - total_len;
7647		ctsio->kern_data_len = total_len;
7648		ctsio->kern_total_len = total_len;
7649	} else {
7650		ctsio->residual = 0;
7651		ctsio->kern_data_len = alloc_len;
7652		ctsio->kern_total_len = alloc_len;
7653	}
7654	ctsio->kern_data_resid = 0;
7655	ctsio->kern_rel_offset = 0;
7656
7657	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7658	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7659	data->byte2 |= RST_ITNRS;
7660
7661	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7662	ctsio->be_move_done = ctl_config_move_done;
7663
7664	ctl_datamove((union ctl_io *)ctsio);
7665	return (retval);
7666}
7667
7668int
7669ctl_report_timestamp(struct ctl_scsiio *ctsio)
7670{
7671	struct ctl_lun *lun;
7672	struct scsi_report_timestamp *cdb;
7673	struct scsi_report_timestamp_data *data;
7674	struct timeval tv;
7675	int64_t timestamp;
7676	int retval;
7677	int alloc_len, total_len;
7678
7679	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7680
7681	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7682	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7683
7684	retval = CTL_RETVAL_COMPLETE;
7685
7686	total_len = sizeof(struct scsi_report_timestamp_data);
7687	alloc_len = scsi_4btoul(cdb->length);
7688
7689	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7690
7691	ctsio->kern_sg_entries = 0;
7692
7693	if (total_len < alloc_len) {
7694		ctsio->residual = alloc_len - total_len;
7695		ctsio->kern_data_len = total_len;
7696		ctsio->kern_total_len = total_len;
7697	} else {
7698		ctsio->residual = 0;
7699		ctsio->kern_data_len = alloc_len;
7700		ctsio->kern_total_len = alloc_len;
7701	}
7702	ctsio->kern_data_resid = 0;
7703	ctsio->kern_rel_offset = 0;
7704
7705	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7706	scsi_ulto2b(sizeof(*data) - 2, data->length);
7707	data->origin = RTS_ORIG_OUTSIDE;
7708	getmicrotime(&tv);
7709	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7710	scsi_ulto4b(timestamp >> 16, data->timestamp);
7711	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7712
7713	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7714	ctsio->be_move_done = ctl_config_move_done;
7715
7716	ctl_datamove((union ctl_io *)ctsio);
7717	return (retval);
7718}
7719
7720int
7721ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7722{
7723	struct scsi_per_res_in *cdb;
7724	int alloc_len, total_len = 0;
7725	/* struct scsi_per_res_in_rsrv in_data; */
7726	struct ctl_lun *lun;
7727	struct ctl_softc *softc;
7728
7729	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7730
7731	softc = control_softc;
7732
7733	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7734
7735	alloc_len = scsi_2btoul(cdb->length);
7736
7737	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7738
7739retry:
7740	mtx_lock(&lun->lun_lock);
7741	switch (cdb->action) {
7742	case SPRI_RK: /* read keys */
7743		total_len = sizeof(struct scsi_per_res_in_keys) +
7744			lun->pr_key_count *
7745			sizeof(struct scsi_per_res_key);
7746		break;
7747	case SPRI_RR: /* read reservation */
7748		if (lun->flags & CTL_LUN_PR_RESERVED)
7749			total_len = sizeof(struct scsi_per_res_in_rsrv);
7750		else
7751			total_len = sizeof(struct scsi_per_res_in_header);
7752		break;
7753	case SPRI_RC: /* report capabilities */
7754		total_len = sizeof(struct scsi_per_res_cap);
7755		break;
7756	case SPRI_RS: /* read full status */
7757		total_len = sizeof(struct scsi_per_res_in_header) +
7758		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7759		    lun->pr_key_count;
7760		break;
7761	default:
7762		panic("Invalid PR type %x", cdb->action);
7763	}
7764	mtx_unlock(&lun->lun_lock);
7765
7766	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7767
7768	if (total_len < alloc_len) {
7769		ctsio->residual = alloc_len - total_len;
7770		ctsio->kern_data_len = total_len;
7771		ctsio->kern_total_len = total_len;
7772	} else {
7773		ctsio->residual = 0;
7774		ctsio->kern_data_len = alloc_len;
7775		ctsio->kern_total_len = alloc_len;
7776	}
7777
7778	ctsio->kern_data_resid = 0;
7779	ctsio->kern_rel_offset = 0;
7780	ctsio->kern_sg_entries = 0;
7781
7782	mtx_lock(&lun->lun_lock);
7783	switch (cdb->action) {
7784	case SPRI_RK: { // read keys
7785        struct scsi_per_res_in_keys *res_keys;
7786		int i, key_count;
7787
7788		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7789
7790		/*
7791		 * We had to drop the lock to allocate our buffer, which
7792		 * leaves time for someone to come in with another
7793		 * persistent reservation.  (That is unlikely, though,
7794		 * since this should be the only persistent reservation
7795		 * command active right now.)
7796		 */
7797		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7798		    (lun->pr_key_count *
7799		     sizeof(struct scsi_per_res_key)))){
7800			mtx_unlock(&lun->lun_lock);
7801			free(ctsio->kern_data_ptr, M_CTL);
7802			printf("%s: reservation length changed, retrying\n",
7803			       __func__);
7804			goto retry;
7805		}
7806
7807		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7808
7809		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7810			     lun->pr_key_count, res_keys->header.length);
7811
7812		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7813			if (!lun->per_res[i].registered)
7814				continue;
7815
7816			/*
7817			 * We used lun->pr_key_count to calculate the
7818			 * size to allocate.  If it turns out the number of
7819			 * initiators with the registered flag set is
7820			 * larger than that (i.e. they haven't been kept in
7821			 * sync), we've got a problem.
7822			 */
7823			if (key_count >= lun->pr_key_count) {
7824#ifdef NEEDTOPORT
7825				csevent_log(CSC_CTL | CSC_SHELF_SW |
7826					    CTL_PR_ERROR,
7827					    csevent_LogType_Fault,
7828					    csevent_AlertLevel_Yellow,
7829					    csevent_FRU_ShelfController,
7830					    csevent_FRU_Firmware,
7831				        csevent_FRU_Unknown,
7832					    "registered keys %d >= key "
7833					    "count %d", key_count,
7834					    lun->pr_key_count);
7835#endif
7836				key_count++;
7837				continue;
7838			}
7839			memcpy(res_keys->keys[key_count].key,
7840			       lun->per_res[i].res_key.key,
7841			       ctl_min(sizeof(res_keys->keys[key_count].key),
7842			       sizeof(lun->per_res[i].res_key)));
7843			key_count++;
7844		}
7845		break;
7846	}
7847	case SPRI_RR: { // read reservation
7848		struct scsi_per_res_in_rsrv *res;
7849		int tmp_len, header_only;
7850
7851		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7852
7853		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7854
7855		if (lun->flags & CTL_LUN_PR_RESERVED)
7856		{
7857			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7858			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7859				    res->header.length);
7860			header_only = 0;
7861		} else {
7862			tmp_len = sizeof(struct scsi_per_res_in_header);
7863			scsi_ulto4b(0, res->header.length);
7864			header_only = 1;
7865		}
7866
7867		/*
7868		 * We had to drop the lock to allocate our buffer, which
7869		 * leaves time for someone to come in with another
7870		 * persistent reservation.  (That is unlikely, though,
7871		 * since this should be the only persistent reservation
7872		 * command active right now.)
7873		 */
7874		if (tmp_len != total_len) {
7875			mtx_unlock(&lun->lun_lock);
7876			free(ctsio->kern_data_ptr, M_CTL);
7877			printf("%s: reservation status changed, retrying\n",
7878			       __func__);
7879			goto retry;
7880		}
7881
7882		/*
7883		 * No reservation held, so we're done.
7884		 */
7885		if (header_only != 0)
7886			break;
7887
7888		/*
7889		 * If the registration is an All Registrants type, the key
7890		 * is 0, since it doesn't really matter.
7891		 */
7892		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7893			memcpy(res->data.reservation,
7894			       &lun->per_res[lun->pr_res_idx].res_key,
7895			       sizeof(struct scsi_per_res_key));
7896		}
7897		res->data.scopetype = lun->res_type;
7898		break;
7899	}
7900	case SPRI_RC:     //report capabilities
7901	{
7902		struct scsi_per_res_cap *res_cap;
7903		uint16_t type_mask;
7904
7905		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7906		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7907		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7908		type_mask = SPRI_TM_WR_EX_AR |
7909			    SPRI_TM_EX_AC_RO |
7910			    SPRI_TM_WR_EX_RO |
7911			    SPRI_TM_EX_AC |
7912			    SPRI_TM_WR_EX |
7913			    SPRI_TM_EX_AC_AR;
7914		scsi_ulto2b(type_mask, res_cap->type_mask);
7915		break;
7916	}
7917	case SPRI_RS: { // read full status
7918		struct scsi_per_res_in_full *res_status;
7919		struct scsi_per_res_in_full_desc *res_desc;
7920		struct ctl_port *port;
7921		int i, len;
7922
7923		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7924
7925		/*
7926		 * We had to drop the lock to allocate our buffer, which
7927		 * leaves time for someone to come in with another
7928		 * persistent reservation.  (That is unlikely, though,
7929		 * since this should be the only persistent reservation
7930		 * command active right now.)
7931		 */
7932		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7933		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7934		     lun->pr_key_count)){
7935			mtx_unlock(&lun->lun_lock);
7936			free(ctsio->kern_data_ptr, M_CTL);
7937			printf("%s: reservation length changed, retrying\n",
7938			       __func__);
7939			goto retry;
7940		}
7941
7942		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7943
7944		res_desc = &res_status->desc[0];
7945		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7946			if (!lun->per_res[i].registered)
7947				continue;
7948
7949			memcpy(&res_desc->res_key, &lun->per_res[i].res_key.key,
7950			    sizeof(res_desc->res_key));
7951			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7952			    (lun->pr_res_idx == i ||
7953			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7954				res_desc->flags = SPRI_FULL_R_HOLDER;
7955				res_desc->scopetype = lun->res_type;
7956			}
7957			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7958			    res_desc->rel_trgt_port_id);
7959			len = 0;
7960			port = softc->ctl_ports[
7961			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7962			if (port != NULL)
7963				len = ctl_create_iid(port,
7964				    i % CTL_MAX_INIT_PER_PORT,
7965				    res_desc->transport_id);
7966			scsi_ulto4b(len, res_desc->additional_length);
7967			res_desc = (struct scsi_per_res_in_full_desc *)
7968			    &res_desc->transport_id[len];
7969		}
7970		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7971		    res_status->header.length);
7972		break;
7973	}
7974	default:
7975		/*
7976		 * This is a bug, because we just checked for this above,
7977		 * and should have returned an error.
7978		 */
7979		panic("Invalid PR type %x", cdb->action);
7980		break; /* NOTREACHED */
7981	}
7982	mtx_unlock(&lun->lun_lock);
7983
7984	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7985	ctsio->be_move_done = ctl_config_move_done;
7986
7987	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7988			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7989			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7990			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7991			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7992
7993	ctl_datamove((union ctl_io *)ctsio);
7994
7995	return (CTL_RETVAL_COMPLETE);
7996}
7997
7998/*
7999 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
8000 * it should return.
8001 */
8002static int
8003ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
8004		uint64_t sa_res_key, uint8_t type, uint32_t residx,
8005		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
8006		struct scsi_per_res_out_parms* param)
8007{
8008	union ctl_ha_msg persis_io;
8009	int retval, i;
8010	int isc_retval;
8011
8012	retval = 0;
8013
8014	mtx_lock(&lun->lun_lock);
8015	if (sa_res_key == 0) {
8016		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8017			/* validate scope and type */
8018			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8019			     SPR_LU_SCOPE) {
8020				mtx_unlock(&lun->lun_lock);
8021				ctl_set_invalid_field(/*ctsio*/ ctsio,
8022						      /*sks_valid*/ 1,
8023						      /*command*/ 1,
8024						      /*field*/ 2,
8025						      /*bit_valid*/ 1,
8026						      /*bit*/ 4);
8027				ctl_done((union ctl_io *)ctsio);
8028				return (1);
8029			}
8030
8031		        if (type>8 || type==2 || type==4 || type==0) {
8032				mtx_unlock(&lun->lun_lock);
8033				ctl_set_invalid_field(/*ctsio*/ ctsio,
8034       	           				      /*sks_valid*/ 1,
8035						      /*command*/ 1,
8036						      /*field*/ 2,
8037						      /*bit_valid*/ 1,
8038						      /*bit*/ 0);
8039				ctl_done((union ctl_io *)ctsio);
8040				return (1);
8041		        }
8042
8043			/* temporarily unregister this nexus */
8044			lun->per_res[residx].registered = 0;
8045
8046			/*
8047			 * Unregister everybody else and build UA for
8048			 * them
8049			 */
8050			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8051				if (lun->per_res[i].registered == 0)
8052					continue;
8053
8054				if (!persis_offset
8055				 && i <CTL_MAX_INITIATORS)
8056					lun->pending_ua[i] |=
8057						CTL_UA_REG_PREEMPT;
8058				else if (persis_offset
8059				      && i >= persis_offset)
8060					lun->pending_ua[i-persis_offset] |=
8061						CTL_UA_REG_PREEMPT;
8062				lun->per_res[i].registered = 0;
8063				memset(&lun->per_res[i].res_key, 0,
8064				       sizeof(struct scsi_per_res_key));
8065			}
8066			lun->per_res[residx].registered = 1;
8067			lun->pr_key_count = 1;
8068			lun->res_type = type;
8069			if (lun->res_type != SPR_TYPE_WR_EX_AR
8070			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8071				lun->pr_res_idx = residx;
8072
8073			/* send msg to other side */
8074			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8075			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8076			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8077			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8078			persis_io.pr.pr_info.res_type = type;
8079			memcpy(persis_io.pr.pr_info.sa_res_key,
8080			       param->serv_act_res_key,
8081			       sizeof(param->serv_act_res_key));
8082			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8083			     &persis_io, sizeof(persis_io), 0)) >
8084			     CTL_HA_STATUS_SUCCESS) {
8085				printf("CTL:Persis Out error returned "
8086				       "from ctl_ha_msg_send %d\n",
8087				       isc_retval);
8088			}
8089		} else {
8090			/* not all registrants */
8091			mtx_unlock(&lun->lun_lock);
8092			free(ctsio->kern_data_ptr, M_CTL);
8093			ctl_set_invalid_field(ctsio,
8094					      /*sks_valid*/ 1,
8095					      /*command*/ 0,
8096					      /*field*/ 8,
8097					      /*bit_valid*/ 0,
8098					      /*bit*/ 0);
8099			ctl_done((union ctl_io *)ctsio);
8100			return (1);
8101		}
8102	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8103		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8104		int found = 0;
8105
8106		if (res_key == sa_res_key) {
8107			/* special case */
8108			/*
8109			 * The spec implies this is not good but doesn't
8110			 * say what to do. There are two choices either
8111			 * generate a res conflict or check condition
8112			 * with illegal field in parameter data. Since
8113			 * that is what is done when the sa_res_key is
8114			 * zero I'll take that approach since this has
8115			 * to do with the sa_res_key.
8116			 */
8117			mtx_unlock(&lun->lun_lock);
8118			free(ctsio->kern_data_ptr, M_CTL);
8119			ctl_set_invalid_field(ctsio,
8120					      /*sks_valid*/ 1,
8121					      /*command*/ 0,
8122					      /*field*/ 8,
8123					      /*bit_valid*/ 0,
8124					      /*bit*/ 0);
8125			ctl_done((union ctl_io *)ctsio);
8126			return (1);
8127		}
8128
8129		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8130			if (lun->per_res[i].registered
8131			 && memcmp(param->serv_act_res_key,
8132			    lun->per_res[i].res_key.key,
8133			    sizeof(struct scsi_per_res_key)) != 0)
8134				continue;
8135
8136			found = 1;
8137			lun->per_res[i].registered = 0;
8138			memset(&lun->per_res[i].res_key, 0,
8139			       sizeof(struct scsi_per_res_key));
8140			lun->pr_key_count--;
8141
8142			if (!persis_offset && i < CTL_MAX_INITIATORS)
8143				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8144			else if (persis_offset && i >= persis_offset)
8145				lun->pending_ua[i-persis_offset] |=
8146					CTL_UA_REG_PREEMPT;
8147		}
8148		if (!found) {
8149			mtx_unlock(&lun->lun_lock);
8150			free(ctsio->kern_data_ptr, M_CTL);
8151			ctl_set_reservation_conflict(ctsio);
8152			ctl_done((union ctl_io *)ctsio);
8153			return (CTL_RETVAL_COMPLETE);
8154		}
8155		/* send msg to other side */
8156		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8157		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8158		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8159		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8160		persis_io.pr.pr_info.res_type = type;
8161		memcpy(persis_io.pr.pr_info.sa_res_key,
8162		       param->serv_act_res_key,
8163		       sizeof(param->serv_act_res_key));
8164		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8165		     &persis_io, sizeof(persis_io), 0)) >
8166		     CTL_HA_STATUS_SUCCESS) {
8167			printf("CTL:Persis Out error returned from "
8168			       "ctl_ha_msg_send %d\n", isc_retval);
8169		}
8170	} else {
8171		/* Reserved but not all registrants */
8172		/* sa_res_key is res holder */
8173		if (memcmp(param->serv_act_res_key,
8174                   lun->per_res[lun->pr_res_idx].res_key.key,
8175                   sizeof(struct scsi_per_res_key)) == 0) {
8176			/* validate scope and type */
8177			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8178			     SPR_LU_SCOPE) {
8179				mtx_unlock(&lun->lun_lock);
8180				ctl_set_invalid_field(/*ctsio*/ ctsio,
8181						      /*sks_valid*/ 1,
8182						      /*command*/ 1,
8183						      /*field*/ 2,
8184						      /*bit_valid*/ 1,
8185						      /*bit*/ 4);
8186				ctl_done((union ctl_io *)ctsio);
8187				return (1);
8188			}
8189
8190			if (type>8 || type==2 || type==4 || type==0) {
8191				mtx_unlock(&lun->lun_lock);
8192				ctl_set_invalid_field(/*ctsio*/ ctsio,
8193						      /*sks_valid*/ 1,
8194						      /*command*/ 1,
8195						      /*field*/ 2,
8196						      /*bit_valid*/ 1,
8197						      /*bit*/ 0);
8198				ctl_done((union ctl_io *)ctsio);
8199				return (1);
8200			}
8201
8202			/*
8203			 * Do the following:
8204			 * if sa_res_key != res_key remove all
8205			 * registrants w/sa_res_key and generate UA
8206			 * for these registrants(Registrations
8207			 * Preempted) if it wasn't an exclusive
8208			 * reservation generate UA(Reservations
8209			 * Preempted) for all other registered nexuses
8210			 * if the type has changed. Establish the new
8211			 * reservation and holder. If res_key and
8212			 * sa_res_key are the same do the above
8213			 * except don't unregister the res holder.
8214			 */
8215
8216			/*
8217			 * Temporarily unregister so it won't get
8218			 * removed or UA generated
8219			 */
8220			lun->per_res[residx].registered = 0;
8221			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8222				if (lun->per_res[i].registered == 0)
8223					continue;
8224
8225				if (memcmp(param->serv_act_res_key,
8226				    lun->per_res[i].res_key.key,
8227				    sizeof(struct scsi_per_res_key)) == 0) {
8228					lun->per_res[i].registered = 0;
8229					memset(&lun->per_res[i].res_key,
8230					       0,
8231					       sizeof(struct scsi_per_res_key));
8232					lun->pr_key_count--;
8233
8234					if (!persis_offset
8235					 && i < CTL_MAX_INITIATORS)
8236						lun->pending_ua[i] |=
8237							CTL_UA_REG_PREEMPT;
8238					else if (persis_offset
8239					      && i >= persis_offset)
8240						lun->pending_ua[i-persis_offset] |=
8241						  CTL_UA_REG_PREEMPT;
8242				} else if (type != lun->res_type
8243					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8244					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8245						if (!persis_offset
8246						 && i < CTL_MAX_INITIATORS)
8247							lun->pending_ua[i] |=
8248							CTL_UA_RES_RELEASE;
8249						else if (persis_offset
8250						      && i >= persis_offset)
8251							lun->pending_ua[
8252							i-persis_offset] |=
8253							CTL_UA_RES_RELEASE;
8254				}
8255			}
8256			lun->per_res[residx].registered = 1;
8257			lun->res_type = type;
8258			if (lun->res_type != SPR_TYPE_WR_EX_AR
8259			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8260				lun->pr_res_idx = residx;
8261			else
8262				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8263
8264			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8265			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8266			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8267			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8268			persis_io.pr.pr_info.res_type = type;
8269			memcpy(persis_io.pr.pr_info.sa_res_key,
8270			       param->serv_act_res_key,
8271			       sizeof(param->serv_act_res_key));
8272			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8273			     &persis_io, sizeof(persis_io), 0)) >
8274			     CTL_HA_STATUS_SUCCESS) {
8275				printf("CTL:Persis Out error returned "
8276				       "from ctl_ha_msg_send %d\n",
8277				       isc_retval);
8278			}
8279		} else {
8280			/*
8281			 * sa_res_key is not the res holder just
8282			 * remove registrants
8283			 */
8284			int found=0;
8285
8286			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8287				if (memcmp(param->serv_act_res_key,
8288				    lun->per_res[i].res_key.key,
8289				    sizeof(struct scsi_per_res_key)) != 0)
8290					continue;
8291
8292				found = 1;
8293				lun->per_res[i].registered = 0;
8294				memset(&lun->per_res[i].res_key, 0,
8295				       sizeof(struct scsi_per_res_key));
8296				lun->pr_key_count--;
8297
8298				if (!persis_offset
8299				 && i < CTL_MAX_INITIATORS)
8300					lun->pending_ua[i] |=
8301						CTL_UA_REG_PREEMPT;
8302				else if (persis_offset
8303				      && i >= persis_offset)
8304					lun->pending_ua[i-persis_offset] |=
8305						CTL_UA_REG_PREEMPT;
8306			}
8307
8308			if (!found) {
8309				mtx_unlock(&lun->lun_lock);
8310				free(ctsio->kern_data_ptr, M_CTL);
8311				ctl_set_reservation_conflict(ctsio);
8312				ctl_done((union ctl_io *)ctsio);
8313		        	return (1);
8314			}
8315			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8316			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8317			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8318			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8319			persis_io.pr.pr_info.res_type = type;
8320			memcpy(persis_io.pr.pr_info.sa_res_key,
8321			       param->serv_act_res_key,
8322			       sizeof(param->serv_act_res_key));
8323			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8324			     &persis_io, sizeof(persis_io), 0)) >
8325			     CTL_HA_STATUS_SUCCESS) {
8326				printf("CTL:Persis Out error returned "
8327				       "from ctl_ha_msg_send %d\n",
8328				isc_retval);
8329			}
8330		}
8331	}
8332
8333	lun->PRGeneration++;
8334	mtx_unlock(&lun->lun_lock);
8335
8336	return (retval);
8337}
8338
8339static void
8340ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8341{
8342	int i;
8343
8344	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8345	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8346	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8347		   msg->pr.pr_info.sa_res_key,
8348		   sizeof(struct scsi_per_res_key)) != 0) {
8349		uint64_t sa_res_key;
8350		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8351
8352		if (sa_res_key == 0) {
8353			/* temporarily unregister this nexus */
8354			lun->per_res[msg->pr.pr_info.residx].registered = 0;
8355
8356			/*
8357			 * Unregister everybody else and build UA for
8358			 * them
8359			 */
8360			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8361				if (lun->per_res[i].registered == 0)
8362					continue;
8363
8364				if (!persis_offset
8365				 && i < CTL_MAX_INITIATORS)
8366					lun->pending_ua[i] |=
8367						CTL_UA_REG_PREEMPT;
8368				else if (persis_offset && i >= persis_offset)
8369					lun->pending_ua[i - persis_offset] |=
8370						CTL_UA_REG_PREEMPT;
8371				lun->per_res[i].registered = 0;
8372				memset(&lun->per_res[i].res_key, 0,
8373				       sizeof(struct scsi_per_res_key));
8374			}
8375
8376			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8377			lun->pr_key_count = 1;
8378			lun->res_type = msg->pr.pr_info.res_type;
8379			if (lun->res_type != SPR_TYPE_WR_EX_AR
8380			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8381				lun->pr_res_idx = msg->pr.pr_info.residx;
8382		} else {
8383		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8384				if (memcmp(msg->pr.pr_info.sa_res_key,
8385		                   lun->per_res[i].res_key.key,
8386		                   sizeof(struct scsi_per_res_key)) != 0)
8387					continue;
8388
8389				lun->per_res[i].registered = 0;
8390				memset(&lun->per_res[i].res_key, 0,
8391				       sizeof(struct scsi_per_res_key));
8392				lun->pr_key_count--;
8393
8394				if (!persis_offset
8395				 && i < persis_offset)
8396					lun->pending_ua[i] |=
8397						CTL_UA_REG_PREEMPT;
8398				else if (persis_offset
8399				      && i >= persis_offset)
8400					lun->pending_ua[i - persis_offset] |=
8401						CTL_UA_REG_PREEMPT;
8402			}
8403		}
8404	} else {
8405		/*
8406		 * Temporarily unregister so it won't get removed
8407		 * or UA generated
8408		 */
8409		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8410		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8411			if (lun->per_res[i].registered == 0)
8412				continue;
8413
8414			if (memcmp(msg->pr.pr_info.sa_res_key,
8415	                   lun->per_res[i].res_key.key,
8416	                   sizeof(struct scsi_per_res_key)) == 0) {
8417				lun->per_res[i].registered = 0;
8418				memset(&lun->per_res[i].res_key, 0,
8419				       sizeof(struct scsi_per_res_key));
8420				lun->pr_key_count--;
8421				if (!persis_offset
8422				 && i < CTL_MAX_INITIATORS)
8423					lun->pending_ua[i] |=
8424						CTL_UA_REG_PREEMPT;
8425				else if (persis_offset
8426				      && i >= persis_offset)
8427					lun->pending_ua[i - persis_offset] |=
8428						CTL_UA_REG_PREEMPT;
8429			} else if (msg->pr.pr_info.res_type != lun->res_type
8430				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8431				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8432					if (!persis_offset
8433					 && i < persis_offset)
8434						lun->pending_ua[i] |=
8435							CTL_UA_RES_RELEASE;
8436					else if (persis_offset
8437					      && i >= persis_offset)
8438					lun->pending_ua[i - persis_offset] |=
8439						CTL_UA_RES_RELEASE;
8440			}
8441		}
8442		lun->per_res[msg->pr.pr_info.residx].registered = 1;
8443		lun->res_type = msg->pr.pr_info.res_type;
8444		if (lun->res_type != SPR_TYPE_WR_EX_AR
8445		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8446			lun->pr_res_idx = msg->pr.pr_info.residx;
8447		else
8448			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8449	}
8450	lun->PRGeneration++;
8451
8452}
8453
8454
8455int
8456ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8457{
8458	int retval;
8459	int isc_retval;
8460	u_int32_t param_len;
8461	struct scsi_per_res_out *cdb;
8462	struct ctl_lun *lun;
8463	struct scsi_per_res_out_parms* param;
8464	struct ctl_softc *softc;
8465	uint32_t residx;
8466	uint64_t res_key, sa_res_key;
8467	uint8_t type;
8468	union ctl_ha_msg persis_io;
8469	int    i;
8470
8471	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8472
8473	retval = CTL_RETVAL_COMPLETE;
8474
8475	softc = control_softc;
8476
8477	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8478	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8479
8480	/*
8481	 * We only support whole-LUN scope.  The scope & type are ignored for
8482	 * register, register and ignore existing key and clear.
8483	 * We sometimes ignore scope and type on preempts too!!
8484	 * Verify reservation type here as well.
8485	 */
8486	type = cdb->scope_type & SPR_TYPE_MASK;
8487	if ((cdb->action == SPRO_RESERVE)
8488	 || (cdb->action == SPRO_RELEASE)) {
8489		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8490			ctl_set_invalid_field(/*ctsio*/ ctsio,
8491					      /*sks_valid*/ 1,
8492					      /*command*/ 1,
8493					      /*field*/ 2,
8494					      /*bit_valid*/ 1,
8495					      /*bit*/ 4);
8496			ctl_done((union ctl_io *)ctsio);
8497			return (CTL_RETVAL_COMPLETE);
8498		}
8499
8500		if (type>8 || type==2 || type==4 || type==0) {
8501			ctl_set_invalid_field(/*ctsio*/ ctsio,
8502					      /*sks_valid*/ 1,
8503					      /*command*/ 1,
8504					      /*field*/ 2,
8505					      /*bit_valid*/ 1,
8506					      /*bit*/ 0);
8507			ctl_done((union ctl_io *)ctsio);
8508			return (CTL_RETVAL_COMPLETE);
8509		}
8510	}
8511
8512	param_len = scsi_4btoul(cdb->length);
8513
8514	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8515		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8516		ctsio->kern_data_len = param_len;
8517		ctsio->kern_total_len = param_len;
8518		ctsio->kern_data_resid = 0;
8519		ctsio->kern_rel_offset = 0;
8520		ctsio->kern_sg_entries = 0;
8521		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8522		ctsio->be_move_done = ctl_config_move_done;
8523		ctl_datamove((union ctl_io *)ctsio);
8524
8525		return (CTL_RETVAL_COMPLETE);
8526	}
8527
8528	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8529
8530	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8531	res_key = scsi_8btou64(param->res_key.key);
8532	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8533
8534	/*
8535	 * Validate the reservation key here except for SPRO_REG_IGNO
8536	 * This must be done for all other service actions
8537	 */
8538	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8539		mtx_lock(&lun->lun_lock);
8540		if (lun->per_res[residx].registered) {
8541		    if (memcmp(param->res_key.key,
8542			       lun->per_res[residx].res_key.key,
8543			       ctl_min(sizeof(param->res_key),
8544			       sizeof(lun->per_res[residx].res_key))) != 0) {
8545				/*
8546				 * The current key passed in doesn't match
8547				 * the one the initiator previously
8548				 * registered.
8549				 */
8550				mtx_unlock(&lun->lun_lock);
8551				free(ctsio->kern_data_ptr, M_CTL);
8552				ctl_set_reservation_conflict(ctsio);
8553				ctl_done((union ctl_io *)ctsio);
8554				return (CTL_RETVAL_COMPLETE);
8555			}
8556		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8557			/*
8558			 * We are not registered
8559			 */
8560			mtx_unlock(&lun->lun_lock);
8561			free(ctsio->kern_data_ptr, M_CTL);
8562			ctl_set_reservation_conflict(ctsio);
8563			ctl_done((union ctl_io *)ctsio);
8564			return (CTL_RETVAL_COMPLETE);
8565		} else if (res_key != 0) {
8566			/*
8567			 * We are not registered and trying to register but
8568			 * the register key isn't zero.
8569			 */
8570			mtx_unlock(&lun->lun_lock);
8571			free(ctsio->kern_data_ptr, M_CTL);
8572			ctl_set_reservation_conflict(ctsio);
8573			ctl_done((union ctl_io *)ctsio);
8574			return (CTL_RETVAL_COMPLETE);
8575		}
8576		mtx_unlock(&lun->lun_lock);
8577	}
8578
8579	switch (cdb->action & SPRO_ACTION_MASK) {
8580	case SPRO_REGISTER:
8581	case SPRO_REG_IGNO: {
8582
8583#if 0
8584		printf("Registration received\n");
8585#endif
8586
8587		/*
8588		 * We don't support any of these options, as we report in
8589		 * the read capabilities request (see
8590		 * ctl_persistent_reserve_in(), above).
8591		 */
8592		if ((param->flags & SPR_SPEC_I_PT)
8593		 || (param->flags & SPR_ALL_TG_PT)
8594		 || (param->flags & SPR_APTPL)) {
8595			int bit_ptr;
8596
8597			if (param->flags & SPR_APTPL)
8598				bit_ptr = 0;
8599			else if (param->flags & SPR_ALL_TG_PT)
8600				bit_ptr = 2;
8601			else /* SPR_SPEC_I_PT */
8602				bit_ptr = 3;
8603
8604			free(ctsio->kern_data_ptr, M_CTL);
8605			ctl_set_invalid_field(ctsio,
8606					      /*sks_valid*/ 1,
8607					      /*command*/ 0,
8608					      /*field*/ 20,
8609					      /*bit_valid*/ 1,
8610					      /*bit*/ bit_ptr);
8611			ctl_done((union ctl_io *)ctsio);
8612			return (CTL_RETVAL_COMPLETE);
8613		}
8614
8615		mtx_lock(&lun->lun_lock);
8616
8617		/*
8618		 * The initiator wants to clear the
8619		 * key/unregister.
8620		 */
8621		if (sa_res_key == 0) {
8622			if ((res_key == 0
8623			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8624			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8625			  && !lun->per_res[residx].registered)) {
8626				mtx_unlock(&lun->lun_lock);
8627				goto done;
8628			}
8629
8630			lun->per_res[residx].registered = 0;
8631			memset(&lun->per_res[residx].res_key,
8632			       0, sizeof(lun->per_res[residx].res_key));
8633			lun->pr_key_count--;
8634
8635			if (residx == lun->pr_res_idx) {
8636				lun->flags &= ~CTL_LUN_PR_RESERVED;
8637				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8638
8639				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8640				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8641				 && lun->pr_key_count) {
8642					/*
8643					 * If the reservation is a registrants
8644					 * only type we need to generate a UA
8645					 * for other registered inits.  The
8646					 * sense code should be RESERVATIONS
8647					 * RELEASED
8648					 */
8649
8650					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8651						if (lun->per_res[
8652						    i+persis_offset].registered
8653						    == 0)
8654							continue;
8655						lun->pending_ua[i] |=
8656							CTL_UA_RES_RELEASE;
8657					}
8658				}
8659				lun->res_type = 0;
8660			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8661				if (lun->pr_key_count==0) {
8662					lun->flags &= ~CTL_LUN_PR_RESERVED;
8663					lun->res_type = 0;
8664					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8665				}
8666			}
8667			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8668			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8669			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8670			persis_io.pr.pr_info.residx = residx;
8671			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8672			     &persis_io, sizeof(persis_io), 0 )) >
8673			     CTL_HA_STATUS_SUCCESS) {
8674				printf("CTL:Persis Out error returned from "
8675				       "ctl_ha_msg_send %d\n", isc_retval);
8676			}
8677		} else /* sa_res_key != 0 */ {
8678
8679			/*
8680			 * If we aren't registered currently then increment
8681			 * the key count and set the registered flag.
8682			 */
8683			if (!lun->per_res[residx].registered) {
8684				lun->pr_key_count++;
8685				lun->per_res[residx].registered = 1;
8686			}
8687
8688			memcpy(&lun->per_res[residx].res_key,
8689			       param->serv_act_res_key,
8690			       ctl_min(sizeof(param->serv_act_res_key),
8691			       sizeof(lun->per_res[residx].res_key)));
8692
8693			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8694			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8695			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8696			persis_io.pr.pr_info.residx = residx;
8697			memcpy(persis_io.pr.pr_info.sa_res_key,
8698			       param->serv_act_res_key,
8699			       sizeof(param->serv_act_res_key));
8700			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8701			     &persis_io, sizeof(persis_io), 0)) >
8702			     CTL_HA_STATUS_SUCCESS) {
8703				printf("CTL:Persis Out error returned from "
8704				       "ctl_ha_msg_send %d\n", isc_retval);
8705			}
8706		}
8707		lun->PRGeneration++;
8708		mtx_unlock(&lun->lun_lock);
8709
8710		break;
8711	}
8712	case SPRO_RESERVE:
8713#if 0
8714                printf("Reserve executed type %d\n", type);
8715#endif
8716		mtx_lock(&lun->lun_lock);
8717		if (lun->flags & CTL_LUN_PR_RESERVED) {
8718			/*
8719			 * if this isn't the reservation holder and it's
8720			 * not a "all registrants" type or if the type is
8721			 * different then we have a conflict
8722			 */
8723			if ((lun->pr_res_idx != residx
8724			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8725			 || lun->res_type != type) {
8726				mtx_unlock(&lun->lun_lock);
8727				free(ctsio->kern_data_ptr, M_CTL);
8728				ctl_set_reservation_conflict(ctsio);
8729				ctl_done((union ctl_io *)ctsio);
8730				return (CTL_RETVAL_COMPLETE);
8731			}
8732			mtx_unlock(&lun->lun_lock);
8733		} else /* create a reservation */ {
8734			/*
8735			 * If it's not an "all registrants" type record
8736			 * reservation holder
8737			 */
8738			if (type != SPR_TYPE_WR_EX_AR
8739			 && type != SPR_TYPE_EX_AC_AR)
8740				lun->pr_res_idx = residx; /* Res holder */
8741			else
8742				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8743
8744			lun->flags |= CTL_LUN_PR_RESERVED;
8745			lun->res_type = type;
8746
8747			mtx_unlock(&lun->lun_lock);
8748
8749			/* send msg to other side */
8750			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8751			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8752			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8753			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8754			persis_io.pr.pr_info.res_type = type;
8755			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8756			     &persis_io, sizeof(persis_io), 0)) >
8757			     CTL_HA_STATUS_SUCCESS) {
8758				printf("CTL:Persis Out error returned from "
8759				       "ctl_ha_msg_send %d\n", isc_retval);
8760			}
8761		}
8762		break;
8763
8764	case SPRO_RELEASE:
8765		mtx_lock(&lun->lun_lock);
8766		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8767			/* No reservation exists return good status */
8768			mtx_unlock(&lun->lun_lock);
8769			goto done;
8770		}
8771		/*
8772		 * Is this nexus a reservation holder?
8773		 */
8774		if (lun->pr_res_idx != residx
8775		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8776			/*
8777			 * not a res holder return good status but
8778			 * do nothing
8779			 */
8780			mtx_unlock(&lun->lun_lock);
8781			goto done;
8782		}
8783
8784		if (lun->res_type != type) {
8785			mtx_unlock(&lun->lun_lock);
8786			free(ctsio->kern_data_ptr, M_CTL);
8787			ctl_set_illegal_pr_release(ctsio);
8788			ctl_done((union ctl_io *)ctsio);
8789			return (CTL_RETVAL_COMPLETE);
8790		}
8791
8792		/* okay to release */
8793		lun->flags &= ~CTL_LUN_PR_RESERVED;
8794		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8795		lun->res_type = 0;
8796
8797		/*
8798		 * if this isn't an exclusive access
8799		 * res generate UA for all other
8800		 * registrants.
8801		 */
8802		if (type != SPR_TYPE_EX_AC
8803		 && type != SPR_TYPE_WR_EX) {
8804			/*
8805			 * temporarily unregister so we don't generate UA
8806			 */
8807			lun->per_res[residx].registered = 0;
8808
8809			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8810				if (lun->per_res[i+persis_offset].registered
8811				    == 0)
8812					continue;
8813				lun->pending_ua[i] |=
8814					CTL_UA_RES_RELEASE;
8815			}
8816
8817			lun->per_res[residx].registered = 1;
8818		}
8819		mtx_unlock(&lun->lun_lock);
8820		/* Send msg to other side */
8821		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8822		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8823		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8824		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8825		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8826			printf("CTL:Persis Out error returned from "
8827			       "ctl_ha_msg_send %d\n", isc_retval);
8828		}
8829		break;
8830
8831	case SPRO_CLEAR:
8832		/* send msg to other side */
8833
8834		mtx_lock(&lun->lun_lock);
8835		lun->flags &= ~CTL_LUN_PR_RESERVED;
8836		lun->res_type = 0;
8837		lun->pr_key_count = 0;
8838		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8839
8840
8841		memset(&lun->per_res[residx].res_key,
8842		       0, sizeof(lun->per_res[residx].res_key));
8843		lun->per_res[residx].registered = 0;
8844
8845		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8846			if (lun->per_res[i].registered) {
8847				if (!persis_offset && i < CTL_MAX_INITIATORS)
8848					lun->pending_ua[i] |=
8849						CTL_UA_RES_PREEMPT;
8850				else if (persis_offset && i >= persis_offset)
8851					lun->pending_ua[i-persis_offset] |=
8852					    CTL_UA_RES_PREEMPT;
8853
8854				memset(&lun->per_res[i].res_key,
8855				       0, sizeof(struct scsi_per_res_key));
8856				lun->per_res[i].registered = 0;
8857			}
8858		lun->PRGeneration++;
8859		mtx_unlock(&lun->lun_lock);
8860		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8861		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8862		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8863		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8864		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8865			printf("CTL:Persis Out error returned from "
8866			       "ctl_ha_msg_send %d\n", isc_retval);
8867		}
8868		break;
8869
8870	case SPRO_PREEMPT: {
8871		int nretval;
8872
8873		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8874					  residx, ctsio, cdb, param);
8875		if (nretval != 0)
8876			return (CTL_RETVAL_COMPLETE);
8877		break;
8878	}
8879	default:
8880		panic("Invalid PR type %x", cdb->action);
8881	}
8882
8883done:
8884	free(ctsio->kern_data_ptr, M_CTL);
8885	ctl_set_success(ctsio);
8886	ctl_done((union ctl_io *)ctsio);
8887
8888	return (retval);
8889}
8890
8891/*
8892 * This routine is for handling a message from the other SC pertaining to
8893 * persistent reserve out. All the error checking will have been done
8894 * so only perorming the action need be done here to keep the two
8895 * in sync.
8896 */
8897static void
8898ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8899{
8900	struct ctl_lun *lun;
8901	struct ctl_softc *softc;
8902	int i;
8903	uint32_t targ_lun;
8904
8905	softc = control_softc;
8906
8907	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8908	lun = softc->ctl_luns[targ_lun];
8909	mtx_lock(&lun->lun_lock);
8910	switch(msg->pr.pr_info.action) {
8911	case CTL_PR_REG_KEY:
8912		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8913			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8914			lun->pr_key_count++;
8915		}
8916		lun->PRGeneration++;
8917		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8918		       msg->pr.pr_info.sa_res_key,
8919		       sizeof(struct scsi_per_res_key));
8920		break;
8921
8922	case CTL_PR_UNREG_KEY:
8923		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8924		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8925		       0, sizeof(struct scsi_per_res_key));
8926		lun->pr_key_count--;
8927
8928		/* XXX Need to see if the reservation has been released */
8929		/* if so do we need to generate UA? */
8930		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8931			lun->flags &= ~CTL_LUN_PR_RESERVED;
8932			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8933
8934			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8935			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8936			 && lun->pr_key_count) {
8937				/*
8938				 * If the reservation is a registrants
8939				 * only type we need to generate a UA
8940				 * for other registered inits.  The
8941				 * sense code should be RESERVATIONS
8942				 * RELEASED
8943				 */
8944
8945				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8946					if (lun->per_res[i+
8947					    persis_offset].registered == 0)
8948						continue;
8949
8950					lun->pending_ua[i] |=
8951						CTL_UA_RES_RELEASE;
8952				}
8953			}
8954			lun->res_type = 0;
8955		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8956			if (lun->pr_key_count==0) {
8957				lun->flags &= ~CTL_LUN_PR_RESERVED;
8958				lun->res_type = 0;
8959				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8960			}
8961		}
8962		lun->PRGeneration++;
8963		break;
8964
8965	case CTL_PR_RESERVE:
8966		lun->flags |= CTL_LUN_PR_RESERVED;
8967		lun->res_type = msg->pr.pr_info.res_type;
8968		lun->pr_res_idx = msg->pr.pr_info.residx;
8969
8970		break;
8971
8972	case CTL_PR_RELEASE:
8973		/*
8974		 * if this isn't an exclusive access res generate UA for all
8975		 * other registrants.
8976		 */
8977		if (lun->res_type != SPR_TYPE_EX_AC
8978		 && lun->res_type != SPR_TYPE_WR_EX) {
8979			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8980				if (lun->per_res[i+persis_offset].registered)
8981					lun->pending_ua[i] |=
8982						CTL_UA_RES_RELEASE;
8983		}
8984
8985		lun->flags &= ~CTL_LUN_PR_RESERVED;
8986		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8987		lun->res_type = 0;
8988		break;
8989
8990	case CTL_PR_PREEMPT:
8991		ctl_pro_preempt_other(lun, msg);
8992		break;
8993	case CTL_PR_CLEAR:
8994		lun->flags &= ~CTL_LUN_PR_RESERVED;
8995		lun->res_type = 0;
8996		lun->pr_key_count = 0;
8997		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8998
8999		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
9000			if (lun->per_res[i].registered == 0)
9001				continue;
9002			if (!persis_offset
9003			 && i < CTL_MAX_INITIATORS)
9004				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
9005			else if (persis_offset
9006			      && i >= persis_offset)
9007				lun->pending_ua[i-persis_offset] |=
9008					CTL_UA_RES_PREEMPT;
9009			memset(&lun->per_res[i].res_key, 0,
9010			       sizeof(struct scsi_per_res_key));
9011			lun->per_res[i].registered = 0;
9012		}
9013		lun->PRGeneration++;
9014		break;
9015	}
9016
9017	mtx_unlock(&lun->lun_lock);
9018}
9019
9020int
9021ctl_read_write(struct ctl_scsiio *ctsio)
9022{
9023	struct ctl_lun *lun;
9024	struct ctl_lba_len_flags *lbalen;
9025	uint64_t lba;
9026	uint32_t num_blocks;
9027	int flags, retval;
9028	int isread;
9029
9030	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9031
9032	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
9033
9034	flags = 0;
9035	retval = CTL_RETVAL_COMPLETE;
9036
9037	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
9038	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
9039	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
9040		uint32_t residx;
9041
9042		/*
9043		 * XXX KDM need a lock here.
9044		 */
9045		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9046		if ((lun->res_type == SPR_TYPE_EX_AC
9047		  && residx != lun->pr_res_idx)
9048		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
9049		   || lun->res_type == SPR_TYPE_EX_AC_AR)
9050		  && !lun->per_res[residx].registered)) {
9051			ctl_set_reservation_conflict(ctsio);
9052			ctl_done((union ctl_io *)ctsio);
9053			return (CTL_RETVAL_COMPLETE);
9054	        }
9055	}
9056
9057	switch (ctsio->cdb[0]) {
9058	case READ_6:
9059	case WRITE_6: {
9060		struct scsi_rw_6 *cdb;
9061
9062		cdb = (struct scsi_rw_6 *)ctsio->cdb;
9063
9064		lba = scsi_3btoul(cdb->addr);
9065		/* only 5 bits are valid in the most significant address byte */
9066		lba &= 0x1fffff;
9067		num_blocks = cdb->length;
9068		/*
9069		 * This is correct according to SBC-2.
9070		 */
9071		if (num_blocks == 0)
9072			num_blocks = 256;
9073		break;
9074	}
9075	case READ_10:
9076	case WRITE_10: {
9077		struct scsi_rw_10 *cdb;
9078
9079		cdb = (struct scsi_rw_10 *)ctsio->cdb;
9080		if (cdb->byte2 & SRW10_FUA)
9081			flags |= CTL_LLF_FUA;
9082		if (cdb->byte2 & SRW10_DPO)
9083			flags |= CTL_LLF_DPO;
9084		lba = scsi_4btoul(cdb->addr);
9085		num_blocks = scsi_2btoul(cdb->length);
9086		break;
9087	}
9088	case WRITE_VERIFY_10: {
9089		struct scsi_write_verify_10 *cdb;
9090
9091		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9092		flags |= CTL_LLF_FUA;
9093		if (cdb->byte2 & SWV_DPO)
9094			flags |= CTL_LLF_DPO;
9095		lba = scsi_4btoul(cdb->addr);
9096		num_blocks = scsi_2btoul(cdb->length);
9097		break;
9098	}
9099	case READ_12:
9100	case WRITE_12: {
9101		struct scsi_rw_12 *cdb;
9102
9103		cdb = (struct scsi_rw_12 *)ctsio->cdb;
9104		if (cdb->byte2 & SRW12_FUA)
9105			flags |= CTL_LLF_FUA;
9106		if (cdb->byte2 & SRW12_DPO)
9107			flags |= CTL_LLF_DPO;
9108		lba = scsi_4btoul(cdb->addr);
9109		num_blocks = scsi_4btoul(cdb->length);
9110		break;
9111	}
9112	case WRITE_VERIFY_12: {
9113		struct scsi_write_verify_12 *cdb;
9114
9115		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9116		flags |= CTL_LLF_FUA;
9117		if (cdb->byte2 & SWV_DPO)
9118			flags |= CTL_LLF_DPO;
9119		lba = scsi_4btoul(cdb->addr);
9120		num_blocks = scsi_4btoul(cdb->length);
9121		break;
9122	}
9123	case READ_16:
9124	case WRITE_16: {
9125		struct scsi_rw_16 *cdb;
9126
9127		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9128		if (cdb->byte2 & SRW12_FUA)
9129			flags |= CTL_LLF_FUA;
9130		if (cdb->byte2 & SRW12_DPO)
9131			flags |= CTL_LLF_DPO;
9132		lba = scsi_8btou64(cdb->addr);
9133		num_blocks = scsi_4btoul(cdb->length);
9134		break;
9135	}
9136	case WRITE_VERIFY_16: {
9137		struct scsi_write_verify_16 *cdb;
9138
9139		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9140		flags |= CTL_LLF_FUA;
9141		if (cdb->byte2 & SWV_DPO)
9142			flags |= CTL_LLF_DPO;
9143		lba = scsi_8btou64(cdb->addr);
9144		num_blocks = scsi_4btoul(cdb->length);
9145		break;
9146	}
9147	default:
9148		/*
9149		 * We got a command we don't support.  This shouldn't
9150		 * happen, commands should be filtered out above us.
9151		 */
9152		ctl_set_invalid_opcode(ctsio);
9153		ctl_done((union ctl_io *)ctsio);
9154
9155		return (CTL_RETVAL_COMPLETE);
9156		break; /* NOTREACHED */
9157	}
9158
9159	/*
9160	 * The first check is to make sure we're in bounds, the second
9161	 * check is to catch wrap-around problems.  If the lba + num blocks
9162	 * is less than the lba, then we've wrapped around and the block
9163	 * range is invalid anyway.
9164	 */
9165	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9166	 || ((lba + num_blocks) < lba)) {
9167		ctl_set_lba_out_of_range(ctsio);
9168		ctl_done((union ctl_io *)ctsio);
9169		return (CTL_RETVAL_COMPLETE);
9170	}
9171
9172	/*
9173	 * According to SBC-3, a transfer length of 0 is not an error.
9174	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9175	 * translates to 256 blocks for those commands.
9176	 */
9177	if (num_blocks == 0) {
9178		ctl_set_success(ctsio);
9179		ctl_done((union ctl_io *)ctsio);
9180		return (CTL_RETVAL_COMPLETE);
9181	}
9182
9183	/* Set FUA and/or DPO if caches are disabled. */
9184	if (isread) {
9185		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9186		    SCP_RCD) != 0)
9187			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9188	} else {
9189		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9190		    SCP_WCE) == 0)
9191			flags |= CTL_LLF_FUA;
9192	}
9193
9194	lbalen = (struct ctl_lba_len_flags *)
9195	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9196	lbalen->lba = lba;
9197	lbalen->len = num_blocks;
9198	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9199
9200	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9201	ctsio->kern_rel_offset = 0;
9202
9203	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9204
9205	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9206
9207	return (retval);
9208}
9209
9210static int
9211ctl_cnw_cont(union ctl_io *io)
9212{
9213	struct ctl_scsiio *ctsio;
9214	struct ctl_lun *lun;
9215	struct ctl_lba_len_flags *lbalen;
9216	int retval;
9217
9218	ctsio = &io->scsiio;
9219	ctsio->io_hdr.status = CTL_STATUS_NONE;
9220	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9221	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9222	lbalen = (struct ctl_lba_len_flags *)
9223	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9224	lbalen->flags &= ~CTL_LLF_COMPARE;
9225	lbalen->flags |= CTL_LLF_WRITE;
9226
9227	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9228	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9229	return (retval);
9230}
9231
9232int
9233ctl_cnw(struct ctl_scsiio *ctsio)
9234{
9235	struct ctl_lun *lun;
9236	struct ctl_lba_len_flags *lbalen;
9237	uint64_t lba;
9238	uint32_t num_blocks;
9239	int flags, retval;
9240
9241	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9242
9243	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9244
9245	flags = 0;
9246	retval = CTL_RETVAL_COMPLETE;
9247
9248	switch (ctsio->cdb[0]) {
9249	case COMPARE_AND_WRITE: {
9250		struct scsi_compare_and_write *cdb;
9251
9252		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9253		if (cdb->byte2 & SRW10_FUA)
9254			flags |= CTL_LLF_FUA;
9255		if (cdb->byte2 & SRW10_DPO)
9256			flags |= CTL_LLF_DPO;
9257		lba = scsi_8btou64(cdb->addr);
9258		num_blocks = cdb->length;
9259		break;
9260	}
9261	default:
9262		/*
9263		 * We got a command we don't support.  This shouldn't
9264		 * happen, commands should be filtered out above us.
9265		 */
9266		ctl_set_invalid_opcode(ctsio);
9267		ctl_done((union ctl_io *)ctsio);
9268
9269		return (CTL_RETVAL_COMPLETE);
9270		break; /* NOTREACHED */
9271	}
9272
9273	/*
9274	 * The first check is to make sure we're in bounds, the second
9275	 * check is to catch wrap-around problems.  If the lba + num blocks
9276	 * is less than the lba, then we've wrapped around and the block
9277	 * range is invalid anyway.
9278	 */
9279	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9280	 || ((lba + num_blocks) < lba)) {
9281		ctl_set_lba_out_of_range(ctsio);
9282		ctl_done((union ctl_io *)ctsio);
9283		return (CTL_RETVAL_COMPLETE);
9284	}
9285
9286	/*
9287	 * According to SBC-3, a transfer length of 0 is not an error.
9288	 */
9289	if (num_blocks == 0) {
9290		ctl_set_success(ctsio);
9291		ctl_done((union ctl_io *)ctsio);
9292		return (CTL_RETVAL_COMPLETE);
9293	}
9294
9295	/* Set FUA if write cache is disabled. */
9296	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9297	    SCP_WCE) == 0)
9298		flags |= CTL_LLF_FUA;
9299
9300	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9301	ctsio->kern_rel_offset = 0;
9302
9303	/*
9304	 * Set the IO_CONT flag, so that if this I/O gets passed to
9305	 * ctl_data_submit_done(), it'll get passed back to
9306	 * ctl_ctl_cnw_cont() for further processing.
9307	 */
9308	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9309	ctsio->io_cont = ctl_cnw_cont;
9310
9311	lbalen = (struct ctl_lba_len_flags *)
9312	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9313	lbalen->lba = lba;
9314	lbalen->len = num_blocks;
9315	lbalen->flags = CTL_LLF_COMPARE | flags;
9316
9317	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9318	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9319	return (retval);
9320}
9321
9322int
9323ctl_verify(struct ctl_scsiio *ctsio)
9324{
9325	struct ctl_lun *lun;
9326	struct ctl_lba_len_flags *lbalen;
9327	uint64_t lba;
9328	uint32_t num_blocks;
9329	int bytchk, flags;
9330	int retval;
9331
9332	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9333
9334	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9335
9336	bytchk = 0;
9337	flags = CTL_LLF_FUA;
9338	retval = CTL_RETVAL_COMPLETE;
9339
9340	switch (ctsio->cdb[0]) {
9341	case VERIFY_10: {
9342		struct scsi_verify_10 *cdb;
9343
9344		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9345		if (cdb->byte2 & SVFY_BYTCHK)
9346			bytchk = 1;
9347		if (cdb->byte2 & SVFY_DPO)
9348			flags |= CTL_LLF_DPO;
9349		lba = scsi_4btoul(cdb->addr);
9350		num_blocks = scsi_2btoul(cdb->length);
9351		break;
9352	}
9353	case VERIFY_12: {
9354		struct scsi_verify_12 *cdb;
9355
9356		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9357		if (cdb->byte2 & SVFY_BYTCHK)
9358			bytchk = 1;
9359		if (cdb->byte2 & SVFY_DPO)
9360			flags |= CTL_LLF_DPO;
9361		lba = scsi_4btoul(cdb->addr);
9362		num_blocks = scsi_4btoul(cdb->length);
9363		break;
9364	}
9365	case VERIFY_16: {
9366		struct scsi_rw_16 *cdb;
9367
9368		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9369		if (cdb->byte2 & SVFY_BYTCHK)
9370			bytchk = 1;
9371		if (cdb->byte2 & SVFY_DPO)
9372			flags |= CTL_LLF_DPO;
9373		lba = scsi_8btou64(cdb->addr);
9374		num_blocks = scsi_4btoul(cdb->length);
9375		break;
9376	}
9377	default:
9378		/*
9379		 * We got a command we don't support.  This shouldn't
9380		 * happen, commands should be filtered out above us.
9381		 */
9382		ctl_set_invalid_opcode(ctsio);
9383		ctl_done((union ctl_io *)ctsio);
9384		return (CTL_RETVAL_COMPLETE);
9385	}
9386
9387	/*
9388	 * The first check is to make sure we're in bounds, the second
9389	 * check is to catch wrap-around problems.  If the lba + num blocks
9390	 * is less than the lba, then we've wrapped around and the block
9391	 * range is invalid anyway.
9392	 */
9393	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9394	 || ((lba + num_blocks) < lba)) {
9395		ctl_set_lba_out_of_range(ctsio);
9396		ctl_done((union ctl_io *)ctsio);
9397		return (CTL_RETVAL_COMPLETE);
9398	}
9399
9400	/*
9401	 * According to SBC-3, a transfer length of 0 is not an error.
9402	 */
9403	if (num_blocks == 0) {
9404		ctl_set_success(ctsio);
9405		ctl_done((union ctl_io *)ctsio);
9406		return (CTL_RETVAL_COMPLETE);
9407	}
9408
9409	lbalen = (struct ctl_lba_len_flags *)
9410	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9411	lbalen->lba = lba;
9412	lbalen->len = num_blocks;
9413	if (bytchk) {
9414		lbalen->flags = CTL_LLF_COMPARE | flags;
9415		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9416	} else {
9417		lbalen->flags = CTL_LLF_VERIFY | flags;
9418		ctsio->kern_total_len = 0;
9419	}
9420	ctsio->kern_rel_offset = 0;
9421
9422	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9423	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9424	return (retval);
9425}
9426
9427int
9428ctl_report_luns(struct ctl_scsiio *ctsio)
9429{
9430	struct scsi_report_luns *cdb;
9431	struct scsi_report_luns_data *lun_data;
9432	struct ctl_lun *lun, *request_lun;
9433	int num_luns, retval;
9434	uint32_t alloc_len, lun_datalen;
9435	int num_filled, well_known;
9436	uint32_t initidx, targ_lun_id, lun_id;
9437
9438	retval = CTL_RETVAL_COMPLETE;
9439	well_known = 0;
9440
9441	cdb = (struct scsi_report_luns *)ctsio->cdb;
9442
9443	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9444
9445	mtx_lock(&control_softc->ctl_lock);
9446	num_luns = control_softc->num_luns;
9447	mtx_unlock(&control_softc->ctl_lock);
9448
9449	switch (cdb->select_report) {
9450	case RPL_REPORT_DEFAULT:
9451	case RPL_REPORT_ALL:
9452		break;
9453	case RPL_REPORT_WELLKNOWN:
9454		well_known = 1;
9455		num_luns = 0;
9456		break;
9457	default:
9458		ctl_set_invalid_field(ctsio,
9459				      /*sks_valid*/ 1,
9460				      /*command*/ 1,
9461				      /*field*/ 2,
9462				      /*bit_valid*/ 0,
9463				      /*bit*/ 0);
9464		ctl_done((union ctl_io *)ctsio);
9465		return (retval);
9466		break; /* NOTREACHED */
9467	}
9468
9469	alloc_len = scsi_4btoul(cdb->length);
9470	/*
9471	 * The initiator has to allocate at least 16 bytes for this request,
9472	 * so he can at least get the header and the first LUN.  Otherwise
9473	 * we reject the request (per SPC-3 rev 14, section 6.21).
9474	 */
9475	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9476	    sizeof(struct scsi_report_luns_lundata))) {
9477		ctl_set_invalid_field(ctsio,
9478				      /*sks_valid*/ 1,
9479				      /*command*/ 1,
9480				      /*field*/ 6,
9481				      /*bit_valid*/ 0,
9482				      /*bit*/ 0);
9483		ctl_done((union ctl_io *)ctsio);
9484		return (retval);
9485	}
9486
9487	request_lun = (struct ctl_lun *)
9488		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9489
9490	lun_datalen = sizeof(*lun_data) +
9491		(num_luns * sizeof(struct scsi_report_luns_lundata));
9492
9493	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9494	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9495	ctsio->kern_sg_entries = 0;
9496
9497	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9498
9499	mtx_lock(&control_softc->ctl_lock);
9500	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9501		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9502		if (lun_id >= CTL_MAX_LUNS)
9503			continue;
9504		lun = control_softc->ctl_luns[lun_id];
9505		if (lun == NULL)
9506			continue;
9507
9508		if (targ_lun_id <= 0xff) {
9509			/*
9510			 * Peripheral addressing method, bus number 0.
9511			 */
9512			lun_data->luns[num_filled].lundata[0] =
9513				RPL_LUNDATA_ATYP_PERIPH;
9514			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9515			num_filled++;
9516		} else if (targ_lun_id <= 0x3fff) {
9517			/*
9518			 * Flat addressing method.
9519			 */
9520			lun_data->luns[num_filled].lundata[0] =
9521				RPL_LUNDATA_ATYP_FLAT |
9522				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9523#ifdef OLDCTLHEADERS
9524				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9525				(targ_lun_id & SRLD_BUS_LUN_MASK);
9526#endif
9527			lun_data->luns[num_filled].lundata[1] =
9528#ifdef OLDCTLHEADERS
9529				targ_lun_id >> SRLD_BUS_LUN_BITS;
9530#endif
9531				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9532			num_filled++;
9533		} else {
9534			printf("ctl_report_luns: bogus LUN number %jd, "
9535			       "skipping\n", (intmax_t)targ_lun_id);
9536		}
9537		/*
9538		 * According to SPC-3, rev 14 section 6.21:
9539		 *
9540		 * "The execution of a REPORT LUNS command to any valid and
9541		 * installed logical unit shall clear the REPORTED LUNS DATA
9542		 * HAS CHANGED unit attention condition for all logical
9543		 * units of that target with respect to the requesting
9544		 * initiator. A valid and installed logical unit is one
9545		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9546		 * INQUIRY data (see 6.4.2)."
9547		 *
9548		 * If request_lun is NULL, the LUN this report luns command
9549		 * was issued to is either disabled or doesn't exist. In that
9550		 * case, we shouldn't clear any pending lun change unit
9551		 * attention.
9552		 */
9553		if (request_lun != NULL) {
9554			mtx_lock(&lun->lun_lock);
9555			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9556			mtx_unlock(&lun->lun_lock);
9557		}
9558	}
9559	mtx_unlock(&control_softc->ctl_lock);
9560
9561	/*
9562	 * It's quite possible that we've returned fewer LUNs than we allocated
9563	 * space for.  Trim it.
9564	 */
9565	lun_datalen = sizeof(*lun_data) +
9566		(num_filled * sizeof(struct scsi_report_luns_lundata));
9567
9568	if (lun_datalen < alloc_len) {
9569		ctsio->residual = alloc_len - lun_datalen;
9570		ctsio->kern_data_len = lun_datalen;
9571		ctsio->kern_total_len = lun_datalen;
9572	} else {
9573		ctsio->residual = 0;
9574		ctsio->kern_data_len = alloc_len;
9575		ctsio->kern_total_len = alloc_len;
9576	}
9577	ctsio->kern_data_resid = 0;
9578	ctsio->kern_rel_offset = 0;
9579	ctsio->kern_sg_entries = 0;
9580
9581	/*
9582	 * We set this to the actual data length, regardless of how much
9583	 * space we actually have to return results.  If the user looks at
9584	 * this value, he'll know whether or not he allocated enough space
9585	 * and reissue the command if necessary.  We don't support well
9586	 * known logical units, so if the user asks for that, return none.
9587	 */
9588	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9589
9590	/*
9591	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9592	 * this request.
9593	 */
9594	ctsio->scsi_status = SCSI_STATUS_OK;
9595
9596	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9597	ctsio->be_move_done = ctl_config_move_done;
9598	ctl_datamove((union ctl_io *)ctsio);
9599
9600	return (retval);
9601}
9602
9603int
9604ctl_request_sense(struct ctl_scsiio *ctsio)
9605{
9606	struct scsi_request_sense *cdb;
9607	struct scsi_sense_data *sense_ptr;
9608	struct ctl_lun *lun;
9609	uint32_t initidx;
9610	int have_error;
9611	scsi_sense_data_type sense_format;
9612
9613	cdb = (struct scsi_request_sense *)ctsio->cdb;
9614
9615	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9616
9617	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9618
9619	/*
9620	 * Determine which sense format the user wants.
9621	 */
9622	if (cdb->byte2 & SRS_DESC)
9623		sense_format = SSD_TYPE_DESC;
9624	else
9625		sense_format = SSD_TYPE_FIXED;
9626
9627	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9628	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9629	ctsio->kern_sg_entries = 0;
9630
9631	/*
9632	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9633	 * larger than the largest allowed value for the length field in the
9634	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9635	 */
9636	ctsio->residual = 0;
9637	ctsio->kern_data_len = cdb->length;
9638	ctsio->kern_total_len = cdb->length;
9639
9640	ctsio->kern_data_resid = 0;
9641	ctsio->kern_rel_offset = 0;
9642	ctsio->kern_sg_entries = 0;
9643
9644	/*
9645	 * If we don't have a LUN, we don't have any pending sense.
9646	 */
9647	if (lun == NULL)
9648		goto no_sense;
9649
9650	have_error = 0;
9651	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9652	/*
9653	 * Check for pending sense, and then for pending unit attentions.
9654	 * Pending sense gets returned first, then pending unit attentions.
9655	 */
9656	mtx_lock(&lun->lun_lock);
9657#ifdef CTL_WITH_CA
9658	if (ctl_is_set(lun->have_ca, initidx)) {
9659		scsi_sense_data_type stored_format;
9660
9661		/*
9662		 * Check to see which sense format was used for the stored
9663		 * sense data.
9664		 */
9665		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9666
9667		/*
9668		 * If the user requested a different sense format than the
9669		 * one we stored, then we need to convert it to the other
9670		 * format.  If we're going from descriptor to fixed format
9671		 * sense data, we may lose things in translation, depending
9672		 * on what options were used.
9673		 *
9674		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9675		 * for some reason we'll just copy it out as-is.
9676		 */
9677		if ((stored_format == SSD_TYPE_FIXED)
9678		 && (sense_format == SSD_TYPE_DESC))
9679			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9680			    &lun->pending_sense[initidx],
9681			    (struct scsi_sense_data_desc *)sense_ptr);
9682		else if ((stored_format == SSD_TYPE_DESC)
9683		      && (sense_format == SSD_TYPE_FIXED))
9684			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9685			    &lun->pending_sense[initidx],
9686			    (struct scsi_sense_data_fixed *)sense_ptr);
9687		else
9688			memcpy(sense_ptr, &lun->pending_sense[initidx],
9689			       ctl_min(sizeof(*sense_ptr),
9690			       sizeof(lun->pending_sense[initidx])));
9691
9692		ctl_clear_mask(lun->have_ca, initidx);
9693		have_error = 1;
9694	} else
9695#endif
9696	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9697		ctl_ua_type ua_type;
9698
9699		ua_type = ctl_build_ua(&lun->pending_ua[initidx],
9700				       sense_ptr, sense_format);
9701		if (ua_type != CTL_UA_NONE)
9702			have_error = 1;
9703	}
9704	mtx_unlock(&lun->lun_lock);
9705
9706	/*
9707	 * We already have a pending error, return it.
9708	 */
9709	if (have_error != 0) {
9710		/*
9711		 * We report the SCSI status as OK, since the status of the
9712		 * request sense command itself is OK.
9713		 */
9714		ctsio->scsi_status = SCSI_STATUS_OK;
9715
9716		/*
9717		 * We report 0 for the sense length, because we aren't doing
9718		 * autosense in this case.  We're reporting sense as
9719		 * parameter data.
9720		 */
9721		ctsio->sense_len = 0;
9722		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9723		ctsio->be_move_done = ctl_config_move_done;
9724		ctl_datamove((union ctl_io *)ctsio);
9725
9726		return (CTL_RETVAL_COMPLETE);
9727	}
9728
9729no_sense:
9730
9731	/*
9732	 * No sense information to report, so we report that everything is
9733	 * okay.
9734	 */
9735	ctl_set_sense_data(sense_ptr,
9736			   lun,
9737			   sense_format,
9738			   /*current_error*/ 1,
9739			   /*sense_key*/ SSD_KEY_NO_SENSE,
9740			   /*asc*/ 0x00,
9741			   /*ascq*/ 0x00,
9742			   SSD_ELEM_NONE);
9743
9744	ctsio->scsi_status = SCSI_STATUS_OK;
9745
9746	/*
9747	 * We report 0 for the sense length, because we aren't doing
9748	 * autosense in this case.  We're reporting sense as parameter data.
9749	 */
9750	ctsio->sense_len = 0;
9751	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9752	ctsio->be_move_done = ctl_config_move_done;
9753	ctl_datamove((union ctl_io *)ctsio);
9754
9755	return (CTL_RETVAL_COMPLETE);
9756}
9757
9758int
9759ctl_tur(struct ctl_scsiio *ctsio)
9760{
9761	struct ctl_lun *lun;
9762
9763	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9764
9765	CTL_DEBUG_PRINT(("ctl_tur\n"));
9766
9767	if (lun == NULL)
9768		return (EINVAL);
9769
9770	ctsio->scsi_status = SCSI_STATUS_OK;
9771	ctsio->io_hdr.status = CTL_SUCCESS;
9772
9773	ctl_done((union ctl_io *)ctsio);
9774
9775	return (CTL_RETVAL_COMPLETE);
9776}
9777
9778#ifdef notyet
9779static int
9780ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9781{
9782
9783}
9784#endif
9785
9786static int
9787ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9788{
9789	struct scsi_vpd_supported_pages *pages;
9790	int sup_page_size;
9791	struct ctl_lun *lun;
9792
9793	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9794
9795	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9796	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9797	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9798	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9799	ctsio->kern_sg_entries = 0;
9800
9801	if (sup_page_size < alloc_len) {
9802		ctsio->residual = alloc_len - sup_page_size;
9803		ctsio->kern_data_len = sup_page_size;
9804		ctsio->kern_total_len = sup_page_size;
9805	} else {
9806		ctsio->residual = 0;
9807		ctsio->kern_data_len = alloc_len;
9808		ctsio->kern_total_len = alloc_len;
9809	}
9810	ctsio->kern_data_resid = 0;
9811	ctsio->kern_rel_offset = 0;
9812	ctsio->kern_sg_entries = 0;
9813
9814	/*
9815	 * The control device is always connected.  The disk device, on the
9816	 * other hand, may not be online all the time.  Need to change this
9817	 * to figure out whether the disk device is actually online or not.
9818	 */
9819	if (lun != NULL)
9820		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9821				lun->be_lun->lun_type;
9822	else
9823		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9824
9825	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9826	/* Supported VPD pages */
9827	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9828	/* Serial Number */
9829	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9830	/* Device Identification */
9831	pages->page_list[2] = SVPD_DEVICE_ID;
9832	/* Extended INQUIRY Data */
9833	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9834	/* Mode Page Policy */
9835	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9836	/* SCSI Ports */
9837	pages->page_list[5] = SVPD_SCSI_PORTS;
9838	/* Third-party Copy */
9839	pages->page_list[6] = SVPD_SCSI_TPC;
9840	/* Block limits */
9841	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9842	/* Block Device Characteristics */
9843	pages->page_list[8] = SVPD_BDC;
9844	/* Logical Block Provisioning */
9845	pages->page_list[9] = SVPD_LBP;
9846
9847	ctsio->scsi_status = SCSI_STATUS_OK;
9848
9849	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9850	ctsio->be_move_done = ctl_config_move_done;
9851	ctl_datamove((union ctl_io *)ctsio);
9852
9853	return (CTL_RETVAL_COMPLETE);
9854}
9855
9856static int
9857ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9858{
9859	struct scsi_vpd_unit_serial_number *sn_ptr;
9860	struct ctl_lun *lun;
9861
9862	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9863
9864	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9865	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9866	ctsio->kern_sg_entries = 0;
9867
9868	if (sizeof(*sn_ptr) < alloc_len) {
9869		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9870		ctsio->kern_data_len = sizeof(*sn_ptr);
9871		ctsio->kern_total_len = sizeof(*sn_ptr);
9872	} else {
9873		ctsio->residual = 0;
9874		ctsio->kern_data_len = alloc_len;
9875		ctsio->kern_total_len = alloc_len;
9876	}
9877	ctsio->kern_data_resid = 0;
9878	ctsio->kern_rel_offset = 0;
9879	ctsio->kern_sg_entries = 0;
9880
9881	/*
9882	 * The control device is always connected.  The disk device, on the
9883	 * other hand, may not be online all the time.  Need to change this
9884	 * to figure out whether the disk device is actually online or not.
9885	 */
9886	if (lun != NULL)
9887		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9888				  lun->be_lun->lun_type;
9889	else
9890		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9891
9892	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9893	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9894	/*
9895	 * If we don't have a LUN, we just leave the serial number as
9896	 * all spaces.
9897	 */
9898	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9899	if (lun != NULL) {
9900		strncpy((char *)sn_ptr->serial_num,
9901			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9902	}
9903	ctsio->scsi_status = SCSI_STATUS_OK;
9904
9905	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9906	ctsio->be_move_done = ctl_config_move_done;
9907	ctl_datamove((union ctl_io *)ctsio);
9908
9909	return (CTL_RETVAL_COMPLETE);
9910}
9911
9912
9913static int
9914ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9915{
9916	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9917	struct ctl_lun *lun;
9918	int data_len;
9919
9920	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9921
9922	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9923	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9924
9925	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9926	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9927	ctsio->kern_sg_entries = 0;
9928
9929	if (data_len < alloc_len) {
9930		ctsio->residual = alloc_len - data_len;
9931		ctsio->kern_data_len = data_len;
9932		ctsio->kern_total_len = data_len;
9933	} else {
9934		ctsio->residual = 0;
9935		ctsio->kern_data_len = alloc_len;
9936		ctsio->kern_total_len = alloc_len;
9937	}
9938	ctsio->kern_data_resid = 0;
9939	ctsio->kern_rel_offset = 0;
9940	ctsio->kern_sg_entries = 0;
9941
9942	/*
9943	 * The control device is always connected.  The disk device, on the
9944	 * other hand, may not be online all the time.
9945	 */
9946	if (lun != NULL)
9947		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9948				     lun->be_lun->lun_type;
9949	else
9950		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9951	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9952	eid_ptr->page_length = data_len - 4;
9953	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9954	eid_ptr->flags3 = SVPD_EID_V_SUP;
9955
9956	ctsio->scsi_status = SCSI_STATUS_OK;
9957	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9958	ctsio->be_move_done = ctl_config_move_done;
9959	ctl_datamove((union ctl_io *)ctsio);
9960
9961	return (CTL_RETVAL_COMPLETE);
9962}
9963
9964static int
9965ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9966{
9967	struct scsi_vpd_mode_page_policy *mpp_ptr;
9968	struct ctl_lun *lun;
9969	int data_len;
9970
9971	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9972
9973	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9974	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9975
9976	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9977	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9978	ctsio->kern_sg_entries = 0;
9979
9980	if (data_len < alloc_len) {
9981		ctsio->residual = alloc_len - data_len;
9982		ctsio->kern_data_len = data_len;
9983		ctsio->kern_total_len = data_len;
9984	} else {
9985		ctsio->residual = 0;
9986		ctsio->kern_data_len = alloc_len;
9987		ctsio->kern_total_len = alloc_len;
9988	}
9989	ctsio->kern_data_resid = 0;
9990	ctsio->kern_rel_offset = 0;
9991	ctsio->kern_sg_entries = 0;
9992
9993	/*
9994	 * The control device is always connected.  The disk device, on the
9995	 * other hand, may not be online all the time.
9996	 */
9997	if (lun != NULL)
9998		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9999				     lun->be_lun->lun_type;
10000	else
10001		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10002	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
10003	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
10004	mpp_ptr->descr[0].page_code = 0x3f;
10005	mpp_ptr->descr[0].subpage_code = 0xff;
10006	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
10007
10008	ctsio->scsi_status = SCSI_STATUS_OK;
10009	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10010	ctsio->be_move_done = ctl_config_move_done;
10011	ctl_datamove((union ctl_io *)ctsio);
10012
10013	return (CTL_RETVAL_COMPLETE);
10014}
10015
10016static int
10017ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
10018{
10019	struct scsi_vpd_device_id *devid_ptr;
10020	struct scsi_vpd_id_descriptor *desc;
10021	struct ctl_softc *ctl_softc;
10022	struct ctl_lun *lun;
10023	struct ctl_port *port;
10024	int data_len;
10025	uint8_t proto;
10026
10027	ctl_softc = control_softc;
10028
10029	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
10030	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10031
10032	data_len = sizeof(struct scsi_vpd_device_id) +
10033	    sizeof(struct scsi_vpd_id_descriptor) +
10034		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
10035	    sizeof(struct scsi_vpd_id_descriptor) +
10036		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
10037	if (lun && lun->lun_devid)
10038		data_len += lun->lun_devid->len;
10039	if (port->port_devid)
10040		data_len += port->port_devid->len;
10041	if (port->target_devid)
10042		data_len += port->target_devid->len;
10043
10044	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10045	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
10046	ctsio->kern_sg_entries = 0;
10047
10048	if (data_len < alloc_len) {
10049		ctsio->residual = alloc_len - data_len;
10050		ctsio->kern_data_len = data_len;
10051		ctsio->kern_total_len = data_len;
10052	} else {
10053		ctsio->residual = 0;
10054		ctsio->kern_data_len = alloc_len;
10055		ctsio->kern_total_len = alloc_len;
10056	}
10057	ctsio->kern_data_resid = 0;
10058	ctsio->kern_rel_offset = 0;
10059	ctsio->kern_sg_entries = 0;
10060
10061	/*
10062	 * The control device is always connected.  The disk device, on the
10063	 * other hand, may not be online all the time.
10064	 */
10065	if (lun != NULL)
10066		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10067				     lun->be_lun->lun_type;
10068	else
10069		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10070	devid_ptr->page_code = SVPD_DEVICE_ID;
10071	scsi_ulto2b(data_len - 4, devid_ptr->length);
10072
10073	if (port->port_type == CTL_PORT_FC)
10074		proto = SCSI_PROTO_FC << 4;
10075	else if (port->port_type == CTL_PORT_ISCSI)
10076		proto = SCSI_PROTO_ISCSI << 4;
10077	else
10078		proto = SCSI_PROTO_SPI << 4;
10079	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
10080
10081	/*
10082	 * We're using a LUN association here.  i.e., this device ID is a
10083	 * per-LUN identifier.
10084	 */
10085	if (lun && lun->lun_devid) {
10086		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
10087		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10088		    lun->lun_devid->len);
10089	}
10090
10091	/*
10092	 * This is for the WWPN which is a port association.
10093	 */
10094	if (port->port_devid) {
10095		memcpy(desc, port->port_devid->data, port->port_devid->len);
10096		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10097		    port->port_devid->len);
10098	}
10099
10100	/*
10101	 * This is for the Relative Target Port(type 4h) identifier
10102	 */
10103	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10104	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10105	    SVPD_ID_TYPE_RELTARG;
10106	desc->length = 4;
10107	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
10108	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10109	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
10110
10111	/*
10112	 * This is for the Target Port Group(type 5h) identifier
10113	 */
10114	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10115	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10116	    SVPD_ID_TYPE_TPORTGRP;
10117	desc->length = 4;
10118	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
10119	    &desc->identifier[2]);
10120	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10121	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10122
10123	/*
10124	 * This is for the Target identifier
10125	 */
10126	if (port->target_devid) {
10127		memcpy(desc, port->target_devid->data, port->target_devid->len);
10128	}
10129
10130	ctsio->scsi_status = SCSI_STATUS_OK;
10131	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10132	ctsio->be_move_done = ctl_config_move_done;
10133	ctl_datamove((union ctl_io *)ctsio);
10134
10135	return (CTL_RETVAL_COMPLETE);
10136}
10137
10138static int
10139ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10140{
10141	struct ctl_softc *softc = control_softc;
10142	struct scsi_vpd_scsi_ports *sp;
10143	struct scsi_vpd_port_designation *pd;
10144	struct scsi_vpd_port_designation_cont *pdc;
10145	struct ctl_lun *lun;
10146	struct ctl_port *port;
10147	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10148	int num_target_port_groups, single;
10149
10150	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10151
10152	single = ctl_is_single;
10153	if (single)
10154		num_target_port_groups = 1;
10155	else
10156		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10157	num_target_ports = 0;
10158	iid_len = 0;
10159	id_len = 0;
10160	mtx_lock(&softc->ctl_lock);
10161	STAILQ_FOREACH(port, &softc->port_list, links) {
10162		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10163			continue;
10164		if (lun != NULL &&
10165		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10166		    CTL_MAX_LUNS)
10167			continue;
10168		num_target_ports++;
10169		if (port->init_devid)
10170			iid_len += port->init_devid->len;
10171		if (port->port_devid)
10172			id_len += port->port_devid->len;
10173	}
10174	mtx_unlock(&softc->ctl_lock);
10175
10176	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10177	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10178	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10179	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10180	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10181	ctsio->kern_sg_entries = 0;
10182
10183	if (data_len < alloc_len) {
10184		ctsio->residual = alloc_len - data_len;
10185		ctsio->kern_data_len = data_len;
10186		ctsio->kern_total_len = data_len;
10187	} else {
10188		ctsio->residual = 0;
10189		ctsio->kern_data_len = alloc_len;
10190		ctsio->kern_total_len = alloc_len;
10191	}
10192	ctsio->kern_data_resid = 0;
10193	ctsio->kern_rel_offset = 0;
10194	ctsio->kern_sg_entries = 0;
10195
10196	/*
10197	 * The control device is always connected.  The disk device, on the
10198	 * other hand, may not be online all the time.  Need to change this
10199	 * to figure out whether the disk device is actually online or not.
10200	 */
10201	if (lun != NULL)
10202		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10203				  lun->be_lun->lun_type;
10204	else
10205		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10206
10207	sp->page_code = SVPD_SCSI_PORTS;
10208	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10209	    sp->page_length);
10210	pd = &sp->design[0];
10211
10212	mtx_lock(&softc->ctl_lock);
10213	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10214		pg = 0;
10215	else
10216		pg = 1;
10217	for (g = 0; g < num_target_port_groups; g++) {
10218		STAILQ_FOREACH(port, &softc->port_list, links) {
10219			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10220				continue;
10221			if (lun != NULL &&
10222			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10223			    CTL_MAX_LUNS)
10224				continue;
10225			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10226			scsi_ulto2b(p, pd->relative_port_id);
10227			if (port->init_devid && g == pg) {
10228				iid_len = port->init_devid->len;
10229				memcpy(pd->initiator_transportid,
10230				    port->init_devid->data, port->init_devid->len);
10231			} else
10232				iid_len = 0;
10233			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10234			pdc = (struct scsi_vpd_port_designation_cont *)
10235			    (&pd->initiator_transportid[iid_len]);
10236			if (port->port_devid && g == pg) {
10237				id_len = port->port_devid->len;
10238				memcpy(pdc->target_port_descriptors,
10239				    port->port_devid->data, port->port_devid->len);
10240			} else
10241				id_len = 0;
10242			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10243			pd = (struct scsi_vpd_port_designation *)
10244			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10245		}
10246	}
10247	mtx_unlock(&softc->ctl_lock);
10248
10249	ctsio->scsi_status = SCSI_STATUS_OK;
10250	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10251	ctsio->be_move_done = ctl_config_move_done;
10252	ctl_datamove((union ctl_io *)ctsio);
10253
10254	return (CTL_RETVAL_COMPLETE);
10255}
10256
10257static int
10258ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10259{
10260	struct scsi_vpd_block_limits *bl_ptr;
10261	struct ctl_lun *lun;
10262	int bs;
10263
10264	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10265
10266	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10267	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10268	ctsio->kern_sg_entries = 0;
10269
10270	if (sizeof(*bl_ptr) < alloc_len) {
10271		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10272		ctsio->kern_data_len = sizeof(*bl_ptr);
10273		ctsio->kern_total_len = sizeof(*bl_ptr);
10274	} else {
10275		ctsio->residual = 0;
10276		ctsio->kern_data_len = alloc_len;
10277		ctsio->kern_total_len = alloc_len;
10278	}
10279	ctsio->kern_data_resid = 0;
10280	ctsio->kern_rel_offset = 0;
10281	ctsio->kern_sg_entries = 0;
10282
10283	/*
10284	 * The control device is always connected.  The disk device, on the
10285	 * other hand, may not be online all the time.  Need to change this
10286	 * to figure out whether the disk device is actually online or not.
10287	 */
10288	if (lun != NULL)
10289		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10290				  lun->be_lun->lun_type;
10291	else
10292		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10293
10294	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10295	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
10296	bl_ptr->max_cmp_write_len = 0xff;
10297	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10298	if (lun != NULL) {
10299		bs = lun->be_lun->blocksize;
10300		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10301		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10302			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10303			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10304			if (lun->be_lun->pblockexp != 0) {
10305				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10306				    bl_ptr->opt_unmap_grain);
10307				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10308				    bl_ptr->unmap_grain_align);
10309			}
10310		}
10311	}
10312	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10313
10314	ctsio->scsi_status = SCSI_STATUS_OK;
10315	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10316	ctsio->be_move_done = ctl_config_move_done;
10317	ctl_datamove((union ctl_io *)ctsio);
10318
10319	return (CTL_RETVAL_COMPLETE);
10320}
10321
10322static int
10323ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10324{
10325	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10326	struct ctl_lun *lun;
10327
10328	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10329
10330	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10331	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10332	ctsio->kern_sg_entries = 0;
10333
10334	if (sizeof(*bdc_ptr) < alloc_len) {
10335		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10336		ctsio->kern_data_len = sizeof(*bdc_ptr);
10337		ctsio->kern_total_len = sizeof(*bdc_ptr);
10338	} else {
10339		ctsio->residual = 0;
10340		ctsio->kern_data_len = alloc_len;
10341		ctsio->kern_total_len = alloc_len;
10342	}
10343	ctsio->kern_data_resid = 0;
10344	ctsio->kern_rel_offset = 0;
10345	ctsio->kern_sg_entries = 0;
10346
10347	/*
10348	 * The control device is always connected.  The disk device, on the
10349	 * other hand, may not be online all the time.  Need to change this
10350	 * to figure out whether the disk device is actually online or not.
10351	 */
10352	if (lun != NULL)
10353		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10354				  lun->be_lun->lun_type;
10355	else
10356		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10357	bdc_ptr->page_code = SVPD_BDC;
10358	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10359	scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10360	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10361
10362	ctsio->scsi_status = SCSI_STATUS_OK;
10363	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10364	ctsio->be_move_done = ctl_config_move_done;
10365	ctl_datamove((union ctl_io *)ctsio);
10366
10367	return (CTL_RETVAL_COMPLETE);
10368}
10369
10370static int
10371ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10372{
10373	struct scsi_vpd_logical_block_prov *lbp_ptr;
10374	struct ctl_lun *lun;
10375
10376	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10377
10378	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10379	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10380	ctsio->kern_sg_entries = 0;
10381
10382	if (sizeof(*lbp_ptr) < alloc_len) {
10383		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10384		ctsio->kern_data_len = sizeof(*lbp_ptr);
10385		ctsio->kern_total_len = sizeof(*lbp_ptr);
10386	} else {
10387		ctsio->residual = 0;
10388		ctsio->kern_data_len = alloc_len;
10389		ctsio->kern_total_len = alloc_len;
10390	}
10391	ctsio->kern_data_resid = 0;
10392	ctsio->kern_rel_offset = 0;
10393	ctsio->kern_sg_entries = 0;
10394
10395	/*
10396	 * The control device is always connected.  The disk device, on the
10397	 * other hand, may not be online all the time.  Need to change this
10398	 * to figure out whether the disk device is actually online or not.
10399	 */
10400	if (lun != NULL)
10401		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10402				  lun->be_lun->lun_type;
10403	else
10404		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10405
10406	lbp_ptr->page_code = SVPD_LBP;
10407	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10408	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10409		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10410		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10411		lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10412	}
10413
10414	ctsio->scsi_status = SCSI_STATUS_OK;
10415	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10416	ctsio->be_move_done = ctl_config_move_done;
10417	ctl_datamove((union ctl_io *)ctsio);
10418
10419	return (CTL_RETVAL_COMPLETE);
10420}
10421
10422static int
10423ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10424{
10425	struct scsi_inquiry *cdb;
10426	struct ctl_lun *lun;
10427	int alloc_len, retval;
10428
10429	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10430	cdb = (struct scsi_inquiry *)ctsio->cdb;
10431
10432	retval = CTL_RETVAL_COMPLETE;
10433
10434	alloc_len = scsi_2btoul(cdb->length);
10435
10436	switch (cdb->page_code) {
10437	case SVPD_SUPPORTED_PAGES:
10438		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10439		break;
10440	case SVPD_UNIT_SERIAL_NUMBER:
10441		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10442		break;
10443	case SVPD_DEVICE_ID:
10444		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10445		break;
10446	case SVPD_EXTENDED_INQUIRY_DATA:
10447		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10448		break;
10449	case SVPD_MODE_PAGE_POLICY:
10450		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10451		break;
10452	case SVPD_SCSI_PORTS:
10453		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10454		break;
10455	case SVPD_SCSI_TPC:
10456		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10457		break;
10458	case SVPD_BLOCK_LIMITS:
10459		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10460		break;
10461	case SVPD_BDC:
10462		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10463		break;
10464	case SVPD_LBP:
10465		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10466		break;
10467	default:
10468		ctl_set_invalid_field(ctsio,
10469				      /*sks_valid*/ 1,
10470				      /*command*/ 1,
10471				      /*field*/ 2,
10472				      /*bit_valid*/ 0,
10473				      /*bit*/ 0);
10474		ctl_done((union ctl_io *)ctsio);
10475		retval = CTL_RETVAL_COMPLETE;
10476		break;
10477	}
10478
10479	return (retval);
10480}
10481
10482static int
10483ctl_inquiry_std(struct ctl_scsiio *ctsio)
10484{
10485	struct scsi_inquiry_data *inq_ptr;
10486	struct scsi_inquiry *cdb;
10487	struct ctl_softc *ctl_softc;
10488	struct ctl_lun *lun;
10489	char *val;
10490	uint32_t alloc_len;
10491	ctl_port_type port_type;
10492
10493	ctl_softc = control_softc;
10494
10495	/*
10496	 * Figure out whether we're talking to a Fibre Channel port or not.
10497	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10498	 * SCSI front ends.
10499	 */
10500	port_type = ctl_softc->ctl_ports[
10501	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10502	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10503		port_type = CTL_PORT_SCSI;
10504
10505	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10506	cdb = (struct scsi_inquiry *)ctsio->cdb;
10507	alloc_len = scsi_2btoul(cdb->length);
10508
10509	/*
10510	 * We malloc the full inquiry data size here and fill it
10511	 * in.  If the user only asks for less, we'll give him
10512	 * that much.
10513	 */
10514	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
10515	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10516	ctsio->kern_sg_entries = 0;
10517	ctsio->kern_data_resid = 0;
10518	ctsio->kern_rel_offset = 0;
10519
10520	if (sizeof(*inq_ptr) < alloc_len) {
10521		ctsio->residual = alloc_len - sizeof(*inq_ptr);
10522		ctsio->kern_data_len = sizeof(*inq_ptr);
10523		ctsio->kern_total_len = sizeof(*inq_ptr);
10524	} else {
10525		ctsio->residual = 0;
10526		ctsio->kern_data_len = alloc_len;
10527		ctsio->kern_total_len = alloc_len;
10528	}
10529
10530	/*
10531	 * If we have a LUN configured, report it as connected.  Otherwise,
10532	 * report that it is offline or no device is supported, depending
10533	 * on the value of inquiry_pq_no_lun.
10534	 *
10535	 * According to the spec (SPC-4 r34), the peripheral qualifier
10536	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10537	 *
10538	 * "A peripheral device having the specified peripheral device type
10539	 * is not connected to this logical unit. However, the device
10540	 * server is capable of supporting the specified peripheral device
10541	 * type on this logical unit."
10542	 *
10543	 * According to the same spec, the peripheral qualifier
10544	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10545	 *
10546	 * "The device server is not capable of supporting a peripheral
10547	 * device on this logical unit. For this peripheral qualifier the
10548	 * peripheral device type shall be set to 1Fh. All other peripheral
10549	 * device type values are reserved for this peripheral qualifier."
10550	 *
10551	 * Given the text, it would seem that we probably want to report that
10552	 * the LUN is offline here.  There is no LUN connected, but we can
10553	 * support a LUN at the given LUN number.
10554	 *
10555	 * In the real world, though, it sounds like things are a little
10556	 * different:
10557	 *
10558	 * - Linux, when presented with a LUN with the offline peripheral
10559	 *   qualifier, will create an sg driver instance for it.  So when
10560	 *   you attach it to CTL, you wind up with a ton of sg driver
10561	 *   instances.  (One for every LUN that Linux bothered to probe.)
10562	 *   Linux does this despite the fact that it issues a REPORT LUNs
10563	 *   to LUN 0 to get the inventory of supported LUNs.
10564	 *
10565	 * - There is other anecdotal evidence (from Emulex folks) about
10566	 *   arrays that use the offline peripheral qualifier for LUNs that
10567	 *   are on the "passive" path in an active/passive array.
10568	 *
10569	 * So the solution is provide a hopefully reasonable default
10570	 * (return bad/no LUN) and allow the user to change the behavior
10571	 * with a tunable/sysctl variable.
10572	 */
10573	if (lun != NULL)
10574		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10575				  lun->be_lun->lun_type;
10576	else if (ctl_softc->inquiry_pq_no_lun == 0)
10577		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10578	else
10579		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10580
10581	/* RMB in byte 2 is 0 */
10582	inq_ptr->version = SCSI_REV_SPC4;
10583
10584	/*
10585	 * According to SAM-3, even if a device only supports a single
10586	 * level of LUN addressing, it should still set the HISUP bit:
10587	 *
10588	 * 4.9.1 Logical unit numbers overview
10589	 *
10590	 * All logical unit number formats described in this standard are
10591	 * hierarchical in structure even when only a single level in that
10592	 * hierarchy is used. The HISUP bit shall be set to one in the
10593	 * standard INQUIRY data (see SPC-2) when any logical unit number
10594	 * format described in this standard is used.  Non-hierarchical
10595	 * formats are outside the scope of this standard.
10596	 *
10597	 * Therefore we set the HiSup bit here.
10598	 *
10599	 * The reponse format is 2, per SPC-3.
10600	 */
10601	inq_ptr->response_format = SID_HiSup | 2;
10602
10603	inq_ptr->additional_length =
10604	    offsetof(struct scsi_inquiry_data, vendor_specific1) -
10605	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10606	CTL_DEBUG_PRINT(("additional_length = %d\n",
10607			 inq_ptr->additional_length));
10608
10609	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10610	/* 16 bit addressing */
10611	if (port_type == CTL_PORT_SCSI)
10612		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10613	/* XXX set the SID_MultiP bit here if we're actually going to
10614	   respond on multiple ports */
10615	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10616
10617	/* 16 bit data bus, synchronous transfers */
10618	if (port_type == CTL_PORT_SCSI)
10619		inq_ptr->flags = SID_WBus16 | SID_Sync;
10620	/*
10621	 * XXX KDM do we want to support tagged queueing on the control
10622	 * device at all?
10623	 */
10624	if ((lun == NULL)
10625	 || (lun->be_lun->lun_type != T_PROCESSOR))
10626		inq_ptr->flags |= SID_CmdQue;
10627	/*
10628	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10629	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10630	 * name and 4 bytes for the revision.
10631	 */
10632	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10633	    "vendor")) == NULL) {
10634		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10635	} else {
10636		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10637		strncpy(inq_ptr->vendor, val,
10638		    min(sizeof(inq_ptr->vendor), strlen(val)));
10639	}
10640	if (lun == NULL) {
10641		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10642		    sizeof(inq_ptr->product));
10643	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10644		switch (lun->be_lun->lun_type) {
10645		case T_DIRECT:
10646			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10647			    sizeof(inq_ptr->product));
10648			break;
10649		case T_PROCESSOR:
10650			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10651			    sizeof(inq_ptr->product));
10652			break;
10653		default:
10654			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10655			    sizeof(inq_ptr->product));
10656			break;
10657		}
10658	} else {
10659		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10660		strncpy(inq_ptr->product, val,
10661		    min(sizeof(inq_ptr->product), strlen(val)));
10662	}
10663
10664	/*
10665	 * XXX make this a macro somewhere so it automatically gets
10666	 * incremented when we make changes.
10667	 */
10668	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10669	    "revision")) == NULL) {
10670		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10671	} else {
10672		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10673		strncpy(inq_ptr->revision, val,
10674		    min(sizeof(inq_ptr->revision), strlen(val)));
10675	}
10676
10677	/*
10678	 * For parallel SCSI, we support double transition and single
10679	 * transition clocking.  We also support QAS (Quick Arbitration
10680	 * and Selection) and Information Unit transfers on both the
10681	 * control and array devices.
10682	 */
10683	if (port_type == CTL_PORT_SCSI)
10684		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10685				    SID_SPI_IUS;
10686
10687	/* SAM-5 (no version claimed) */
10688	scsi_ulto2b(0x00A0, inq_ptr->version1);
10689	/* SPC-4 (no version claimed) */
10690	scsi_ulto2b(0x0460, inq_ptr->version2);
10691	if (port_type == CTL_PORT_FC) {
10692		/* FCP-2 ANSI INCITS.350:2003 */
10693		scsi_ulto2b(0x0917, inq_ptr->version3);
10694	} else if (port_type == CTL_PORT_SCSI) {
10695		/* SPI-4 ANSI INCITS.362:200x */
10696		scsi_ulto2b(0x0B56, inq_ptr->version3);
10697	} else if (port_type == CTL_PORT_ISCSI) {
10698		/* iSCSI (no version claimed) */
10699		scsi_ulto2b(0x0960, inq_ptr->version3);
10700	} else if (port_type == CTL_PORT_SAS) {
10701		/* SAS (no version claimed) */
10702		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10703	}
10704
10705	if (lun == NULL) {
10706		/* SBC-3 (no version claimed) */
10707		scsi_ulto2b(0x04C0, inq_ptr->version4);
10708	} else {
10709		switch (lun->be_lun->lun_type) {
10710		case T_DIRECT:
10711			/* SBC-3 (no version claimed) */
10712			scsi_ulto2b(0x04C0, inq_ptr->version4);
10713			break;
10714		case T_PROCESSOR:
10715		default:
10716			break;
10717		}
10718	}
10719
10720	ctsio->scsi_status = SCSI_STATUS_OK;
10721	if (ctsio->kern_data_len > 0) {
10722		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10723		ctsio->be_move_done = ctl_config_move_done;
10724		ctl_datamove((union ctl_io *)ctsio);
10725	} else {
10726		ctsio->io_hdr.status = CTL_SUCCESS;
10727		ctl_done((union ctl_io *)ctsio);
10728	}
10729
10730	return (CTL_RETVAL_COMPLETE);
10731}
10732
10733int
10734ctl_inquiry(struct ctl_scsiio *ctsio)
10735{
10736	struct scsi_inquiry *cdb;
10737	int retval;
10738
10739	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10740
10741	cdb = (struct scsi_inquiry *)ctsio->cdb;
10742	if (cdb->byte2 & SI_EVPD)
10743		retval = ctl_inquiry_evpd(ctsio);
10744	else if (cdb->page_code == 0)
10745		retval = ctl_inquiry_std(ctsio);
10746	else {
10747		ctl_set_invalid_field(ctsio,
10748				      /*sks_valid*/ 1,
10749				      /*command*/ 1,
10750				      /*field*/ 2,
10751				      /*bit_valid*/ 0,
10752				      /*bit*/ 0);
10753		ctl_done((union ctl_io *)ctsio);
10754		return (CTL_RETVAL_COMPLETE);
10755	}
10756
10757	return (retval);
10758}
10759
10760/*
10761 * For known CDB types, parse the LBA and length.
10762 */
10763static int
10764ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10765{
10766	if (io->io_hdr.io_type != CTL_IO_SCSI)
10767		return (1);
10768
10769	switch (io->scsiio.cdb[0]) {
10770	case COMPARE_AND_WRITE: {
10771		struct scsi_compare_and_write *cdb;
10772
10773		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10774
10775		*lba = scsi_8btou64(cdb->addr);
10776		*len = cdb->length;
10777		break;
10778	}
10779	case READ_6:
10780	case WRITE_6: {
10781		struct scsi_rw_6 *cdb;
10782
10783		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10784
10785		*lba = scsi_3btoul(cdb->addr);
10786		/* only 5 bits are valid in the most significant address byte */
10787		*lba &= 0x1fffff;
10788		*len = cdb->length;
10789		break;
10790	}
10791	case READ_10:
10792	case WRITE_10: {
10793		struct scsi_rw_10 *cdb;
10794
10795		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10796
10797		*lba = scsi_4btoul(cdb->addr);
10798		*len = scsi_2btoul(cdb->length);
10799		break;
10800	}
10801	case WRITE_VERIFY_10: {
10802		struct scsi_write_verify_10 *cdb;
10803
10804		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10805
10806		*lba = scsi_4btoul(cdb->addr);
10807		*len = scsi_2btoul(cdb->length);
10808		break;
10809	}
10810	case READ_12:
10811	case WRITE_12: {
10812		struct scsi_rw_12 *cdb;
10813
10814		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10815
10816		*lba = scsi_4btoul(cdb->addr);
10817		*len = scsi_4btoul(cdb->length);
10818		break;
10819	}
10820	case WRITE_VERIFY_12: {
10821		struct scsi_write_verify_12 *cdb;
10822
10823		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10824
10825		*lba = scsi_4btoul(cdb->addr);
10826		*len = scsi_4btoul(cdb->length);
10827		break;
10828	}
10829	case READ_16:
10830	case WRITE_16: {
10831		struct scsi_rw_16 *cdb;
10832
10833		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10834
10835		*lba = scsi_8btou64(cdb->addr);
10836		*len = scsi_4btoul(cdb->length);
10837		break;
10838	}
10839	case WRITE_VERIFY_16: {
10840		struct scsi_write_verify_16 *cdb;
10841
10842		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10843
10844
10845		*lba = scsi_8btou64(cdb->addr);
10846		*len = scsi_4btoul(cdb->length);
10847		break;
10848	}
10849	case WRITE_SAME_10: {
10850		struct scsi_write_same_10 *cdb;
10851
10852		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10853
10854		*lba = scsi_4btoul(cdb->addr);
10855		*len = scsi_2btoul(cdb->length);
10856		break;
10857	}
10858	case WRITE_SAME_16: {
10859		struct scsi_write_same_16 *cdb;
10860
10861		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10862
10863		*lba = scsi_8btou64(cdb->addr);
10864		*len = scsi_4btoul(cdb->length);
10865		break;
10866	}
10867	case VERIFY_10: {
10868		struct scsi_verify_10 *cdb;
10869
10870		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10871
10872		*lba = scsi_4btoul(cdb->addr);
10873		*len = scsi_2btoul(cdb->length);
10874		break;
10875	}
10876	case VERIFY_12: {
10877		struct scsi_verify_12 *cdb;
10878
10879		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10880
10881		*lba = scsi_4btoul(cdb->addr);
10882		*len = scsi_4btoul(cdb->length);
10883		break;
10884	}
10885	case VERIFY_16: {
10886		struct scsi_verify_16 *cdb;
10887
10888		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10889
10890		*lba = scsi_8btou64(cdb->addr);
10891		*len = scsi_4btoul(cdb->length);
10892		break;
10893	}
10894	case UNMAP: {
10895		*lba = 0;
10896		*len = UINT64_MAX;
10897		break;
10898	}
10899	default:
10900		return (1);
10901		break; /* NOTREACHED */
10902	}
10903
10904	return (0);
10905}
10906
10907static ctl_action
10908ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10909{
10910	uint64_t endlba1, endlba2;
10911
10912	endlba1 = lba1 + len1 - 1;
10913	endlba2 = lba2 + len2 - 1;
10914
10915	if ((endlba1 < lba2)
10916	 || (endlba2 < lba1))
10917		return (CTL_ACTION_PASS);
10918	else
10919		return (CTL_ACTION_BLOCK);
10920}
10921
10922static int
10923ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10924{
10925	struct ctl_ptr_len_flags *ptrlen;
10926	struct scsi_unmap_desc *buf, *end, *range;
10927	uint64_t lba;
10928	uint32_t len;
10929
10930	/* If not UNMAP -- go other way. */
10931	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10932	    io->scsiio.cdb[0] != UNMAP)
10933		return (CTL_ACTION_ERROR);
10934
10935	/* If UNMAP without data -- block and wait for data. */
10936	ptrlen = (struct ctl_ptr_len_flags *)
10937	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10938	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10939	    ptrlen->ptr == NULL)
10940		return (CTL_ACTION_BLOCK);
10941
10942	/* UNMAP with data -- check for collision. */
10943	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10944	end = buf + ptrlen->len / sizeof(*buf);
10945	for (range = buf; range < end; range++) {
10946		lba = scsi_8btou64(range->lba);
10947		len = scsi_4btoul(range->length);
10948		if ((lba < lba2 + len2) && (lba + len > lba2))
10949			return (CTL_ACTION_BLOCK);
10950	}
10951	return (CTL_ACTION_PASS);
10952}
10953
10954static ctl_action
10955ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10956{
10957	uint64_t lba1, lba2;
10958	uint64_t len1, len2;
10959	int retval;
10960
10961	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10962		return (CTL_ACTION_ERROR);
10963
10964	retval = ctl_extent_check_unmap(io2, lba1, len1);
10965	if (retval != CTL_ACTION_ERROR)
10966		return (retval);
10967
10968	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10969		return (CTL_ACTION_ERROR);
10970
10971	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10972}
10973
10974static ctl_action
10975ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10976    union ctl_io *ooa_io)
10977{
10978	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10979	ctl_serialize_action *serialize_row;
10980
10981	/*
10982	 * The initiator attempted multiple untagged commands at the same
10983	 * time.  Can't do that.
10984	 */
10985	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10986	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10987	 && ((pending_io->io_hdr.nexus.targ_port ==
10988	      ooa_io->io_hdr.nexus.targ_port)
10989	  && (pending_io->io_hdr.nexus.initid.id ==
10990	      ooa_io->io_hdr.nexus.initid.id))
10991	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10992		return (CTL_ACTION_OVERLAP);
10993
10994	/*
10995	 * The initiator attempted to send multiple tagged commands with
10996	 * the same ID.  (It's fine if different initiators have the same
10997	 * tag ID.)
10998	 *
10999	 * Even if all of those conditions are true, we don't kill the I/O
11000	 * if the command ahead of us has been aborted.  We won't end up
11001	 * sending it to the FETD, and it's perfectly legal to resend a
11002	 * command with the same tag number as long as the previous
11003	 * instance of this tag number has been aborted somehow.
11004	 */
11005	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11006	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11007	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11008	 && ((pending_io->io_hdr.nexus.targ_port ==
11009	      ooa_io->io_hdr.nexus.targ_port)
11010	  && (pending_io->io_hdr.nexus.initid.id ==
11011	      ooa_io->io_hdr.nexus.initid.id))
11012	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
11013		return (CTL_ACTION_OVERLAP_TAG);
11014
11015	/*
11016	 * If we get a head of queue tag, SAM-3 says that we should
11017	 * immediately execute it.
11018	 *
11019	 * What happens if this command would normally block for some other
11020	 * reason?  e.g. a request sense with a head of queue tag
11021	 * immediately after a write.  Normally that would block, but this
11022	 * will result in its getting executed immediately...
11023	 *
11024	 * We currently return "pass" instead of "skip", so we'll end up
11025	 * going through the rest of the queue to check for overlapped tags.
11026	 *
11027	 * XXX KDM check for other types of blockage first??
11028	 */
11029	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11030		return (CTL_ACTION_PASS);
11031
11032	/*
11033	 * Ordered tags have to block until all items ahead of them
11034	 * have completed.  If we get called with an ordered tag, we always
11035	 * block, if something else is ahead of us in the queue.
11036	 */
11037	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11038		return (CTL_ACTION_BLOCK);
11039
11040	/*
11041	 * Simple tags get blocked until all head of queue and ordered tags
11042	 * ahead of them have completed.  I'm lumping untagged commands in
11043	 * with simple tags here.  XXX KDM is that the right thing to do?
11044	 */
11045	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11046	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11047	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11048	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11049		return (CTL_ACTION_BLOCK);
11050
11051	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11052	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11053
11054	serialize_row = ctl_serialize_table[ooa_entry->seridx];
11055
11056	switch (serialize_row[pending_entry->seridx]) {
11057	case CTL_SER_BLOCK:
11058		return (CTL_ACTION_BLOCK);
11059	case CTL_SER_EXTENT:
11060		return (ctl_extent_check(pending_io, ooa_io));
11061	case CTL_SER_EXTENTOPT:
11062		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11063		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11064			return (ctl_extent_check(pending_io, ooa_io));
11065		/* FALLTHROUGH */
11066	case CTL_SER_PASS:
11067		return (CTL_ACTION_PASS);
11068	case CTL_SER_BLOCKOPT:
11069		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11070		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11071			return (CTL_ACTION_BLOCK);
11072		return (CTL_ACTION_PASS);
11073	case CTL_SER_SKIP:
11074		return (CTL_ACTION_SKIP);
11075	default:
11076		panic("invalid serialization value %d",
11077		      serialize_row[pending_entry->seridx]);
11078	}
11079
11080	return (CTL_ACTION_ERROR);
11081}
11082
11083/*
11084 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11085 * Assumptions:
11086 * - pending_io is generally either incoming, or on the blocked queue
11087 * - starting I/O is the I/O we want to start the check with.
11088 */
11089static ctl_action
11090ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11091	      union ctl_io *starting_io)
11092{
11093	union ctl_io *ooa_io;
11094	ctl_action action;
11095
11096	mtx_assert(&lun->lun_lock, MA_OWNED);
11097
11098	/*
11099	 * Run back along the OOA queue, starting with the current
11100	 * blocked I/O and going through every I/O before it on the
11101	 * queue.  If starting_io is NULL, we'll just end up returning
11102	 * CTL_ACTION_PASS.
11103	 */
11104	for (ooa_io = starting_io; ooa_io != NULL;
11105	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11106	     ooa_links)){
11107
11108		/*
11109		 * This routine just checks to see whether
11110		 * cur_blocked is blocked by ooa_io, which is ahead
11111		 * of it in the queue.  It doesn't queue/dequeue
11112		 * cur_blocked.
11113		 */
11114		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11115		switch (action) {
11116		case CTL_ACTION_BLOCK:
11117		case CTL_ACTION_OVERLAP:
11118		case CTL_ACTION_OVERLAP_TAG:
11119		case CTL_ACTION_SKIP:
11120		case CTL_ACTION_ERROR:
11121			return (action);
11122			break; /* NOTREACHED */
11123		case CTL_ACTION_PASS:
11124			break;
11125		default:
11126			panic("invalid action %d", action);
11127			break;  /* NOTREACHED */
11128		}
11129	}
11130
11131	return (CTL_ACTION_PASS);
11132}
11133
11134/*
11135 * Assumptions:
11136 * - An I/O has just completed, and has been removed from the per-LUN OOA
11137 *   queue, so some items on the blocked queue may now be unblocked.
11138 */
11139static int
11140ctl_check_blocked(struct ctl_lun *lun)
11141{
11142	union ctl_io *cur_blocked, *next_blocked;
11143
11144	mtx_assert(&lun->lun_lock, MA_OWNED);
11145
11146	/*
11147	 * Run forward from the head of the blocked queue, checking each
11148	 * entry against the I/Os prior to it on the OOA queue to see if
11149	 * there is still any blockage.
11150	 *
11151	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11152	 * with our removing a variable on it while it is traversing the
11153	 * list.
11154	 */
11155	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11156	     cur_blocked != NULL; cur_blocked = next_blocked) {
11157		union ctl_io *prev_ooa;
11158		ctl_action action;
11159
11160		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11161							  blocked_links);
11162
11163		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11164						      ctl_ooaq, ooa_links);
11165
11166		/*
11167		 * If cur_blocked happens to be the first item in the OOA
11168		 * queue now, prev_ooa will be NULL, and the action
11169		 * returned will just be CTL_ACTION_PASS.
11170		 */
11171		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11172
11173		switch (action) {
11174		case CTL_ACTION_BLOCK:
11175			/* Nothing to do here, still blocked */
11176			break;
11177		case CTL_ACTION_OVERLAP:
11178		case CTL_ACTION_OVERLAP_TAG:
11179			/*
11180			 * This shouldn't happen!  In theory we've already
11181			 * checked this command for overlap...
11182			 */
11183			break;
11184		case CTL_ACTION_PASS:
11185		case CTL_ACTION_SKIP: {
11186			struct ctl_softc *softc;
11187			const struct ctl_cmd_entry *entry;
11188			uint32_t initidx;
11189			int isc_retval;
11190
11191			/*
11192			 * The skip case shouldn't happen, this transaction
11193			 * should have never made it onto the blocked queue.
11194			 */
11195			/*
11196			 * This I/O is no longer blocked, we can remove it
11197			 * from the blocked queue.  Since this is a TAILQ
11198			 * (doubly linked list), we can do O(1) removals
11199			 * from any place on the list.
11200			 */
11201			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11202				     blocked_links);
11203			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11204
11205			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11206				/*
11207				 * Need to send IO back to original side to
11208				 * run
11209				 */
11210				union ctl_ha_msg msg_info;
11211
11212				msg_info.hdr.original_sc =
11213					cur_blocked->io_hdr.original_sc;
11214				msg_info.hdr.serializing_sc = cur_blocked;
11215				msg_info.hdr.msg_type = CTL_MSG_R2R;
11216				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11217				     &msg_info, sizeof(msg_info), 0)) >
11218				     CTL_HA_STATUS_SUCCESS) {
11219					printf("CTL:Check Blocked error from "
11220					       "ctl_ha_msg_send %d\n",
11221					       isc_retval);
11222				}
11223				break;
11224			}
11225			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11226			softc = control_softc;
11227
11228			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11229
11230			/*
11231			 * Check this I/O for LUN state changes that may
11232			 * have happened while this command was blocked.
11233			 * The LUN state may have been changed by a command
11234			 * ahead of us in the queue, so we need to re-check
11235			 * for any states that can be caused by SCSI
11236			 * commands.
11237			 */
11238			if (ctl_scsiio_lun_check(softc, lun, entry,
11239						 &cur_blocked->scsiio) == 0) {
11240				cur_blocked->io_hdr.flags |=
11241				                      CTL_FLAG_IS_WAS_ON_RTR;
11242				ctl_enqueue_rtr(cur_blocked);
11243			} else
11244				ctl_done(cur_blocked);
11245			break;
11246		}
11247		default:
11248			/*
11249			 * This probably shouldn't happen -- we shouldn't
11250			 * get CTL_ACTION_ERROR, or anything else.
11251			 */
11252			break;
11253		}
11254	}
11255
11256	return (CTL_RETVAL_COMPLETE);
11257}
11258
11259/*
11260 * This routine (with one exception) checks LUN flags that can be set by
11261 * commands ahead of us in the OOA queue.  These flags have to be checked
11262 * when a command initially comes in, and when we pull a command off the
11263 * blocked queue and are preparing to execute it.  The reason we have to
11264 * check these flags for commands on the blocked queue is that the LUN
11265 * state may have been changed by a command ahead of us while we're on the
11266 * blocked queue.
11267 *
11268 * Ordering is somewhat important with these checks, so please pay
11269 * careful attention to the placement of any new checks.
11270 */
11271static int
11272ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11273    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11274{
11275	int retval;
11276
11277	retval = 0;
11278
11279	mtx_assert(&lun->lun_lock, MA_OWNED);
11280
11281	/*
11282	 * If this shelf is a secondary shelf controller, we have to reject
11283	 * any media access commands.
11284	 */
11285#if 0
11286	/* No longer needed for HA */
11287	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11288	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11289		ctl_set_lun_standby(ctsio);
11290		retval = 1;
11291		goto bailout;
11292	}
11293#endif
11294
11295	/*
11296	 * Check for a reservation conflict.  If this command isn't allowed
11297	 * even on reserved LUNs, and if this initiator isn't the one who
11298	 * reserved us, reject the command with a reservation conflict.
11299	 */
11300	if ((lun->flags & CTL_LUN_RESERVED)
11301	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11302		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
11303		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
11304		 || (ctsio->io_hdr.nexus.targ_target.id !=
11305		     lun->rsv_nexus.targ_target.id)) {
11306			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11307			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11308			retval = 1;
11309			goto bailout;
11310		}
11311	}
11312
11313	if ( (lun->flags & CTL_LUN_PR_RESERVED)
11314	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11315		uint32_t residx;
11316
11317		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11318		/*
11319		 * if we aren't registered or it's a res holder type
11320		 * reservation and this isn't the res holder then set a
11321		 * conflict.
11322		 * NOTE: Commands which might be allowed on write exclusive
11323		 * type reservations are checked in the particular command
11324		 * for a conflict. Read and SSU are the only ones.
11325		 */
11326		if (!lun->per_res[residx].registered
11327		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11328			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11329			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11330			retval = 1;
11331			goto bailout;
11332		}
11333
11334	}
11335
11336	if ((lun->flags & CTL_LUN_OFFLINE)
11337	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11338		ctl_set_lun_not_ready(ctsio);
11339		retval = 1;
11340		goto bailout;
11341	}
11342
11343	/*
11344	 * If the LUN is stopped, see if this particular command is allowed
11345	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11346	 */
11347	if ((lun->flags & CTL_LUN_STOPPED)
11348	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11349		/* "Logical unit not ready, initializing cmd. required" */
11350		ctl_set_lun_stopped(ctsio);
11351		retval = 1;
11352		goto bailout;
11353	}
11354
11355	if ((lun->flags & CTL_LUN_INOPERABLE)
11356	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11357		/* "Medium format corrupted" */
11358		ctl_set_medium_format_corrupted(ctsio);
11359		retval = 1;
11360		goto bailout;
11361	}
11362
11363bailout:
11364	return (retval);
11365
11366}
11367
11368static void
11369ctl_failover_io(union ctl_io *io, int have_lock)
11370{
11371	ctl_set_busy(&io->scsiio);
11372	ctl_done(io);
11373}
11374
11375static void
11376ctl_failover(void)
11377{
11378	struct ctl_lun *lun;
11379	struct ctl_softc *ctl_softc;
11380	union ctl_io *next_io, *pending_io;
11381	union ctl_io *io;
11382	int lun_idx;
11383	int i;
11384
11385	ctl_softc = control_softc;
11386
11387	mtx_lock(&ctl_softc->ctl_lock);
11388	/*
11389	 * Remove any cmds from the other SC from the rtr queue.  These
11390	 * will obviously only be for LUNs for which we're the primary.
11391	 * We can't send status or get/send data for these commands.
11392	 * Since they haven't been executed yet, we can just remove them.
11393	 * We'll either abort them or delete them below, depending on
11394	 * which HA mode we're in.
11395	 */
11396#ifdef notyet
11397	mtx_lock(&ctl_softc->queue_lock);
11398	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11399	     io != NULL; io = next_io) {
11400		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11401		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11402			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11403				      ctl_io_hdr, links);
11404	}
11405	mtx_unlock(&ctl_softc->queue_lock);
11406#endif
11407
11408	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11409		lun = ctl_softc->ctl_luns[lun_idx];
11410		if (lun==NULL)
11411			continue;
11412
11413		/*
11414		 * Processor LUNs are primary on both sides.
11415		 * XXX will this always be true?
11416		 */
11417		if (lun->be_lun->lun_type == T_PROCESSOR)
11418			continue;
11419
11420		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11421		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11422			printf("FAILOVER: primary lun %d\n", lun_idx);
11423		        /*
11424			 * Remove all commands from the other SC. First from the
11425			 * blocked queue then from the ooa queue. Once we have
11426			 * removed them. Call ctl_check_blocked to see if there
11427			 * is anything that can run.
11428			 */
11429			for (io = (union ctl_io *)TAILQ_FIRST(
11430			     &lun->blocked_queue); io != NULL; io = next_io) {
11431
11432		        	next_io = (union ctl_io *)TAILQ_NEXT(
11433				    &io->io_hdr, blocked_links);
11434
11435				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11436					TAILQ_REMOVE(&lun->blocked_queue,
11437						     &io->io_hdr,blocked_links);
11438					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11439					TAILQ_REMOVE(&lun->ooa_queue,
11440						     &io->io_hdr, ooa_links);
11441
11442					ctl_free_io(io);
11443				}
11444			}
11445
11446			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11447	     		     io != NULL; io = next_io) {
11448
11449		        	next_io = (union ctl_io *)TAILQ_NEXT(
11450				    &io->io_hdr, ooa_links);
11451
11452				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11453
11454					TAILQ_REMOVE(&lun->ooa_queue,
11455						&io->io_hdr,
11456					     	ooa_links);
11457
11458					ctl_free_io(io);
11459				}
11460			}
11461			ctl_check_blocked(lun);
11462		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11463			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11464
11465			printf("FAILOVER: primary lun %d\n", lun_idx);
11466			/*
11467			 * Abort all commands from the other SC.  We can't
11468			 * send status back for them now.  These should get
11469			 * cleaned up when they are completed or come out
11470			 * for a datamove operation.
11471			 */
11472			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11473	     		     io != NULL; io = next_io) {
11474		        	next_io = (union ctl_io *)TAILQ_NEXT(
11475					&io->io_hdr, ooa_links);
11476
11477				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11478					io->io_hdr.flags |= CTL_FLAG_ABORT;
11479			}
11480		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11481			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11482
11483			printf("FAILOVER: secondary lun %d\n", lun_idx);
11484
11485			lun->flags |= CTL_LUN_PRIMARY_SC;
11486
11487			/*
11488			 * We send all I/O that was sent to this controller
11489			 * and redirected to the other side back with
11490			 * busy status, and have the initiator retry it.
11491			 * Figuring out how much data has been transferred,
11492			 * etc. and picking up where we left off would be
11493			 * very tricky.
11494			 *
11495			 * XXX KDM need to remove I/O from the blocked
11496			 * queue as well!
11497			 */
11498			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11499			     &lun->ooa_queue); pending_io != NULL;
11500			     pending_io = next_io) {
11501
11502				next_io =  (union ctl_io *)TAILQ_NEXT(
11503					&pending_io->io_hdr, ooa_links);
11504
11505				pending_io->io_hdr.flags &=
11506					~CTL_FLAG_SENT_2OTHER_SC;
11507
11508				if (pending_io->io_hdr.flags &
11509				    CTL_FLAG_IO_ACTIVE) {
11510					pending_io->io_hdr.flags |=
11511						CTL_FLAG_FAILOVER;
11512				} else {
11513					ctl_set_busy(&pending_io->scsiio);
11514					ctl_done(pending_io);
11515				}
11516			}
11517
11518			/*
11519			 * Build Unit Attention
11520			 */
11521			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11522				lun->pending_ua[i] |=
11523				                     CTL_UA_ASYM_ACC_CHANGE;
11524			}
11525		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11526			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11527			printf("FAILOVER: secondary lun %d\n", lun_idx);
11528			/*
11529			 * if the first io on the OOA is not on the RtR queue
11530			 * add it.
11531			 */
11532			lun->flags |= CTL_LUN_PRIMARY_SC;
11533
11534			pending_io = (union ctl_io *)TAILQ_FIRST(
11535			    &lun->ooa_queue);
11536			if (pending_io==NULL) {
11537				printf("Nothing on OOA queue\n");
11538				continue;
11539			}
11540
11541			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11542			if ((pending_io->io_hdr.flags &
11543			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11544				pending_io->io_hdr.flags |=
11545				    CTL_FLAG_IS_WAS_ON_RTR;
11546				ctl_enqueue_rtr(pending_io);
11547			}
11548#if 0
11549			else
11550			{
11551				printf("Tag 0x%04x is running\n",
11552				      pending_io->scsiio.tag_num);
11553			}
11554#endif
11555
11556			next_io = (union ctl_io *)TAILQ_NEXT(
11557			    &pending_io->io_hdr, ooa_links);
11558			for (pending_io=next_io; pending_io != NULL;
11559			     pending_io = next_io) {
11560				pending_io->io_hdr.flags &=
11561				    ~CTL_FLAG_SENT_2OTHER_SC;
11562				next_io = (union ctl_io *)TAILQ_NEXT(
11563					&pending_io->io_hdr, ooa_links);
11564				if (pending_io->io_hdr.flags &
11565				    CTL_FLAG_IS_WAS_ON_RTR) {
11566#if 0
11567				        printf("Tag 0x%04x is running\n",
11568				      		pending_io->scsiio.tag_num);
11569#endif
11570					continue;
11571				}
11572
11573				switch (ctl_check_ooa(lun, pending_io,
11574			            (union ctl_io *)TAILQ_PREV(
11575				    &pending_io->io_hdr, ctl_ooaq,
11576				    ooa_links))) {
11577
11578				case CTL_ACTION_BLOCK:
11579					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11580							  &pending_io->io_hdr,
11581							  blocked_links);
11582					pending_io->io_hdr.flags |=
11583					    CTL_FLAG_BLOCKED;
11584					break;
11585				case CTL_ACTION_PASS:
11586				case CTL_ACTION_SKIP:
11587					pending_io->io_hdr.flags |=
11588					    CTL_FLAG_IS_WAS_ON_RTR;
11589					ctl_enqueue_rtr(pending_io);
11590					break;
11591				case CTL_ACTION_OVERLAP:
11592					ctl_set_overlapped_cmd(
11593					    (struct ctl_scsiio *)pending_io);
11594					ctl_done(pending_io);
11595					break;
11596				case CTL_ACTION_OVERLAP_TAG:
11597					ctl_set_overlapped_tag(
11598					    (struct ctl_scsiio *)pending_io,
11599					    pending_io->scsiio.tag_num & 0xff);
11600					ctl_done(pending_io);
11601					break;
11602				case CTL_ACTION_ERROR:
11603				default:
11604					ctl_set_internal_failure(
11605						(struct ctl_scsiio *)pending_io,
11606						0,  // sks_valid
11607						0); //retry count
11608					ctl_done(pending_io);
11609					break;
11610				}
11611			}
11612
11613			/*
11614			 * Build Unit Attention
11615			 */
11616			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11617				lun->pending_ua[i] |=
11618				                     CTL_UA_ASYM_ACC_CHANGE;
11619			}
11620		} else {
11621			panic("Unhandled HA mode failover, LUN flags = %#x, "
11622			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11623		}
11624	}
11625	ctl_pause_rtr = 0;
11626	mtx_unlock(&ctl_softc->ctl_lock);
11627}
11628
11629static int
11630ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11631{
11632	struct ctl_lun *lun;
11633	const struct ctl_cmd_entry *entry;
11634	uint32_t initidx, targ_lun;
11635	int retval;
11636
11637	retval = 0;
11638
11639	lun = NULL;
11640
11641	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11642	if ((targ_lun < CTL_MAX_LUNS)
11643	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11644		lun = ctl_softc->ctl_luns[targ_lun];
11645		/*
11646		 * If the LUN is invalid, pretend that it doesn't exist.
11647		 * It will go away as soon as all pending I/O has been
11648		 * completed.
11649		 */
11650		if (lun->flags & CTL_LUN_DISABLED) {
11651			lun = NULL;
11652		} else {
11653			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11654			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11655				lun->be_lun;
11656			if (lun->be_lun->lun_type == T_PROCESSOR) {
11657				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11658			}
11659
11660			/*
11661			 * Every I/O goes into the OOA queue for a
11662			 * particular LUN, and stays there until completion.
11663			 */
11664			mtx_lock(&lun->lun_lock);
11665			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11666			    ooa_links);
11667		}
11668	} else {
11669		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11670		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11671	}
11672
11673	/* Get command entry and return error if it is unsuppotyed. */
11674	entry = ctl_validate_command(ctsio);
11675	if (entry == NULL) {
11676		if (lun)
11677			mtx_unlock(&lun->lun_lock);
11678		return (retval);
11679	}
11680
11681	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11682	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11683
11684	/*
11685	 * Check to see whether we can send this command to LUNs that don't
11686	 * exist.  This should pretty much only be the case for inquiry
11687	 * and request sense.  Further checks, below, really require having
11688	 * a LUN, so we can't really check the command anymore.  Just put
11689	 * it on the rtr queue.
11690	 */
11691	if (lun == NULL) {
11692		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11693			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11694			ctl_enqueue_rtr((union ctl_io *)ctsio);
11695			return (retval);
11696		}
11697
11698		ctl_set_unsupported_lun(ctsio);
11699		ctl_done((union ctl_io *)ctsio);
11700		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11701		return (retval);
11702	} else {
11703		/*
11704		 * Make sure we support this particular command on this LUN.
11705		 * e.g., we don't support writes to the control LUN.
11706		 */
11707		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11708			mtx_unlock(&lun->lun_lock);
11709			ctl_set_invalid_opcode(ctsio);
11710			ctl_done((union ctl_io *)ctsio);
11711			return (retval);
11712		}
11713	}
11714
11715	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11716
11717#ifdef CTL_WITH_CA
11718	/*
11719	 * If we've got a request sense, it'll clear the contingent
11720	 * allegiance condition.  Otherwise, if we have a CA condition for
11721	 * this initiator, clear it, because it sent down a command other
11722	 * than request sense.
11723	 */
11724	if ((ctsio->cdb[0] != REQUEST_SENSE)
11725	 && (ctl_is_set(lun->have_ca, initidx)))
11726		ctl_clear_mask(lun->have_ca, initidx);
11727#endif
11728
11729	/*
11730	 * If the command has this flag set, it handles its own unit
11731	 * attention reporting, we shouldn't do anything.  Otherwise we
11732	 * check for any pending unit attentions, and send them back to the
11733	 * initiator.  We only do this when a command initially comes in,
11734	 * not when we pull it off the blocked queue.
11735	 *
11736	 * According to SAM-3, section 5.3.2, the order that things get
11737	 * presented back to the host is basically unit attentions caused
11738	 * by some sort of reset event, busy status, reservation conflicts
11739	 * or task set full, and finally any other status.
11740	 *
11741	 * One issue here is that some of the unit attentions we report
11742	 * don't fall into the "reset" category (e.g. "reported luns data
11743	 * has changed").  So reporting it here, before the reservation
11744	 * check, may be technically wrong.  I guess the only thing to do
11745	 * would be to check for and report the reset events here, and then
11746	 * check for the other unit attention types after we check for a
11747	 * reservation conflict.
11748	 *
11749	 * XXX KDM need to fix this
11750	 */
11751	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11752		ctl_ua_type ua_type;
11753
11754		if (lun->pending_ua[initidx] != CTL_UA_NONE) {
11755			scsi_sense_data_type sense_format;
11756
11757			if (lun != NULL)
11758				sense_format = (lun->flags &
11759				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11760				    SSD_TYPE_FIXED;
11761			else
11762				sense_format = SSD_TYPE_FIXED;
11763
11764			ua_type = ctl_build_ua(&lun->pending_ua[initidx],
11765			    &ctsio->sense_data, sense_format);
11766			if (ua_type != CTL_UA_NONE) {
11767				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11768				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11769						       CTL_AUTOSENSE;
11770				ctsio->sense_len = SSD_FULL_SIZE;
11771				mtx_unlock(&lun->lun_lock);
11772				ctl_done((union ctl_io *)ctsio);
11773				return (retval);
11774			}
11775		}
11776	}
11777
11778
11779	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11780		mtx_unlock(&lun->lun_lock);
11781		ctl_done((union ctl_io *)ctsio);
11782		return (retval);
11783	}
11784
11785	/*
11786	 * XXX CHD this is where we want to send IO to other side if
11787	 * this LUN is secondary on this SC. We will need to make a copy
11788	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11789	 * the copy we send as FROM_OTHER.
11790	 * We also need to stuff the address of the original IO so we can
11791	 * find it easily. Something similar will need be done on the other
11792	 * side so when we are done we can find the copy.
11793	 */
11794	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11795		union ctl_ha_msg msg_info;
11796		int isc_retval;
11797
11798		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11799
11800		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11801		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11802#if 0
11803		printf("1. ctsio %p\n", ctsio);
11804#endif
11805		msg_info.hdr.serializing_sc = NULL;
11806		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11807		msg_info.scsi.tag_num = ctsio->tag_num;
11808		msg_info.scsi.tag_type = ctsio->tag_type;
11809		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11810
11811		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11812
11813		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11814		    (void *)&msg_info, sizeof(msg_info), 0)) >
11815		    CTL_HA_STATUS_SUCCESS) {
11816			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11817			       isc_retval);
11818			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11819		} else {
11820#if 0
11821			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11822#endif
11823		}
11824
11825		/*
11826		 * XXX KDM this I/O is off the incoming queue, but hasn't
11827		 * been inserted on any other queue.  We may need to come
11828		 * up with a holding queue while we wait for serialization
11829		 * so that we have an idea of what we're waiting for from
11830		 * the other side.
11831		 */
11832		mtx_unlock(&lun->lun_lock);
11833		return (retval);
11834	}
11835
11836	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11837			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11838			      ctl_ooaq, ooa_links))) {
11839	case CTL_ACTION_BLOCK:
11840		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11841		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11842				  blocked_links);
11843		mtx_unlock(&lun->lun_lock);
11844		return (retval);
11845	case CTL_ACTION_PASS:
11846	case CTL_ACTION_SKIP:
11847		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11848		mtx_unlock(&lun->lun_lock);
11849		ctl_enqueue_rtr((union ctl_io *)ctsio);
11850		break;
11851	case CTL_ACTION_OVERLAP:
11852		mtx_unlock(&lun->lun_lock);
11853		ctl_set_overlapped_cmd(ctsio);
11854		ctl_done((union ctl_io *)ctsio);
11855		break;
11856	case CTL_ACTION_OVERLAP_TAG:
11857		mtx_unlock(&lun->lun_lock);
11858		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11859		ctl_done((union ctl_io *)ctsio);
11860		break;
11861	case CTL_ACTION_ERROR:
11862	default:
11863		mtx_unlock(&lun->lun_lock);
11864		ctl_set_internal_failure(ctsio,
11865					 /*sks_valid*/ 0,
11866					 /*retry_count*/ 0);
11867		ctl_done((union ctl_io *)ctsio);
11868		break;
11869	}
11870	return (retval);
11871}
11872
11873const struct ctl_cmd_entry *
11874ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11875{
11876	const struct ctl_cmd_entry *entry;
11877	int service_action;
11878
11879	entry = &ctl_cmd_table[ctsio->cdb[0]];
11880	if (sa)
11881		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11882	if (entry->flags & CTL_CMD_FLAG_SA5) {
11883		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11884		entry = &((const struct ctl_cmd_entry *)
11885		    entry->execute)[service_action];
11886	}
11887	return (entry);
11888}
11889
11890const struct ctl_cmd_entry *
11891ctl_validate_command(struct ctl_scsiio *ctsio)
11892{
11893	const struct ctl_cmd_entry *entry;
11894	int i, sa;
11895	uint8_t diff;
11896
11897	entry = ctl_get_cmd_entry(ctsio, &sa);
11898	if (entry->execute == NULL) {
11899		if (sa)
11900			ctl_set_invalid_field(ctsio,
11901					      /*sks_valid*/ 1,
11902					      /*command*/ 1,
11903					      /*field*/ 1,
11904					      /*bit_valid*/ 1,
11905					      /*bit*/ 4);
11906		else
11907			ctl_set_invalid_opcode(ctsio);
11908		ctl_done((union ctl_io *)ctsio);
11909		return (NULL);
11910	}
11911	KASSERT(entry->length > 0,
11912	    ("Not defined length for command 0x%02x/0x%02x",
11913	     ctsio->cdb[0], ctsio->cdb[1]));
11914	for (i = 1; i < entry->length; i++) {
11915		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11916		if (diff == 0)
11917			continue;
11918		ctl_set_invalid_field(ctsio,
11919				      /*sks_valid*/ 1,
11920				      /*command*/ 1,
11921				      /*field*/ i,
11922				      /*bit_valid*/ 1,
11923				      /*bit*/ fls(diff) - 1);
11924		ctl_done((union ctl_io *)ctsio);
11925		return (NULL);
11926	}
11927	return (entry);
11928}
11929
11930static int
11931ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11932{
11933
11934	switch (lun_type) {
11935	case T_PROCESSOR:
11936		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11937		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11938			return (0);
11939		break;
11940	case T_DIRECT:
11941		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11942		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11943			return (0);
11944		break;
11945	default:
11946		return (0);
11947	}
11948	return (1);
11949}
11950
11951static int
11952ctl_scsiio(struct ctl_scsiio *ctsio)
11953{
11954	int retval;
11955	const struct ctl_cmd_entry *entry;
11956
11957	retval = CTL_RETVAL_COMPLETE;
11958
11959	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11960
11961	entry = ctl_get_cmd_entry(ctsio, NULL);
11962
11963	/*
11964	 * If this I/O has been aborted, just send it straight to
11965	 * ctl_done() without executing it.
11966	 */
11967	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11968		ctl_done((union ctl_io *)ctsio);
11969		goto bailout;
11970	}
11971
11972	/*
11973	 * All the checks should have been handled by ctl_scsiio_precheck().
11974	 * We should be clear now to just execute the I/O.
11975	 */
11976	retval = entry->execute(ctsio);
11977
11978bailout:
11979	return (retval);
11980}
11981
11982/*
11983 * Since we only implement one target right now, a bus reset simply resets
11984 * our single target.
11985 */
11986static int
11987ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11988{
11989	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11990}
11991
11992static int
11993ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11994		 ctl_ua_type ua_type)
11995{
11996	struct ctl_lun *lun;
11997	int retval;
11998
11999	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12000		union ctl_ha_msg msg_info;
12001
12002		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
12003		msg_info.hdr.nexus = io->io_hdr.nexus;
12004		if (ua_type==CTL_UA_TARG_RESET)
12005			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
12006		else
12007			msg_info.task.task_action = CTL_TASK_BUS_RESET;
12008		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12009		msg_info.hdr.original_sc = NULL;
12010		msg_info.hdr.serializing_sc = NULL;
12011		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12012		    (void *)&msg_info, sizeof(msg_info), 0)) {
12013		}
12014	}
12015	retval = 0;
12016
12017	mtx_lock(&ctl_softc->ctl_lock);
12018	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
12019		retval += ctl_lun_reset(lun, io, ua_type);
12020	mtx_unlock(&ctl_softc->ctl_lock);
12021
12022	return (retval);
12023}
12024
12025/*
12026 * The LUN should always be set.  The I/O is optional, and is used to
12027 * distinguish between I/Os sent by this initiator, and by other
12028 * initiators.  We set unit attention for initiators other than this one.
12029 * SAM-3 is vague on this point.  It does say that a unit attention should
12030 * be established for other initiators when a LUN is reset (see section
12031 * 5.7.3), but it doesn't specifically say that the unit attention should
12032 * be established for this particular initiator when a LUN is reset.  Here
12033 * is the relevant text, from SAM-3 rev 8:
12034 *
12035 * 5.7.2 When a SCSI initiator port aborts its own tasks
12036 *
12037 * When a SCSI initiator port causes its own task(s) to be aborted, no
12038 * notification that the task(s) have been aborted shall be returned to
12039 * the SCSI initiator port other than the completion response for the
12040 * command or task management function action that caused the task(s) to
12041 * be aborted and notification(s) associated with related effects of the
12042 * action (e.g., a reset unit attention condition).
12043 *
12044 * XXX KDM for now, we're setting unit attention for all initiators.
12045 */
12046static int
12047ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
12048{
12049	union ctl_io *xio;
12050#if 0
12051	uint32_t initindex;
12052#endif
12053	int i;
12054
12055	mtx_lock(&lun->lun_lock);
12056	/*
12057	 * Run through the OOA queue and abort each I/O.
12058	 */
12059#if 0
12060	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12061#endif
12062	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12063	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12064		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
12065	}
12066
12067	/*
12068	 * This version sets unit attention for every
12069	 */
12070#if 0
12071	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12072	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12073		if (initindex == i)
12074			continue;
12075		lun->pending_ua[i] |= ua_type;
12076	}
12077#endif
12078
12079	/*
12080	 * A reset (any kind, really) clears reservations established with
12081	 * RESERVE/RELEASE.  It does not clear reservations established
12082	 * with PERSISTENT RESERVE OUT, but we don't support that at the
12083	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
12084	 * reservations made with the RESERVE/RELEASE commands, because
12085	 * those commands are obsolete in SPC-3.
12086	 */
12087	lun->flags &= ~CTL_LUN_RESERVED;
12088
12089	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12090#ifdef CTL_WITH_CA
12091		ctl_clear_mask(lun->have_ca, i);
12092#endif
12093		lun->pending_ua[i] |= ua_type;
12094	}
12095	mtx_unlock(&lun->lun_lock);
12096
12097	return (0);
12098}
12099
12100static void
12101ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
12102    int other_sc)
12103{
12104	union ctl_io *xio;
12105
12106	mtx_assert(&lun->lun_lock, MA_OWNED);
12107
12108	/*
12109	 * Run through the OOA queue and attempt to find the given I/O.
12110	 * The target port, initiator ID, tag type and tag number have to
12111	 * match the values that we got from the initiator.  If we have an
12112	 * untagged command to abort, simply abort the first untagged command
12113	 * we come to.  We only allow one untagged command at a time of course.
12114	 */
12115	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12116	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12117
12118		if ((targ_port == UINT32_MAX ||
12119		     targ_port == xio->io_hdr.nexus.targ_port) &&
12120		    (init_id == UINT32_MAX ||
12121		     init_id == xio->io_hdr.nexus.initid.id)) {
12122			if (targ_port != xio->io_hdr.nexus.targ_port ||
12123			    init_id != xio->io_hdr.nexus.initid.id)
12124				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12125			xio->io_hdr.flags |= CTL_FLAG_ABORT;
12126			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12127				union ctl_ha_msg msg_info;
12128
12129				msg_info.hdr.nexus = xio->io_hdr.nexus;
12130				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12131				msg_info.task.tag_num = xio->scsiio.tag_num;
12132				msg_info.task.tag_type = xio->scsiio.tag_type;
12133				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12134				msg_info.hdr.original_sc = NULL;
12135				msg_info.hdr.serializing_sc = NULL;
12136				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12137				    (void *)&msg_info, sizeof(msg_info), 0);
12138			}
12139		}
12140	}
12141}
12142
12143static int
12144ctl_abort_task_set(union ctl_io *io)
12145{
12146	struct ctl_softc *softc = control_softc;
12147	struct ctl_lun *lun;
12148	uint32_t targ_lun;
12149
12150	/*
12151	 * Look up the LUN.
12152	 */
12153	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12154	mtx_lock(&softc->ctl_lock);
12155	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12156		lun = softc->ctl_luns[targ_lun];
12157	else {
12158		mtx_unlock(&softc->ctl_lock);
12159		return (1);
12160	}
12161
12162	mtx_lock(&lun->lun_lock);
12163	mtx_unlock(&softc->ctl_lock);
12164	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12165		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12166		    io->io_hdr.nexus.initid.id,
12167		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12168	} else { /* CTL_TASK_CLEAR_TASK_SET */
12169		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12170		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12171	}
12172	mtx_unlock(&lun->lun_lock);
12173	return (0);
12174}
12175
12176static int
12177ctl_i_t_nexus_reset(union ctl_io *io)
12178{
12179	struct ctl_softc *softc = control_softc;
12180	struct ctl_lun *lun;
12181	uint32_t initindex;
12182
12183	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12184	mtx_lock(&softc->ctl_lock);
12185	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12186		mtx_lock(&lun->lun_lock);
12187		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12188		    io->io_hdr.nexus.initid.id,
12189		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12190#ifdef CTL_WITH_CA
12191		ctl_clear_mask(lun->have_ca, initindex);
12192#endif
12193		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12194		mtx_unlock(&lun->lun_lock);
12195	}
12196	mtx_unlock(&softc->ctl_lock);
12197	return (0);
12198}
12199
12200static int
12201ctl_abort_task(union ctl_io *io)
12202{
12203	union ctl_io *xio;
12204	struct ctl_lun *lun;
12205	struct ctl_softc *ctl_softc;
12206#if 0
12207	struct sbuf sb;
12208	char printbuf[128];
12209#endif
12210	int found;
12211	uint32_t targ_lun;
12212
12213	ctl_softc = control_softc;
12214	found = 0;
12215
12216	/*
12217	 * Look up the LUN.
12218	 */
12219	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12220	mtx_lock(&ctl_softc->ctl_lock);
12221	if ((targ_lun < CTL_MAX_LUNS)
12222	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12223		lun = ctl_softc->ctl_luns[targ_lun];
12224	else {
12225		mtx_unlock(&ctl_softc->ctl_lock);
12226		return (1);
12227	}
12228
12229#if 0
12230	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12231	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12232#endif
12233
12234	mtx_lock(&lun->lun_lock);
12235	mtx_unlock(&ctl_softc->ctl_lock);
12236	/*
12237	 * Run through the OOA queue and attempt to find the given I/O.
12238	 * The target port, initiator ID, tag type and tag number have to
12239	 * match the values that we got from the initiator.  If we have an
12240	 * untagged command to abort, simply abort the first untagged command
12241	 * we come to.  We only allow one untagged command at a time of course.
12242	 */
12243#if 0
12244	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12245#endif
12246	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12247	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12248#if 0
12249		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12250
12251		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12252			    lun->lun, xio->scsiio.tag_num,
12253			    xio->scsiio.tag_type,
12254			    (xio->io_hdr.blocked_links.tqe_prev
12255			    == NULL) ? "" : " BLOCKED",
12256			    (xio->io_hdr.flags &
12257			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12258			    (xio->io_hdr.flags &
12259			    CTL_FLAG_ABORT) ? " ABORT" : "",
12260			    (xio->io_hdr.flags &
12261			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12262		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12263		sbuf_finish(&sb);
12264		printf("%s\n", sbuf_data(&sb));
12265#endif
12266
12267		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12268		 && (xio->io_hdr.nexus.initid.id ==
12269		     io->io_hdr.nexus.initid.id)) {
12270			/*
12271			 * If the abort says that the task is untagged, the
12272			 * task in the queue must be untagged.  Otherwise,
12273			 * we just check to see whether the tag numbers
12274			 * match.  This is because the QLogic firmware
12275			 * doesn't pass back the tag type in an abort
12276			 * request.
12277			 */
12278#if 0
12279			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12280			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12281			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12282#endif
12283			/*
12284			 * XXX KDM we've got problems with FC, because it
12285			 * doesn't send down a tag type with aborts.  So we
12286			 * can only really go by the tag number...
12287			 * This may cause problems with parallel SCSI.
12288			 * Need to figure that out!!
12289			 */
12290			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12291				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12292				found = 1;
12293				if ((io->io_hdr.flags &
12294				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12295				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12296					union ctl_ha_msg msg_info;
12297
12298					io->io_hdr.flags |=
12299					                CTL_FLAG_SENT_2OTHER_SC;
12300					msg_info.hdr.nexus = io->io_hdr.nexus;
12301					msg_info.task.task_action =
12302						CTL_TASK_ABORT_TASK;
12303					msg_info.task.tag_num =
12304						io->taskio.tag_num;
12305					msg_info.task.tag_type =
12306						io->taskio.tag_type;
12307					msg_info.hdr.msg_type =
12308						CTL_MSG_MANAGE_TASKS;
12309					msg_info.hdr.original_sc = NULL;
12310					msg_info.hdr.serializing_sc = NULL;
12311#if 0
12312					printf("Sent Abort to other side\n");
12313#endif
12314					if (CTL_HA_STATUS_SUCCESS !=
12315					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12316		    				(void *)&msg_info,
12317						sizeof(msg_info), 0)) {
12318					}
12319				}
12320#if 0
12321				printf("ctl_abort_task: found I/O to abort\n");
12322#endif
12323				break;
12324			}
12325		}
12326	}
12327	mtx_unlock(&lun->lun_lock);
12328
12329	if (found == 0) {
12330		/*
12331		 * This isn't really an error.  It's entirely possible for
12332		 * the abort and command completion to cross on the wire.
12333		 * This is more of an informative/diagnostic error.
12334		 */
12335#if 0
12336		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12337		       "%d:%d:%d:%d tag %d type %d\n",
12338		       io->io_hdr.nexus.initid.id,
12339		       io->io_hdr.nexus.targ_port,
12340		       io->io_hdr.nexus.targ_target.id,
12341		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12342		       io->taskio.tag_type);
12343#endif
12344	}
12345	return (0);
12346}
12347
12348static void
12349ctl_run_task(union ctl_io *io)
12350{
12351	struct ctl_softc *ctl_softc = control_softc;
12352	int retval = 1;
12353	const char *task_desc;
12354
12355	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12356
12357	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12358	    ("ctl_run_task: Unextected io_type %d\n",
12359	     io->io_hdr.io_type));
12360
12361	task_desc = ctl_scsi_task_string(&io->taskio);
12362	if (task_desc != NULL) {
12363#ifdef NEEDTOPORT
12364		csevent_log(CSC_CTL | CSC_SHELF_SW |
12365			    CTL_TASK_REPORT,
12366			    csevent_LogType_Trace,
12367			    csevent_Severity_Information,
12368			    csevent_AlertLevel_Green,
12369			    csevent_FRU_Firmware,
12370			    csevent_FRU_Unknown,
12371			    "CTL: received task: %s",task_desc);
12372#endif
12373	} else {
12374#ifdef NEEDTOPORT
12375		csevent_log(CSC_CTL | CSC_SHELF_SW |
12376			    CTL_TASK_REPORT,
12377			    csevent_LogType_Trace,
12378			    csevent_Severity_Information,
12379			    csevent_AlertLevel_Green,
12380			    csevent_FRU_Firmware,
12381			    csevent_FRU_Unknown,
12382			    "CTL: received unknown task "
12383			    "type: %d (%#x)",
12384			    io->taskio.task_action,
12385			    io->taskio.task_action);
12386#endif
12387	}
12388	switch (io->taskio.task_action) {
12389	case CTL_TASK_ABORT_TASK:
12390		retval = ctl_abort_task(io);
12391		break;
12392	case CTL_TASK_ABORT_TASK_SET:
12393	case CTL_TASK_CLEAR_TASK_SET:
12394		retval = ctl_abort_task_set(io);
12395		break;
12396	case CTL_TASK_CLEAR_ACA:
12397		break;
12398	case CTL_TASK_I_T_NEXUS_RESET:
12399		retval = ctl_i_t_nexus_reset(io);
12400		break;
12401	case CTL_TASK_LUN_RESET: {
12402		struct ctl_lun *lun;
12403		uint32_t targ_lun;
12404
12405		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12406		mtx_lock(&ctl_softc->ctl_lock);
12407		if ((targ_lun < CTL_MAX_LUNS)
12408		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12409			lun = ctl_softc->ctl_luns[targ_lun];
12410		else {
12411			mtx_unlock(&ctl_softc->ctl_lock);
12412			retval = 1;
12413			break;
12414		}
12415
12416		if (!(io->io_hdr.flags &
12417		    CTL_FLAG_FROM_OTHER_SC)) {
12418			union ctl_ha_msg msg_info;
12419
12420			io->io_hdr.flags |=
12421				CTL_FLAG_SENT_2OTHER_SC;
12422			msg_info.hdr.msg_type =
12423				CTL_MSG_MANAGE_TASKS;
12424			msg_info.hdr.nexus = io->io_hdr.nexus;
12425			msg_info.task.task_action =
12426				CTL_TASK_LUN_RESET;
12427			msg_info.hdr.original_sc = NULL;
12428			msg_info.hdr.serializing_sc = NULL;
12429			if (CTL_HA_STATUS_SUCCESS !=
12430			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12431			    (void *)&msg_info,
12432			    sizeof(msg_info), 0)) {
12433			}
12434		}
12435
12436		retval = ctl_lun_reset(lun, io,
12437				       CTL_UA_LUN_RESET);
12438		mtx_unlock(&ctl_softc->ctl_lock);
12439		break;
12440	}
12441	case CTL_TASK_TARGET_RESET:
12442		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12443		break;
12444	case CTL_TASK_BUS_RESET:
12445		retval = ctl_bus_reset(ctl_softc, io);
12446		break;
12447	case CTL_TASK_PORT_LOGIN:
12448		break;
12449	case CTL_TASK_PORT_LOGOUT:
12450		break;
12451	default:
12452		printf("ctl_run_task: got unknown task management event %d\n",
12453		       io->taskio.task_action);
12454		break;
12455	}
12456	if (retval == 0)
12457		io->io_hdr.status = CTL_SUCCESS;
12458	else
12459		io->io_hdr.status = CTL_ERROR;
12460	ctl_done(io);
12461}
12462
12463/*
12464 * For HA operation.  Handle commands that come in from the other
12465 * controller.
12466 */
12467static void
12468ctl_handle_isc(union ctl_io *io)
12469{
12470	int free_io;
12471	struct ctl_lun *lun;
12472	struct ctl_softc *ctl_softc;
12473	uint32_t targ_lun;
12474
12475	ctl_softc = control_softc;
12476
12477	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12478	lun = ctl_softc->ctl_luns[targ_lun];
12479
12480	switch (io->io_hdr.msg_type) {
12481	case CTL_MSG_SERIALIZE:
12482		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12483		break;
12484	case CTL_MSG_R2R: {
12485		const struct ctl_cmd_entry *entry;
12486
12487		/*
12488		 * This is only used in SER_ONLY mode.
12489		 */
12490		free_io = 0;
12491		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12492		mtx_lock(&lun->lun_lock);
12493		if (ctl_scsiio_lun_check(ctl_softc, lun,
12494		    entry, (struct ctl_scsiio *)io) != 0) {
12495			mtx_unlock(&lun->lun_lock);
12496			ctl_done(io);
12497			break;
12498		}
12499		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12500		mtx_unlock(&lun->lun_lock);
12501		ctl_enqueue_rtr(io);
12502		break;
12503	}
12504	case CTL_MSG_FINISH_IO:
12505		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12506			free_io = 0;
12507			ctl_done(io);
12508		} else {
12509			free_io = 1;
12510			mtx_lock(&lun->lun_lock);
12511			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12512				     ooa_links);
12513			ctl_check_blocked(lun);
12514			mtx_unlock(&lun->lun_lock);
12515		}
12516		break;
12517	case CTL_MSG_PERS_ACTION:
12518		ctl_hndl_per_res_out_on_other_sc(
12519			(union ctl_ha_msg *)&io->presio.pr_msg);
12520		free_io = 1;
12521		break;
12522	case CTL_MSG_BAD_JUJU:
12523		free_io = 0;
12524		ctl_done(io);
12525		break;
12526	case CTL_MSG_DATAMOVE:
12527		/* Only used in XFER mode */
12528		free_io = 0;
12529		ctl_datamove_remote(io);
12530		break;
12531	case CTL_MSG_DATAMOVE_DONE:
12532		/* Only used in XFER mode */
12533		free_io = 0;
12534		io->scsiio.be_move_done(io);
12535		break;
12536	default:
12537		free_io = 1;
12538		printf("%s: Invalid message type %d\n",
12539		       __func__, io->io_hdr.msg_type);
12540		break;
12541	}
12542	if (free_io)
12543		ctl_free_io(io);
12544
12545}
12546
12547
12548/*
12549 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12550 * there is no match.
12551 */
12552static ctl_lun_error_pattern
12553ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12554{
12555	const struct ctl_cmd_entry *entry;
12556	ctl_lun_error_pattern filtered_pattern, pattern;
12557
12558	pattern = desc->error_pattern;
12559
12560	/*
12561	 * XXX KDM we need more data passed into this function to match a
12562	 * custom pattern, and we actually need to implement custom pattern
12563	 * matching.
12564	 */
12565	if (pattern & CTL_LUN_PAT_CMD)
12566		return (CTL_LUN_PAT_CMD);
12567
12568	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12569		return (CTL_LUN_PAT_ANY);
12570
12571	entry = ctl_get_cmd_entry(ctsio, NULL);
12572
12573	filtered_pattern = entry->pattern & pattern;
12574
12575	/*
12576	 * If the user requested specific flags in the pattern (e.g.
12577	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12578	 * flags.
12579	 *
12580	 * If the user did not specify any flags, it doesn't matter whether
12581	 * or not the command supports the flags.
12582	 */
12583	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12584	     (pattern & ~CTL_LUN_PAT_MASK))
12585		return (CTL_LUN_PAT_NONE);
12586
12587	/*
12588	 * If the user asked for a range check, see if the requested LBA
12589	 * range overlaps with this command's LBA range.
12590	 */
12591	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12592		uint64_t lba1;
12593		uint64_t len1;
12594		ctl_action action;
12595		int retval;
12596
12597		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12598		if (retval != 0)
12599			return (CTL_LUN_PAT_NONE);
12600
12601		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12602					      desc->lba_range.len);
12603		/*
12604		 * A "pass" means that the LBA ranges don't overlap, so
12605		 * this doesn't match the user's range criteria.
12606		 */
12607		if (action == CTL_ACTION_PASS)
12608			return (CTL_LUN_PAT_NONE);
12609	}
12610
12611	return (filtered_pattern);
12612}
12613
12614static void
12615ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12616{
12617	struct ctl_error_desc *desc, *desc2;
12618
12619	mtx_assert(&lun->lun_lock, MA_OWNED);
12620
12621	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12622		ctl_lun_error_pattern pattern;
12623		/*
12624		 * Check to see whether this particular command matches
12625		 * the pattern in the descriptor.
12626		 */
12627		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12628		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12629			continue;
12630
12631		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12632		case CTL_LUN_INJ_ABORTED:
12633			ctl_set_aborted(&io->scsiio);
12634			break;
12635		case CTL_LUN_INJ_MEDIUM_ERR:
12636			ctl_set_medium_error(&io->scsiio);
12637			break;
12638		case CTL_LUN_INJ_UA:
12639			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12640			 * OCCURRED */
12641			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12642			break;
12643		case CTL_LUN_INJ_CUSTOM:
12644			/*
12645			 * We're assuming the user knows what he is doing.
12646			 * Just copy the sense information without doing
12647			 * checks.
12648			 */
12649			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12650			      ctl_min(sizeof(desc->custom_sense),
12651				      sizeof(io->scsiio.sense_data)));
12652			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12653			io->scsiio.sense_len = SSD_FULL_SIZE;
12654			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12655			break;
12656		case CTL_LUN_INJ_NONE:
12657		default:
12658			/*
12659			 * If this is an error injection type we don't know
12660			 * about, clear the continuous flag (if it is set)
12661			 * so it will get deleted below.
12662			 */
12663			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12664			break;
12665		}
12666		/*
12667		 * By default, each error injection action is a one-shot
12668		 */
12669		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12670			continue;
12671
12672		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12673
12674		free(desc, M_CTL);
12675	}
12676}
12677
12678#ifdef CTL_IO_DELAY
12679static void
12680ctl_datamove_timer_wakeup(void *arg)
12681{
12682	union ctl_io *io;
12683
12684	io = (union ctl_io *)arg;
12685
12686	ctl_datamove(io);
12687}
12688#endif /* CTL_IO_DELAY */
12689
12690void
12691ctl_datamove(union ctl_io *io)
12692{
12693	void (*fe_datamove)(union ctl_io *io);
12694
12695	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12696
12697	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12698
12699#ifdef CTL_TIME_IO
12700	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12701		char str[256];
12702		char path_str[64];
12703		struct sbuf sb;
12704
12705		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12706		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12707
12708		sbuf_cat(&sb, path_str);
12709		switch (io->io_hdr.io_type) {
12710		case CTL_IO_SCSI:
12711			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12712			sbuf_printf(&sb, "\n");
12713			sbuf_cat(&sb, path_str);
12714			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12715				    io->scsiio.tag_num, io->scsiio.tag_type);
12716			break;
12717		case CTL_IO_TASK:
12718			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12719				    "Tag Type: %d\n", io->taskio.task_action,
12720				    io->taskio.tag_num, io->taskio.tag_type);
12721			break;
12722		default:
12723			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12724			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12725			break;
12726		}
12727		sbuf_cat(&sb, path_str);
12728		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12729			    (intmax_t)time_uptime - io->io_hdr.start_time);
12730		sbuf_finish(&sb);
12731		printf("%s", sbuf_data(&sb));
12732	}
12733#endif /* CTL_TIME_IO */
12734
12735#ifdef CTL_IO_DELAY
12736	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12737		struct ctl_lun *lun;
12738
12739		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12740
12741		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12742	} else {
12743		struct ctl_lun *lun;
12744
12745		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12746		if ((lun != NULL)
12747		 && (lun->delay_info.datamove_delay > 0)) {
12748			struct callout *callout;
12749
12750			callout = (struct callout *)&io->io_hdr.timer_bytes;
12751			callout_init(callout, /*mpsafe*/ 1);
12752			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12753			callout_reset(callout,
12754				      lun->delay_info.datamove_delay * hz,
12755				      ctl_datamove_timer_wakeup, io);
12756			if (lun->delay_info.datamove_type ==
12757			    CTL_DELAY_TYPE_ONESHOT)
12758				lun->delay_info.datamove_delay = 0;
12759			return;
12760		}
12761	}
12762#endif
12763
12764	/*
12765	 * This command has been aborted.  Set the port status, so we fail
12766	 * the data move.
12767	 */
12768	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12769		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12770		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12771		       io->io_hdr.nexus.targ_port,
12772		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12773		       io->io_hdr.nexus.targ_lun);
12774		io->io_hdr.port_status = 31337;
12775		/*
12776		 * Note that the backend, in this case, will get the
12777		 * callback in its context.  In other cases it may get
12778		 * called in the frontend's interrupt thread context.
12779		 */
12780		io->scsiio.be_move_done(io);
12781		return;
12782	}
12783
12784	/*
12785	 * If we're in XFER mode and this I/O is from the other shelf
12786	 * controller, we need to send the DMA to the other side to
12787	 * actually transfer the data to/from the host.  In serialize only
12788	 * mode the transfer happens below CTL and ctl_datamove() is only
12789	 * called on the machine that originally received the I/O.
12790	 */
12791	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12792	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12793		union ctl_ha_msg msg;
12794		uint32_t sg_entries_sent;
12795		int do_sg_copy;
12796		int i;
12797
12798		memset(&msg, 0, sizeof(msg));
12799		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12800		msg.hdr.original_sc = io->io_hdr.original_sc;
12801		msg.hdr.serializing_sc = io;
12802		msg.hdr.nexus = io->io_hdr.nexus;
12803		msg.dt.flags = io->io_hdr.flags;
12804		/*
12805		 * We convert everything into a S/G list here.  We can't
12806		 * pass by reference, only by value between controllers.
12807		 * So we can't pass a pointer to the S/G list, only as many
12808		 * S/G entries as we can fit in here.  If it's possible for
12809		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12810		 * then we need to break this up into multiple transfers.
12811		 */
12812		if (io->scsiio.kern_sg_entries == 0) {
12813			msg.dt.kern_sg_entries = 1;
12814			/*
12815			 * If this is in cached memory, flush the cache
12816			 * before we send the DMA request to the other
12817			 * controller.  We want to do this in either the
12818			 * read or the write case.  The read case is
12819			 * straightforward.  In the write case, we want to
12820			 * make sure nothing is in the local cache that
12821			 * could overwrite the DMAed data.
12822			 */
12823			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12824				/*
12825				 * XXX KDM use bus_dmamap_sync() here.
12826				 */
12827			}
12828
12829			/*
12830			 * Convert to a physical address if this is a
12831			 * virtual address.
12832			 */
12833			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12834				msg.dt.sg_list[0].addr =
12835					io->scsiio.kern_data_ptr;
12836			} else {
12837				/*
12838				 * XXX KDM use busdma here!
12839				 */
12840#if 0
12841				msg.dt.sg_list[0].addr = (void *)
12842					vtophys(io->scsiio.kern_data_ptr);
12843#endif
12844			}
12845
12846			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12847			do_sg_copy = 0;
12848		} else {
12849			struct ctl_sg_entry *sgl;
12850
12851			do_sg_copy = 1;
12852			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12853			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12854			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12855				/*
12856				 * XXX KDM use bus_dmamap_sync() here.
12857				 */
12858			}
12859		}
12860
12861		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12862		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12863		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12864		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12865		msg.dt.sg_sequence = 0;
12866
12867		/*
12868		 * Loop until we've sent all of the S/G entries.  On the
12869		 * other end, we'll recompose these S/G entries into one
12870		 * contiguous list before passing it to the
12871		 */
12872		for (sg_entries_sent = 0; sg_entries_sent <
12873		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12874			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12875				sizeof(msg.dt.sg_list[0])),
12876				msg.dt.kern_sg_entries - sg_entries_sent);
12877
12878			if (do_sg_copy != 0) {
12879				struct ctl_sg_entry *sgl;
12880				int j;
12881
12882				sgl = (struct ctl_sg_entry *)
12883					io->scsiio.kern_data_ptr;
12884				/*
12885				 * If this is in cached memory, flush the cache
12886				 * before we send the DMA request to the other
12887				 * controller.  We want to do this in either
12888				 * the * read or the write case.  The read
12889				 * case is straightforward.  In the write
12890				 * case, we want to make sure nothing is
12891				 * in the local cache that could overwrite
12892				 * the DMAed data.
12893				 */
12894
12895				for (i = sg_entries_sent, j = 0;
12896				     i < msg.dt.cur_sg_entries; i++, j++) {
12897					if ((io->io_hdr.flags &
12898					     CTL_FLAG_NO_DATASYNC) == 0) {
12899						/*
12900						 * XXX KDM use bus_dmamap_sync()
12901						 */
12902					}
12903					if ((io->io_hdr.flags &
12904					     CTL_FLAG_BUS_ADDR) == 0) {
12905						/*
12906						 * XXX KDM use busdma.
12907						 */
12908#if 0
12909						msg.dt.sg_list[j].addr =(void *)
12910						       vtophys(sgl[i].addr);
12911#endif
12912					} else {
12913						msg.dt.sg_list[j].addr =
12914							sgl[i].addr;
12915					}
12916					msg.dt.sg_list[j].len = sgl[i].len;
12917				}
12918			}
12919
12920			sg_entries_sent += msg.dt.cur_sg_entries;
12921			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12922				msg.dt.sg_last = 1;
12923			else
12924				msg.dt.sg_last = 0;
12925
12926			/*
12927			 * XXX KDM drop and reacquire the lock here?
12928			 */
12929			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12930			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12931				/*
12932				 * XXX do something here.
12933				 */
12934			}
12935
12936			msg.dt.sent_sg_entries = sg_entries_sent;
12937		}
12938		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12939		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12940			ctl_failover_io(io, /*have_lock*/ 0);
12941
12942	} else {
12943
12944		/*
12945		 * Lookup the fe_datamove() function for this particular
12946		 * front end.
12947		 */
12948		fe_datamove =
12949		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12950
12951		fe_datamove(io);
12952	}
12953}
12954
12955static void
12956ctl_send_datamove_done(union ctl_io *io, int have_lock)
12957{
12958	union ctl_ha_msg msg;
12959	int isc_status;
12960
12961	memset(&msg, 0, sizeof(msg));
12962
12963	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12964	msg.hdr.original_sc = io;
12965	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12966	msg.hdr.nexus = io->io_hdr.nexus;
12967	msg.hdr.status = io->io_hdr.status;
12968	msg.scsi.tag_num = io->scsiio.tag_num;
12969	msg.scsi.tag_type = io->scsiio.tag_type;
12970	msg.scsi.scsi_status = io->scsiio.scsi_status;
12971	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12972	       sizeof(io->scsiio.sense_data));
12973	msg.scsi.sense_len = io->scsiio.sense_len;
12974	msg.scsi.sense_residual = io->scsiio.sense_residual;
12975	msg.scsi.fetd_status = io->io_hdr.port_status;
12976	msg.scsi.residual = io->scsiio.residual;
12977	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12978
12979	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12980		ctl_failover_io(io, /*have_lock*/ have_lock);
12981		return;
12982	}
12983
12984	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12985	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12986		/* XXX do something if this fails */
12987	}
12988
12989}
12990
12991/*
12992 * The DMA to the remote side is done, now we need to tell the other side
12993 * we're done so it can continue with its data movement.
12994 */
12995static void
12996ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12997{
12998	union ctl_io *io;
12999
13000	io = rq->context;
13001
13002	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13003		printf("%s: ISC DMA write failed with error %d", __func__,
13004		       rq->ret);
13005		ctl_set_internal_failure(&io->scsiio,
13006					 /*sks_valid*/ 1,
13007					 /*retry_count*/ rq->ret);
13008	}
13009
13010	ctl_dt_req_free(rq);
13011
13012	/*
13013	 * In this case, we had to malloc the memory locally.  Free it.
13014	 */
13015	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13016		int i;
13017		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13018			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13019	}
13020	/*
13021	 * The data is in local and remote memory, so now we need to send
13022	 * status (good or back) back to the other side.
13023	 */
13024	ctl_send_datamove_done(io, /*have_lock*/ 0);
13025}
13026
13027/*
13028 * We've moved the data from the host/controller into local memory.  Now we
13029 * need to push it over to the remote controller's memory.
13030 */
13031static int
13032ctl_datamove_remote_dm_write_cb(union ctl_io *io)
13033{
13034	int retval;
13035
13036	retval = 0;
13037
13038	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
13039					  ctl_datamove_remote_write_cb);
13040
13041	return (retval);
13042}
13043
13044static void
13045ctl_datamove_remote_write(union ctl_io *io)
13046{
13047	int retval;
13048	void (*fe_datamove)(union ctl_io *io);
13049
13050	/*
13051	 * - Get the data from the host/HBA into local memory.
13052	 * - DMA memory from the local controller to the remote controller.
13053	 * - Send status back to the remote controller.
13054	 */
13055
13056	retval = ctl_datamove_remote_sgl_setup(io);
13057	if (retval != 0)
13058		return;
13059
13060	/* Switch the pointer over so the FETD knows what to do */
13061	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13062
13063	/*
13064	 * Use a custom move done callback, since we need to send completion
13065	 * back to the other controller, not to the backend on this side.
13066	 */
13067	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
13068
13069	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13070
13071	fe_datamove(io);
13072
13073	return;
13074
13075}
13076
13077static int
13078ctl_datamove_remote_dm_read_cb(union ctl_io *io)
13079{
13080#if 0
13081	char str[256];
13082	char path_str[64];
13083	struct sbuf sb;
13084#endif
13085
13086	/*
13087	 * In this case, we had to malloc the memory locally.  Free it.
13088	 */
13089	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13090		int i;
13091		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13092			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13093	}
13094
13095#if 0
13096	scsi_path_string(io, path_str, sizeof(path_str));
13097	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13098	sbuf_cat(&sb, path_str);
13099	scsi_command_string(&io->scsiio, NULL, &sb);
13100	sbuf_printf(&sb, "\n");
13101	sbuf_cat(&sb, path_str);
13102	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13103		    io->scsiio.tag_num, io->scsiio.tag_type);
13104	sbuf_cat(&sb, path_str);
13105	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
13106		    io->io_hdr.flags, io->io_hdr.status);
13107	sbuf_finish(&sb);
13108	printk("%s", sbuf_data(&sb));
13109#endif
13110
13111
13112	/*
13113	 * The read is done, now we need to send status (good or bad) back
13114	 * to the other side.
13115	 */
13116	ctl_send_datamove_done(io, /*have_lock*/ 0);
13117
13118	return (0);
13119}
13120
13121static void
13122ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
13123{
13124	union ctl_io *io;
13125	void (*fe_datamove)(union ctl_io *io);
13126
13127	io = rq->context;
13128
13129	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13130		printf("%s: ISC DMA read failed with error %d", __func__,
13131		       rq->ret);
13132		ctl_set_internal_failure(&io->scsiio,
13133					 /*sks_valid*/ 1,
13134					 /*retry_count*/ rq->ret);
13135	}
13136
13137	ctl_dt_req_free(rq);
13138
13139	/* Switch the pointer over so the FETD knows what to do */
13140	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13141
13142	/*
13143	 * Use a custom move done callback, since we need to send completion
13144	 * back to the other controller, not to the backend on this side.
13145	 */
13146	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13147
13148	/* XXX KDM add checks like the ones in ctl_datamove? */
13149
13150	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13151
13152	fe_datamove(io);
13153}
13154
13155static int
13156ctl_datamove_remote_sgl_setup(union ctl_io *io)
13157{
13158	struct ctl_sg_entry *local_sglist, *remote_sglist;
13159	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13160	struct ctl_softc *softc;
13161	int retval;
13162	int i;
13163
13164	retval = 0;
13165	softc = control_softc;
13166
13167	local_sglist = io->io_hdr.local_sglist;
13168	local_dma_sglist = io->io_hdr.local_dma_sglist;
13169	remote_sglist = io->io_hdr.remote_sglist;
13170	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13171
13172	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13173		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13174			local_sglist[i].len = remote_sglist[i].len;
13175
13176			/*
13177			 * XXX Detect the situation where the RS-level I/O
13178			 * redirector on the other side has already read the
13179			 * data off of the AOR RS on this side, and
13180			 * transferred it to remote (mirror) memory on the
13181			 * other side.  Since we already have the data in
13182			 * memory here, we just need to use it.
13183			 *
13184			 * XXX KDM this can probably be removed once we
13185			 * get the cache device code in and take the
13186			 * current AOR implementation out.
13187			 */
13188#ifdef NEEDTOPORT
13189			if ((remote_sglist[i].addr >=
13190			     (void *)vtophys(softc->mirr->addr))
13191			 && (remote_sglist[i].addr <
13192			     ((void *)vtophys(softc->mirr->addr) +
13193			     CacheMirrorOffset))) {
13194				local_sglist[i].addr = remote_sglist[i].addr -
13195					CacheMirrorOffset;
13196				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13197				     CTL_FLAG_DATA_IN)
13198					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13199			} else {
13200				local_sglist[i].addr = remote_sglist[i].addr +
13201					CacheMirrorOffset;
13202			}
13203#endif
13204#if 0
13205			printf("%s: local %p, remote %p, len %d\n",
13206			       __func__, local_sglist[i].addr,
13207			       remote_sglist[i].addr, local_sglist[i].len);
13208#endif
13209		}
13210	} else {
13211		uint32_t len_to_go;
13212
13213		/*
13214		 * In this case, we don't have automatically allocated
13215		 * memory for this I/O on this controller.  This typically
13216		 * happens with internal CTL I/O -- e.g. inquiry, mode
13217		 * sense, etc.  Anything coming from RAIDCore will have
13218		 * a mirror area available.
13219		 */
13220		len_to_go = io->scsiio.kern_data_len;
13221
13222		/*
13223		 * Clear the no datasync flag, we have to use malloced
13224		 * buffers.
13225		 */
13226		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13227
13228		/*
13229		 * The difficult thing here is that the size of the various
13230		 * S/G segments may be different than the size from the
13231		 * remote controller.  That'll make it harder when DMAing
13232		 * the data back to the other side.
13233		 */
13234		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13235		     sizeof(io->io_hdr.remote_sglist[0])) &&
13236		     (len_to_go > 0); i++) {
13237			local_sglist[i].len = ctl_min(len_to_go, 131072);
13238			CTL_SIZE_8B(local_dma_sglist[i].len,
13239				    local_sglist[i].len);
13240			local_sglist[i].addr =
13241				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13242
13243			local_dma_sglist[i].addr = local_sglist[i].addr;
13244
13245			if (local_sglist[i].addr == NULL) {
13246				int j;
13247
13248				printf("malloc failed for %zd bytes!",
13249				       local_dma_sglist[i].len);
13250				for (j = 0; j < i; j++) {
13251					free(local_sglist[j].addr, M_CTL);
13252				}
13253				ctl_set_internal_failure(&io->scsiio,
13254							 /*sks_valid*/ 1,
13255							 /*retry_count*/ 4857);
13256				retval = 1;
13257				goto bailout_error;
13258
13259			}
13260			/* XXX KDM do we need a sync here? */
13261
13262			len_to_go -= local_sglist[i].len;
13263		}
13264		/*
13265		 * Reset the number of S/G entries accordingly.  The
13266		 * original number of S/G entries is available in
13267		 * rem_sg_entries.
13268		 */
13269		io->scsiio.kern_sg_entries = i;
13270
13271#if 0
13272		printf("%s: kern_sg_entries = %d\n", __func__,
13273		       io->scsiio.kern_sg_entries);
13274		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13275			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13276			       local_sglist[i].addr, local_sglist[i].len,
13277			       local_dma_sglist[i].len);
13278#endif
13279	}
13280
13281
13282	return (retval);
13283
13284bailout_error:
13285
13286	ctl_send_datamove_done(io, /*have_lock*/ 0);
13287
13288	return (retval);
13289}
13290
13291static int
13292ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13293			 ctl_ha_dt_cb callback)
13294{
13295	struct ctl_ha_dt_req *rq;
13296	struct ctl_sg_entry *remote_sglist, *local_sglist;
13297	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13298	uint32_t local_used, remote_used, total_used;
13299	int retval;
13300	int i, j;
13301
13302	retval = 0;
13303
13304	rq = ctl_dt_req_alloc();
13305
13306	/*
13307	 * If we failed to allocate the request, and if the DMA didn't fail
13308	 * anyway, set busy status.  This is just a resource allocation
13309	 * failure.
13310	 */
13311	if ((rq == NULL)
13312	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13313		ctl_set_busy(&io->scsiio);
13314
13315	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13316
13317		if (rq != NULL)
13318			ctl_dt_req_free(rq);
13319
13320		/*
13321		 * The data move failed.  We need to return status back
13322		 * to the other controller.  No point in trying to DMA
13323		 * data to the remote controller.
13324		 */
13325
13326		ctl_send_datamove_done(io, /*have_lock*/ 0);
13327
13328		retval = 1;
13329
13330		goto bailout;
13331	}
13332
13333	local_sglist = io->io_hdr.local_sglist;
13334	local_dma_sglist = io->io_hdr.local_dma_sglist;
13335	remote_sglist = io->io_hdr.remote_sglist;
13336	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13337	local_used = 0;
13338	remote_used = 0;
13339	total_used = 0;
13340
13341	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13342		rq->ret = CTL_HA_STATUS_SUCCESS;
13343		rq->context = io;
13344		callback(rq);
13345		goto bailout;
13346	}
13347
13348	/*
13349	 * Pull/push the data over the wire from/to the other controller.
13350	 * This takes into account the possibility that the local and
13351	 * remote sglists may not be identical in terms of the size of
13352	 * the elements and the number of elements.
13353	 *
13354	 * One fundamental assumption here is that the length allocated for
13355	 * both the local and remote sglists is identical.  Otherwise, we've
13356	 * essentially got a coding error of some sort.
13357	 */
13358	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13359		int isc_ret;
13360		uint32_t cur_len, dma_length;
13361		uint8_t *tmp_ptr;
13362
13363		rq->id = CTL_HA_DATA_CTL;
13364		rq->command = command;
13365		rq->context = io;
13366
13367		/*
13368		 * Both pointers should be aligned.  But it is possible
13369		 * that the allocation length is not.  They should both
13370		 * also have enough slack left over at the end, though,
13371		 * to round up to the next 8 byte boundary.
13372		 */
13373		cur_len = ctl_min(local_sglist[i].len - local_used,
13374				  remote_sglist[j].len - remote_used);
13375
13376		/*
13377		 * In this case, we have a size issue and need to decrease
13378		 * the size, except in the case where we actually have less
13379		 * than 8 bytes left.  In that case, we need to increase
13380		 * the DMA length to get the last bit.
13381		 */
13382		if ((cur_len & 0x7) != 0) {
13383			if (cur_len > 0x7) {
13384				cur_len = cur_len - (cur_len & 0x7);
13385				dma_length = cur_len;
13386			} else {
13387				CTL_SIZE_8B(dma_length, cur_len);
13388			}
13389
13390		} else
13391			dma_length = cur_len;
13392
13393		/*
13394		 * If we had to allocate memory for this I/O, instead of using
13395		 * the non-cached mirror memory, we'll need to flush the cache
13396		 * before trying to DMA to the other controller.
13397		 *
13398		 * We could end up doing this multiple times for the same
13399		 * segment if we have a larger local segment than remote
13400		 * segment.  That shouldn't be an issue.
13401		 */
13402		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13403			/*
13404			 * XXX KDM use bus_dmamap_sync() here.
13405			 */
13406		}
13407
13408		rq->size = dma_length;
13409
13410		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13411		tmp_ptr += local_used;
13412
13413		/* Use physical addresses when talking to ISC hardware */
13414		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13415			/* XXX KDM use busdma */
13416#if 0
13417			rq->local = vtophys(tmp_ptr);
13418#endif
13419		} else
13420			rq->local = tmp_ptr;
13421
13422		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13423		tmp_ptr += remote_used;
13424		rq->remote = tmp_ptr;
13425
13426		rq->callback = NULL;
13427
13428		local_used += cur_len;
13429		if (local_used >= local_sglist[i].len) {
13430			i++;
13431			local_used = 0;
13432		}
13433
13434		remote_used += cur_len;
13435		if (remote_used >= remote_sglist[j].len) {
13436			j++;
13437			remote_used = 0;
13438		}
13439		total_used += cur_len;
13440
13441		if (total_used >= io->scsiio.kern_data_len)
13442			rq->callback = callback;
13443
13444		if ((rq->size & 0x7) != 0) {
13445			printf("%s: warning: size %d is not on 8b boundary\n",
13446			       __func__, rq->size);
13447		}
13448		if (((uintptr_t)rq->local & 0x7) != 0) {
13449			printf("%s: warning: local %p not on 8b boundary\n",
13450			       __func__, rq->local);
13451		}
13452		if (((uintptr_t)rq->remote & 0x7) != 0) {
13453			printf("%s: warning: remote %p not on 8b boundary\n",
13454			       __func__, rq->local);
13455		}
13456#if 0
13457		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13458		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13459		       rq->local, rq->remote, rq->size);
13460#endif
13461
13462		isc_ret = ctl_dt_single(rq);
13463		if (isc_ret == CTL_HA_STATUS_WAIT)
13464			continue;
13465
13466		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13467			rq->ret = CTL_HA_STATUS_SUCCESS;
13468		} else {
13469			rq->ret = isc_ret;
13470		}
13471		callback(rq);
13472		goto bailout;
13473	}
13474
13475bailout:
13476	return (retval);
13477
13478}
13479
13480static void
13481ctl_datamove_remote_read(union ctl_io *io)
13482{
13483	int retval;
13484	int i;
13485
13486	/*
13487	 * This will send an error to the other controller in the case of a
13488	 * failure.
13489	 */
13490	retval = ctl_datamove_remote_sgl_setup(io);
13491	if (retval != 0)
13492		return;
13493
13494	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13495					  ctl_datamove_remote_read_cb);
13496	if ((retval != 0)
13497	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13498		/*
13499		 * Make sure we free memory if there was an error..  The
13500		 * ctl_datamove_remote_xfer() function will send the
13501		 * datamove done message, or call the callback with an
13502		 * error if there is a problem.
13503		 */
13504		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13505			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13506	}
13507
13508	return;
13509}
13510
13511/*
13512 * Process a datamove request from the other controller.  This is used for
13513 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13514 * first.  Once that is complete, the data gets DMAed into the remote
13515 * controller's memory.  For reads, we DMA from the remote controller's
13516 * memory into our memory first, and then move it out to the FETD.
13517 */
13518static void
13519ctl_datamove_remote(union ctl_io *io)
13520{
13521	struct ctl_softc *softc;
13522
13523	softc = control_softc;
13524
13525	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13526
13527	/*
13528	 * Note that we look for an aborted I/O here, but don't do some of
13529	 * the other checks that ctl_datamove() normally does.
13530	 * We don't need to run the datamove delay code, since that should
13531	 * have been done if need be on the other controller.
13532	 */
13533	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13534		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13535		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13536		       io->io_hdr.nexus.targ_port,
13537		       io->io_hdr.nexus.targ_target.id,
13538		       io->io_hdr.nexus.targ_lun);
13539		io->io_hdr.port_status = 31338;
13540		ctl_send_datamove_done(io, /*have_lock*/ 0);
13541		return;
13542	}
13543
13544	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13545		ctl_datamove_remote_write(io);
13546	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13547		ctl_datamove_remote_read(io);
13548	} else {
13549		union ctl_ha_msg msg;
13550		struct scsi_sense_data *sense;
13551		uint8_t sks[3];
13552		int retry_count;
13553
13554		memset(&msg, 0, sizeof(msg));
13555
13556		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13557		msg.hdr.status = CTL_SCSI_ERROR;
13558		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13559
13560		retry_count = 4243;
13561
13562		sense = &msg.scsi.sense_data;
13563		sks[0] = SSD_SCS_VALID;
13564		sks[1] = (retry_count >> 8) & 0xff;
13565		sks[2] = retry_count & 0xff;
13566
13567		/* "Internal target failure" */
13568		scsi_set_sense_data(sense,
13569				    /*sense_format*/ SSD_TYPE_NONE,
13570				    /*current_error*/ 1,
13571				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13572				    /*asc*/ 0x44,
13573				    /*ascq*/ 0x00,
13574				    /*type*/ SSD_ELEM_SKS,
13575				    /*size*/ sizeof(sks),
13576				    /*data*/ sks,
13577				    SSD_ELEM_NONE);
13578
13579		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13580		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13581			ctl_failover_io(io, /*have_lock*/ 1);
13582			return;
13583		}
13584
13585		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13586		    CTL_HA_STATUS_SUCCESS) {
13587			/* XXX KDM what to do if this fails? */
13588		}
13589		return;
13590	}
13591
13592}
13593
13594static int
13595ctl_process_done(union ctl_io *io)
13596{
13597	struct ctl_lun *lun;
13598	struct ctl_softc *ctl_softc;
13599	void (*fe_done)(union ctl_io *io);
13600	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13601
13602	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13603
13604	fe_done =
13605	    control_softc->ctl_ports[targ_port]->fe_done;
13606
13607#ifdef CTL_TIME_IO
13608	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13609		char str[256];
13610		char path_str[64];
13611		struct sbuf sb;
13612
13613		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13614		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13615
13616		sbuf_cat(&sb, path_str);
13617		switch (io->io_hdr.io_type) {
13618		case CTL_IO_SCSI:
13619			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13620			sbuf_printf(&sb, "\n");
13621			sbuf_cat(&sb, path_str);
13622			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13623				    io->scsiio.tag_num, io->scsiio.tag_type);
13624			break;
13625		case CTL_IO_TASK:
13626			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13627				    "Tag Type: %d\n", io->taskio.task_action,
13628				    io->taskio.tag_num, io->taskio.tag_type);
13629			break;
13630		default:
13631			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13632			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13633			break;
13634		}
13635		sbuf_cat(&sb, path_str);
13636		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13637			    (intmax_t)time_uptime - io->io_hdr.start_time);
13638		sbuf_finish(&sb);
13639		printf("%s", sbuf_data(&sb));
13640	}
13641#endif /* CTL_TIME_IO */
13642
13643	switch (io->io_hdr.io_type) {
13644	case CTL_IO_SCSI:
13645		break;
13646	case CTL_IO_TASK:
13647		if (bootverbose || verbose > 0)
13648			ctl_io_error_print(io, NULL);
13649		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13650			ctl_free_io(io);
13651		else
13652			fe_done(io);
13653		return (CTL_RETVAL_COMPLETE);
13654		break;
13655	default:
13656		printf("ctl_process_done: invalid io type %d\n",
13657		       io->io_hdr.io_type);
13658		panic("ctl_process_done: invalid io type %d\n",
13659		      io->io_hdr.io_type);
13660		break; /* NOTREACHED */
13661	}
13662
13663	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13664	if (lun == NULL) {
13665		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13666				 io->io_hdr.nexus.targ_mapped_lun));
13667		fe_done(io);
13668		goto bailout;
13669	}
13670	ctl_softc = lun->ctl_softc;
13671
13672	mtx_lock(&lun->lun_lock);
13673
13674	/*
13675	 * Check to see if we have any errors to inject here.  We only
13676	 * inject errors for commands that don't already have errors set.
13677	 */
13678	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13679	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13680		ctl_inject_error(lun, io);
13681
13682	/*
13683	 * XXX KDM how do we treat commands that aren't completed
13684	 * successfully?
13685	 *
13686	 * XXX KDM should we also track I/O latency?
13687	 */
13688	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13689	    io->io_hdr.io_type == CTL_IO_SCSI) {
13690#ifdef CTL_TIME_IO
13691		struct bintime cur_bt;
13692#endif
13693		int type;
13694
13695		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13696		    CTL_FLAG_DATA_IN)
13697			type = CTL_STATS_READ;
13698		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13699		    CTL_FLAG_DATA_OUT)
13700			type = CTL_STATS_WRITE;
13701		else
13702			type = CTL_STATS_NO_IO;
13703
13704		lun->stats.ports[targ_port].bytes[type] +=
13705		    io->scsiio.kern_total_len;
13706		lun->stats.ports[targ_port].operations[type]++;
13707#ifdef CTL_TIME_IO
13708		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13709		   &io->io_hdr.dma_bt);
13710		lun->stats.ports[targ_port].num_dmas[type] +=
13711		    io->io_hdr.num_dmas;
13712		getbintime(&cur_bt);
13713		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13714		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13715#endif
13716	}
13717
13718	/*
13719	 * Remove this from the OOA queue.
13720	 */
13721	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13722
13723	/*
13724	 * Run through the blocked queue on this LUN and see if anything
13725	 * has become unblocked, now that this transaction is done.
13726	 */
13727	ctl_check_blocked(lun);
13728
13729	/*
13730	 * If the LUN has been invalidated, free it if there is nothing
13731	 * left on its OOA queue.
13732	 */
13733	if ((lun->flags & CTL_LUN_INVALID)
13734	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13735		mtx_unlock(&lun->lun_lock);
13736		mtx_lock(&ctl_softc->ctl_lock);
13737		ctl_free_lun(lun);
13738		mtx_unlock(&ctl_softc->ctl_lock);
13739	} else
13740		mtx_unlock(&lun->lun_lock);
13741
13742	/*
13743	 * If this command has been aborted, make sure we set the status
13744	 * properly.  The FETD is responsible for freeing the I/O and doing
13745	 * whatever it needs to do to clean up its state.
13746	 */
13747	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13748		ctl_set_task_aborted(&io->scsiio);
13749
13750	/*
13751	 * We print out status for every task management command.  For SCSI
13752	 * commands, we filter out any unit attention errors; they happen
13753	 * on every boot, and would clutter up the log.  Note:  task
13754	 * management commands aren't printed here, they are printed above,
13755	 * since they should never even make it down here.
13756	 */
13757	switch (io->io_hdr.io_type) {
13758	case CTL_IO_SCSI: {
13759		int error_code, sense_key, asc, ascq;
13760
13761		sense_key = 0;
13762
13763		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13764		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13765			/*
13766			 * Since this is just for printing, no need to
13767			 * show errors here.
13768			 */
13769			scsi_extract_sense_len(&io->scsiio.sense_data,
13770					       io->scsiio.sense_len,
13771					       &error_code,
13772					       &sense_key,
13773					       &asc,
13774					       &ascq,
13775					       /*show_errors*/ 0);
13776		}
13777
13778		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13779		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13780		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13781		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13782
13783			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13784				ctl_softc->skipped_prints++;
13785			} else {
13786				uint32_t skipped_prints;
13787
13788				skipped_prints = ctl_softc->skipped_prints;
13789
13790				ctl_softc->skipped_prints = 0;
13791				ctl_softc->last_print_jiffies = time_uptime;
13792
13793				if (skipped_prints > 0) {
13794#ifdef NEEDTOPORT
13795					csevent_log(CSC_CTL | CSC_SHELF_SW |
13796					    CTL_ERROR_REPORT,
13797					    csevent_LogType_Trace,
13798					    csevent_Severity_Information,
13799					    csevent_AlertLevel_Green,
13800					    csevent_FRU_Firmware,
13801					    csevent_FRU_Unknown,
13802					    "High CTL error volume, %d prints "
13803					    "skipped", skipped_prints);
13804#endif
13805				}
13806				if (bootverbose || verbose > 0)
13807					ctl_io_error_print(io, NULL);
13808			}
13809		}
13810		break;
13811	}
13812	case CTL_IO_TASK:
13813		if (bootverbose || verbose > 0)
13814			ctl_io_error_print(io, NULL);
13815		break;
13816	default:
13817		break;
13818	}
13819
13820	/*
13821	 * Tell the FETD or the other shelf controller we're done with this
13822	 * command.  Note that only SCSI commands get to this point.  Task
13823	 * management commands are completed above.
13824	 *
13825	 * We only send status to the other controller if we're in XFER
13826	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13827	 * received the I/O (from CTL's perspective), and so the status is
13828	 * generated there.
13829	 *
13830	 * XXX KDM if we hold the lock here, we could cause a deadlock
13831	 * if the frontend comes back in in this context to queue
13832	 * something.
13833	 */
13834	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13835	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13836		union ctl_ha_msg msg;
13837
13838		memset(&msg, 0, sizeof(msg));
13839		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13840		msg.hdr.original_sc = io->io_hdr.original_sc;
13841		msg.hdr.nexus = io->io_hdr.nexus;
13842		msg.hdr.status = io->io_hdr.status;
13843		msg.scsi.scsi_status = io->scsiio.scsi_status;
13844		msg.scsi.tag_num = io->scsiio.tag_num;
13845		msg.scsi.tag_type = io->scsiio.tag_type;
13846		msg.scsi.sense_len = io->scsiio.sense_len;
13847		msg.scsi.sense_residual = io->scsiio.sense_residual;
13848		msg.scsi.residual = io->scsiio.residual;
13849		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13850		       sizeof(io->scsiio.sense_data));
13851		/*
13852		 * We copy this whether or not this is an I/O-related
13853		 * command.  Otherwise, we'd have to go and check to see
13854		 * whether it's a read/write command, and it really isn't
13855		 * worth it.
13856		 */
13857		memcpy(&msg.scsi.lbalen,
13858		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13859		       sizeof(msg.scsi.lbalen));
13860
13861		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13862				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13863			/* XXX do something here */
13864		}
13865
13866		ctl_free_io(io);
13867	} else
13868		fe_done(io);
13869
13870bailout:
13871
13872	return (CTL_RETVAL_COMPLETE);
13873}
13874
13875#ifdef CTL_WITH_CA
13876/*
13877 * Front end should call this if it doesn't do autosense.  When the request
13878 * sense comes back in from the initiator, we'll dequeue this and send it.
13879 */
13880int
13881ctl_queue_sense(union ctl_io *io)
13882{
13883	struct ctl_lun *lun;
13884	struct ctl_softc *ctl_softc;
13885	uint32_t initidx, targ_lun;
13886
13887	ctl_softc = control_softc;
13888
13889	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13890
13891	/*
13892	 * LUN lookup will likely move to the ctl_work_thread() once we
13893	 * have our new queueing infrastructure (that doesn't put things on
13894	 * a per-LUN queue initially).  That is so that we can handle
13895	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13896	 * can't deal with that right now.
13897	 */
13898	mtx_lock(&ctl_softc->ctl_lock);
13899
13900	/*
13901	 * If we don't have a LUN for this, just toss the sense
13902	 * information.
13903	 */
13904	targ_lun = io->io_hdr.nexus.targ_lun;
13905	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13906	if ((targ_lun < CTL_MAX_LUNS)
13907	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13908		lun = ctl_softc->ctl_luns[targ_lun];
13909	else
13910		goto bailout;
13911
13912	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13913
13914	mtx_lock(&lun->lun_lock);
13915	/*
13916	 * Already have CA set for this LUN...toss the sense information.
13917	 */
13918	if (ctl_is_set(lun->have_ca, initidx)) {
13919		mtx_unlock(&lun->lun_lock);
13920		goto bailout;
13921	}
13922
13923	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13924	       ctl_min(sizeof(lun->pending_sense[initidx]),
13925	       sizeof(io->scsiio.sense_data)));
13926	ctl_set_mask(lun->have_ca, initidx);
13927	mtx_unlock(&lun->lun_lock);
13928
13929bailout:
13930	mtx_unlock(&ctl_softc->ctl_lock);
13931
13932	ctl_free_io(io);
13933
13934	return (CTL_RETVAL_COMPLETE);
13935}
13936#endif
13937
13938/*
13939 * Primary command inlet from frontend ports.  All SCSI and task I/O
13940 * requests must go through this function.
13941 */
13942int
13943ctl_queue(union ctl_io *io)
13944{
13945	struct ctl_softc *ctl_softc;
13946
13947	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13948
13949	ctl_softc = control_softc;
13950
13951#ifdef CTL_TIME_IO
13952	io->io_hdr.start_time = time_uptime;
13953	getbintime(&io->io_hdr.start_bt);
13954#endif /* CTL_TIME_IO */
13955
13956	/* Map FE-specific LUN ID into global one. */
13957	io->io_hdr.nexus.targ_mapped_lun =
13958	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13959
13960	switch (io->io_hdr.io_type) {
13961	case CTL_IO_SCSI:
13962	case CTL_IO_TASK:
13963		ctl_enqueue_incoming(io);
13964		break;
13965	default:
13966		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13967		return (EINVAL);
13968	}
13969
13970	return (CTL_RETVAL_COMPLETE);
13971}
13972
13973#ifdef CTL_IO_DELAY
13974static void
13975ctl_done_timer_wakeup(void *arg)
13976{
13977	union ctl_io *io;
13978
13979	io = (union ctl_io *)arg;
13980	ctl_done(io);
13981}
13982#endif /* CTL_IO_DELAY */
13983
13984void
13985ctl_done(union ctl_io *io)
13986{
13987	struct ctl_softc *ctl_softc;
13988
13989	ctl_softc = control_softc;
13990
13991	/*
13992	 * Enable this to catch duplicate completion issues.
13993	 */
13994#if 0
13995	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13996		printf("%s: type %d msg %d cdb %x iptl: "
13997		       "%d:%d:%d:%d tag 0x%04x "
13998		       "flag %#x status %x\n",
13999			__func__,
14000			io->io_hdr.io_type,
14001			io->io_hdr.msg_type,
14002			io->scsiio.cdb[0],
14003			io->io_hdr.nexus.initid.id,
14004			io->io_hdr.nexus.targ_port,
14005			io->io_hdr.nexus.targ_target.id,
14006			io->io_hdr.nexus.targ_lun,
14007			(io->io_hdr.io_type ==
14008			CTL_IO_TASK) ?
14009			io->taskio.tag_num :
14010			io->scsiio.tag_num,
14011		        io->io_hdr.flags,
14012			io->io_hdr.status);
14013	} else
14014		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
14015#endif
14016
14017	/*
14018	 * This is an internal copy of an I/O, and should not go through
14019	 * the normal done processing logic.
14020	 */
14021	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
14022		return;
14023
14024	/*
14025	 * We need to send a msg to the serializing shelf to finish the IO
14026	 * as well.  We don't send a finish message to the other shelf if
14027	 * this is a task management command.  Task management commands
14028	 * aren't serialized in the OOA queue, but rather just executed on
14029	 * both shelf controllers for commands that originated on that
14030	 * controller.
14031	 */
14032	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
14033	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
14034		union ctl_ha_msg msg_io;
14035
14036		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
14037		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
14038		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
14039		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
14040		}
14041		/* continue on to finish IO */
14042	}
14043#ifdef CTL_IO_DELAY
14044	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
14045		struct ctl_lun *lun;
14046
14047		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14048
14049		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
14050	} else {
14051		struct ctl_lun *lun;
14052
14053		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14054
14055		if ((lun != NULL)
14056		 && (lun->delay_info.done_delay > 0)) {
14057			struct callout *callout;
14058
14059			callout = (struct callout *)&io->io_hdr.timer_bytes;
14060			callout_init(callout, /*mpsafe*/ 1);
14061			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
14062			callout_reset(callout,
14063				      lun->delay_info.done_delay * hz,
14064				      ctl_done_timer_wakeup, io);
14065			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
14066				lun->delay_info.done_delay = 0;
14067			return;
14068		}
14069	}
14070#endif /* CTL_IO_DELAY */
14071
14072	ctl_enqueue_done(io);
14073}
14074
14075int
14076ctl_isc(struct ctl_scsiio *ctsio)
14077{
14078	struct ctl_lun *lun;
14079	int retval;
14080
14081	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14082
14083	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
14084
14085	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
14086
14087	retval = lun->backend->data_submit((union ctl_io *)ctsio);
14088
14089	return (retval);
14090}
14091
14092
14093static void
14094ctl_work_thread(void *arg)
14095{
14096	struct ctl_thread *thr = (struct ctl_thread *)arg;
14097	struct ctl_softc *softc = thr->ctl_softc;
14098	union ctl_io *io;
14099	int retval;
14100
14101	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
14102
14103	for (;;) {
14104		retval = 0;
14105
14106		/*
14107		 * We handle the queues in this order:
14108		 * - ISC
14109		 * - done queue (to free up resources, unblock other commands)
14110		 * - RtR queue
14111		 * - incoming queue
14112		 *
14113		 * If those queues are empty, we break out of the loop and
14114		 * go to sleep.
14115		 */
14116		mtx_lock(&thr->queue_lock);
14117		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
14118		if (io != NULL) {
14119			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
14120			mtx_unlock(&thr->queue_lock);
14121			ctl_handle_isc(io);
14122			continue;
14123		}
14124		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
14125		if (io != NULL) {
14126			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
14127			/* clear any blocked commands, call fe_done */
14128			mtx_unlock(&thr->queue_lock);
14129			retval = ctl_process_done(io);
14130			continue;
14131		}
14132		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
14133		if (io != NULL) {
14134			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
14135			mtx_unlock(&thr->queue_lock);
14136			if (io->io_hdr.io_type == CTL_IO_TASK)
14137				ctl_run_task(io);
14138			else
14139				ctl_scsiio_precheck(softc, &io->scsiio);
14140			continue;
14141		}
14142		if (!ctl_pause_rtr) {
14143			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
14144			if (io != NULL) {
14145				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
14146				mtx_unlock(&thr->queue_lock);
14147				retval = ctl_scsiio(&io->scsiio);
14148				if (retval != CTL_RETVAL_COMPLETE)
14149					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
14150				continue;
14151			}
14152		}
14153
14154		/* Sleep until we have something to do. */
14155		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
14156	}
14157}
14158
14159static void
14160ctl_lun_thread(void *arg)
14161{
14162	struct ctl_softc *softc = (struct ctl_softc *)arg;
14163	struct ctl_be_lun *be_lun;
14164	int retval;
14165
14166	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14167
14168	for (;;) {
14169		retval = 0;
14170		mtx_lock(&softc->ctl_lock);
14171		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14172		if (be_lun != NULL) {
14173			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14174			mtx_unlock(&softc->ctl_lock);
14175			ctl_create_lun(be_lun);
14176			continue;
14177		}
14178
14179		/* Sleep until we have something to do. */
14180		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14181		    PDROP | PRIBIO, "-", 0);
14182	}
14183}
14184
14185static void
14186ctl_enqueue_incoming(union ctl_io *io)
14187{
14188	struct ctl_softc *softc = control_softc;
14189	struct ctl_thread *thr;
14190	u_int idx;
14191
14192	idx = (io->io_hdr.nexus.targ_port * 127 +
14193	       io->io_hdr.nexus.initid.id) % worker_threads;
14194	thr = &softc->threads[idx];
14195	mtx_lock(&thr->queue_lock);
14196	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14197	mtx_unlock(&thr->queue_lock);
14198	wakeup(thr);
14199}
14200
14201static void
14202ctl_enqueue_rtr(union ctl_io *io)
14203{
14204	struct ctl_softc *softc = control_softc;
14205	struct ctl_thread *thr;
14206
14207	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14208	mtx_lock(&thr->queue_lock);
14209	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14210	mtx_unlock(&thr->queue_lock);
14211	wakeup(thr);
14212}
14213
14214static void
14215ctl_enqueue_done(union ctl_io *io)
14216{
14217	struct ctl_softc *softc = control_softc;
14218	struct ctl_thread *thr;
14219
14220	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14221	mtx_lock(&thr->queue_lock);
14222	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14223	mtx_unlock(&thr->queue_lock);
14224	wakeup(thr);
14225}
14226
14227static void
14228ctl_enqueue_isc(union ctl_io *io)
14229{
14230	struct ctl_softc *softc = control_softc;
14231	struct ctl_thread *thr;
14232
14233	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14234	mtx_lock(&thr->queue_lock);
14235	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14236	mtx_unlock(&thr->queue_lock);
14237	wakeup(thr);
14238}
14239
14240/* Initialization and failover */
14241
14242void
14243ctl_init_isc_msg(void)
14244{
14245	printf("CTL: Still calling this thing\n");
14246}
14247
14248/*
14249 * Init component
14250 * 	Initializes component into configuration defined by bootMode
14251 *	(see hasc-sv.c)
14252 *  	returns hasc_Status:
14253 * 		OK
14254 *		ERROR - fatal error
14255 */
14256static ctl_ha_comp_status
14257ctl_isc_init(struct ctl_ha_component *c)
14258{
14259	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14260
14261	c->status = ret;
14262	return ret;
14263}
14264
14265/* Start component
14266 * 	Starts component in state requested. If component starts successfully,
14267 *	it must set its own state to the requestrd state
14268 *	When requested state is HASC_STATE_HA, the component may refine it
14269 * 	by adding _SLAVE or _MASTER flags.
14270 *	Currently allowed state transitions are:
14271 *	UNKNOWN->HA		- initial startup
14272 *	UNKNOWN->SINGLE - initial startup when no parter detected
14273 *	HA->SINGLE		- failover
14274 * returns ctl_ha_comp_status:
14275 * 		OK	- component successfully started in requested state
14276 *		FAILED  - could not start the requested state, failover may
14277 * 			  be possible
14278 *		ERROR	- fatal error detected, no future startup possible
14279 */
14280static ctl_ha_comp_status
14281ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14282{
14283	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14284
14285	printf("%s: go\n", __func__);
14286
14287	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14288	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14289		ctl_is_single = 0;
14290		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14291		    != CTL_HA_STATUS_SUCCESS) {
14292			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14293			ret = CTL_HA_COMP_STATUS_ERROR;
14294		}
14295	} else if (CTL_HA_STATE_IS_HA(c->state)
14296		&& CTL_HA_STATE_IS_SINGLE(state)){
14297		// HA->SINGLE transition
14298	        ctl_failover();
14299		ctl_is_single = 1;
14300	} else {
14301		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14302		       c->state, state);
14303		ret = CTL_HA_COMP_STATUS_ERROR;
14304	}
14305	if (CTL_HA_STATE_IS_SINGLE(state))
14306		ctl_is_single = 1;
14307
14308	c->state = state;
14309	c->status = ret;
14310	return ret;
14311}
14312
14313/*
14314 * Quiesce component
14315 * The component must clear any error conditions (set status to OK) and
14316 * prepare itself to another Start call
14317 * returns ctl_ha_comp_status:
14318 * 	OK
14319 *	ERROR
14320 */
14321static ctl_ha_comp_status
14322ctl_isc_quiesce(struct ctl_ha_component *c)
14323{
14324	int ret = CTL_HA_COMP_STATUS_OK;
14325
14326	ctl_pause_rtr = 1;
14327	c->status = ret;
14328	return ret;
14329}
14330
14331struct ctl_ha_component ctl_ha_component_ctlisc =
14332{
14333	.name = "CTL ISC",
14334	.state = CTL_HA_STATE_UNKNOWN,
14335	.init = ctl_isc_init,
14336	.start = ctl_isc_start,
14337	.quiesce = ctl_isc_quiesce
14338};
14339
14340/*
14341 *  vim: ts=8
14342 */
14343