ctl.c revision 268686
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 268686 2014-07-15 17:09:02Z 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*/ 0,
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*/0,
284	/*eca_and_aen*/0,
285	/*reserved*/0,
286	/*aen_holdoff_period*/{0, 0}
287};
288
289static struct scsi_control_page control_page_changeable = {
290	/*page_code*/SMS_CONTROL_MODE_PAGE,
291	/*page_length*/sizeof(struct scsi_control_page) - 2,
292	/*rlec*/SCP_DSENSE,
293	/*queue_flags*/0,
294	/*eca_and_aen*/0,
295	/*reserved*/0,
296	/*aen_holdoff_period*/{0, 0}
297};
298
299
300/*
301 * XXX KDM move these into the softc.
302 */
303static int rcv_sync_msg;
304static int persis_offset;
305static uint8_t ctl_pause_rtr;
306static int     ctl_is_single = 1;
307static int     index_to_aps_page;
308
309SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
310static int worker_threads = -1;
311TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
312SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
313    &worker_threads, 1, "Number of worker threads");
314static int verbose = 0;
315TUNABLE_INT("kern.cam.ctl.verbose", &verbose);
316SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
317    &verbose, 0, "Show SCSI errors returned to initiator");
318
319/*
320 * Serial number (0x80), device id (0x83), supported pages (0x00),
321 * Block limits (0xB0) and Logical Block Provisioning (0xB2)
322 */
323#define SCSI_EVPD_NUM_SUPPORTED_PAGES	5
324
325static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
326				  int param);
327static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
328static int ctl_init(void);
329void ctl_shutdown(void);
330static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
331static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
332static void ctl_ioctl_online(void *arg);
333static void ctl_ioctl_offline(void *arg);
334static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
335static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
336static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
337static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
338static int ctl_ioctl_submit_wait(union ctl_io *io);
339static void ctl_ioctl_datamove(union ctl_io *io);
340static void ctl_ioctl_done(union ctl_io *io);
341static void ctl_ioctl_hard_startstop_callback(void *arg,
342					      struct cfi_metatask *metatask);
343static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
344static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
345			      struct ctl_ooa *ooa_hdr,
346			      struct ctl_ooa_entry *kern_entries);
347static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
348		     struct thread *td);
349uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
350uint32_t ctl_port_idx(int port_num);
351static uint32_t ctl_map_lun(int port_num, uint32_t lun);
352static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
353#ifdef unused
354static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
355				   uint32_t targ_target, uint32_t targ_lun,
356				   int can_wait);
357static void ctl_kfree_io(union ctl_io *io);
358#endif /* unused */
359static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
360			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
361static int ctl_free_lun(struct ctl_lun *lun);
362static void ctl_create_lun(struct ctl_be_lun *be_lun);
363/**
364static void ctl_failover_change_pages(struct ctl_softc *softc,
365				      struct ctl_scsiio *ctsio, int master);
366**/
367
368static int ctl_do_mode_select(union ctl_io *io);
369static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
370			   uint64_t res_key, uint64_t sa_res_key,
371			   uint8_t type, uint32_t residx,
372			   struct ctl_scsiio *ctsio,
373			   struct scsi_per_res_out *cdb,
374			   struct scsi_per_res_out_parms* param);
375static void ctl_pro_preempt_other(struct ctl_lun *lun,
376				  union ctl_ha_msg *msg);
377static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
378static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
379static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
380static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
381static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
382					 int alloc_len);
383static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
384static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
385static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
386static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
387static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
388static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
389					 union ctl_io *ooa_io);
390static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
391				union ctl_io *starting_io);
392static int ctl_check_blocked(struct ctl_lun *lun);
393static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
394				struct ctl_lun *lun,
395				const struct ctl_cmd_entry *entry,
396				struct ctl_scsiio *ctsio);
397//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
398static void ctl_failover(void);
399static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
400			       struct ctl_scsiio *ctsio);
401static int ctl_scsiio(struct ctl_scsiio *ctsio);
402
403static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
404static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
405			    ctl_ua_type ua_type);
406static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
407			 ctl_ua_type ua_type);
408static int ctl_abort_task(union ctl_io *io);
409static void ctl_run_task(union ctl_io *io);
410#ifdef CTL_IO_DELAY
411static void ctl_datamove_timer_wakeup(void *arg);
412static void ctl_done_timer_wakeup(void *arg);
413#endif /* CTL_IO_DELAY */
414
415static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
416static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
417static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
418static void ctl_datamove_remote_write(union ctl_io *io);
419static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
420static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
421static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
422static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
423				    ctl_ha_dt_cb callback);
424static void ctl_datamove_remote_read(union ctl_io *io);
425static void ctl_datamove_remote(union ctl_io *io);
426static int ctl_process_done(union ctl_io *io);
427static void ctl_lun_thread(void *arg);
428static void ctl_work_thread(void *arg);
429static void ctl_enqueue_incoming(union ctl_io *io);
430static void ctl_enqueue_rtr(union ctl_io *io);
431static void ctl_enqueue_done(union ctl_io *io);
432static void ctl_enqueue_isc(union ctl_io *io);
433static const struct ctl_cmd_entry *
434    ctl_get_cmd_entry(struct ctl_scsiio *ctsio);
435static const struct ctl_cmd_entry *
436    ctl_validate_command(struct ctl_scsiio *ctsio);
437static int ctl_cmd_applicable(uint8_t lun_type,
438    const struct ctl_cmd_entry *entry);
439
440/*
441 * Load the serialization table.  This isn't very pretty, but is probably
442 * the easiest way to do it.
443 */
444#include "ctl_ser_table.c"
445
446/*
447 * We only need to define open, close and ioctl routines for this driver.
448 */
449static struct cdevsw ctl_cdevsw = {
450	.d_version =	D_VERSION,
451	.d_flags =	0,
452	.d_open =	ctl_open,
453	.d_close =	ctl_close,
454	.d_ioctl =	ctl_ioctl,
455	.d_name =	"ctl",
456};
457
458
459MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
460MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
461
462static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
463
464static moduledata_t ctl_moduledata = {
465	"ctl",
466	ctl_module_event_handler,
467	NULL
468};
469
470DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
471MODULE_VERSION(ctl, 1);
472
473static struct ctl_frontend ioctl_frontend =
474{
475	.name = "ioctl",
476};
477
478static void
479ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
480			    union ctl_ha_msg *msg_info)
481{
482	struct ctl_scsiio *ctsio;
483
484	if (msg_info->hdr.original_sc == NULL) {
485		printf("%s: original_sc == NULL!\n", __func__);
486		/* XXX KDM now what? */
487		return;
488	}
489
490	ctsio = &msg_info->hdr.original_sc->scsiio;
491	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
492	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
493	ctsio->io_hdr.status = msg_info->hdr.status;
494	ctsio->scsi_status = msg_info->scsi.scsi_status;
495	ctsio->sense_len = msg_info->scsi.sense_len;
496	ctsio->sense_residual = msg_info->scsi.sense_residual;
497	ctsio->residual = msg_info->scsi.residual;
498	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
499	       sizeof(ctsio->sense_data));
500	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
501	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
502	ctl_enqueue_isc((union ctl_io *)ctsio);
503}
504
505static void
506ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
507				union ctl_ha_msg *msg_info)
508{
509	struct ctl_scsiio *ctsio;
510
511	if (msg_info->hdr.serializing_sc == NULL) {
512		printf("%s: serializing_sc == NULL!\n", __func__);
513		/* XXX KDM now what? */
514		return;
515	}
516
517	ctsio = &msg_info->hdr.serializing_sc->scsiio;
518#if 0
519	/*
520	 * Attempt to catch the situation where an I/O has
521	 * been freed, and we're using it again.
522	 */
523	if (ctsio->io_hdr.io_type == 0xff) {
524		union ctl_io *tmp_io;
525		tmp_io = (union ctl_io *)ctsio;
526		printf("%s: %p use after free!\n", __func__,
527		       ctsio);
528		printf("%s: type %d msg %d cdb %x iptl: "
529		       "%d:%d:%d:%d tag 0x%04x "
530		       "flag %#x status %x\n",
531			__func__,
532			tmp_io->io_hdr.io_type,
533			tmp_io->io_hdr.msg_type,
534			tmp_io->scsiio.cdb[0],
535			tmp_io->io_hdr.nexus.initid.id,
536			tmp_io->io_hdr.nexus.targ_port,
537			tmp_io->io_hdr.nexus.targ_target.id,
538			tmp_io->io_hdr.nexus.targ_lun,
539			(tmp_io->io_hdr.io_type ==
540			CTL_IO_TASK) ?
541			tmp_io->taskio.tag_num :
542			tmp_io->scsiio.tag_num,
543		        tmp_io->io_hdr.flags,
544			tmp_io->io_hdr.status);
545	}
546#endif
547	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
548	ctl_enqueue_isc((union ctl_io *)ctsio);
549}
550
551/*
552 * ISC (Inter Shelf Communication) event handler.  Events from the HA
553 * subsystem come in here.
554 */
555static void
556ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
557{
558	struct ctl_softc *ctl_softc;
559	union ctl_io *io;
560	struct ctl_prio *presio;
561	ctl_ha_status isc_status;
562
563	ctl_softc = control_softc;
564	io = NULL;
565
566
567#if 0
568	printf("CTL: Isc Msg event %d\n", event);
569#endif
570	if (event == CTL_HA_EVT_MSG_RECV) {
571		union ctl_ha_msg msg_info;
572
573		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
574					     sizeof(msg_info), /*wait*/ 0);
575#if 0
576		printf("CTL: msg_type %d\n", msg_info.msg_type);
577#endif
578		if (isc_status != 0) {
579			printf("Error receiving message, status = %d\n",
580			       isc_status);
581			return;
582		}
583
584		switch (msg_info.hdr.msg_type) {
585		case CTL_MSG_SERIALIZE:
586#if 0
587			printf("Serialize\n");
588#endif
589			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
590			if (io == NULL) {
591				printf("ctl_isc_event_handler: can't allocate "
592				       "ctl_io!\n");
593				/* Bad Juju */
594				/* Need to set busy and send msg back */
595				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
596				msg_info.hdr.status = CTL_SCSI_ERROR;
597				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
598				msg_info.scsi.sense_len = 0;
599			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
600				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
601				}
602				goto bailout;
603			}
604			ctl_zero_io(io);
605			// populate ctsio from msg_info
606			io->io_hdr.io_type = CTL_IO_SCSI;
607			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
608			io->io_hdr.original_sc = msg_info.hdr.original_sc;
609#if 0
610			printf("pOrig %x\n", (int)msg_info.original_sc);
611#endif
612			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
613					    CTL_FLAG_IO_ACTIVE;
614			/*
615			 * If we're in serialization-only mode, we don't
616			 * want to go through full done processing.  Thus
617			 * the COPY flag.
618			 *
619			 * XXX KDM add another flag that is more specific.
620			 */
621			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
622				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
623			io->io_hdr.nexus = msg_info.hdr.nexus;
624#if 0
625			printf("targ %d, port %d, iid %d, lun %d\n",
626			       io->io_hdr.nexus.targ_target.id,
627			       io->io_hdr.nexus.targ_port,
628			       io->io_hdr.nexus.initid.id,
629			       io->io_hdr.nexus.targ_lun);
630#endif
631			io->scsiio.tag_num = msg_info.scsi.tag_num;
632			io->scsiio.tag_type = msg_info.scsi.tag_type;
633			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
634			       CTL_MAX_CDBLEN);
635			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
636				const struct ctl_cmd_entry *entry;
637
638				entry = ctl_get_cmd_entry(&io->scsiio);
639				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
640				io->io_hdr.flags |=
641					entry->flags & CTL_FLAG_DATA_MASK;
642			}
643			ctl_enqueue_isc(io);
644			break;
645
646		/* Performed on the Originating SC, XFER mode only */
647		case CTL_MSG_DATAMOVE: {
648			struct ctl_sg_entry *sgl;
649			int i, j;
650
651			io = msg_info.hdr.original_sc;
652			if (io == NULL) {
653				printf("%s: original_sc == NULL!\n", __func__);
654				/* XXX KDM do something here */
655				break;
656			}
657			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
658			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
659			/*
660			 * Keep track of this, we need to send it back over
661			 * when the datamove is complete.
662			 */
663			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
664
665			if (msg_info.dt.sg_sequence == 0) {
666				/*
667				 * XXX KDM we use the preallocated S/G list
668				 * here, but we'll need to change this to
669				 * dynamic allocation if we need larger S/G
670				 * lists.
671				 */
672				if (msg_info.dt.kern_sg_entries >
673				    sizeof(io->io_hdr.remote_sglist) /
674				    sizeof(io->io_hdr.remote_sglist[0])) {
675					printf("%s: number of S/G entries "
676					    "needed %u > allocated num %zd\n",
677					    __func__,
678					    msg_info.dt.kern_sg_entries,
679					    sizeof(io->io_hdr.remote_sglist)/
680					    sizeof(io->io_hdr.remote_sglist[0]));
681
682					/*
683					 * XXX KDM send a message back to
684					 * the other side to shut down the
685					 * DMA.  The error will come back
686					 * through via the normal channel.
687					 */
688					break;
689				}
690				sgl = io->io_hdr.remote_sglist;
691				memset(sgl, 0,
692				       sizeof(io->io_hdr.remote_sglist));
693
694				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
695
696				io->scsiio.kern_sg_entries =
697					msg_info.dt.kern_sg_entries;
698				io->scsiio.rem_sg_entries =
699					msg_info.dt.kern_sg_entries;
700				io->scsiio.kern_data_len =
701					msg_info.dt.kern_data_len;
702				io->scsiio.kern_total_len =
703					msg_info.dt.kern_total_len;
704				io->scsiio.kern_data_resid =
705					msg_info.dt.kern_data_resid;
706				io->scsiio.kern_rel_offset =
707					msg_info.dt.kern_rel_offset;
708				/*
709				 * Clear out per-DMA flags.
710				 */
711				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
712				/*
713				 * Add per-DMA flags that are set for this
714				 * particular DMA request.
715				 */
716				io->io_hdr.flags |= msg_info.dt.flags &
717						    CTL_FLAG_RDMA_MASK;
718			} else
719				sgl = (struct ctl_sg_entry *)
720					io->scsiio.kern_data_ptr;
721
722			for (i = msg_info.dt.sent_sg_entries, j = 0;
723			     i < (msg_info.dt.sent_sg_entries +
724			     msg_info.dt.cur_sg_entries); i++, j++) {
725				sgl[i].addr = msg_info.dt.sg_list[j].addr;
726				sgl[i].len = msg_info.dt.sg_list[j].len;
727
728#if 0
729				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
730				       __func__,
731				       msg_info.dt.sg_list[j].addr,
732				       msg_info.dt.sg_list[j].len,
733				       sgl[i].addr, sgl[i].len, j, i);
734#endif
735			}
736#if 0
737			memcpy(&sgl[msg_info.dt.sent_sg_entries],
738			       msg_info.dt.sg_list,
739			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
740#endif
741
742			/*
743			 * If this is the last piece of the I/O, we've got
744			 * the full S/G list.  Queue processing in the thread.
745			 * Otherwise wait for the next piece.
746			 */
747			if (msg_info.dt.sg_last != 0)
748				ctl_enqueue_isc(io);
749			break;
750		}
751		/* Performed on the Serializing (primary) SC, XFER mode only */
752		case CTL_MSG_DATAMOVE_DONE: {
753			if (msg_info.hdr.serializing_sc == NULL) {
754				printf("%s: serializing_sc == NULL!\n",
755				       __func__);
756				/* XXX KDM now what? */
757				break;
758			}
759			/*
760			 * We grab the sense information here in case
761			 * there was a failure, so we can return status
762			 * back to the initiator.
763			 */
764			io = msg_info.hdr.serializing_sc;
765			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
766			io->io_hdr.status = msg_info.hdr.status;
767			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
768			io->scsiio.sense_len = msg_info.scsi.sense_len;
769			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
770			io->io_hdr.port_status = msg_info.scsi.fetd_status;
771			io->scsiio.residual = msg_info.scsi.residual;
772			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
773			       sizeof(io->scsiio.sense_data));
774			ctl_enqueue_isc(io);
775			break;
776		}
777
778		/* Preformed on Originating SC, SER_ONLY mode */
779		case CTL_MSG_R2R:
780			io = msg_info.hdr.original_sc;
781			if (io == NULL) {
782				printf("%s: Major Bummer\n", __func__);
783				return;
784			} else {
785#if 0
786				printf("pOrig %x\n",(int) ctsio);
787#endif
788			}
789			io->io_hdr.msg_type = CTL_MSG_R2R;
790			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
791			ctl_enqueue_isc(io);
792			break;
793
794		/*
795		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
796		 * mode.
797		 * Performed on the Originating (i.e. secondary) SC in XFER
798		 * mode
799		 */
800		case CTL_MSG_FINISH_IO:
801			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
802				ctl_isc_handler_finish_xfer(ctl_softc,
803							    &msg_info);
804			else
805				ctl_isc_handler_finish_ser_only(ctl_softc,
806								&msg_info);
807			break;
808
809		/* Preformed on Originating SC */
810		case CTL_MSG_BAD_JUJU:
811			io = msg_info.hdr.original_sc;
812			if (io == NULL) {
813				printf("%s: Bad JUJU!, original_sc is NULL!\n",
814				       __func__);
815				break;
816			}
817			ctl_copy_sense_data(&msg_info, io);
818			/*
819			 * IO should have already been cleaned up on other
820			 * SC so clear this flag so we won't send a message
821			 * back to finish the IO there.
822			 */
823			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
824			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
825
826			/* io = msg_info.hdr.serializing_sc; */
827			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
828			ctl_enqueue_isc(io);
829			break;
830
831		/* Handle resets sent from the other side */
832		case CTL_MSG_MANAGE_TASKS: {
833			struct ctl_taskio *taskio;
834			taskio = (struct ctl_taskio *)ctl_alloc_io(
835				(void *)ctl_softc->othersc_pool);
836			if (taskio == NULL) {
837				printf("ctl_isc_event_handler: can't allocate "
838				       "ctl_io!\n");
839				/* Bad Juju */
840				/* should I just call the proper reset func
841				   here??? */
842				goto bailout;
843			}
844			ctl_zero_io((union ctl_io *)taskio);
845			taskio->io_hdr.io_type = CTL_IO_TASK;
846			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
847			taskio->io_hdr.nexus = msg_info.hdr.nexus;
848			taskio->task_action = msg_info.task.task_action;
849			taskio->tag_num = msg_info.task.tag_num;
850			taskio->tag_type = msg_info.task.tag_type;
851#ifdef CTL_TIME_IO
852			taskio->io_hdr.start_time = time_uptime;
853			getbintime(&taskio->io_hdr.start_bt);
854#if 0
855			cs_prof_gettime(&taskio->io_hdr.start_ticks);
856#endif
857#endif /* CTL_TIME_IO */
858			ctl_run_task((union ctl_io *)taskio);
859			break;
860		}
861		/* Persistent Reserve action which needs attention */
862		case CTL_MSG_PERS_ACTION:
863			presio = (struct ctl_prio *)ctl_alloc_io(
864				(void *)ctl_softc->othersc_pool);
865			if (presio == NULL) {
866				printf("ctl_isc_event_handler: can't allocate "
867				       "ctl_io!\n");
868				/* Bad Juju */
869				/* Need to set busy and send msg back */
870				goto bailout;
871			}
872			ctl_zero_io((union ctl_io *)presio);
873			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
874			presio->pr_msg = msg_info.pr;
875			ctl_enqueue_isc((union ctl_io *)presio);
876			break;
877		case CTL_MSG_SYNC_FE:
878			rcv_sync_msg = 1;
879			break;
880		case CTL_MSG_APS_LOCK: {
881			// It's quicker to execute this then to
882			// queue it.
883			struct ctl_lun *lun;
884			struct ctl_page_index *page_index;
885			struct copan_aps_subpage *current_sp;
886			uint32_t targ_lun;
887
888			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
889			lun = ctl_softc->ctl_luns[targ_lun];
890			mtx_lock(&lun->lun_lock);
891			page_index = &lun->mode_pages.index[index_to_aps_page];
892			current_sp = (struct copan_aps_subpage *)
893				     (page_index->page_data +
894				     (page_index->page_len * CTL_PAGE_CURRENT));
895
896			current_sp->lock_active = msg_info.aps.lock_flag;
897			mtx_unlock(&lun->lun_lock);
898		        break;
899		}
900		default:
901		        printf("How did I get here?\n");
902		}
903	} else if (event == CTL_HA_EVT_MSG_SENT) {
904		if (param != CTL_HA_STATUS_SUCCESS) {
905			printf("Bad status from ctl_ha_msg_send status %d\n",
906			       param);
907		}
908		return;
909	} else if (event == CTL_HA_EVT_DISCONNECT) {
910		printf("CTL: Got a disconnect from Isc\n");
911		return;
912	} else {
913		printf("ctl_isc_event_handler: Unknown event %d\n", event);
914		return;
915	}
916
917bailout:
918	return;
919}
920
921static void
922ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
923{
924	struct scsi_sense_data *sense;
925
926	sense = &dest->scsiio.sense_data;
927	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
928	dest->scsiio.scsi_status = src->scsi.scsi_status;
929	dest->scsiio.sense_len = src->scsi.sense_len;
930	dest->io_hdr.status = src->hdr.status;
931}
932
933static int
934ctl_init(void)
935{
936	struct ctl_softc *softc;
937	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
938	struct ctl_port *port;
939        uint8_t sc_id =0;
940	int i, error, retval;
941	//int isc_retval;
942
943	retval = 0;
944	ctl_pause_rtr = 0;
945        rcv_sync_msg = 0;
946
947	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
948			       M_WAITOK | M_ZERO);
949	softc = control_softc;
950
951	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
952			      "cam/ctl");
953
954	softc->dev->si_drv1 = softc;
955
956	/*
957	 * By default, return a "bad LUN" peripheral qualifier for unknown
958	 * LUNs.  The user can override this default using the tunable or
959	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
960	 */
961	softc->inquiry_pq_no_lun = 1;
962	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
963			  &softc->inquiry_pq_no_lun);
964	sysctl_ctx_init(&softc->sysctl_ctx);
965	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
966		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
967		CTLFLAG_RD, 0, "CAM Target Layer");
968
969	if (softc->sysctl_tree == NULL) {
970		printf("%s: unable to allocate sysctl tree\n", __func__);
971		destroy_dev(softc->dev);
972		free(control_softc, M_DEVBUF);
973		control_softc = NULL;
974		return (ENOMEM);
975	}
976
977	SYSCTL_ADD_INT(&softc->sysctl_ctx,
978		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
979		       "inquiry_pq_no_lun", CTLFLAG_RW,
980		       &softc->inquiry_pq_no_lun, 0,
981		       "Report no lun possible for invalid LUNs");
982
983	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
984	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
985	softc->open_count = 0;
986
987	/*
988	 * Default to actually sending a SYNCHRONIZE CACHE command down to
989	 * the drive.
990	 */
991	softc->flags = CTL_FLAG_REAL_SYNC;
992
993	/*
994	 * In Copan's HA scheme, the "master" and "slave" roles are
995	 * figured out through the slot the controller is in.  Although it
996	 * is an active/active system, someone has to be in charge.
997 	 */
998#ifdef NEEDTOPORT
999        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1000#endif
1001
1002        if (sc_id == 0) {
1003		softc->flags |= CTL_FLAG_MASTER_SHELF;
1004		persis_offset = 0;
1005	} else
1006		persis_offset = CTL_MAX_INITIATORS;
1007
1008	/*
1009	 * XXX KDM need to figure out where we want to get our target ID
1010	 * and WWID.  Is it different on each port?
1011	 */
1012	softc->target.id = 0;
1013	softc->target.wwid[0] = 0x12345678;
1014	softc->target.wwid[1] = 0x87654321;
1015	STAILQ_INIT(&softc->lun_list);
1016	STAILQ_INIT(&softc->pending_lun_queue);
1017	STAILQ_INIT(&softc->fe_list);
1018	STAILQ_INIT(&softc->port_list);
1019	STAILQ_INIT(&softc->be_list);
1020	STAILQ_INIT(&softc->io_pools);
1021
1022	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1023			    &internal_pool)!= 0){
1024		printf("ctl: can't allocate %d entry internal pool, "
1025		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1026		return (ENOMEM);
1027	}
1028
1029	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1030			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1031		printf("ctl: can't allocate %d entry emergency pool, "
1032		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1033		ctl_pool_free(internal_pool);
1034		return (ENOMEM);
1035	}
1036
1037	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1038	                    &other_pool) != 0)
1039	{
1040		printf("ctl: can't allocate %d entry other SC pool, "
1041		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1042		ctl_pool_free(internal_pool);
1043		ctl_pool_free(emergency_pool);
1044		return (ENOMEM);
1045	}
1046
1047	softc->internal_pool = internal_pool;
1048	softc->emergency_pool = emergency_pool;
1049	softc->othersc_pool = other_pool;
1050
1051	if (worker_threads <= 0)
1052		worker_threads = max(1, mp_ncpus / 4);
1053	if (worker_threads > CTL_MAX_THREADS)
1054		worker_threads = CTL_MAX_THREADS;
1055
1056	for (i = 0; i < worker_threads; i++) {
1057		struct ctl_thread *thr = &softc->threads[i];
1058
1059		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1060		thr->ctl_softc = softc;
1061		STAILQ_INIT(&thr->incoming_queue);
1062		STAILQ_INIT(&thr->rtr_queue);
1063		STAILQ_INIT(&thr->done_queue);
1064		STAILQ_INIT(&thr->isc_queue);
1065
1066		error = kproc_kthread_add(ctl_work_thread, thr,
1067		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1068		if (error != 0) {
1069			printf("error creating CTL work thread!\n");
1070			ctl_pool_free(internal_pool);
1071			ctl_pool_free(emergency_pool);
1072			ctl_pool_free(other_pool);
1073			return (error);
1074		}
1075	}
1076	error = kproc_kthread_add(ctl_lun_thread, softc,
1077	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1078	if (error != 0) {
1079		printf("error creating CTL lun thread!\n");
1080		ctl_pool_free(internal_pool);
1081		ctl_pool_free(emergency_pool);
1082		ctl_pool_free(other_pool);
1083		return (error);
1084	}
1085	if (bootverbose)
1086		printf("ctl: CAM Target Layer loaded\n");
1087
1088	/*
1089	 * Initialize the initiator and portname mappings
1090	 */
1091	memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1092
1093	/*
1094	 * Initialize the ioctl front end.
1095	 */
1096	ctl_frontend_register(&ioctl_frontend);
1097	port = &softc->ioctl_info.port;
1098	port->frontend = &ioctl_frontend;
1099	sprintf(softc->ioctl_info.port_name, "ioctl");
1100	port->port_type = CTL_PORT_IOCTL;
1101	port->num_requested_ctl_io = 100;
1102	port->port_name = softc->ioctl_info.port_name;
1103	port->port_online = ctl_ioctl_online;
1104	port->port_offline = ctl_ioctl_offline;
1105	port->onoff_arg = &softc->ioctl_info;
1106	port->lun_enable = ctl_ioctl_lun_enable;
1107	port->lun_disable = ctl_ioctl_lun_disable;
1108	port->targ_lun_arg = &softc->ioctl_info;
1109	port->fe_datamove = ctl_ioctl_datamove;
1110	port->fe_done = ctl_ioctl_done;
1111	port->max_targets = 15;
1112	port->max_target_id = 15;
1113
1114	if (ctl_port_register(&softc->ioctl_info.port,
1115	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1116		printf("ctl: ioctl front end registration failed, will "
1117		       "continue anyway\n");
1118	}
1119
1120#ifdef CTL_IO_DELAY
1121	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1122		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1123		       sizeof(struct callout), CTL_TIMER_BYTES);
1124		return (EINVAL);
1125	}
1126#endif /* CTL_IO_DELAY */
1127
1128	return (0);
1129}
1130
1131void
1132ctl_shutdown(void)
1133{
1134	struct ctl_softc *softc;
1135	struct ctl_lun *lun, *next_lun;
1136	struct ctl_io_pool *pool;
1137
1138	softc = (struct ctl_softc *)control_softc;
1139
1140	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1141		printf("ctl: ioctl front end deregistration failed\n");
1142
1143	mtx_lock(&softc->ctl_lock);
1144
1145	/*
1146	 * Free up each LUN.
1147	 */
1148	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1149		next_lun = STAILQ_NEXT(lun, links);
1150		ctl_free_lun(lun);
1151	}
1152
1153	mtx_unlock(&softc->ctl_lock);
1154
1155	ctl_frontend_deregister(&ioctl_frontend);
1156
1157	/*
1158	 * This will rip the rug out from under any FETDs or anyone else
1159	 * that has a pool allocated.  Since we increment our module
1160	 * refcount any time someone outside the main CTL module allocates
1161	 * a pool, we shouldn't have any problems here.  The user won't be
1162	 * able to unload the CTL module until client modules have
1163	 * successfully unloaded.
1164	 */
1165	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1166		ctl_pool_free(pool);
1167
1168#if 0
1169	ctl_shutdown_thread(softc->work_thread);
1170	mtx_destroy(&softc->queue_lock);
1171#endif
1172
1173	mtx_destroy(&softc->pool_lock);
1174	mtx_destroy(&softc->ctl_lock);
1175
1176	destroy_dev(softc->dev);
1177
1178	sysctl_ctx_free(&softc->sysctl_ctx);
1179
1180	free(control_softc, M_DEVBUF);
1181	control_softc = NULL;
1182
1183	if (bootverbose)
1184		printf("ctl: CAM Target Layer unloaded\n");
1185}
1186
1187static int
1188ctl_module_event_handler(module_t mod, int what, void *arg)
1189{
1190
1191	switch (what) {
1192	case MOD_LOAD:
1193		return (ctl_init());
1194	case MOD_UNLOAD:
1195		return (EBUSY);
1196	default:
1197		return (EOPNOTSUPP);
1198	}
1199}
1200
1201/*
1202 * XXX KDM should we do some access checks here?  Bump a reference count to
1203 * prevent a CTL module from being unloaded while someone has it open?
1204 */
1205static int
1206ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1207{
1208	return (0);
1209}
1210
1211static int
1212ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1213{
1214	return (0);
1215}
1216
1217int
1218ctl_port_enable(ctl_port_type port_type)
1219{
1220	struct ctl_softc *softc;
1221	struct ctl_port *port;
1222
1223	if (ctl_is_single == 0) {
1224		union ctl_ha_msg msg_info;
1225		int isc_retval;
1226
1227#if 0
1228		printf("%s: HA mode, synchronizing frontend enable\n",
1229		        __func__);
1230#endif
1231		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1232	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1233		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1234			printf("Sync msg send error retval %d\n", isc_retval);
1235		}
1236		if (!rcv_sync_msg) {
1237			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1238			        sizeof(msg_info), 1);
1239		}
1240#if 0
1241        	printf("CTL:Frontend Enable\n");
1242	} else {
1243		printf("%s: single mode, skipping frontend synchronization\n",
1244		        __func__);
1245#endif
1246	}
1247
1248	softc = control_softc;
1249
1250	STAILQ_FOREACH(port, &softc->port_list, links) {
1251		if (port_type & port->port_type)
1252		{
1253#if 0
1254			printf("port %d\n", port->targ_port);
1255#endif
1256			ctl_port_online(port);
1257		}
1258	}
1259
1260	return (0);
1261}
1262
1263int
1264ctl_port_disable(ctl_port_type port_type)
1265{
1266	struct ctl_softc *softc;
1267	struct ctl_port *port;
1268
1269	softc = control_softc;
1270
1271	STAILQ_FOREACH(port, &softc->port_list, links) {
1272		if (port_type & port->port_type)
1273			ctl_port_offline(port);
1274	}
1275
1276	return (0);
1277}
1278
1279/*
1280 * Returns 0 for success, 1 for failure.
1281 * Currently the only failure mode is if there aren't enough entries
1282 * allocated.  So, in case of a failure, look at num_entries_dropped,
1283 * reallocate and try again.
1284 */
1285int
1286ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1287	      int *num_entries_filled, int *num_entries_dropped,
1288	      ctl_port_type port_type, int no_virtual)
1289{
1290	struct ctl_softc *softc;
1291	struct ctl_port *port;
1292	int entries_dropped, entries_filled;
1293	int retval;
1294	int i;
1295
1296	softc = control_softc;
1297
1298	retval = 0;
1299	entries_filled = 0;
1300	entries_dropped = 0;
1301
1302	i = 0;
1303	mtx_lock(&softc->ctl_lock);
1304	STAILQ_FOREACH(port, &softc->port_list, links) {
1305		struct ctl_port_entry *entry;
1306
1307		if ((port->port_type & port_type) == 0)
1308			continue;
1309
1310		if ((no_virtual != 0)
1311		 && (port->virtual_port != 0))
1312			continue;
1313
1314		if (entries_filled >= num_entries_alloced) {
1315			entries_dropped++;
1316			continue;
1317		}
1318		entry = &entries[i];
1319
1320		entry->port_type = port->port_type;
1321		strlcpy(entry->port_name, port->port_name,
1322			sizeof(entry->port_name));
1323		entry->physical_port = port->physical_port;
1324		entry->virtual_port = port->virtual_port;
1325		entry->wwnn = port->wwnn;
1326		entry->wwpn = port->wwpn;
1327
1328		i++;
1329		entries_filled++;
1330	}
1331
1332	mtx_unlock(&softc->ctl_lock);
1333
1334	if (entries_dropped > 0)
1335		retval = 1;
1336
1337	*num_entries_dropped = entries_dropped;
1338	*num_entries_filled = entries_filled;
1339
1340	return (retval);
1341}
1342
1343static void
1344ctl_ioctl_online(void *arg)
1345{
1346	struct ctl_ioctl_info *ioctl_info;
1347
1348	ioctl_info = (struct ctl_ioctl_info *)arg;
1349
1350	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1351}
1352
1353static void
1354ctl_ioctl_offline(void *arg)
1355{
1356	struct ctl_ioctl_info *ioctl_info;
1357
1358	ioctl_info = (struct ctl_ioctl_info *)arg;
1359
1360	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1361}
1362
1363/*
1364 * Remove an initiator by port number and initiator ID.
1365 * Returns 0 for success, 1 for failure.
1366 */
1367int
1368ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1369{
1370	struct ctl_softc *softc;
1371
1372	softc = control_softc;
1373
1374	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1375
1376	if ((targ_port < 0)
1377	 || (targ_port > CTL_MAX_PORTS)) {
1378		printf("%s: invalid port number %d\n", __func__, targ_port);
1379		return (1);
1380	}
1381	if (iid > CTL_MAX_INIT_PER_PORT) {
1382		printf("%s: initiator ID %u > maximun %u!\n",
1383		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1384		return (1);
1385	}
1386
1387	mtx_lock(&softc->ctl_lock);
1388
1389	softc->wwpn_iid[targ_port][iid].in_use = 0;
1390
1391	mtx_unlock(&softc->ctl_lock);
1392
1393	return (0);
1394}
1395
1396/*
1397 * Add an initiator to the initiator map.
1398 * Returns 0 for success, 1 for failure.
1399 */
1400int
1401ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1402{
1403	struct ctl_softc *softc;
1404	int retval;
1405
1406	softc = control_softc;
1407
1408	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1409
1410	retval = 0;
1411
1412	if ((targ_port < 0)
1413	 || (targ_port > CTL_MAX_PORTS)) {
1414		printf("%s: invalid port number %d\n", __func__, targ_port);
1415		return (1);
1416	}
1417	if (iid > CTL_MAX_INIT_PER_PORT) {
1418		printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1419		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1420		return (1);
1421	}
1422
1423	mtx_lock(&softc->ctl_lock);
1424
1425	if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1426		/*
1427		 * We don't treat this as an error.
1428		 */
1429		if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1430			printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1431			       __func__, targ_port, iid, (uintmax_t)wwpn);
1432			goto bailout;
1433		}
1434
1435		/*
1436		 * This is an error, but what do we do about it?  The
1437		 * driver is telling us we have a new WWPN for this
1438		 * initiator ID, so we pretty much need to use it.
1439		 */
1440		printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1441		       "still at that address\n", __func__, targ_port, iid,
1442		       (uintmax_t)wwpn,
1443		       (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1444
1445		/*
1446		 * XXX KDM clear have_ca and ua_pending on each LUN for
1447		 * this initiator.
1448		 */
1449	}
1450	softc->wwpn_iid[targ_port][iid].in_use = 1;
1451	softc->wwpn_iid[targ_port][iid].iid = iid;
1452	softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1453	softc->wwpn_iid[targ_port][iid].port = targ_port;
1454
1455bailout:
1456
1457	mtx_unlock(&softc->ctl_lock);
1458
1459	return (retval);
1460}
1461
1462static int
1463ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1464{
1465	return (0);
1466}
1467
1468static int
1469ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1470{
1471	return (0);
1472}
1473
1474/*
1475 * Data movement routine for the CTL ioctl frontend port.
1476 */
1477static int
1478ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1479{
1480	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1481	struct ctl_sg_entry ext_entry, kern_entry;
1482	int ext_sglen, ext_sg_entries, kern_sg_entries;
1483	int ext_sg_start, ext_offset;
1484	int len_to_copy, len_copied;
1485	int kern_watermark, ext_watermark;
1486	int ext_sglist_malloced;
1487	int i, j;
1488
1489	ext_sglist_malloced = 0;
1490	ext_sg_start = 0;
1491	ext_offset = 0;
1492
1493	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1494
1495	/*
1496	 * If this flag is set, fake the data transfer.
1497	 */
1498	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1499		ctsio->ext_data_filled = ctsio->ext_data_len;
1500		goto bailout;
1501	}
1502
1503	/*
1504	 * To simplify things here, if we have a single buffer, stick it in
1505	 * a S/G entry and just make it a single entry S/G list.
1506	 */
1507	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1508		int len_seen;
1509
1510		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1511
1512		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1513							   M_WAITOK);
1514		ext_sglist_malloced = 1;
1515		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1516				   ext_sglen) != 0) {
1517			ctl_set_internal_failure(ctsio,
1518						 /*sks_valid*/ 0,
1519						 /*retry_count*/ 0);
1520			goto bailout;
1521		}
1522		ext_sg_entries = ctsio->ext_sg_entries;
1523		len_seen = 0;
1524		for (i = 0; i < ext_sg_entries; i++) {
1525			if ((len_seen + ext_sglist[i].len) >=
1526			     ctsio->ext_data_filled) {
1527				ext_sg_start = i;
1528				ext_offset = ctsio->ext_data_filled - len_seen;
1529				break;
1530			}
1531			len_seen += ext_sglist[i].len;
1532		}
1533	} else {
1534		ext_sglist = &ext_entry;
1535		ext_sglist->addr = ctsio->ext_data_ptr;
1536		ext_sglist->len = ctsio->ext_data_len;
1537		ext_sg_entries = 1;
1538		ext_sg_start = 0;
1539		ext_offset = ctsio->ext_data_filled;
1540	}
1541
1542	if (ctsio->kern_sg_entries > 0) {
1543		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1544		kern_sg_entries = ctsio->kern_sg_entries;
1545	} else {
1546		kern_sglist = &kern_entry;
1547		kern_sglist->addr = ctsio->kern_data_ptr;
1548		kern_sglist->len = ctsio->kern_data_len;
1549		kern_sg_entries = 1;
1550	}
1551
1552
1553	kern_watermark = 0;
1554	ext_watermark = ext_offset;
1555	len_copied = 0;
1556	for (i = ext_sg_start, j = 0;
1557	     i < ext_sg_entries && j < kern_sg_entries;) {
1558		uint8_t *ext_ptr, *kern_ptr;
1559
1560		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1561				      kern_sglist[j].len - kern_watermark);
1562
1563		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1564		ext_ptr = ext_ptr + ext_watermark;
1565		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1566			/*
1567			 * XXX KDM fix this!
1568			 */
1569			panic("need to implement bus address support");
1570#if 0
1571			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1572#endif
1573		} else
1574			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1575		kern_ptr = kern_ptr + kern_watermark;
1576
1577		kern_watermark += len_to_copy;
1578		ext_watermark += len_to_copy;
1579
1580		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1581		     CTL_FLAG_DATA_IN) {
1582			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1583					 "bytes to user\n", len_to_copy));
1584			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1585					 "to %p\n", kern_ptr, ext_ptr));
1586			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1587				ctl_set_internal_failure(ctsio,
1588							 /*sks_valid*/ 0,
1589							 /*retry_count*/ 0);
1590				goto bailout;
1591			}
1592		} else {
1593			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1594					 "bytes from user\n", len_to_copy));
1595			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1596					 "to %p\n", ext_ptr, kern_ptr));
1597			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1598				ctl_set_internal_failure(ctsio,
1599							 /*sks_valid*/ 0,
1600							 /*retry_count*/0);
1601				goto bailout;
1602			}
1603		}
1604
1605		len_copied += len_to_copy;
1606
1607		if (ext_sglist[i].len == ext_watermark) {
1608			i++;
1609			ext_watermark = 0;
1610		}
1611
1612		if (kern_sglist[j].len == kern_watermark) {
1613			j++;
1614			kern_watermark = 0;
1615		}
1616	}
1617
1618	ctsio->ext_data_filled += len_copied;
1619
1620	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1621			 "kern_sg_entries: %d\n", ext_sg_entries,
1622			 kern_sg_entries));
1623	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1624			 "kern_data_len = %d\n", ctsio->ext_data_len,
1625			 ctsio->kern_data_len));
1626
1627
1628	/* XXX KDM set residual?? */
1629bailout:
1630
1631	if (ext_sglist_malloced != 0)
1632		free(ext_sglist, M_CTL);
1633
1634	return (CTL_RETVAL_COMPLETE);
1635}
1636
1637/*
1638 * Serialize a command that went down the "wrong" side, and so was sent to
1639 * this controller for execution.  The logic is a little different than the
1640 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1641 * sent back to the other side, but in the success case, we execute the
1642 * command on this side (XFER mode) or tell the other side to execute it
1643 * (SER_ONLY mode).
1644 */
1645static int
1646ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1647{
1648	struct ctl_softc *ctl_softc;
1649	union ctl_ha_msg msg_info;
1650	struct ctl_lun *lun;
1651	int retval = 0;
1652	uint32_t targ_lun;
1653
1654	ctl_softc = control_softc;
1655
1656	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1657	lun = ctl_softc->ctl_luns[targ_lun];
1658	if (lun==NULL)
1659	{
1660		/*
1661		 * Why isn't LUN defined? The other side wouldn't
1662		 * send a cmd if the LUN is undefined.
1663		 */
1664		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1665
1666		/* "Logical unit not supported" */
1667		ctl_set_sense_data(&msg_info.scsi.sense_data,
1668				   lun,
1669				   /*sense_format*/SSD_TYPE_NONE,
1670				   /*current_error*/ 1,
1671				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1672				   /*asc*/ 0x25,
1673				   /*ascq*/ 0x00,
1674				   SSD_ELEM_NONE);
1675
1676		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1677		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1678		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1679		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1680		msg_info.hdr.serializing_sc = NULL;
1681		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1682	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1683				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1684		}
1685		return(1);
1686
1687	}
1688
1689	mtx_lock(&lun->lun_lock);
1690    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1691
1692	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1693		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1694		 ooa_links))) {
1695	case CTL_ACTION_BLOCK:
1696		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1697		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1698				  blocked_links);
1699		break;
1700	case CTL_ACTION_PASS:
1701	case CTL_ACTION_SKIP:
1702		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1703			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1704			ctl_enqueue_rtr((union ctl_io *)ctsio);
1705		} else {
1706
1707			/* send msg back to other side */
1708			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1709			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1710			msg_info.hdr.msg_type = CTL_MSG_R2R;
1711#if 0
1712			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1713#endif
1714		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1715			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1716			}
1717		}
1718		break;
1719	case CTL_ACTION_OVERLAP:
1720		/* OVERLAPPED COMMANDS ATTEMPTED */
1721		ctl_set_sense_data(&msg_info.scsi.sense_data,
1722				   lun,
1723				   /*sense_format*/SSD_TYPE_NONE,
1724				   /*current_error*/ 1,
1725				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1726				   /*asc*/ 0x4E,
1727				   /*ascq*/ 0x00,
1728				   SSD_ELEM_NONE);
1729
1730		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1731		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1732		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1733		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1734		msg_info.hdr.serializing_sc = NULL;
1735		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1736#if 0
1737		printf("BAD JUJU:Major Bummer Overlap\n");
1738#endif
1739		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1740		retval = 1;
1741		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1742		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1743		}
1744		break;
1745	case CTL_ACTION_OVERLAP_TAG:
1746		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1747		ctl_set_sense_data(&msg_info.scsi.sense_data,
1748				   lun,
1749				   /*sense_format*/SSD_TYPE_NONE,
1750				   /*current_error*/ 1,
1751				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1752				   /*asc*/ 0x4D,
1753				   /*ascq*/ ctsio->tag_num & 0xff,
1754				   SSD_ELEM_NONE);
1755
1756		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1757		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1758		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1759		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1760		msg_info.hdr.serializing_sc = NULL;
1761		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1762#if 0
1763		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1764#endif
1765		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1766		retval = 1;
1767		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1768		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1769		}
1770		break;
1771	case CTL_ACTION_ERROR:
1772	default:
1773		/* "Internal target failure" */
1774		ctl_set_sense_data(&msg_info.scsi.sense_data,
1775				   lun,
1776				   /*sense_format*/SSD_TYPE_NONE,
1777				   /*current_error*/ 1,
1778				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1779				   /*asc*/ 0x44,
1780				   /*ascq*/ 0x00,
1781				   SSD_ELEM_NONE);
1782
1783		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1784		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1785		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1786		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1787		msg_info.hdr.serializing_sc = NULL;
1788		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1789#if 0
1790		printf("BAD JUJU:Major Bummer HW Error\n");
1791#endif
1792		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1793		retval = 1;
1794		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1795		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1796		}
1797		break;
1798	}
1799	mtx_unlock(&lun->lun_lock);
1800	return (retval);
1801}
1802
1803static int
1804ctl_ioctl_submit_wait(union ctl_io *io)
1805{
1806	struct ctl_fe_ioctl_params params;
1807	ctl_fe_ioctl_state last_state;
1808	int done, retval;
1809
1810	retval = 0;
1811
1812	bzero(&params, sizeof(params));
1813
1814	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1815	cv_init(&params.sem, "ctlioccv");
1816	params.state = CTL_IOCTL_INPROG;
1817	last_state = params.state;
1818
1819	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1820
1821	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1822
1823	/* This shouldn't happen */
1824	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1825		return (retval);
1826
1827	done = 0;
1828
1829	do {
1830		mtx_lock(&params.ioctl_mtx);
1831		/*
1832		 * Check the state here, and don't sleep if the state has
1833		 * already changed (i.e. wakeup has already occured, but we
1834		 * weren't waiting yet).
1835		 */
1836		if (params.state == last_state) {
1837			/* XXX KDM cv_wait_sig instead? */
1838			cv_wait(&params.sem, &params.ioctl_mtx);
1839		}
1840		last_state = params.state;
1841
1842		switch (params.state) {
1843		case CTL_IOCTL_INPROG:
1844			/* Why did we wake up? */
1845			/* XXX KDM error here? */
1846			mtx_unlock(&params.ioctl_mtx);
1847			break;
1848		case CTL_IOCTL_DATAMOVE:
1849			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1850
1851			/*
1852			 * change last_state back to INPROG to avoid
1853			 * deadlock on subsequent data moves.
1854			 */
1855			params.state = last_state = CTL_IOCTL_INPROG;
1856
1857			mtx_unlock(&params.ioctl_mtx);
1858			ctl_ioctl_do_datamove(&io->scsiio);
1859			/*
1860			 * Note that in some cases, most notably writes,
1861			 * this will queue the I/O and call us back later.
1862			 * In other cases, generally reads, this routine
1863			 * will immediately call back and wake us up,
1864			 * probably using our own context.
1865			 */
1866			io->scsiio.be_move_done(io);
1867			break;
1868		case CTL_IOCTL_DONE:
1869			mtx_unlock(&params.ioctl_mtx);
1870			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1871			done = 1;
1872			break;
1873		default:
1874			mtx_unlock(&params.ioctl_mtx);
1875			/* XXX KDM error here? */
1876			break;
1877		}
1878	} while (done == 0);
1879
1880	mtx_destroy(&params.ioctl_mtx);
1881	cv_destroy(&params.sem);
1882
1883	return (CTL_RETVAL_COMPLETE);
1884}
1885
1886static void
1887ctl_ioctl_datamove(union ctl_io *io)
1888{
1889	struct ctl_fe_ioctl_params *params;
1890
1891	params = (struct ctl_fe_ioctl_params *)
1892		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1893
1894	mtx_lock(&params->ioctl_mtx);
1895	params->state = CTL_IOCTL_DATAMOVE;
1896	cv_broadcast(&params->sem);
1897	mtx_unlock(&params->ioctl_mtx);
1898}
1899
1900static void
1901ctl_ioctl_done(union ctl_io *io)
1902{
1903	struct ctl_fe_ioctl_params *params;
1904
1905	params = (struct ctl_fe_ioctl_params *)
1906		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1907
1908	mtx_lock(&params->ioctl_mtx);
1909	params->state = CTL_IOCTL_DONE;
1910	cv_broadcast(&params->sem);
1911	mtx_unlock(&params->ioctl_mtx);
1912}
1913
1914static void
1915ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1916{
1917	struct ctl_fe_ioctl_startstop_info *sd_info;
1918
1919	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1920
1921	sd_info->hs_info.status = metatask->status;
1922	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1923	sd_info->hs_info.luns_complete =
1924		metatask->taskinfo.startstop.luns_complete;
1925	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1926
1927	cv_broadcast(&sd_info->sem);
1928}
1929
1930static void
1931ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1932{
1933	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1934
1935	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1936
1937	mtx_lock(fe_bbr_info->lock);
1938	fe_bbr_info->bbr_info->status = metatask->status;
1939	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1940	fe_bbr_info->wakeup_done = 1;
1941	mtx_unlock(fe_bbr_info->lock);
1942
1943	cv_broadcast(&fe_bbr_info->sem);
1944}
1945
1946/*
1947 * Returns 0 for success, errno for failure.
1948 */
1949static int
1950ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1951		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1952{
1953	union ctl_io *io;
1954	int retval;
1955
1956	retval = 0;
1957
1958	mtx_lock(&lun->lun_lock);
1959	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
1960	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
1961	     ooa_links)) {
1962		struct ctl_ooa_entry *entry;
1963
1964		/*
1965		 * If we've got more than we can fit, just count the
1966		 * remaining entries.
1967		 */
1968		if (*cur_fill_num >= ooa_hdr->alloc_num)
1969			continue;
1970
1971		entry = &kern_entries[*cur_fill_num];
1972
1973		entry->tag_num = io->scsiio.tag_num;
1974		entry->lun_num = lun->lun;
1975#ifdef CTL_TIME_IO
1976		entry->start_bt = io->io_hdr.start_bt;
1977#endif
1978		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
1979		entry->cdb_len = io->scsiio.cdb_len;
1980		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
1981			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
1982
1983		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
1984			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
1985
1986		if (io->io_hdr.flags & CTL_FLAG_ABORT)
1987			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
1988
1989		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
1990			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
1991
1992		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
1993			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
1994	}
1995	mtx_unlock(&lun->lun_lock);
1996
1997	return (retval);
1998}
1999
2000static void *
2001ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2002		 size_t error_str_len)
2003{
2004	void *kptr;
2005
2006	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2007
2008	if (copyin(user_addr, kptr, len) != 0) {
2009		snprintf(error_str, error_str_len, "Error copying %d bytes "
2010			 "from user address %p to kernel address %p", len,
2011			 user_addr, kptr);
2012		free(kptr, M_CTL);
2013		return (NULL);
2014	}
2015
2016	return (kptr);
2017}
2018
2019static void
2020ctl_free_args(int num_args, struct ctl_be_arg *args)
2021{
2022	int i;
2023
2024	if (args == NULL)
2025		return;
2026
2027	for (i = 0; i < num_args; i++) {
2028		free(args[i].kname, M_CTL);
2029		free(args[i].kvalue, M_CTL);
2030	}
2031
2032	free(args, M_CTL);
2033}
2034
2035static struct ctl_be_arg *
2036ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2037		char *error_str, size_t error_str_len)
2038{
2039	struct ctl_be_arg *args;
2040	int i;
2041
2042	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2043				error_str, error_str_len);
2044
2045	if (args == NULL)
2046		goto bailout;
2047
2048	for (i = 0; i < num_args; i++) {
2049		args[i].kname = NULL;
2050		args[i].kvalue = NULL;
2051	}
2052
2053	for (i = 0; i < num_args; i++) {
2054		uint8_t *tmpptr;
2055
2056		args[i].kname = ctl_copyin_alloc(args[i].name,
2057			args[i].namelen, error_str, error_str_len);
2058		if (args[i].kname == NULL)
2059			goto bailout;
2060
2061		if (args[i].kname[args[i].namelen - 1] != '\0') {
2062			snprintf(error_str, error_str_len, "Argument %d "
2063				 "name is not NUL-terminated", i);
2064			goto bailout;
2065		}
2066
2067		if (args[i].flags & CTL_BEARG_RD) {
2068			tmpptr = ctl_copyin_alloc(args[i].value,
2069				args[i].vallen, error_str, error_str_len);
2070			if (tmpptr == NULL)
2071				goto bailout;
2072			if ((args[i].flags & CTL_BEARG_ASCII)
2073			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2074				snprintf(error_str, error_str_len, "Argument "
2075				    "%d value is not NUL-terminated", i);
2076				goto bailout;
2077			}
2078			args[i].kvalue = tmpptr;
2079		} else {
2080			args[i].kvalue = malloc(args[i].vallen,
2081			    M_CTL, M_WAITOK | M_ZERO);
2082		}
2083	}
2084
2085	return (args);
2086bailout:
2087
2088	ctl_free_args(num_args, args);
2089
2090	return (NULL);
2091}
2092
2093static void
2094ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2095{
2096	int i;
2097
2098	for (i = 0; i < num_args; i++) {
2099		if (args[i].flags & CTL_BEARG_WR)
2100			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2101	}
2102}
2103
2104/*
2105 * Escape characters that are illegal or not recommended in XML.
2106 */
2107int
2108ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2109{
2110	int retval;
2111
2112	retval = 0;
2113
2114	for (; *str; str++) {
2115		switch (*str) {
2116		case '&':
2117			retval = sbuf_printf(sb, "&amp;");
2118			break;
2119		case '>':
2120			retval = sbuf_printf(sb, "&gt;");
2121			break;
2122		case '<':
2123			retval = sbuf_printf(sb, "&lt;");
2124			break;
2125		default:
2126			retval = sbuf_putc(sb, *str);
2127			break;
2128		}
2129
2130		if (retval != 0)
2131			break;
2132
2133	}
2134
2135	return (retval);
2136}
2137
2138static int
2139ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2140	  struct thread *td)
2141{
2142	struct ctl_softc *softc;
2143	int retval;
2144
2145	softc = control_softc;
2146
2147	retval = 0;
2148
2149	switch (cmd) {
2150	case CTL_IO: {
2151		union ctl_io *io;
2152		void *pool_tmp;
2153
2154		/*
2155		 * If we haven't been "enabled", don't allow any SCSI I/O
2156		 * to this FETD.
2157		 */
2158		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2159			retval = EPERM;
2160			break;
2161		}
2162
2163		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2164		if (io == NULL) {
2165			printf("ctl_ioctl: can't allocate ctl_io!\n");
2166			retval = ENOSPC;
2167			break;
2168		}
2169
2170		/*
2171		 * Need to save the pool reference so it doesn't get
2172		 * spammed by the user's ctl_io.
2173		 */
2174		pool_tmp = io->io_hdr.pool;
2175
2176		memcpy(io, (void *)addr, sizeof(*io));
2177
2178		io->io_hdr.pool = pool_tmp;
2179		/*
2180		 * No status yet, so make sure the status is set properly.
2181		 */
2182		io->io_hdr.status = CTL_STATUS_NONE;
2183
2184		/*
2185		 * The user sets the initiator ID, target and LUN IDs.
2186		 */
2187		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2188		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2189		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2190		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2191			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2192
2193		retval = ctl_ioctl_submit_wait(io);
2194
2195		if (retval != 0) {
2196			ctl_free_io(io);
2197			break;
2198		}
2199
2200		memcpy((void *)addr, io, sizeof(*io));
2201
2202		/* return this to our pool */
2203		ctl_free_io(io);
2204
2205		break;
2206	}
2207	case CTL_ENABLE_PORT:
2208	case CTL_DISABLE_PORT:
2209	case CTL_SET_PORT_WWNS: {
2210		struct ctl_port *port;
2211		struct ctl_port_entry *entry;
2212
2213		entry = (struct ctl_port_entry *)addr;
2214
2215		mtx_lock(&softc->ctl_lock);
2216		STAILQ_FOREACH(port, &softc->port_list, links) {
2217			int action, done;
2218
2219			action = 0;
2220			done = 0;
2221
2222			if ((entry->port_type == CTL_PORT_NONE)
2223			 && (entry->targ_port == port->targ_port)) {
2224				/*
2225				 * If the user only wants to enable or
2226				 * disable or set WWNs on a specific port,
2227				 * do the operation and we're done.
2228				 */
2229				action = 1;
2230				done = 1;
2231			} else if (entry->port_type & port->port_type) {
2232				/*
2233				 * Compare the user's type mask with the
2234				 * particular frontend type to see if we
2235				 * have a match.
2236				 */
2237				action = 1;
2238				done = 0;
2239
2240				/*
2241				 * Make sure the user isn't trying to set
2242				 * WWNs on multiple ports at the same time.
2243				 */
2244				if (cmd == CTL_SET_PORT_WWNS) {
2245					printf("%s: Can't set WWNs on "
2246					       "multiple ports\n", __func__);
2247					retval = EINVAL;
2248					break;
2249				}
2250			}
2251			if (action != 0) {
2252				/*
2253				 * XXX KDM we have to drop the lock here,
2254				 * because the online/offline operations
2255				 * can potentially block.  We need to
2256				 * reference count the frontends so they
2257				 * can't go away,
2258				 */
2259				mtx_unlock(&softc->ctl_lock);
2260
2261				if (cmd == CTL_ENABLE_PORT) {
2262					struct ctl_lun *lun;
2263
2264					STAILQ_FOREACH(lun, &softc->lun_list,
2265						       links) {
2266						port->lun_enable(port->targ_lun_arg,
2267						    lun->target,
2268						    lun->lun);
2269					}
2270
2271					ctl_port_online(port);
2272				} else if (cmd == CTL_DISABLE_PORT) {
2273					struct ctl_lun *lun;
2274
2275					ctl_port_offline(port);
2276
2277					STAILQ_FOREACH(lun, &softc->lun_list,
2278						       links) {
2279						port->lun_disable(
2280						    port->targ_lun_arg,
2281						    lun->target,
2282						    lun->lun);
2283					}
2284				}
2285
2286				mtx_lock(&softc->ctl_lock);
2287
2288				if (cmd == CTL_SET_PORT_WWNS)
2289					ctl_port_set_wwns(port,
2290					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2291					    1 : 0, entry->wwnn,
2292					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2293					    1 : 0, entry->wwpn);
2294			}
2295			if (done != 0)
2296				break;
2297		}
2298		mtx_unlock(&softc->ctl_lock);
2299		break;
2300	}
2301	case CTL_GET_PORT_LIST: {
2302		struct ctl_port *port;
2303		struct ctl_port_list *list;
2304		int i;
2305
2306		list = (struct ctl_port_list *)addr;
2307
2308		if (list->alloc_len != (list->alloc_num *
2309		    sizeof(struct ctl_port_entry))) {
2310			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2311			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2312			       "%zu\n", __func__, list->alloc_len,
2313			       list->alloc_num, sizeof(struct ctl_port_entry));
2314			retval = EINVAL;
2315			break;
2316		}
2317		list->fill_len = 0;
2318		list->fill_num = 0;
2319		list->dropped_num = 0;
2320		i = 0;
2321		mtx_lock(&softc->ctl_lock);
2322		STAILQ_FOREACH(port, &softc->port_list, links) {
2323			struct ctl_port_entry entry, *list_entry;
2324
2325			if (list->fill_num >= list->alloc_num) {
2326				list->dropped_num++;
2327				continue;
2328			}
2329
2330			entry.port_type = port->port_type;
2331			strlcpy(entry.port_name, port->port_name,
2332				sizeof(entry.port_name));
2333			entry.targ_port = port->targ_port;
2334			entry.physical_port = port->physical_port;
2335			entry.virtual_port = port->virtual_port;
2336			entry.wwnn = port->wwnn;
2337			entry.wwpn = port->wwpn;
2338			if (port->status & CTL_PORT_STATUS_ONLINE)
2339				entry.online = 1;
2340			else
2341				entry.online = 0;
2342
2343			list_entry = &list->entries[i];
2344
2345			retval = copyout(&entry, list_entry, sizeof(entry));
2346			if (retval != 0) {
2347				printf("%s: CTL_GET_PORT_LIST: copyout "
2348				       "returned %d\n", __func__, retval);
2349				break;
2350			}
2351			i++;
2352			list->fill_num++;
2353			list->fill_len += sizeof(entry);
2354		}
2355		mtx_unlock(&softc->ctl_lock);
2356
2357		/*
2358		 * If this is non-zero, we had a copyout fault, so there's
2359		 * probably no point in attempting to set the status inside
2360		 * the structure.
2361		 */
2362		if (retval != 0)
2363			break;
2364
2365		if (list->dropped_num > 0)
2366			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2367		else
2368			list->status = CTL_PORT_LIST_OK;
2369		break;
2370	}
2371	case CTL_DUMP_OOA: {
2372		struct ctl_lun *lun;
2373		union ctl_io *io;
2374		char printbuf[128];
2375		struct sbuf sb;
2376
2377		mtx_lock(&softc->ctl_lock);
2378		printf("Dumping OOA queues:\n");
2379		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2380			mtx_lock(&lun->lun_lock);
2381			for (io = (union ctl_io *)TAILQ_FIRST(
2382			     &lun->ooa_queue); io != NULL;
2383			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2384			     ooa_links)) {
2385				sbuf_new(&sb, printbuf, sizeof(printbuf),
2386					 SBUF_FIXEDLEN);
2387				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2388					    (intmax_t)lun->lun,
2389					    io->scsiio.tag_num,
2390					    (io->io_hdr.flags &
2391					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2392					    (io->io_hdr.flags &
2393					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2394					    (io->io_hdr.flags &
2395					    CTL_FLAG_ABORT) ? " ABORT" : "",
2396			                    (io->io_hdr.flags &
2397		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2398				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2399				sbuf_finish(&sb);
2400				printf("%s\n", sbuf_data(&sb));
2401			}
2402			mtx_unlock(&lun->lun_lock);
2403		}
2404		printf("OOA queues dump done\n");
2405		mtx_unlock(&softc->ctl_lock);
2406		break;
2407	}
2408	case CTL_GET_OOA: {
2409		struct ctl_lun *lun;
2410		struct ctl_ooa *ooa_hdr;
2411		struct ctl_ooa_entry *entries;
2412		uint32_t cur_fill_num;
2413
2414		ooa_hdr = (struct ctl_ooa *)addr;
2415
2416		if ((ooa_hdr->alloc_len == 0)
2417		 || (ooa_hdr->alloc_num == 0)) {
2418			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2419			       "must be non-zero\n", __func__,
2420			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2421			retval = EINVAL;
2422			break;
2423		}
2424
2425		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2426		    sizeof(struct ctl_ooa_entry))) {
2427			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2428			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2429			       __func__, ooa_hdr->alloc_len,
2430			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2431			retval = EINVAL;
2432			break;
2433		}
2434
2435		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2436		if (entries == NULL) {
2437			printf("%s: could not allocate %d bytes for OOA "
2438			       "dump\n", __func__, ooa_hdr->alloc_len);
2439			retval = ENOMEM;
2440			break;
2441		}
2442
2443		mtx_lock(&softc->ctl_lock);
2444		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2445		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2446		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2447			mtx_unlock(&softc->ctl_lock);
2448			free(entries, M_CTL);
2449			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2450			       __func__, (uintmax_t)ooa_hdr->lun_num);
2451			retval = EINVAL;
2452			break;
2453		}
2454
2455		cur_fill_num = 0;
2456
2457		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2458			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2459				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2460					ooa_hdr, entries);
2461				if (retval != 0)
2462					break;
2463			}
2464			if (retval != 0) {
2465				mtx_unlock(&softc->ctl_lock);
2466				free(entries, M_CTL);
2467				break;
2468			}
2469		} else {
2470			lun = softc->ctl_luns[ooa_hdr->lun_num];
2471
2472			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2473						    entries);
2474		}
2475		mtx_unlock(&softc->ctl_lock);
2476
2477		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2478		ooa_hdr->fill_len = ooa_hdr->fill_num *
2479			sizeof(struct ctl_ooa_entry);
2480		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2481		if (retval != 0) {
2482			printf("%s: error copying out %d bytes for OOA dump\n",
2483			       __func__, ooa_hdr->fill_len);
2484		}
2485
2486		getbintime(&ooa_hdr->cur_bt);
2487
2488		if (cur_fill_num > ooa_hdr->alloc_num) {
2489			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2490			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2491		} else {
2492			ooa_hdr->dropped_num = 0;
2493			ooa_hdr->status = CTL_OOA_OK;
2494		}
2495
2496		free(entries, M_CTL);
2497		break;
2498	}
2499	case CTL_CHECK_OOA: {
2500		union ctl_io *io;
2501		struct ctl_lun *lun;
2502		struct ctl_ooa_info *ooa_info;
2503
2504
2505		ooa_info = (struct ctl_ooa_info *)addr;
2506
2507		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2508			ooa_info->status = CTL_OOA_INVALID_LUN;
2509			break;
2510		}
2511		mtx_lock(&softc->ctl_lock);
2512		lun = softc->ctl_luns[ooa_info->lun_id];
2513		if (lun == NULL) {
2514			mtx_unlock(&softc->ctl_lock);
2515			ooa_info->status = CTL_OOA_INVALID_LUN;
2516			break;
2517		}
2518		mtx_lock(&lun->lun_lock);
2519		mtx_unlock(&softc->ctl_lock);
2520		ooa_info->num_entries = 0;
2521		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2522		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2523		     &io->io_hdr, ooa_links)) {
2524			ooa_info->num_entries++;
2525		}
2526		mtx_unlock(&lun->lun_lock);
2527
2528		ooa_info->status = CTL_OOA_SUCCESS;
2529
2530		break;
2531	}
2532	case CTL_HARD_START:
2533	case CTL_HARD_STOP: {
2534		struct ctl_fe_ioctl_startstop_info ss_info;
2535		struct cfi_metatask *metatask;
2536		struct mtx hs_mtx;
2537
2538		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2539
2540		cv_init(&ss_info.sem, "hard start/stop cv" );
2541
2542		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2543		if (metatask == NULL) {
2544			retval = ENOMEM;
2545			mtx_destroy(&hs_mtx);
2546			break;
2547		}
2548
2549		if (cmd == CTL_HARD_START)
2550			metatask->tasktype = CFI_TASK_STARTUP;
2551		else
2552			metatask->tasktype = CFI_TASK_SHUTDOWN;
2553
2554		metatask->callback = ctl_ioctl_hard_startstop_callback;
2555		metatask->callback_arg = &ss_info;
2556
2557		cfi_action(metatask);
2558
2559		/* Wait for the callback */
2560		mtx_lock(&hs_mtx);
2561		cv_wait_sig(&ss_info.sem, &hs_mtx);
2562		mtx_unlock(&hs_mtx);
2563
2564		/*
2565		 * All information has been copied from the metatask by the
2566		 * time cv_broadcast() is called, so we free the metatask here.
2567		 */
2568		cfi_free_metatask(metatask);
2569
2570		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2571
2572		mtx_destroy(&hs_mtx);
2573		break;
2574	}
2575	case CTL_BBRREAD: {
2576		struct ctl_bbrread_info *bbr_info;
2577		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2578		struct mtx bbr_mtx;
2579		struct cfi_metatask *metatask;
2580
2581		bbr_info = (struct ctl_bbrread_info *)addr;
2582
2583		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2584
2585		bzero(&bbr_mtx, sizeof(bbr_mtx));
2586		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2587
2588		fe_bbr_info.bbr_info = bbr_info;
2589		fe_bbr_info.lock = &bbr_mtx;
2590
2591		cv_init(&fe_bbr_info.sem, "BBR read cv");
2592		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2593
2594		if (metatask == NULL) {
2595			mtx_destroy(&bbr_mtx);
2596			cv_destroy(&fe_bbr_info.sem);
2597			retval = ENOMEM;
2598			break;
2599		}
2600		metatask->tasktype = CFI_TASK_BBRREAD;
2601		metatask->callback = ctl_ioctl_bbrread_callback;
2602		metatask->callback_arg = &fe_bbr_info;
2603		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2604		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2605		metatask->taskinfo.bbrread.len = bbr_info->len;
2606
2607		cfi_action(metatask);
2608
2609		mtx_lock(&bbr_mtx);
2610		while (fe_bbr_info.wakeup_done == 0)
2611			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2612		mtx_unlock(&bbr_mtx);
2613
2614		bbr_info->status = metatask->status;
2615		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2616		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2617		memcpy(&bbr_info->sense_data,
2618		       &metatask->taskinfo.bbrread.sense_data,
2619		       ctl_min(sizeof(bbr_info->sense_data),
2620			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2621
2622		cfi_free_metatask(metatask);
2623
2624		mtx_destroy(&bbr_mtx);
2625		cv_destroy(&fe_bbr_info.sem);
2626
2627		break;
2628	}
2629	case CTL_DELAY_IO: {
2630		struct ctl_io_delay_info *delay_info;
2631#ifdef CTL_IO_DELAY
2632		struct ctl_lun *lun;
2633#endif /* CTL_IO_DELAY */
2634
2635		delay_info = (struct ctl_io_delay_info *)addr;
2636
2637#ifdef CTL_IO_DELAY
2638		mtx_lock(&softc->ctl_lock);
2639
2640		if ((delay_info->lun_id > CTL_MAX_LUNS)
2641		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2642			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2643		} else {
2644			lun = softc->ctl_luns[delay_info->lun_id];
2645			mtx_lock(&lun->lun_lock);
2646
2647			delay_info->status = CTL_DELAY_STATUS_OK;
2648
2649			switch (delay_info->delay_type) {
2650			case CTL_DELAY_TYPE_CONT:
2651				break;
2652			case CTL_DELAY_TYPE_ONESHOT:
2653				break;
2654			default:
2655				delay_info->status =
2656					CTL_DELAY_STATUS_INVALID_TYPE;
2657				break;
2658			}
2659
2660			switch (delay_info->delay_loc) {
2661			case CTL_DELAY_LOC_DATAMOVE:
2662				lun->delay_info.datamove_type =
2663					delay_info->delay_type;
2664				lun->delay_info.datamove_delay =
2665					delay_info->delay_secs;
2666				break;
2667			case CTL_DELAY_LOC_DONE:
2668				lun->delay_info.done_type =
2669					delay_info->delay_type;
2670				lun->delay_info.done_delay =
2671					delay_info->delay_secs;
2672				break;
2673			default:
2674				delay_info->status =
2675					CTL_DELAY_STATUS_INVALID_LOC;
2676				break;
2677			}
2678			mtx_unlock(&lun->lun_lock);
2679		}
2680
2681		mtx_unlock(&softc->ctl_lock);
2682#else
2683		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2684#endif /* CTL_IO_DELAY */
2685		break;
2686	}
2687	case CTL_REALSYNC_SET: {
2688		int *syncstate;
2689
2690		syncstate = (int *)addr;
2691
2692		mtx_lock(&softc->ctl_lock);
2693		switch (*syncstate) {
2694		case 0:
2695			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2696			break;
2697		case 1:
2698			softc->flags |= CTL_FLAG_REAL_SYNC;
2699			break;
2700		default:
2701			retval = EINVAL;
2702			break;
2703		}
2704		mtx_unlock(&softc->ctl_lock);
2705		break;
2706	}
2707	case CTL_REALSYNC_GET: {
2708		int *syncstate;
2709
2710		syncstate = (int*)addr;
2711
2712		mtx_lock(&softc->ctl_lock);
2713		if (softc->flags & CTL_FLAG_REAL_SYNC)
2714			*syncstate = 1;
2715		else
2716			*syncstate = 0;
2717		mtx_unlock(&softc->ctl_lock);
2718
2719		break;
2720	}
2721	case CTL_SETSYNC:
2722	case CTL_GETSYNC: {
2723		struct ctl_sync_info *sync_info;
2724		struct ctl_lun *lun;
2725
2726		sync_info = (struct ctl_sync_info *)addr;
2727
2728		mtx_lock(&softc->ctl_lock);
2729		lun = softc->ctl_luns[sync_info->lun_id];
2730		if (lun == NULL) {
2731			mtx_unlock(&softc->ctl_lock);
2732			sync_info->status = CTL_GS_SYNC_NO_LUN;
2733		}
2734		/*
2735		 * Get or set the sync interval.  We're not bounds checking
2736		 * in the set case, hopefully the user won't do something
2737		 * silly.
2738		 */
2739		mtx_lock(&lun->lun_lock);
2740		mtx_unlock(&softc->ctl_lock);
2741		if (cmd == CTL_GETSYNC)
2742			sync_info->sync_interval = lun->sync_interval;
2743		else
2744			lun->sync_interval = sync_info->sync_interval;
2745		mtx_unlock(&lun->lun_lock);
2746
2747		sync_info->status = CTL_GS_SYNC_OK;
2748
2749		break;
2750	}
2751	case CTL_GETSTATS: {
2752		struct ctl_stats *stats;
2753		struct ctl_lun *lun;
2754		int i;
2755
2756		stats = (struct ctl_stats *)addr;
2757
2758		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2759		     stats->alloc_len) {
2760			stats->status = CTL_SS_NEED_MORE_SPACE;
2761			stats->num_luns = softc->num_luns;
2762			break;
2763		}
2764		/*
2765		 * XXX KDM no locking here.  If the LUN list changes,
2766		 * things can blow up.
2767		 */
2768		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2769		     i++, lun = STAILQ_NEXT(lun, links)) {
2770			retval = copyout(&lun->stats, &stats->lun_stats[i],
2771					 sizeof(lun->stats));
2772			if (retval != 0)
2773				break;
2774		}
2775		stats->num_luns = softc->num_luns;
2776		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2777				 softc->num_luns;
2778		stats->status = CTL_SS_OK;
2779#ifdef CTL_TIME_IO
2780		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2781#else
2782		stats->flags = CTL_STATS_FLAG_NONE;
2783#endif
2784		getnanouptime(&stats->timestamp);
2785		break;
2786	}
2787	case CTL_ERROR_INJECT: {
2788		struct ctl_error_desc *err_desc, *new_err_desc;
2789		struct ctl_lun *lun;
2790
2791		err_desc = (struct ctl_error_desc *)addr;
2792
2793		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2794				      M_WAITOK | M_ZERO);
2795		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2796
2797		mtx_lock(&softc->ctl_lock);
2798		lun = softc->ctl_luns[err_desc->lun_id];
2799		if (lun == NULL) {
2800			mtx_unlock(&softc->ctl_lock);
2801			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2802			       __func__, (uintmax_t)err_desc->lun_id);
2803			retval = EINVAL;
2804			break;
2805		}
2806		mtx_lock(&lun->lun_lock);
2807		mtx_unlock(&softc->ctl_lock);
2808
2809		/*
2810		 * We could do some checking here to verify the validity
2811		 * of the request, but given the complexity of error
2812		 * injection requests, the checking logic would be fairly
2813		 * complex.
2814		 *
2815		 * For now, if the request is invalid, it just won't get
2816		 * executed and might get deleted.
2817		 */
2818		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2819
2820		/*
2821		 * XXX KDM check to make sure the serial number is unique,
2822		 * in case we somehow manage to wrap.  That shouldn't
2823		 * happen for a very long time, but it's the right thing to
2824		 * do.
2825		 */
2826		new_err_desc->serial = lun->error_serial;
2827		err_desc->serial = lun->error_serial;
2828		lun->error_serial++;
2829
2830		mtx_unlock(&lun->lun_lock);
2831		break;
2832	}
2833	case CTL_ERROR_INJECT_DELETE: {
2834		struct ctl_error_desc *delete_desc, *desc, *desc2;
2835		struct ctl_lun *lun;
2836		int delete_done;
2837
2838		delete_desc = (struct ctl_error_desc *)addr;
2839		delete_done = 0;
2840
2841		mtx_lock(&softc->ctl_lock);
2842		lun = softc->ctl_luns[delete_desc->lun_id];
2843		if (lun == NULL) {
2844			mtx_unlock(&softc->ctl_lock);
2845			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2846			       __func__, (uintmax_t)delete_desc->lun_id);
2847			retval = EINVAL;
2848			break;
2849		}
2850		mtx_lock(&lun->lun_lock);
2851		mtx_unlock(&softc->ctl_lock);
2852		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2853			if (desc->serial != delete_desc->serial)
2854				continue;
2855
2856			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2857				      links);
2858			free(desc, M_CTL);
2859			delete_done = 1;
2860		}
2861		mtx_unlock(&lun->lun_lock);
2862		if (delete_done == 0) {
2863			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2864			       "error serial %ju on LUN %u\n", __func__,
2865			       delete_desc->serial, delete_desc->lun_id);
2866			retval = EINVAL;
2867			break;
2868		}
2869		break;
2870	}
2871	case CTL_DUMP_STRUCTS: {
2872		int i, j, k;
2873		struct ctl_port *port;
2874		struct ctl_frontend *fe;
2875
2876		printf("CTL IID to WWPN map start:\n");
2877		for (i = 0; i < CTL_MAX_PORTS; i++) {
2878			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2879				if (softc->wwpn_iid[i][j].in_use == 0)
2880					continue;
2881
2882				printf("port %d iid %u WWPN %#jx\n",
2883				       softc->wwpn_iid[i][j].port,
2884				       softc->wwpn_iid[i][j].iid,
2885				       (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2886			}
2887		}
2888		printf("CTL IID to WWPN map end\n");
2889		printf("CTL Persistent Reservation information start:\n");
2890		for (i = 0; i < CTL_MAX_LUNS; i++) {
2891			struct ctl_lun *lun;
2892
2893			lun = softc->ctl_luns[i];
2894
2895			if ((lun == NULL)
2896			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2897				continue;
2898
2899			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2900				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2901					if (lun->per_res[j+k].registered == 0)
2902						continue;
2903					printf("LUN %d port %d iid %d key "
2904					       "%#jx\n", i, j, k,
2905					       (uintmax_t)scsi_8btou64(
2906					       lun->per_res[j+k].res_key.key));
2907				}
2908			}
2909		}
2910		printf("CTL Persistent Reservation information end\n");
2911		printf("CTL Ports:\n");
2912		/*
2913		 * XXX KDM calling this without a lock.  We'd likely want
2914		 * to drop the lock before calling the frontend's dump
2915		 * routine anyway.
2916		 */
2917		STAILQ_FOREACH(port, &softc->port_list, links) {
2918			printf("Port %s Frontend %s Type %u pport %d vport %d WWNN "
2919			       "%#jx WWPN %#jx\n", port->port_name,
2920			       port->frontend->name, port->port_type,
2921			       port->physical_port, port->virtual_port,
2922			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2923		}
2924		printf("CTL Port information end\n");
2925		printf("CTL Frontends:\n");
2926		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2927			printf("Frontend %s\n", fe->name);
2928			if (fe->fe_dump != NULL)
2929				fe->fe_dump();
2930		}
2931		printf("CTL Frontend information end\n");
2932		break;
2933	}
2934	case CTL_LUN_REQ: {
2935		struct ctl_lun_req *lun_req;
2936		struct ctl_backend_driver *backend;
2937
2938		lun_req = (struct ctl_lun_req *)addr;
2939
2940		backend = ctl_backend_find(lun_req->backend);
2941		if (backend == NULL) {
2942			lun_req->status = CTL_LUN_ERROR;
2943			snprintf(lun_req->error_str,
2944				 sizeof(lun_req->error_str),
2945				 "Backend \"%s\" not found.",
2946				 lun_req->backend);
2947			break;
2948		}
2949		if (lun_req->num_be_args > 0) {
2950			lun_req->kern_be_args = ctl_copyin_args(
2951				lun_req->num_be_args,
2952				lun_req->be_args,
2953				lun_req->error_str,
2954				sizeof(lun_req->error_str));
2955			if (lun_req->kern_be_args == NULL) {
2956				lun_req->status = CTL_LUN_ERROR;
2957				break;
2958			}
2959		}
2960
2961		retval = backend->ioctl(dev, cmd, addr, flag, td);
2962
2963		if (lun_req->num_be_args > 0) {
2964			ctl_copyout_args(lun_req->num_be_args,
2965				      lun_req->kern_be_args);
2966			ctl_free_args(lun_req->num_be_args,
2967				      lun_req->kern_be_args);
2968		}
2969		break;
2970	}
2971	case CTL_LUN_LIST: {
2972		struct sbuf *sb;
2973		struct ctl_lun *lun;
2974		struct ctl_lun_list *list;
2975		struct ctl_option *opt;
2976
2977		list = (struct ctl_lun_list *)addr;
2978
2979		/*
2980		 * Allocate a fixed length sbuf here, based on the length
2981		 * of the user's buffer.  We could allocate an auto-extending
2982		 * buffer, and then tell the user how much larger our
2983		 * amount of data is than his buffer, but that presents
2984		 * some problems:
2985		 *
2986		 * 1.  The sbuf(9) routines use a blocking malloc, and so
2987		 *     we can't hold a lock while calling them with an
2988		 *     auto-extending buffer.
2989 		 *
2990		 * 2.  There is not currently a LUN reference counting
2991		 *     mechanism, outside of outstanding transactions on
2992		 *     the LUN's OOA queue.  So a LUN could go away on us
2993		 *     while we're getting the LUN number, backend-specific
2994		 *     information, etc.  Thus, given the way things
2995		 *     currently work, we need to hold the CTL lock while
2996		 *     grabbing LUN information.
2997		 *
2998		 * So, from the user's standpoint, the best thing to do is
2999		 * allocate what he thinks is a reasonable buffer length,
3000		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3001		 * double the buffer length and try again.  (And repeat
3002		 * that until he succeeds.)
3003		 */
3004		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3005		if (sb == NULL) {
3006			list->status = CTL_LUN_LIST_ERROR;
3007			snprintf(list->error_str, sizeof(list->error_str),
3008				 "Unable to allocate %d bytes for LUN list",
3009				 list->alloc_len);
3010			break;
3011		}
3012
3013		sbuf_printf(sb, "<ctllunlist>\n");
3014
3015		mtx_lock(&softc->ctl_lock);
3016		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3017			mtx_lock(&lun->lun_lock);
3018			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3019					     (uintmax_t)lun->lun);
3020
3021			/*
3022			 * Bail out as soon as we see that we've overfilled
3023			 * the buffer.
3024			 */
3025			if (retval != 0)
3026				break;
3027
3028			retval = sbuf_printf(sb, "\t<backend_type>%s"
3029					     "</backend_type>\n",
3030					     (lun->backend == NULL) ?  "none" :
3031					     lun->backend->name);
3032
3033			if (retval != 0)
3034				break;
3035
3036			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3037					     lun->be_lun->lun_type);
3038
3039			if (retval != 0)
3040				break;
3041
3042			if (lun->backend == NULL) {
3043				retval = sbuf_printf(sb, "</lun>\n");
3044				if (retval != 0)
3045					break;
3046				continue;
3047			}
3048
3049			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3050					     (lun->be_lun->maxlba > 0) ?
3051					     lun->be_lun->maxlba + 1 : 0);
3052
3053			if (retval != 0)
3054				break;
3055
3056			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3057					     lun->be_lun->blocksize);
3058
3059			if (retval != 0)
3060				break;
3061
3062			retval = sbuf_printf(sb, "\t<serial_number>");
3063
3064			if (retval != 0)
3065				break;
3066
3067			retval = ctl_sbuf_printf_esc(sb,
3068						     lun->be_lun->serial_num);
3069
3070			if (retval != 0)
3071				break;
3072
3073			retval = sbuf_printf(sb, "</serial_number>\n");
3074
3075			if (retval != 0)
3076				break;
3077
3078			retval = sbuf_printf(sb, "\t<device_id>");
3079
3080			if (retval != 0)
3081				break;
3082
3083			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3084
3085			if (retval != 0)
3086				break;
3087
3088			retval = sbuf_printf(sb, "</device_id>\n");
3089
3090			if (retval != 0)
3091				break;
3092
3093			if (lun->backend->lun_info != NULL) {
3094				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3095				if (retval != 0)
3096					break;
3097			}
3098			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3099				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3100				    opt->name, opt->value, opt->name);
3101				if (retval != 0)
3102					break;
3103			}
3104
3105			retval = sbuf_printf(sb, "</lun>\n");
3106
3107			if (retval != 0)
3108				break;
3109			mtx_unlock(&lun->lun_lock);
3110		}
3111		if (lun != NULL)
3112			mtx_unlock(&lun->lun_lock);
3113		mtx_unlock(&softc->ctl_lock);
3114
3115		if ((retval != 0)
3116		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3117			retval = 0;
3118			sbuf_delete(sb);
3119			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3120			snprintf(list->error_str, sizeof(list->error_str),
3121				 "Out of space, %d bytes is too small",
3122				 list->alloc_len);
3123			break;
3124		}
3125
3126		sbuf_finish(sb);
3127
3128		retval = copyout(sbuf_data(sb), list->lun_xml,
3129				 sbuf_len(sb) + 1);
3130
3131		list->fill_len = sbuf_len(sb) + 1;
3132		list->status = CTL_LUN_LIST_OK;
3133		sbuf_delete(sb);
3134		break;
3135	}
3136	case CTL_ISCSI: {
3137		struct ctl_iscsi *ci;
3138		struct ctl_frontend *fe;
3139
3140		ci = (struct ctl_iscsi *)addr;
3141
3142		fe = ctl_frontend_find("iscsi");
3143		if (fe == NULL) {
3144			ci->status = CTL_ISCSI_ERROR;
3145			snprintf(ci->error_str, sizeof(ci->error_str),
3146			    "Frontend \"iscsi\" not found.");
3147			break;
3148		}
3149
3150		retval = fe->ioctl(dev, cmd, addr, flag, td);
3151		break;
3152	}
3153	case CTL_PORT_REQ: {
3154		struct ctl_req *req;
3155		struct ctl_frontend *fe;
3156
3157		req = (struct ctl_req *)addr;
3158
3159		fe = ctl_frontend_find(req->driver);
3160		if (fe == NULL) {
3161			req->status = CTL_LUN_ERROR;
3162			snprintf(req->error_str, sizeof(req->error_str),
3163			    "Frontend \"%s\" not found.", req->driver);
3164			break;
3165		}
3166		if (req->num_args > 0) {
3167			req->kern_args = ctl_copyin_args(req->num_args,
3168			    req->args, req->error_str, sizeof(req->error_str));
3169			if (req->kern_args == NULL) {
3170				req->status = CTL_LUN_ERROR;
3171				break;
3172			}
3173		}
3174
3175		retval = fe->ioctl(dev, cmd, addr, flag, td);
3176
3177		if (req->num_args > 0) {
3178			ctl_copyout_args(req->num_args, req->kern_args);
3179			ctl_free_args(req->num_args, req->kern_args);
3180		}
3181		break;
3182	}
3183	case CTL_PORT_LIST: {
3184		struct sbuf *sb;
3185		struct ctl_port *port;
3186		struct ctl_lun_list *list;
3187		struct ctl_option *opt;
3188
3189		list = (struct ctl_lun_list *)addr;
3190
3191		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3192		if (sb == NULL) {
3193			list->status = CTL_LUN_LIST_ERROR;
3194			snprintf(list->error_str, sizeof(list->error_str),
3195				 "Unable to allocate %d bytes for LUN list",
3196				 list->alloc_len);
3197			break;
3198		}
3199
3200		sbuf_printf(sb, "<ctlportlist>\n");
3201
3202		mtx_lock(&softc->ctl_lock);
3203		STAILQ_FOREACH(port, &softc->port_list, links) {
3204			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3205					     (uintmax_t)port->targ_port);
3206
3207			/*
3208			 * Bail out as soon as we see that we've overfilled
3209			 * the buffer.
3210			 */
3211			if (retval != 0)
3212				break;
3213
3214			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3215			    "</frontend_type>\n", port->frontend->name);
3216			if (retval != 0)
3217				break;
3218
3219			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3220					     port->port_type);
3221			if (retval != 0)
3222				break;
3223
3224			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3225			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3226			if (retval != 0)
3227				break;
3228
3229			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3230			    port->port_name);
3231			if (retval != 0)
3232				break;
3233
3234			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3235			    port->physical_port);
3236			if (retval != 0)
3237				break;
3238
3239			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3240			    port->virtual_port);
3241			if (retval != 0)
3242				break;
3243
3244			retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3245			    (uintmax_t)port->wwnn);
3246			if (retval != 0)
3247				break;
3248
3249			retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3250			    (uintmax_t)port->wwpn);
3251			if (retval != 0)
3252				break;
3253
3254			STAILQ_FOREACH(opt, &port->options, links) {
3255				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3256				    opt->name, opt->value, opt->name);
3257				if (retval != 0)
3258					break;
3259			}
3260
3261			retval = sbuf_printf(sb, "</targ_port>\n");
3262			if (retval != 0)
3263				break;
3264		}
3265		mtx_unlock(&softc->ctl_lock);
3266
3267		if ((retval != 0)
3268		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3269			retval = 0;
3270			sbuf_delete(sb);
3271			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3272			snprintf(list->error_str, sizeof(list->error_str),
3273				 "Out of space, %d bytes is too small",
3274				 list->alloc_len);
3275			break;
3276		}
3277
3278		sbuf_finish(sb);
3279
3280		retval = copyout(sbuf_data(sb), list->lun_xml,
3281				 sbuf_len(sb) + 1);
3282
3283		list->fill_len = sbuf_len(sb) + 1;
3284		list->status = CTL_LUN_LIST_OK;
3285		sbuf_delete(sb);
3286		break;
3287	}
3288	default: {
3289		/* XXX KDM should we fix this? */
3290#if 0
3291		struct ctl_backend_driver *backend;
3292		unsigned int type;
3293		int found;
3294
3295		found = 0;
3296
3297		/*
3298		 * We encode the backend type as the ioctl type for backend
3299		 * ioctls.  So parse it out here, and then search for a
3300		 * backend of this type.
3301		 */
3302		type = _IOC_TYPE(cmd);
3303
3304		STAILQ_FOREACH(backend, &softc->be_list, links) {
3305			if (backend->type == type) {
3306				found = 1;
3307				break;
3308			}
3309		}
3310		if (found == 0) {
3311			printf("ctl: unknown ioctl command %#lx or backend "
3312			       "%d\n", cmd, type);
3313			retval = EINVAL;
3314			break;
3315		}
3316		retval = backend->ioctl(dev, cmd, addr, flag, td);
3317#endif
3318		retval = ENOTTY;
3319		break;
3320	}
3321	}
3322	return (retval);
3323}
3324
3325uint32_t
3326ctl_get_initindex(struct ctl_nexus *nexus)
3327{
3328	if (nexus->targ_port < CTL_MAX_PORTS)
3329		return (nexus->initid.id +
3330			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3331	else
3332		return (nexus->initid.id +
3333		       ((nexus->targ_port - CTL_MAX_PORTS) *
3334			CTL_MAX_INIT_PER_PORT));
3335}
3336
3337uint32_t
3338ctl_get_resindex(struct ctl_nexus *nexus)
3339{
3340	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3341}
3342
3343uint32_t
3344ctl_port_idx(int port_num)
3345{
3346	if (port_num < CTL_MAX_PORTS)
3347		return(port_num);
3348	else
3349		return(port_num - CTL_MAX_PORTS);
3350}
3351
3352static uint32_t
3353ctl_map_lun(int port_num, uint32_t lun_id)
3354{
3355	struct ctl_port *port;
3356
3357	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3358	if (port == NULL)
3359		return (UINT32_MAX);
3360	if (port->lun_map == NULL)
3361		return (lun_id);
3362	return (port->lun_map(port->targ_lun_arg, lun_id));
3363}
3364
3365static uint32_t
3366ctl_map_lun_back(int port_num, uint32_t lun_id)
3367{
3368	struct ctl_port *port;
3369	uint32_t i;
3370
3371	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3372	if (port->lun_map == NULL)
3373		return (lun_id);
3374	for (i = 0; i < CTL_MAX_LUNS; i++) {
3375		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3376			return (i);
3377	}
3378	return (UINT32_MAX);
3379}
3380
3381/*
3382 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3383 * that are a power of 2.
3384 */
3385int
3386ctl_ffz(uint32_t *mask, uint32_t size)
3387{
3388	uint32_t num_chunks, num_pieces;
3389	int i, j;
3390
3391	num_chunks = (size >> 5);
3392	if (num_chunks == 0)
3393		num_chunks++;
3394	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3395
3396	for (i = 0; i < num_chunks; i++) {
3397		for (j = 0; j < num_pieces; j++) {
3398			if ((mask[i] & (1 << j)) == 0)
3399				return ((i << 5) + j);
3400		}
3401	}
3402
3403	return (-1);
3404}
3405
3406int
3407ctl_set_mask(uint32_t *mask, uint32_t bit)
3408{
3409	uint32_t chunk, piece;
3410
3411	chunk = bit >> 5;
3412	piece = bit % (sizeof(uint32_t) * 8);
3413
3414	if ((mask[chunk] & (1 << piece)) != 0)
3415		return (-1);
3416	else
3417		mask[chunk] |= (1 << piece);
3418
3419	return (0);
3420}
3421
3422int
3423ctl_clear_mask(uint32_t *mask, uint32_t bit)
3424{
3425	uint32_t chunk, piece;
3426
3427	chunk = bit >> 5;
3428	piece = bit % (sizeof(uint32_t) * 8);
3429
3430	if ((mask[chunk] & (1 << piece)) == 0)
3431		return (-1);
3432	else
3433		mask[chunk] &= ~(1 << piece);
3434
3435	return (0);
3436}
3437
3438int
3439ctl_is_set(uint32_t *mask, uint32_t bit)
3440{
3441	uint32_t chunk, piece;
3442
3443	chunk = bit >> 5;
3444	piece = bit % (sizeof(uint32_t) * 8);
3445
3446	if ((mask[chunk] & (1 << piece)) == 0)
3447		return (0);
3448	else
3449		return (1);
3450}
3451
3452#ifdef unused
3453/*
3454 * The bus, target and lun are optional, they can be filled in later.
3455 * can_wait is used to determine whether we can wait on the malloc or not.
3456 */
3457union ctl_io*
3458ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3459	      uint32_t targ_lun, int can_wait)
3460{
3461	union ctl_io *io;
3462
3463	if (can_wait)
3464		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3465	else
3466		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3467
3468	if (io != NULL) {
3469		io->io_hdr.io_type = io_type;
3470		io->io_hdr.targ_port = targ_port;
3471		/*
3472		 * XXX KDM this needs to change/go away.  We need to move
3473		 * to a preallocated pool of ctl_scsiio structures.
3474		 */
3475		io->io_hdr.nexus.targ_target.id = targ_target;
3476		io->io_hdr.nexus.targ_lun = targ_lun;
3477	}
3478
3479	return (io);
3480}
3481
3482void
3483ctl_kfree_io(union ctl_io *io)
3484{
3485	free(io, M_CTL);
3486}
3487#endif /* unused */
3488
3489/*
3490 * ctl_softc, pool_type, total_ctl_io are passed in.
3491 * npool is passed out.
3492 */
3493int
3494ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3495		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3496{
3497	uint32_t i;
3498	union ctl_io *cur_io, *next_io;
3499	struct ctl_io_pool *pool;
3500	int retval;
3501
3502	retval = 0;
3503
3504	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3505					    M_NOWAIT | M_ZERO);
3506	if (pool == NULL) {
3507		retval = ENOMEM;
3508		goto bailout;
3509	}
3510
3511	pool->type = pool_type;
3512	pool->ctl_softc = ctl_softc;
3513
3514	mtx_lock(&ctl_softc->pool_lock);
3515	pool->id = ctl_softc->cur_pool_id++;
3516	mtx_unlock(&ctl_softc->pool_lock);
3517
3518	pool->flags = CTL_POOL_FLAG_NONE;
3519	pool->refcount = 1;		/* Reference for validity. */
3520	STAILQ_INIT(&pool->free_queue);
3521
3522	/*
3523	 * XXX KDM other options here:
3524	 * - allocate a page at a time
3525	 * - allocate one big chunk of memory.
3526	 * Page allocation might work well, but would take a little more
3527	 * tracking.
3528	 */
3529	for (i = 0; i < total_ctl_io; i++) {
3530		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3531						M_NOWAIT);
3532		if (cur_io == NULL) {
3533			retval = ENOMEM;
3534			break;
3535		}
3536		cur_io->io_hdr.pool = pool;
3537		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3538		pool->total_ctl_io++;
3539		pool->free_ctl_io++;
3540	}
3541
3542	if (retval != 0) {
3543		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3544		     cur_io != NULL; cur_io = next_io) {
3545			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3546							      links);
3547			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3548				      ctl_io_hdr, links);
3549			free(cur_io, M_CTLIO);
3550		}
3551
3552		free(pool, M_CTL);
3553		goto bailout;
3554	}
3555	mtx_lock(&ctl_softc->pool_lock);
3556	ctl_softc->num_pools++;
3557	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3558	/*
3559	 * Increment our usage count if this is an external consumer, so we
3560	 * can't get unloaded until the external consumer (most likely a
3561	 * FETD) unloads and frees his pool.
3562	 *
3563	 * XXX KDM will this increment the caller's module use count, or
3564	 * mine?
3565	 */
3566#if 0
3567	if ((pool_type != CTL_POOL_EMERGENCY)
3568	 && (pool_type != CTL_POOL_INTERNAL)
3569	 && (pool_type != CTL_POOL_4OTHERSC))
3570		MOD_INC_USE_COUNT;
3571#endif
3572
3573	mtx_unlock(&ctl_softc->pool_lock);
3574
3575	*npool = pool;
3576
3577bailout:
3578
3579	return (retval);
3580}
3581
3582static int
3583ctl_pool_acquire(struct ctl_io_pool *pool)
3584{
3585
3586	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3587
3588	if (pool->flags & CTL_POOL_FLAG_INVALID)
3589		return (EINVAL);
3590
3591	pool->refcount++;
3592
3593	return (0);
3594}
3595
3596static void
3597ctl_pool_release(struct ctl_io_pool *pool)
3598{
3599	struct ctl_softc *ctl_softc = pool->ctl_softc;
3600	union ctl_io *io;
3601
3602	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3603
3604	if (--pool->refcount != 0)
3605		return;
3606
3607	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3608		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3609			      links);
3610		free(io, M_CTLIO);
3611	}
3612
3613	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3614	ctl_softc->num_pools--;
3615
3616	/*
3617	 * XXX KDM will this decrement the caller's usage count or mine?
3618	 */
3619#if 0
3620	if ((pool->type != CTL_POOL_EMERGENCY)
3621	 && (pool->type != CTL_POOL_INTERNAL)
3622	 && (pool->type != CTL_POOL_4OTHERSC))
3623		MOD_DEC_USE_COUNT;
3624#endif
3625
3626	free(pool, M_CTL);
3627}
3628
3629void
3630ctl_pool_free(struct ctl_io_pool *pool)
3631{
3632	struct ctl_softc *ctl_softc;
3633
3634	if (pool == NULL)
3635		return;
3636
3637	ctl_softc = pool->ctl_softc;
3638	mtx_lock(&ctl_softc->pool_lock);
3639	pool->flags |= CTL_POOL_FLAG_INVALID;
3640	ctl_pool_release(pool);
3641	mtx_unlock(&ctl_softc->pool_lock);
3642}
3643
3644/*
3645 * This routine does not block (except for spinlocks of course).
3646 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3647 * possible.
3648 */
3649union ctl_io *
3650ctl_alloc_io(void *pool_ref)
3651{
3652	union ctl_io *io;
3653	struct ctl_softc *ctl_softc;
3654	struct ctl_io_pool *pool, *npool;
3655	struct ctl_io_pool *emergency_pool;
3656
3657	pool = (struct ctl_io_pool *)pool_ref;
3658
3659	if (pool == NULL) {
3660		printf("%s: pool is NULL\n", __func__);
3661		return (NULL);
3662	}
3663
3664	emergency_pool = NULL;
3665
3666	ctl_softc = pool->ctl_softc;
3667
3668	mtx_lock(&ctl_softc->pool_lock);
3669	/*
3670	 * First, try to get the io structure from the user's pool.
3671	 */
3672	if (ctl_pool_acquire(pool) == 0) {
3673		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3674		if (io != NULL) {
3675			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3676			pool->total_allocated++;
3677			pool->free_ctl_io--;
3678			mtx_unlock(&ctl_softc->pool_lock);
3679			return (io);
3680		} else
3681			ctl_pool_release(pool);
3682	}
3683	/*
3684	 * If he doesn't have any io structures left, search for an
3685	 * emergency pool and grab one from there.
3686	 */
3687	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3688		if (npool->type != CTL_POOL_EMERGENCY)
3689			continue;
3690
3691		if (ctl_pool_acquire(npool) != 0)
3692			continue;
3693
3694		emergency_pool = npool;
3695
3696		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3697		if (io != NULL) {
3698			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3699			npool->total_allocated++;
3700			npool->free_ctl_io--;
3701			mtx_unlock(&ctl_softc->pool_lock);
3702			return (io);
3703		} else
3704			ctl_pool_release(npool);
3705	}
3706
3707	/* Drop the spinlock before we malloc */
3708	mtx_unlock(&ctl_softc->pool_lock);
3709
3710	/*
3711	 * The emergency pool (if it exists) didn't have one, so try an
3712	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3713	 */
3714	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3715	if (io != NULL) {
3716		/*
3717		 * If the emergency pool exists but is empty, add this
3718		 * ctl_io to its list when it gets freed.
3719		 */
3720		if (emergency_pool != NULL) {
3721			mtx_lock(&ctl_softc->pool_lock);
3722			if (ctl_pool_acquire(emergency_pool) == 0) {
3723				io->io_hdr.pool = emergency_pool;
3724				emergency_pool->total_ctl_io++;
3725				/*
3726				 * Need to bump this, otherwise
3727				 * total_allocated and total_freed won't
3728				 * match when we no longer have anything
3729				 * outstanding.
3730				 */
3731				emergency_pool->total_allocated++;
3732			}
3733			mtx_unlock(&ctl_softc->pool_lock);
3734		} else
3735			io->io_hdr.pool = NULL;
3736	}
3737
3738	return (io);
3739}
3740
3741void
3742ctl_free_io(union ctl_io *io)
3743{
3744	if (io == NULL)
3745		return;
3746
3747	/*
3748	 * If this ctl_io has a pool, return it to that pool.
3749	 */
3750	if (io->io_hdr.pool != NULL) {
3751		struct ctl_io_pool *pool;
3752
3753		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3754		mtx_lock(&pool->ctl_softc->pool_lock);
3755		io->io_hdr.io_type = 0xff;
3756		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3757		pool->total_freed++;
3758		pool->free_ctl_io++;
3759		ctl_pool_release(pool);
3760		mtx_unlock(&pool->ctl_softc->pool_lock);
3761	} else {
3762		/*
3763		 * Otherwise, just free it.  We probably malloced it and
3764		 * the emergency pool wasn't available.
3765		 */
3766		free(io, M_CTLIO);
3767	}
3768
3769}
3770
3771void
3772ctl_zero_io(union ctl_io *io)
3773{
3774	void *pool_ref;
3775
3776	if (io == NULL)
3777		return;
3778
3779	/*
3780	 * May need to preserve linked list pointers at some point too.
3781	 */
3782	pool_ref = io->io_hdr.pool;
3783
3784	memset(io, 0, sizeof(*io));
3785
3786	io->io_hdr.pool = pool_ref;
3787}
3788
3789/*
3790 * This routine is currently used for internal copies of ctl_ios that need
3791 * to persist for some reason after we've already returned status to the
3792 * FETD.  (Thus the flag set.)
3793 *
3794 * XXX XXX
3795 * Note that this makes a blind copy of all fields in the ctl_io, except
3796 * for the pool reference.  This includes any memory that has been
3797 * allocated!  That memory will no longer be valid after done has been
3798 * called, so this would be VERY DANGEROUS for command that actually does
3799 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3800 * start and stop commands, which don't transfer any data, so this is not a
3801 * problem.  If it is used for anything else, the caller would also need to
3802 * allocate data buffer space and this routine would need to be modified to
3803 * copy the data buffer(s) as well.
3804 */
3805void
3806ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3807{
3808	void *pool_ref;
3809
3810	if ((src == NULL)
3811	 || (dest == NULL))
3812		return;
3813
3814	/*
3815	 * May need to preserve linked list pointers at some point too.
3816	 */
3817	pool_ref = dest->io_hdr.pool;
3818
3819	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3820
3821	dest->io_hdr.pool = pool_ref;
3822	/*
3823	 * We need to know that this is an internal copy, and doesn't need
3824	 * to get passed back to the FETD that allocated it.
3825	 */
3826	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3827}
3828
3829#ifdef NEEDTOPORT
3830static void
3831ctl_update_power_subpage(struct copan_power_subpage *page)
3832{
3833	int num_luns, num_partitions, config_type;
3834	struct ctl_softc *softc;
3835	cs_BOOL_t aor_present, shelf_50pct_power;
3836	cs_raidset_personality_t rs_type;
3837	int max_active_luns;
3838
3839	softc = control_softc;
3840
3841	/* subtract out the processor LUN */
3842	num_luns = softc->num_luns - 1;
3843	/*
3844	 * Default to 7 LUNs active, which was the only number we allowed
3845	 * in the past.
3846	 */
3847	max_active_luns = 7;
3848
3849	num_partitions = config_GetRsPartitionInfo();
3850	config_type = config_GetConfigType();
3851	shelf_50pct_power = config_GetShelfPowerMode();
3852	aor_present = config_IsAorRsPresent();
3853
3854	rs_type = ddb_GetRsRaidType(1);
3855	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3856	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3857		EPRINT(0, "Unsupported RS type %d!", rs_type);
3858	}
3859
3860
3861	page->total_luns = num_luns;
3862
3863	switch (config_type) {
3864	case 40:
3865		/*
3866		 * In a 40 drive configuration, it doesn't matter what DC
3867		 * cards we have, whether we have AOR enabled or not,
3868		 * partitioning or not, or what type of RAIDset we have.
3869		 * In that scenario, we can power up every LUN we present
3870		 * to the user.
3871		 */
3872		max_active_luns = num_luns;
3873
3874		break;
3875	case 64:
3876		if (shelf_50pct_power == CS_FALSE) {
3877			/* 25% power */
3878			if (aor_present == CS_TRUE) {
3879				if (rs_type ==
3880				     CS_RAIDSET_PERSONALITY_RAID5) {
3881					max_active_luns = 7;
3882				} else if (rs_type ==
3883					 CS_RAIDSET_PERSONALITY_RAID1){
3884					max_active_luns = 14;
3885				} else {
3886					/* XXX KDM now what?? */
3887				}
3888			} else {
3889				if (rs_type ==
3890				     CS_RAIDSET_PERSONALITY_RAID5) {
3891					max_active_luns = 8;
3892				} else if (rs_type ==
3893					 CS_RAIDSET_PERSONALITY_RAID1){
3894					max_active_luns = 16;
3895				} else {
3896					/* XXX KDM now what?? */
3897				}
3898			}
3899		} else {
3900			/* 50% power */
3901			/*
3902			 * With 50% power in a 64 drive configuration, we
3903			 * can power all LUNs we present.
3904			 */
3905			max_active_luns = num_luns;
3906		}
3907		break;
3908	case 112:
3909		if (shelf_50pct_power == CS_FALSE) {
3910			/* 25% power */
3911			if (aor_present == CS_TRUE) {
3912				if (rs_type ==
3913				     CS_RAIDSET_PERSONALITY_RAID5) {
3914					max_active_luns = 7;
3915				} else if (rs_type ==
3916					 CS_RAIDSET_PERSONALITY_RAID1){
3917					max_active_luns = 14;
3918				} else {
3919					/* XXX KDM now what?? */
3920				}
3921			} else {
3922				if (rs_type ==
3923				     CS_RAIDSET_PERSONALITY_RAID5) {
3924					max_active_luns = 8;
3925				} else if (rs_type ==
3926					 CS_RAIDSET_PERSONALITY_RAID1){
3927					max_active_luns = 16;
3928				} else {
3929					/* XXX KDM now what?? */
3930				}
3931			}
3932		} else {
3933			/* 50% power */
3934			if (aor_present == CS_TRUE) {
3935				if (rs_type ==
3936				     CS_RAIDSET_PERSONALITY_RAID5) {
3937					max_active_luns = 14;
3938				} else if (rs_type ==
3939					 CS_RAIDSET_PERSONALITY_RAID1){
3940					/*
3941					 * We're assuming here that disk
3942					 * caching is enabled, and so we're
3943					 * able to power up half of each
3944					 * LUN, and cache all writes.
3945					 */
3946					max_active_luns = num_luns;
3947				} else {
3948					/* XXX KDM now what?? */
3949				}
3950			} else {
3951				if (rs_type ==
3952				     CS_RAIDSET_PERSONALITY_RAID5) {
3953					max_active_luns = 15;
3954				} else if (rs_type ==
3955					 CS_RAIDSET_PERSONALITY_RAID1){
3956					max_active_luns = 30;
3957				} else {
3958					/* XXX KDM now what?? */
3959				}
3960			}
3961		}
3962		break;
3963	default:
3964		/*
3965		 * In this case, we have an unknown configuration, so we
3966		 * just use the default from above.
3967		 */
3968		break;
3969	}
3970
3971	page->max_active_luns = max_active_luns;
3972#if 0
3973	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3974	       page->total_luns, page->max_active_luns);
3975#endif
3976}
3977#endif /* NEEDTOPORT */
3978
3979/*
3980 * This routine could be used in the future to load default and/or saved
3981 * mode page parameters for a particuar lun.
3982 */
3983static int
3984ctl_init_page_index(struct ctl_lun *lun)
3985{
3986	int i;
3987	struct ctl_page_index *page_index;
3988	struct ctl_softc *softc;
3989
3990	memcpy(&lun->mode_pages.index, page_index_template,
3991	       sizeof(page_index_template));
3992
3993	softc = lun->ctl_softc;
3994
3995	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3996
3997		page_index = &lun->mode_pages.index[i];
3998		/*
3999		 * If this is a disk-only mode page, there's no point in
4000		 * setting it up.  For some pages, we have to have some
4001		 * basic information about the disk in order to calculate the
4002		 * mode page data.
4003		 */
4004		if ((lun->be_lun->lun_type != T_DIRECT)
4005		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4006			continue;
4007
4008		switch (page_index->page_code & SMPH_PC_MASK) {
4009		case SMS_FORMAT_DEVICE_PAGE: {
4010			struct scsi_format_page *format_page;
4011
4012			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4013				panic("subpage is incorrect!");
4014
4015			/*
4016			 * Sectors per track are set above.  Bytes per
4017			 * sector need to be set here on a per-LUN basis.
4018			 */
4019			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4020			       &format_page_default,
4021			       sizeof(format_page_default));
4022			memcpy(&lun->mode_pages.format_page[
4023			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4024			       sizeof(format_page_changeable));
4025			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4026			       &format_page_default,
4027			       sizeof(format_page_default));
4028			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4029			       &format_page_default,
4030			       sizeof(format_page_default));
4031
4032			format_page = &lun->mode_pages.format_page[
4033				CTL_PAGE_CURRENT];
4034			scsi_ulto2b(lun->be_lun->blocksize,
4035				    format_page->bytes_per_sector);
4036
4037			format_page = &lun->mode_pages.format_page[
4038				CTL_PAGE_DEFAULT];
4039			scsi_ulto2b(lun->be_lun->blocksize,
4040				    format_page->bytes_per_sector);
4041
4042			format_page = &lun->mode_pages.format_page[
4043				CTL_PAGE_SAVED];
4044			scsi_ulto2b(lun->be_lun->blocksize,
4045				    format_page->bytes_per_sector);
4046
4047			page_index->page_data =
4048				(uint8_t *)lun->mode_pages.format_page;
4049			break;
4050		}
4051		case SMS_RIGID_DISK_PAGE: {
4052			struct scsi_rigid_disk_page *rigid_disk_page;
4053			uint32_t sectors_per_cylinder;
4054			uint64_t cylinders;
4055#ifndef	__XSCALE__
4056			int shift;
4057#endif /* !__XSCALE__ */
4058
4059			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4060				panic("invalid subpage value %d",
4061				      page_index->subpage);
4062
4063			/*
4064			 * Rotation rate and sectors per track are set
4065			 * above.  We calculate the cylinders here based on
4066			 * capacity.  Due to the number of heads and
4067			 * sectors per track we're using, smaller arrays
4068			 * may turn out to have 0 cylinders.  Linux and
4069			 * FreeBSD don't pay attention to these mode pages
4070			 * to figure out capacity, but Solaris does.  It
4071			 * seems to deal with 0 cylinders just fine, and
4072			 * works out a fake geometry based on the capacity.
4073			 */
4074			memcpy(&lun->mode_pages.rigid_disk_page[
4075			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4076			       sizeof(rigid_disk_page_default));
4077			memcpy(&lun->mode_pages.rigid_disk_page[
4078			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4079			       sizeof(rigid_disk_page_changeable));
4080			memcpy(&lun->mode_pages.rigid_disk_page[
4081			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4082			       sizeof(rigid_disk_page_default));
4083			memcpy(&lun->mode_pages.rigid_disk_page[
4084			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4085			       sizeof(rigid_disk_page_default));
4086
4087			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4088				CTL_DEFAULT_HEADS;
4089
4090			/*
4091			 * The divide method here will be more accurate,
4092			 * probably, but results in floating point being
4093			 * used in the kernel on i386 (__udivdi3()).  On the
4094			 * XScale, though, __udivdi3() is implemented in
4095			 * software.
4096			 *
4097			 * The shift method for cylinder calculation is
4098			 * accurate if sectors_per_cylinder is a power of
4099			 * 2.  Otherwise it might be slightly off -- you
4100			 * might have a bit of a truncation problem.
4101			 */
4102#ifdef	__XSCALE__
4103			cylinders = (lun->be_lun->maxlba + 1) /
4104				sectors_per_cylinder;
4105#else
4106			for (shift = 31; shift > 0; shift--) {
4107				if (sectors_per_cylinder & (1 << shift))
4108					break;
4109			}
4110			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4111#endif
4112
4113			/*
4114			 * We've basically got 3 bytes, or 24 bits for the
4115			 * cylinder size in the mode page.  If we're over,
4116			 * just round down to 2^24.
4117			 */
4118			if (cylinders > 0xffffff)
4119				cylinders = 0xffffff;
4120
4121			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4122				CTL_PAGE_CURRENT];
4123			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4124
4125			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4126				CTL_PAGE_DEFAULT];
4127			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4128
4129			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4130				CTL_PAGE_SAVED];
4131			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4132
4133			page_index->page_data =
4134				(uint8_t *)lun->mode_pages.rigid_disk_page;
4135			break;
4136		}
4137		case SMS_CACHING_PAGE: {
4138
4139			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4140				panic("invalid subpage value %d",
4141				      page_index->subpage);
4142			/*
4143			 * Defaults should be okay here, no calculations
4144			 * needed.
4145			 */
4146			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4147			       &caching_page_default,
4148			       sizeof(caching_page_default));
4149			memcpy(&lun->mode_pages.caching_page[
4150			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4151			       sizeof(caching_page_changeable));
4152			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4153			       &caching_page_default,
4154			       sizeof(caching_page_default));
4155			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4156			       &caching_page_default,
4157			       sizeof(caching_page_default));
4158			page_index->page_data =
4159				(uint8_t *)lun->mode_pages.caching_page;
4160			break;
4161		}
4162		case SMS_CONTROL_MODE_PAGE: {
4163
4164			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4165				panic("invalid subpage value %d",
4166				      page_index->subpage);
4167
4168			/*
4169			 * Defaults should be okay here, no calculations
4170			 * needed.
4171			 */
4172			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4173			       &control_page_default,
4174			       sizeof(control_page_default));
4175			memcpy(&lun->mode_pages.control_page[
4176			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4177			       sizeof(control_page_changeable));
4178			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4179			       &control_page_default,
4180			       sizeof(control_page_default));
4181			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4182			       &control_page_default,
4183			       sizeof(control_page_default));
4184			page_index->page_data =
4185				(uint8_t *)lun->mode_pages.control_page;
4186			break;
4187
4188		}
4189		case SMS_VENDOR_SPECIFIC_PAGE:{
4190			switch (page_index->subpage) {
4191			case PWR_SUBPAGE_CODE: {
4192				struct copan_power_subpage *current_page,
4193							   *saved_page;
4194
4195				memcpy(&lun->mode_pages.power_subpage[
4196				       CTL_PAGE_CURRENT],
4197				       &power_page_default,
4198				       sizeof(power_page_default));
4199				memcpy(&lun->mode_pages.power_subpage[
4200				       CTL_PAGE_CHANGEABLE],
4201				       &power_page_changeable,
4202				       sizeof(power_page_changeable));
4203				memcpy(&lun->mode_pages.power_subpage[
4204				       CTL_PAGE_DEFAULT],
4205				       &power_page_default,
4206				       sizeof(power_page_default));
4207				memcpy(&lun->mode_pages.power_subpage[
4208				       CTL_PAGE_SAVED],
4209				       &power_page_default,
4210				       sizeof(power_page_default));
4211				page_index->page_data =
4212				    (uint8_t *)lun->mode_pages.power_subpage;
4213
4214				current_page = (struct copan_power_subpage *)
4215					(page_index->page_data +
4216					 (page_index->page_len *
4217					  CTL_PAGE_CURRENT));
4218			        saved_page = (struct copan_power_subpage *)
4219				        (page_index->page_data +
4220					 (page_index->page_len *
4221					  CTL_PAGE_SAVED));
4222				break;
4223			}
4224			case APS_SUBPAGE_CODE: {
4225				struct copan_aps_subpage *current_page,
4226							 *saved_page;
4227
4228				// This gets set multiple times but
4229				// it should always be the same. It's
4230				// only done during init so who cares.
4231				index_to_aps_page = i;
4232
4233				memcpy(&lun->mode_pages.aps_subpage[
4234				       CTL_PAGE_CURRENT],
4235				       &aps_page_default,
4236				       sizeof(aps_page_default));
4237				memcpy(&lun->mode_pages.aps_subpage[
4238				       CTL_PAGE_CHANGEABLE],
4239				       &aps_page_changeable,
4240				       sizeof(aps_page_changeable));
4241				memcpy(&lun->mode_pages.aps_subpage[
4242				       CTL_PAGE_DEFAULT],
4243				       &aps_page_default,
4244				       sizeof(aps_page_default));
4245				memcpy(&lun->mode_pages.aps_subpage[
4246				       CTL_PAGE_SAVED],
4247				       &aps_page_default,
4248				       sizeof(aps_page_default));
4249				page_index->page_data =
4250					(uint8_t *)lun->mode_pages.aps_subpage;
4251
4252				current_page = (struct copan_aps_subpage *)
4253					(page_index->page_data +
4254					 (page_index->page_len *
4255					  CTL_PAGE_CURRENT));
4256				saved_page = (struct copan_aps_subpage *)
4257					(page_index->page_data +
4258					 (page_index->page_len *
4259					  CTL_PAGE_SAVED));
4260				break;
4261			}
4262			case DBGCNF_SUBPAGE_CODE: {
4263				struct copan_debugconf_subpage *current_page,
4264							       *saved_page;
4265
4266				memcpy(&lun->mode_pages.debugconf_subpage[
4267				       CTL_PAGE_CURRENT],
4268				       &debugconf_page_default,
4269				       sizeof(debugconf_page_default));
4270				memcpy(&lun->mode_pages.debugconf_subpage[
4271				       CTL_PAGE_CHANGEABLE],
4272				       &debugconf_page_changeable,
4273				       sizeof(debugconf_page_changeable));
4274				memcpy(&lun->mode_pages.debugconf_subpage[
4275				       CTL_PAGE_DEFAULT],
4276				       &debugconf_page_default,
4277				       sizeof(debugconf_page_default));
4278				memcpy(&lun->mode_pages.debugconf_subpage[
4279				       CTL_PAGE_SAVED],
4280				       &debugconf_page_default,
4281				       sizeof(debugconf_page_default));
4282				page_index->page_data =
4283					(uint8_t *)lun->mode_pages.debugconf_subpage;
4284
4285				current_page = (struct copan_debugconf_subpage *)
4286					(page_index->page_data +
4287					 (page_index->page_len *
4288					  CTL_PAGE_CURRENT));
4289				saved_page = (struct copan_debugconf_subpage *)
4290					(page_index->page_data +
4291					 (page_index->page_len *
4292					  CTL_PAGE_SAVED));
4293				break;
4294			}
4295			default:
4296				panic("invalid subpage value %d",
4297				      page_index->subpage);
4298				break;
4299			}
4300   			break;
4301		}
4302		default:
4303			panic("invalid page value %d",
4304			      page_index->page_code & SMPH_PC_MASK);
4305			break;
4306    	}
4307	}
4308
4309	return (CTL_RETVAL_COMPLETE);
4310}
4311
4312/*
4313 * LUN allocation.
4314 *
4315 * Requirements:
4316 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4317 *   wants us to allocate the LUN and he can block.
4318 * - ctl_softc is always set
4319 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4320 *
4321 * Returns 0 for success, non-zero (errno) for failure.
4322 */
4323static int
4324ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4325	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4326{
4327	struct ctl_lun *nlun, *lun;
4328	struct ctl_port *port;
4329	struct scsi_vpd_id_descriptor *desc;
4330	struct scsi_vpd_id_t10 *t10id;
4331	const char *scsiname, *vendor;
4332	int lun_number, i, lun_malloced;
4333	int devidlen, idlen1, idlen2, len;
4334
4335	if (be_lun == NULL)
4336		return (EINVAL);
4337
4338	/*
4339	 * We currently only support Direct Access or Processor LUN types.
4340	 */
4341	switch (be_lun->lun_type) {
4342	case T_DIRECT:
4343		break;
4344	case T_PROCESSOR:
4345		break;
4346	case T_SEQUENTIAL:
4347	case T_CHANGER:
4348	default:
4349		be_lun->lun_config_status(be_lun->be_lun,
4350					  CTL_LUN_CONFIG_FAILURE);
4351		break;
4352	}
4353	if (ctl_lun == NULL) {
4354		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4355		lun_malloced = 1;
4356	} else {
4357		lun_malloced = 0;
4358		lun = ctl_lun;
4359	}
4360
4361	memset(lun, 0, sizeof(*lun));
4362	if (lun_malloced)
4363		lun->flags = CTL_LUN_MALLOCED;
4364
4365	/* Generate LUN ID. */
4366	devidlen = max(CTL_DEVID_MIN_LEN,
4367	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4368	idlen1 = sizeof(*t10id) + devidlen;
4369	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4370	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4371	if (scsiname != NULL) {
4372		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4373		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4374	}
4375	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4376	    M_CTL, M_WAITOK | M_ZERO);
4377	lun->lun_devid->len = len;
4378	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4379	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4380	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4381	desc->length = idlen1;
4382	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4383	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4384	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4385		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4386	} else {
4387		strncpy(t10id->vendor, vendor,
4388		    min(sizeof(t10id->vendor), strlen(vendor)));
4389	}
4390	strncpy((char *)t10id->vendor_spec_id,
4391	    (char *)be_lun->device_id, devidlen);
4392	if (scsiname != NULL) {
4393		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4394		    desc->length);
4395		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4396		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4397		    SVPD_ID_TYPE_SCSI_NAME;
4398		desc->length = idlen2;
4399		strlcpy(desc->identifier, scsiname, idlen2);
4400	}
4401
4402	mtx_lock(&ctl_softc->ctl_lock);
4403	/*
4404	 * See if the caller requested a particular LUN number.  If so, see
4405	 * if it is available.  Otherwise, allocate the first available LUN.
4406	 */
4407	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4408		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4409		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4410			mtx_unlock(&ctl_softc->ctl_lock);
4411			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4412				printf("ctl: requested LUN ID %d is higher "
4413				       "than CTL_MAX_LUNS - 1 (%d)\n",
4414				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4415			} else {
4416				/*
4417				 * XXX KDM return an error, or just assign
4418				 * another LUN ID in this case??
4419				 */
4420				printf("ctl: requested LUN ID %d is already "
4421				       "in use\n", be_lun->req_lun_id);
4422			}
4423			if (lun->flags & CTL_LUN_MALLOCED)
4424				free(lun, M_CTL);
4425			be_lun->lun_config_status(be_lun->be_lun,
4426						  CTL_LUN_CONFIG_FAILURE);
4427			return (ENOSPC);
4428		}
4429		lun_number = be_lun->req_lun_id;
4430	} else {
4431		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4432		if (lun_number == -1) {
4433			mtx_unlock(&ctl_softc->ctl_lock);
4434			printf("ctl: can't allocate LUN on target %ju, out of "
4435			       "LUNs\n", (uintmax_t)target_id.id);
4436			if (lun->flags & CTL_LUN_MALLOCED)
4437				free(lun, M_CTL);
4438			be_lun->lun_config_status(be_lun->be_lun,
4439						  CTL_LUN_CONFIG_FAILURE);
4440			return (ENOSPC);
4441		}
4442	}
4443	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4444
4445	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4446	lun->target = target_id;
4447	lun->lun = lun_number;
4448	lun->be_lun = be_lun;
4449	/*
4450	 * The processor LUN is always enabled.  Disk LUNs come on line
4451	 * disabled, and must be enabled by the backend.
4452	 */
4453	lun->flags |= CTL_LUN_DISABLED;
4454	lun->backend = be_lun->be;
4455	be_lun->ctl_lun = lun;
4456	be_lun->lun_id = lun_number;
4457	atomic_add_int(&be_lun->be->num_luns, 1);
4458	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4459		lun->flags |= CTL_LUN_STOPPED;
4460
4461	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4462		lun->flags |= CTL_LUN_INOPERABLE;
4463
4464	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4465		lun->flags |= CTL_LUN_PRIMARY_SC;
4466
4467	lun->ctl_softc = ctl_softc;
4468	TAILQ_INIT(&lun->ooa_queue);
4469	TAILQ_INIT(&lun->blocked_queue);
4470	STAILQ_INIT(&lun->error_list);
4471
4472	/*
4473	 * Initialize the mode page index.
4474	 */
4475	ctl_init_page_index(lun);
4476
4477	/*
4478	 * Set the poweron UA for all initiators on this LUN only.
4479	 */
4480	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4481		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4482
4483	/*
4484	 * Now, before we insert this lun on the lun list, set the lun
4485	 * inventory changed UA for all other luns.
4486	 */
4487	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4488		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4489			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4490		}
4491	}
4492
4493	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4494
4495	ctl_softc->ctl_luns[lun_number] = lun;
4496
4497	ctl_softc->num_luns++;
4498
4499	/* Setup statistics gathering */
4500	lun->stats.device_type = be_lun->lun_type;
4501	lun->stats.lun_number = lun_number;
4502	if (lun->stats.device_type == T_DIRECT)
4503		lun->stats.blocksize = be_lun->blocksize;
4504	else
4505		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4506	for (i = 0;i < CTL_MAX_PORTS;i++)
4507		lun->stats.ports[i].targ_port = i;
4508
4509	mtx_unlock(&ctl_softc->ctl_lock);
4510
4511	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4512
4513	/*
4514	 * Run through each registered FETD and bring it online if it isn't
4515	 * already.  Enable the target ID if it hasn't been enabled, and
4516	 * enable this particular LUN.
4517	 */
4518	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4519		int retval;
4520
4521		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4522		if (retval != 0) {
4523			printf("ctl_alloc_lun: FETD %s port %d returned error "
4524			       "%d for lun_enable on target %ju lun %d\n",
4525			       port->port_name, port->targ_port, retval,
4526			       (uintmax_t)target_id.id, lun_number);
4527		} else
4528			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4529	}
4530	return (0);
4531}
4532
4533/*
4534 * Delete a LUN.
4535 * Assumptions:
4536 * - LUN has already been marked invalid and any pending I/O has been taken
4537 *   care of.
4538 */
4539static int
4540ctl_free_lun(struct ctl_lun *lun)
4541{
4542	struct ctl_softc *softc;
4543#if 0
4544	struct ctl_port *port;
4545#endif
4546	struct ctl_lun *nlun;
4547	int i;
4548
4549	softc = lun->ctl_softc;
4550
4551	mtx_assert(&softc->ctl_lock, MA_OWNED);
4552
4553	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4554
4555	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4556
4557	softc->ctl_luns[lun->lun] = NULL;
4558
4559	if (!TAILQ_EMPTY(&lun->ooa_queue))
4560		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4561
4562	softc->num_luns--;
4563
4564	/*
4565	 * XXX KDM this scheme only works for a single target/multiple LUN
4566	 * setup.  It needs to be revamped for a multiple target scheme.
4567	 *
4568	 * XXX KDM this results in port->lun_disable() getting called twice,
4569	 * once when ctl_disable_lun() is called, and a second time here.
4570	 * We really need to re-think the LUN disable semantics.  There
4571	 * should probably be several steps/levels to LUN removal:
4572	 *  - disable
4573	 *  - invalidate
4574	 *  - free
4575 	 *
4576	 * Right now we only have a disable method when communicating to
4577	 * the front end ports, at least for individual LUNs.
4578	 */
4579#if 0
4580	STAILQ_FOREACH(port, &softc->port_list, links) {
4581		int retval;
4582
4583		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4584					 lun->lun);
4585		if (retval != 0) {
4586			printf("ctl_free_lun: FETD %s port %d returned error "
4587			       "%d for lun_disable on target %ju lun %jd\n",
4588			       port->port_name, port->targ_port, retval,
4589			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4590		}
4591
4592		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4593			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4594
4595			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4596			if (retval != 0) {
4597				printf("ctl_free_lun: FETD %s port %d "
4598				       "returned error %d for targ_disable on "
4599				       "target %ju\n", port->port_name,
4600				       port->targ_port, retval,
4601				       (uintmax_t)lun->target.id);
4602			} else
4603				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4604
4605			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4606				continue;
4607
4608#if 0
4609			port->port_offline(port->onoff_arg);
4610			port->status &= ~CTL_PORT_STATUS_ONLINE;
4611#endif
4612		}
4613	}
4614#endif
4615
4616	/*
4617	 * Tell the backend to free resources, if this LUN has a backend.
4618	 */
4619	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4620	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4621
4622	mtx_destroy(&lun->lun_lock);
4623	free(lun->lun_devid, M_CTL);
4624	if (lun->flags & CTL_LUN_MALLOCED)
4625		free(lun, M_CTL);
4626
4627	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4628		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4629			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4630		}
4631	}
4632
4633	return (0);
4634}
4635
4636static void
4637ctl_create_lun(struct ctl_be_lun *be_lun)
4638{
4639	struct ctl_softc *ctl_softc;
4640
4641	ctl_softc = control_softc;
4642
4643	/*
4644	 * ctl_alloc_lun() should handle all potential failure cases.
4645	 */
4646	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4647}
4648
4649int
4650ctl_add_lun(struct ctl_be_lun *be_lun)
4651{
4652	struct ctl_softc *ctl_softc = control_softc;
4653
4654	mtx_lock(&ctl_softc->ctl_lock);
4655	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4656	mtx_unlock(&ctl_softc->ctl_lock);
4657	wakeup(&ctl_softc->pending_lun_queue);
4658
4659	return (0);
4660}
4661
4662int
4663ctl_enable_lun(struct ctl_be_lun *be_lun)
4664{
4665	struct ctl_softc *ctl_softc;
4666	struct ctl_port *port, *nport;
4667	struct ctl_lun *lun;
4668	int retval;
4669
4670	ctl_softc = control_softc;
4671
4672	lun = (struct ctl_lun *)be_lun->ctl_lun;
4673
4674	mtx_lock(&ctl_softc->ctl_lock);
4675	mtx_lock(&lun->lun_lock);
4676	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4677		/*
4678		 * eh?  Why did we get called if the LUN is already
4679		 * enabled?
4680		 */
4681		mtx_unlock(&lun->lun_lock);
4682		mtx_unlock(&ctl_softc->ctl_lock);
4683		return (0);
4684	}
4685	lun->flags &= ~CTL_LUN_DISABLED;
4686	mtx_unlock(&lun->lun_lock);
4687
4688	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4689		nport = STAILQ_NEXT(port, links);
4690
4691		/*
4692		 * Drop the lock while we call the FETD's enable routine.
4693		 * This can lead to a callback into CTL (at least in the
4694		 * case of the internal initiator frontend.
4695		 */
4696		mtx_unlock(&ctl_softc->ctl_lock);
4697		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4698		mtx_lock(&ctl_softc->ctl_lock);
4699		if (retval != 0) {
4700			printf("%s: FETD %s port %d returned error "
4701			       "%d for lun_enable on target %ju lun %jd\n",
4702			       __func__, port->port_name, port->targ_port, retval,
4703			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4704		}
4705#if 0
4706		 else {
4707            /* NOTE:  TODO:  why does lun enable affect port status? */
4708			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4709		}
4710#endif
4711	}
4712
4713	mtx_unlock(&ctl_softc->ctl_lock);
4714
4715	return (0);
4716}
4717
4718int
4719ctl_disable_lun(struct ctl_be_lun *be_lun)
4720{
4721	struct ctl_softc *ctl_softc;
4722	struct ctl_port *port;
4723	struct ctl_lun *lun;
4724	int retval;
4725
4726	ctl_softc = control_softc;
4727
4728	lun = (struct ctl_lun *)be_lun->ctl_lun;
4729
4730	mtx_lock(&ctl_softc->ctl_lock);
4731	mtx_lock(&lun->lun_lock);
4732	if (lun->flags & CTL_LUN_DISABLED) {
4733		mtx_unlock(&lun->lun_lock);
4734		mtx_unlock(&ctl_softc->ctl_lock);
4735		return (0);
4736	}
4737	lun->flags |= CTL_LUN_DISABLED;
4738	mtx_unlock(&lun->lun_lock);
4739
4740	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4741		mtx_unlock(&ctl_softc->ctl_lock);
4742		/*
4743		 * Drop the lock before we call the frontend's disable
4744		 * routine, to avoid lock order reversals.
4745		 *
4746		 * XXX KDM what happens if the frontend list changes while
4747		 * we're traversing it?  It's unlikely, but should be handled.
4748		 */
4749		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4750					 lun->lun);
4751		mtx_lock(&ctl_softc->ctl_lock);
4752		if (retval != 0) {
4753			printf("ctl_alloc_lun: FETD %s port %d returned error "
4754			       "%d for lun_disable on target %ju lun %jd\n",
4755			       port->port_name, port->targ_port, retval,
4756			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4757		}
4758	}
4759
4760	mtx_unlock(&ctl_softc->ctl_lock);
4761
4762	return (0);
4763}
4764
4765int
4766ctl_start_lun(struct ctl_be_lun *be_lun)
4767{
4768	struct ctl_softc *ctl_softc;
4769	struct ctl_lun *lun;
4770
4771	ctl_softc = control_softc;
4772
4773	lun = (struct ctl_lun *)be_lun->ctl_lun;
4774
4775	mtx_lock(&lun->lun_lock);
4776	lun->flags &= ~CTL_LUN_STOPPED;
4777	mtx_unlock(&lun->lun_lock);
4778
4779	return (0);
4780}
4781
4782int
4783ctl_stop_lun(struct ctl_be_lun *be_lun)
4784{
4785	struct ctl_softc *ctl_softc;
4786	struct ctl_lun *lun;
4787
4788	ctl_softc = control_softc;
4789
4790	lun = (struct ctl_lun *)be_lun->ctl_lun;
4791
4792	mtx_lock(&lun->lun_lock);
4793	lun->flags |= CTL_LUN_STOPPED;
4794	mtx_unlock(&lun->lun_lock);
4795
4796	return (0);
4797}
4798
4799int
4800ctl_lun_offline(struct ctl_be_lun *be_lun)
4801{
4802	struct ctl_softc *ctl_softc;
4803	struct ctl_lun *lun;
4804
4805	ctl_softc = control_softc;
4806
4807	lun = (struct ctl_lun *)be_lun->ctl_lun;
4808
4809	mtx_lock(&lun->lun_lock);
4810	lun->flags |= CTL_LUN_OFFLINE;
4811	mtx_unlock(&lun->lun_lock);
4812
4813	return (0);
4814}
4815
4816int
4817ctl_lun_online(struct ctl_be_lun *be_lun)
4818{
4819	struct ctl_softc *ctl_softc;
4820	struct ctl_lun *lun;
4821
4822	ctl_softc = control_softc;
4823
4824	lun = (struct ctl_lun *)be_lun->ctl_lun;
4825
4826	mtx_lock(&lun->lun_lock);
4827	lun->flags &= ~CTL_LUN_OFFLINE;
4828	mtx_unlock(&lun->lun_lock);
4829
4830	return (0);
4831}
4832
4833int
4834ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4835{
4836	struct ctl_softc *ctl_softc;
4837	struct ctl_lun *lun;
4838
4839	ctl_softc = control_softc;
4840
4841	lun = (struct ctl_lun *)be_lun->ctl_lun;
4842
4843	mtx_lock(&lun->lun_lock);
4844
4845	/*
4846	 * The LUN needs to be disabled before it can be marked invalid.
4847	 */
4848	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4849		mtx_unlock(&lun->lun_lock);
4850		return (-1);
4851	}
4852	/*
4853	 * Mark the LUN invalid.
4854	 */
4855	lun->flags |= CTL_LUN_INVALID;
4856
4857	/*
4858	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4859	 * If we have something in the OOA queue, we'll free it when the
4860	 * last I/O completes.
4861	 */
4862	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4863		mtx_unlock(&lun->lun_lock);
4864		mtx_lock(&ctl_softc->ctl_lock);
4865		ctl_free_lun(lun);
4866		mtx_unlock(&ctl_softc->ctl_lock);
4867	} else
4868		mtx_unlock(&lun->lun_lock);
4869
4870	return (0);
4871}
4872
4873int
4874ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4875{
4876	struct ctl_softc *ctl_softc;
4877	struct ctl_lun *lun;
4878
4879	ctl_softc = control_softc;
4880	lun = (struct ctl_lun *)be_lun->ctl_lun;
4881
4882	mtx_lock(&lun->lun_lock);
4883	lun->flags |= CTL_LUN_INOPERABLE;
4884	mtx_unlock(&lun->lun_lock);
4885
4886	return (0);
4887}
4888
4889int
4890ctl_lun_operable(struct ctl_be_lun *be_lun)
4891{
4892	struct ctl_softc *ctl_softc;
4893	struct ctl_lun *lun;
4894
4895	ctl_softc = control_softc;
4896	lun = (struct ctl_lun *)be_lun->ctl_lun;
4897
4898	mtx_lock(&lun->lun_lock);
4899	lun->flags &= ~CTL_LUN_INOPERABLE;
4900	mtx_unlock(&lun->lun_lock);
4901
4902	return (0);
4903}
4904
4905int
4906ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4907		   int lock)
4908{
4909	struct ctl_softc *softc;
4910	struct ctl_lun *lun;
4911	struct copan_aps_subpage *current_sp;
4912	struct ctl_page_index *page_index;
4913	int i;
4914
4915	softc = control_softc;
4916
4917	mtx_lock(&softc->ctl_lock);
4918
4919	lun = (struct ctl_lun *)be_lun->ctl_lun;
4920	mtx_lock(&lun->lun_lock);
4921
4922	page_index = NULL;
4923	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4924		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4925		     APS_PAGE_CODE)
4926			continue;
4927
4928		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4929			continue;
4930		page_index = &lun->mode_pages.index[i];
4931	}
4932
4933	if (page_index == NULL) {
4934		mtx_unlock(&lun->lun_lock);
4935		mtx_unlock(&softc->ctl_lock);
4936		printf("%s: APS subpage not found for lun %ju!\n", __func__,
4937		       (uintmax_t)lun->lun);
4938		return (1);
4939	}
4940#if 0
4941	if ((softc->aps_locked_lun != 0)
4942	 && (softc->aps_locked_lun != lun->lun)) {
4943		printf("%s: attempt to lock LUN %llu when %llu is already "
4944		       "locked\n");
4945		mtx_unlock(&lun->lun_lock);
4946		mtx_unlock(&softc->ctl_lock);
4947		return (1);
4948	}
4949#endif
4950
4951	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4952		(page_index->page_len * CTL_PAGE_CURRENT));
4953
4954	if (lock != 0) {
4955		current_sp->lock_active = APS_LOCK_ACTIVE;
4956		softc->aps_locked_lun = lun->lun;
4957	} else {
4958		current_sp->lock_active = 0;
4959		softc->aps_locked_lun = 0;
4960	}
4961
4962
4963	/*
4964	 * If we're in HA mode, try to send the lock message to the other
4965	 * side.
4966	 */
4967	if (ctl_is_single == 0) {
4968		int isc_retval;
4969		union ctl_ha_msg lock_msg;
4970
4971		lock_msg.hdr.nexus = *nexus;
4972		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4973		if (lock != 0)
4974			lock_msg.aps.lock_flag = 1;
4975		else
4976			lock_msg.aps.lock_flag = 0;
4977		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4978					 sizeof(lock_msg), 0);
4979		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4980			printf("%s: APS (lock=%d) error returned from "
4981			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4982			mtx_unlock(&lun->lun_lock);
4983			mtx_unlock(&softc->ctl_lock);
4984			return (1);
4985		}
4986	}
4987
4988	mtx_unlock(&lun->lun_lock);
4989	mtx_unlock(&softc->ctl_lock);
4990
4991	return (0);
4992}
4993
4994void
4995ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4996{
4997	struct ctl_lun *lun;
4998	struct ctl_softc *softc;
4999	int i;
5000
5001	softc = control_softc;
5002
5003	lun = (struct ctl_lun *)be_lun->ctl_lun;
5004
5005	mtx_lock(&lun->lun_lock);
5006
5007	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5008		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
5009
5010	mtx_unlock(&lun->lun_lock);
5011}
5012
5013/*
5014 * Backend "memory move is complete" callback for requests that never
5015 * make it down to say RAIDCore's configuration code.
5016 */
5017int
5018ctl_config_move_done(union ctl_io *io)
5019{
5020	int retval;
5021
5022	retval = CTL_RETVAL_COMPLETE;
5023
5024
5025	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5026	/*
5027	 * XXX KDM this shouldn't happen, but what if it does?
5028	 */
5029	if (io->io_hdr.io_type != CTL_IO_SCSI)
5030		panic("I/O type isn't CTL_IO_SCSI!");
5031
5032	if ((io->io_hdr.port_status == 0)
5033	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5034	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5035		io->io_hdr.status = CTL_SUCCESS;
5036	else if ((io->io_hdr.port_status != 0)
5037	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5038	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5039		/*
5040		 * For hardware error sense keys, the sense key
5041		 * specific value is defined to be a retry count,
5042		 * but we use it to pass back an internal FETD
5043		 * error code.  XXX KDM  Hopefully the FETD is only
5044		 * using 16 bits for an error code, since that's
5045		 * all the space we have in the sks field.
5046		 */
5047		ctl_set_internal_failure(&io->scsiio,
5048					 /*sks_valid*/ 1,
5049					 /*retry_count*/
5050					 io->io_hdr.port_status);
5051		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5052			free(io->scsiio.kern_data_ptr, M_CTL);
5053		ctl_done(io);
5054		goto bailout;
5055	}
5056
5057	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5058	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5059	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5060		/*
5061		 * XXX KDM just assuming a single pointer here, and not a
5062		 * S/G list.  If we start using S/G lists for config data,
5063		 * we'll need to know how to clean them up here as well.
5064		 */
5065		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5066			free(io->scsiio.kern_data_ptr, M_CTL);
5067		/* Hopefully the user has already set the status... */
5068		ctl_done(io);
5069	} else {
5070		/*
5071		 * XXX KDM now we need to continue data movement.  Some
5072		 * options:
5073		 * - call ctl_scsiio() again?  We don't do this for data
5074		 *   writes, because for those at least we know ahead of
5075		 *   time where the write will go and how long it is.  For
5076		 *   config writes, though, that information is largely
5077		 *   contained within the write itself, thus we need to
5078		 *   parse out the data again.
5079		 *
5080		 * - Call some other function once the data is in?
5081		 */
5082
5083		/*
5084		 * XXX KDM call ctl_scsiio() again for now, and check flag
5085		 * bits to see whether we're allocated or not.
5086		 */
5087		retval = ctl_scsiio(&io->scsiio);
5088	}
5089bailout:
5090	return (retval);
5091}
5092
5093/*
5094 * This gets called by a backend driver when it is done with a
5095 * data_submit method.
5096 */
5097void
5098ctl_data_submit_done(union ctl_io *io)
5099{
5100	/*
5101	 * If the IO_CONT flag is set, we need to call the supplied
5102	 * function to continue processing the I/O, instead of completing
5103	 * the I/O just yet.
5104	 *
5105	 * If there is an error, though, we don't want to keep processing.
5106	 * Instead, just send status back to the initiator.
5107	 */
5108	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5109	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5110	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5111	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5112		io->scsiio.io_cont(io);
5113		return;
5114	}
5115	ctl_done(io);
5116}
5117
5118/*
5119 * This gets called by a backend driver when it is done with a
5120 * configuration write.
5121 */
5122void
5123ctl_config_write_done(union ctl_io *io)
5124{
5125	/*
5126	 * If the IO_CONT flag is set, we need to call the supplied
5127	 * function to continue processing the I/O, instead of completing
5128	 * the I/O just yet.
5129	 *
5130	 * If there is an error, though, we don't want to keep processing.
5131	 * Instead, just send status back to the initiator.
5132	 */
5133	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5134	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5135	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5136		io->scsiio.io_cont(io);
5137		return;
5138	}
5139	/*
5140	 * Since a configuration write can be done for commands that actually
5141	 * have data allocated, like write buffer, and commands that have
5142	 * no data, like start/stop unit, we need to check here.
5143	 */
5144	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5145		free(io->scsiio.kern_data_ptr, M_CTL);
5146	ctl_done(io);
5147}
5148
5149/*
5150 * SCSI release command.
5151 */
5152int
5153ctl_scsi_release(struct ctl_scsiio *ctsio)
5154{
5155	int length, longid, thirdparty_id, resv_id;
5156	struct ctl_softc *ctl_softc;
5157	struct ctl_lun *lun;
5158
5159	length = 0;
5160	resv_id = 0;
5161
5162	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5163
5164	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5165	ctl_softc = control_softc;
5166
5167	switch (ctsio->cdb[0]) {
5168	case RELEASE_10: {
5169		struct scsi_release_10 *cdb;
5170
5171		cdb = (struct scsi_release_10 *)ctsio->cdb;
5172
5173		if (cdb->byte2 & SR10_LONGID)
5174			longid = 1;
5175		else
5176			thirdparty_id = cdb->thirdparty_id;
5177
5178		resv_id = cdb->resv_id;
5179		length = scsi_2btoul(cdb->length);
5180		break;
5181	}
5182	}
5183
5184
5185	/*
5186	 * XXX KDM right now, we only support LUN reservation.  We don't
5187	 * support 3rd party reservations, or extent reservations, which
5188	 * might actually need the parameter list.  If we've gotten this
5189	 * far, we've got a LUN reservation.  Anything else got kicked out
5190	 * above.  So, according to SPC, ignore the length.
5191	 */
5192	length = 0;
5193
5194	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5195	 && (length > 0)) {
5196		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5197		ctsio->kern_data_len = length;
5198		ctsio->kern_total_len = length;
5199		ctsio->kern_data_resid = 0;
5200		ctsio->kern_rel_offset = 0;
5201		ctsio->kern_sg_entries = 0;
5202		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5203		ctsio->be_move_done = ctl_config_move_done;
5204		ctl_datamove((union ctl_io *)ctsio);
5205
5206		return (CTL_RETVAL_COMPLETE);
5207	}
5208
5209	if (length > 0)
5210		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5211
5212	mtx_lock(&lun->lun_lock);
5213
5214	/*
5215	 * According to SPC, it is not an error for an intiator to attempt
5216	 * to release a reservation on a LUN that isn't reserved, or that
5217	 * is reserved by another initiator.  The reservation can only be
5218	 * released, though, by the initiator who made it or by one of
5219	 * several reset type events.
5220	 */
5221	if (lun->flags & CTL_LUN_RESERVED) {
5222		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5223		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5224		 && (ctsio->io_hdr.nexus.targ_target.id ==
5225		     lun->rsv_nexus.targ_target.id)) {
5226			lun->flags &= ~CTL_LUN_RESERVED;
5227		}
5228	}
5229
5230	mtx_unlock(&lun->lun_lock);
5231
5232	ctsio->scsi_status = SCSI_STATUS_OK;
5233	ctsio->io_hdr.status = CTL_SUCCESS;
5234
5235	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5236		free(ctsio->kern_data_ptr, M_CTL);
5237		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5238	}
5239
5240	ctl_done((union ctl_io *)ctsio);
5241	return (CTL_RETVAL_COMPLETE);
5242}
5243
5244int
5245ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5246{
5247	int extent, thirdparty, longid;
5248	int resv_id, length;
5249	uint64_t thirdparty_id;
5250	struct ctl_softc *ctl_softc;
5251	struct ctl_lun *lun;
5252
5253	extent = 0;
5254	thirdparty = 0;
5255	longid = 0;
5256	resv_id = 0;
5257	length = 0;
5258	thirdparty_id = 0;
5259
5260	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5261
5262	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5263	ctl_softc = control_softc;
5264
5265	switch (ctsio->cdb[0]) {
5266	case RESERVE_10: {
5267		struct scsi_reserve_10 *cdb;
5268
5269		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5270
5271		if (cdb->byte2 & SR10_LONGID)
5272			longid = 1;
5273		else
5274			thirdparty_id = cdb->thirdparty_id;
5275
5276		resv_id = cdb->resv_id;
5277		length = scsi_2btoul(cdb->length);
5278		break;
5279	}
5280	}
5281
5282	/*
5283	 * XXX KDM right now, we only support LUN reservation.  We don't
5284	 * support 3rd party reservations, or extent reservations, which
5285	 * might actually need the parameter list.  If we've gotten this
5286	 * far, we've got a LUN reservation.  Anything else got kicked out
5287	 * above.  So, according to SPC, ignore the length.
5288	 */
5289	length = 0;
5290
5291	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5292	 && (length > 0)) {
5293		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5294		ctsio->kern_data_len = length;
5295		ctsio->kern_total_len = length;
5296		ctsio->kern_data_resid = 0;
5297		ctsio->kern_rel_offset = 0;
5298		ctsio->kern_sg_entries = 0;
5299		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5300		ctsio->be_move_done = ctl_config_move_done;
5301		ctl_datamove((union ctl_io *)ctsio);
5302
5303		return (CTL_RETVAL_COMPLETE);
5304	}
5305
5306	if (length > 0)
5307		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5308
5309	mtx_lock(&lun->lun_lock);
5310	if (lun->flags & CTL_LUN_RESERVED) {
5311		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5312		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5313		 || (ctsio->io_hdr.nexus.targ_target.id !=
5314		     lun->rsv_nexus.targ_target.id)) {
5315			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5316			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5317			goto bailout;
5318		}
5319	}
5320
5321	lun->flags |= CTL_LUN_RESERVED;
5322	lun->rsv_nexus = ctsio->io_hdr.nexus;
5323
5324	ctsio->scsi_status = SCSI_STATUS_OK;
5325	ctsio->io_hdr.status = CTL_SUCCESS;
5326
5327bailout:
5328	mtx_unlock(&lun->lun_lock);
5329
5330	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5331		free(ctsio->kern_data_ptr, M_CTL);
5332		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5333	}
5334
5335	ctl_done((union ctl_io *)ctsio);
5336	return (CTL_RETVAL_COMPLETE);
5337}
5338
5339int
5340ctl_start_stop(struct ctl_scsiio *ctsio)
5341{
5342	struct scsi_start_stop_unit *cdb;
5343	struct ctl_lun *lun;
5344	struct ctl_softc *ctl_softc;
5345	int retval;
5346
5347	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5348
5349	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5350	ctl_softc = control_softc;
5351	retval = 0;
5352
5353	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5354
5355	/*
5356	 * XXX KDM
5357	 * We don't support the immediate bit on a stop unit.  In order to
5358	 * do that, we would need to code up a way to know that a stop is
5359	 * pending, and hold off any new commands until it completes, one
5360	 * way or another.  Then we could accept or reject those commands
5361	 * depending on its status.  We would almost need to do the reverse
5362	 * of what we do below for an immediate start -- return the copy of
5363	 * the ctl_io to the FETD with status to send to the host (and to
5364	 * free the copy!) and then free the original I/O once the stop
5365	 * actually completes.  That way, the OOA queue mechanism can work
5366	 * to block commands that shouldn't proceed.  Another alternative
5367	 * would be to put the copy in the queue in place of the original,
5368	 * and return the original back to the caller.  That could be
5369	 * slightly safer..
5370	 */
5371	if ((cdb->byte2 & SSS_IMMED)
5372	 && ((cdb->how & SSS_START) == 0)) {
5373		ctl_set_invalid_field(ctsio,
5374				      /*sks_valid*/ 1,
5375				      /*command*/ 1,
5376				      /*field*/ 1,
5377				      /*bit_valid*/ 1,
5378				      /*bit*/ 0);
5379		ctl_done((union ctl_io *)ctsio);
5380		return (CTL_RETVAL_COMPLETE);
5381	}
5382
5383	if ((lun->flags & CTL_LUN_PR_RESERVED)
5384	 && ((cdb->how & SSS_START)==0)) {
5385		uint32_t residx;
5386
5387		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5388		if (!lun->per_res[residx].registered
5389		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5390
5391			ctl_set_reservation_conflict(ctsio);
5392			ctl_done((union ctl_io *)ctsio);
5393			return (CTL_RETVAL_COMPLETE);
5394		}
5395	}
5396
5397	/*
5398	 * If there is no backend on this device, we can't start or stop
5399	 * it.  In theory we shouldn't get any start/stop commands in the
5400	 * first place at this level if the LUN doesn't have a backend.
5401	 * That should get stopped by the command decode code.
5402	 */
5403	if (lun->backend == NULL) {
5404		ctl_set_invalid_opcode(ctsio);
5405		ctl_done((union ctl_io *)ctsio);
5406		return (CTL_RETVAL_COMPLETE);
5407	}
5408
5409	/*
5410	 * XXX KDM Copan-specific offline behavior.
5411	 * Figure out a reasonable way to port this?
5412	 */
5413#ifdef NEEDTOPORT
5414	mtx_lock(&lun->lun_lock);
5415
5416	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5417	 && (lun->flags & CTL_LUN_OFFLINE)) {
5418		/*
5419		 * If the LUN is offline, and the on/offline bit isn't set,
5420		 * reject the start or stop.  Otherwise, let it through.
5421		 */
5422		mtx_unlock(&lun->lun_lock);
5423		ctl_set_lun_not_ready(ctsio);
5424		ctl_done((union ctl_io *)ctsio);
5425	} else {
5426		mtx_unlock(&lun->lun_lock);
5427#endif /* NEEDTOPORT */
5428		/*
5429		 * This could be a start or a stop when we're online,
5430		 * or a stop/offline or start/online.  A start or stop when
5431		 * we're offline is covered in the case above.
5432		 */
5433		/*
5434		 * In the non-immediate case, we send the request to
5435		 * the backend and return status to the user when
5436		 * it is done.
5437		 *
5438		 * In the immediate case, we allocate a new ctl_io
5439		 * to hold a copy of the request, and send that to
5440		 * the backend.  We then set good status on the
5441		 * user's request and return it immediately.
5442		 */
5443		if (cdb->byte2 & SSS_IMMED) {
5444			union ctl_io *new_io;
5445
5446			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5447			if (new_io == NULL) {
5448				ctl_set_busy(ctsio);
5449				ctl_done((union ctl_io *)ctsio);
5450			} else {
5451				ctl_copy_io((union ctl_io *)ctsio,
5452					    new_io);
5453				retval = lun->backend->config_write(new_io);
5454				ctl_set_success(ctsio);
5455				ctl_done((union ctl_io *)ctsio);
5456			}
5457		} else {
5458			retval = lun->backend->config_write(
5459				(union ctl_io *)ctsio);
5460		}
5461#ifdef NEEDTOPORT
5462	}
5463#endif
5464	return (retval);
5465}
5466
5467/*
5468 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5469 * we don't really do anything with the LBA and length fields if the user
5470 * passes them in.  Instead we'll just flush out the cache for the entire
5471 * LUN.
5472 */
5473int
5474ctl_sync_cache(struct ctl_scsiio *ctsio)
5475{
5476	struct ctl_lun *lun;
5477	struct ctl_softc *ctl_softc;
5478	uint64_t starting_lba;
5479	uint32_t block_count;
5480	int retval;
5481
5482	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5483
5484	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5485	ctl_softc = control_softc;
5486	retval = 0;
5487
5488	switch (ctsio->cdb[0]) {
5489	case SYNCHRONIZE_CACHE: {
5490		struct scsi_sync_cache *cdb;
5491		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5492
5493		starting_lba = scsi_4btoul(cdb->begin_lba);
5494		block_count = scsi_2btoul(cdb->lb_count);
5495		break;
5496	}
5497	case SYNCHRONIZE_CACHE_16: {
5498		struct scsi_sync_cache_16 *cdb;
5499		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5500
5501		starting_lba = scsi_8btou64(cdb->begin_lba);
5502		block_count = scsi_4btoul(cdb->lb_count);
5503		break;
5504	}
5505	default:
5506		ctl_set_invalid_opcode(ctsio);
5507		ctl_done((union ctl_io *)ctsio);
5508		goto bailout;
5509		break; /* NOTREACHED */
5510	}
5511
5512	/*
5513	 * We check the LBA and length, but don't do anything with them.
5514	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5515	 * get flushed.  This check will just help satisfy anyone who wants
5516	 * to see an error for an out of range LBA.
5517	 */
5518	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5519		ctl_set_lba_out_of_range(ctsio);
5520		ctl_done((union ctl_io *)ctsio);
5521		goto bailout;
5522	}
5523
5524	/*
5525	 * If this LUN has no backend, we can't flush the cache anyway.
5526	 */
5527	if (lun->backend == NULL) {
5528		ctl_set_invalid_opcode(ctsio);
5529		ctl_done((union ctl_io *)ctsio);
5530		goto bailout;
5531	}
5532
5533	/*
5534	 * Check to see whether we're configured to send the SYNCHRONIZE
5535	 * CACHE command directly to the back end.
5536	 */
5537	mtx_lock(&lun->lun_lock);
5538	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5539	 && (++(lun->sync_count) >= lun->sync_interval)) {
5540		lun->sync_count = 0;
5541		mtx_unlock(&lun->lun_lock);
5542		retval = lun->backend->config_write((union ctl_io *)ctsio);
5543	} else {
5544		mtx_unlock(&lun->lun_lock);
5545		ctl_set_success(ctsio);
5546		ctl_done((union ctl_io *)ctsio);
5547	}
5548
5549bailout:
5550
5551	return (retval);
5552}
5553
5554int
5555ctl_format(struct ctl_scsiio *ctsio)
5556{
5557	struct scsi_format *cdb;
5558	struct ctl_lun *lun;
5559	struct ctl_softc *ctl_softc;
5560	int length, defect_list_len;
5561
5562	CTL_DEBUG_PRINT(("ctl_format\n"));
5563
5564	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5565	ctl_softc = control_softc;
5566
5567	cdb = (struct scsi_format *)ctsio->cdb;
5568
5569	length = 0;
5570	if (cdb->byte2 & SF_FMTDATA) {
5571		if (cdb->byte2 & SF_LONGLIST)
5572			length = sizeof(struct scsi_format_header_long);
5573		else
5574			length = sizeof(struct scsi_format_header_short);
5575	}
5576
5577	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5578	 && (length > 0)) {
5579		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5580		ctsio->kern_data_len = length;
5581		ctsio->kern_total_len = length;
5582		ctsio->kern_data_resid = 0;
5583		ctsio->kern_rel_offset = 0;
5584		ctsio->kern_sg_entries = 0;
5585		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5586		ctsio->be_move_done = ctl_config_move_done;
5587		ctl_datamove((union ctl_io *)ctsio);
5588
5589		return (CTL_RETVAL_COMPLETE);
5590	}
5591
5592	defect_list_len = 0;
5593
5594	if (cdb->byte2 & SF_FMTDATA) {
5595		if (cdb->byte2 & SF_LONGLIST) {
5596			struct scsi_format_header_long *header;
5597
5598			header = (struct scsi_format_header_long *)
5599				ctsio->kern_data_ptr;
5600
5601			defect_list_len = scsi_4btoul(header->defect_list_len);
5602			if (defect_list_len != 0) {
5603				ctl_set_invalid_field(ctsio,
5604						      /*sks_valid*/ 1,
5605						      /*command*/ 0,
5606						      /*field*/ 2,
5607						      /*bit_valid*/ 0,
5608						      /*bit*/ 0);
5609				goto bailout;
5610			}
5611		} else {
5612			struct scsi_format_header_short *header;
5613
5614			header = (struct scsi_format_header_short *)
5615				ctsio->kern_data_ptr;
5616
5617			defect_list_len = scsi_2btoul(header->defect_list_len);
5618			if (defect_list_len != 0) {
5619				ctl_set_invalid_field(ctsio,
5620						      /*sks_valid*/ 1,
5621						      /*command*/ 0,
5622						      /*field*/ 2,
5623						      /*bit_valid*/ 0,
5624						      /*bit*/ 0);
5625				goto bailout;
5626			}
5627		}
5628	}
5629
5630	/*
5631	 * The format command will clear out the "Medium format corrupted"
5632	 * status if set by the configuration code.  That status is really
5633	 * just a way to notify the host that we have lost the media, and
5634	 * get them to issue a command that will basically make them think
5635	 * they're blowing away the media.
5636	 */
5637	mtx_lock(&lun->lun_lock);
5638	lun->flags &= ~CTL_LUN_INOPERABLE;
5639	mtx_unlock(&lun->lun_lock);
5640
5641	ctsio->scsi_status = SCSI_STATUS_OK;
5642	ctsio->io_hdr.status = CTL_SUCCESS;
5643bailout:
5644
5645	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5646		free(ctsio->kern_data_ptr, M_CTL);
5647		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5648	}
5649
5650	ctl_done((union ctl_io *)ctsio);
5651	return (CTL_RETVAL_COMPLETE);
5652}
5653
5654int
5655ctl_read_buffer(struct ctl_scsiio *ctsio)
5656{
5657	struct scsi_read_buffer *cdb;
5658	struct ctl_lun *lun;
5659	int buffer_offset, len;
5660	static uint8_t descr[4];
5661	static uint8_t echo_descr[4] = { 0 };
5662
5663	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5664
5665	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5666	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5667
5668	if (lun->flags & CTL_LUN_PR_RESERVED) {
5669		uint32_t residx;
5670
5671		/*
5672		 * XXX KDM need a lock here.
5673		 */
5674		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5675		if ((lun->res_type == SPR_TYPE_EX_AC
5676		  && residx != lun->pr_res_idx)
5677		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5678		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5679		  && !lun->per_res[residx].registered)) {
5680			ctl_set_reservation_conflict(ctsio);
5681			ctl_done((union ctl_io *)ctsio);
5682			return (CTL_RETVAL_COMPLETE);
5683	        }
5684	}
5685
5686	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5687	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5688	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5689		ctl_set_invalid_field(ctsio,
5690				      /*sks_valid*/ 1,
5691				      /*command*/ 1,
5692				      /*field*/ 1,
5693				      /*bit_valid*/ 1,
5694				      /*bit*/ 4);
5695		ctl_done((union ctl_io *)ctsio);
5696		return (CTL_RETVAL_COMPLETE);
5697	}
5698
5699	len = scsi_3btoul(cdb->length);
5700	buffer_offset = scsi_3btoul(cdb->offset);
5701
5702	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5703		ctl_set_invalid_field(ctsio,
5704				      /*sks_valid*/ 1,
5705				      /*command*/ 1,
5706				      /*field*/ 6,
5707				      /*bit_valid*/ 0,
5708				      /*bit*/ 0);
5709		ctl_done((union ctl_io *)ctsio);
5710		return (CTL_RETVAL_COMPLETE);
5711	}
5712
5713	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5714		descr[0] = 0;
5715		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5716		ctsio->kern_data_ptr = descr;
5717		len = min(len, sizeof(descr));
5718	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5719		ctsio->kern_data_ptr = echo_descr;
5720		len = min(len, sizeof(echo_descr));
5721	} else
5722		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5723	ctsio->kern_data_len = len;
5724	ctsio->kern_total_len = len;
5725	ctsio->kern_data_resid = 0;
5726	ctsio->kern_rel_offset = 0;
5727	ctsio->kern_sg_entries = 0;
5728	ctsio->be_move_done = ctl_config_move_done;
5729	ctl_datamove((union ctl_io *)ctsio);
5730
5731	return (CTL_RETVAL_COMPLETE);
5732}
5733
5734int
5735ctl_write_buffer(struct ctl_scsiio *ctsio)
5736{
5737	struct scsi_write_buffer *cdb;
5738	struct ctl_lun *lun;
5739	int buffer_offset, len;
5740
5741	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5742
5743	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5744	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5745
5746	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5747		ctl_set_invalid_field(ctsio,
5748				      /*sks_valid*/ 1,
5749				      /*command*/ 1,
5750				      /*field*/ 1,
5751				      /*bit_valid*/ 1,
5752				      /*bit*/ 4);
5753		ctl_done((union ctl_io *)ctsio);
5754		return (CTL_RETVAL_COMPLETE);
5755	}
5756
5757	len = scsi_3btoul(cdb->length);
5758	buffer_offset = scsi_3btoul(cdb->offset);
5759
5760	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5761		ctl_set_invalid_field(ctsio,
5762				      /*sks_valid*/ 1,
5763				      /*command*/ 1,
5764				      /*field*/ 6,
5765				      /*bit_valid*/ 0,
5766				      /*bit*/ 0);
5767		ctl_done((union ctl_io *)ctsio);
5768		return (CTL_RETVAL_COMPLETE);
5769	}
5770
5771	/*
5772	 * If we've got a kernel request that hasn't been malloced yet,
5773	 * malloc it and tell the caller the data buffer is here.
5774	 */
5775	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5776		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5777		ctsio->kern_data_len = len;
5778		ctsio->kern_total_len = len;
5779		ctsio->kern_data_resid = 0;
5780		ctsio->kern_rel_offset = 0;
5781		ctsio->kern_sg_entries = 0;
5782		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5783		ctsio->be_move_done = ctl_config_move_done;
5784		ctl_datamove((union ctl_io *)ctsio);
5785
5786		return (CTL_RETVAL_COMPLETE);
5787	}
5788
5789	ctl_done((union ctl_io *)ctsio);
5790
5791	return (CTL_RETVAL_COMPLETE);
5792}
5793
5794int
5795ctl_write_same(struct ctl_scsiio *ctsio)
5796{
5797	struct ctl_lun *lun;
5798	struct ctl_lba_len_flags *lbalen;
5799	uint64_t lba;
5800	uint32_t num_blocks;
5801	int len, retval;
5802	uint8_t byte2;
5803
5804	retval = CTL_RETVAL_COMPLETE;
5805
5806	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5807
5808	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5809
5810	switch (ctsio->cdb[0]) {
5811	case WRITE_SAME_10: {
5812		struct scsi_write_same_10 *cdb;
5813
5814		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5815
5816		lba = scsi_4btoul(cdb->addr);
5817		num_blocks = scsi_2btoul(cdb->length);
5818		byte2 = cdb->byte2;
5819		break;
5820	}
5821	case WRITE_SAME_16: {
5822		struct scsi_write_same_16 *cdb;
5823
5824		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5825
5826		lba = scsi_8btou64(cdb->addr);
5827		num_blocks = scsi_4btoul(cdb->length);
5828		byte2 = cdb->byte2;
5829		break;
5830	}
5831	default:
5832		/*
5833		 * We got a command we don't support.  This shouldn't
5834		 * happen, commands should be filtered out above us.
5835		 */
5836		ctl_set_invalid_opcode(ctsio);
5837		ctl_done((union ctl_io *)ctsio);
5838
5839		return (CTL_RETVAL_COMPLETE);
5840		break; /* NOTREACHED */
5841	}
5842
5843	/*
5844	 * The first check is to make sure we're in bounds, the second
5845	 * check is to catch wrap-around problems.  If the lba + num blocks
5846	 * is less than the lba, then we've wrapped around and the block
5847	 * range is invalid anyway.
5848	 */
5849	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5850	 || ((lba + num_blocks) < lba)) {
5851		ctl_set_lba_out_of_range(ctsio);
5852		ctl_done((union ctl_io *)ctsio);
5853		return (CTL_RETVAL_COMPLETE);
5854	}
5855
5856	/* Zero number of blocks means "to the last logical block" */
5857	if (num_blocks == 0) {
5858		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5859			ctl_set_invalid_field(ctsio,
5860					      /*sks_valid*/ 0,
5861					      /*command*/ 1,
5862					      /*field*/ 0,
5863					      /*bit_valid*/ 0,
5864					      /*bit*/ 0);
5865			ctl_done((union ctl_io *)ctsio);
5866			return (CTL_RETVAL_COMPLETE);
5867		}
5868		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5869	}
5870
5871	len = lun->be_lun->blocksize;
5872
5873	/*
5874	 * If we've got a kernel request that hasn't been malloced yet,
5875	 * malloc it and tell the caller the data buffer is here.
5876	 */
5877	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5878		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5879		ctsio->kern_data_len = len;
5880		ctsio->kern_total_len = len;
5881		ctsio->kern_data_resid = 0;
5882		ctsio->kern_rel_offset = 0;
5883		ctsio->kern_sg_entries = 0;
5884		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5885		ctsio->be_move_done = ctl_config_move_done;
5886		ctl_datamove((union ctl_io *)ctsio);
5887
5888		return (CTL_RETVAL_COMPLETE);
5889	}
5890
5891	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5892	lbalen->lba = lba;
5893	lbalen->len = num_blocks;
5894	lbalen->flags = byte2;
5895	retval = lun->backend->config_write((union ctl_io *)ctsio);
5896
5897	return (retval);
5898}
5899
5900int
5901ctl_unmap(struct ctl_scsiio *ctsio)
5902{
5903	struct ctl_lun *lun;
5904	struct scsi_unmap *cdb;
5905	struct ctl_ptr_len_flags *ptrlen;
5906	struct scsi_unmap_header *hdr;
5907	struct scsi_unmap_desc *buf, *end;
5908	uint64_t lba;
5909	uint32_t num_blocks;
5910	int len, retval;
5911	uint8_t byte2;
5912
5913	retval = CTL_RETVAL_COMPLETE;
5914
5915	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5916
5917	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5918	cdb = (struct scsi_unmap *)ctsio->cdb;
5919
5920	len = scsi_2btoul(cdb->length);
5921	byte2 = cdb->byte2;
5922
5923	/*
5924	 * If we've got a kernel request that hasn't been malloced yet,
5925	 * malloc it and tell the caller the data buffer is here.
5926	 */
5927	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5928		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5929		ctsio->kern_data_len = len;
5930		ctsio->kern_total_len = len;
5931		ctsio->kern_data_resid = 0;
5932		ctsio->kern_rel_offset = 0;
5933		ctsio->kern_sg_entries = 0;
5934		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5935		ctsio->be_move_done = ctl_config_move_done;
5936		ctl_datamove((union ctl_io *)ctsio);
5937
5938		return (CTL_RETVAL_COMPLETE);
5939	}
5940
5941	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5942	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5943	if (len < sizeof (*hdr) ||
5944	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5945	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5946	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5947		ctl_set_invalid_field(ctsio,
5948				      /*sks_valid*/ 0,
5949				      /*command*/ 0,
5950				      /*field*/ 0,
5951				      /*bit_valid*/ 0,
5952				      /*bit*/ 0);
5953		ctl_done((union ctl_io *)ctsio);
5954		return (CTL_RETVAL_COMPLETE);
5955	}
5956	len = scsi_2btoul(hdr->desc_length);
5957	buf = (struct scsi_unmap_desc *)(hdr + 1);
5958	end = buf + len / sizeof(*buf);
5959
5960	ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5961	ptrlen->ptr = (void *)buf;
5962	ptrlen->len = len;
5963	ptrlen->flags = byte2;
5964
5965	for (; buf < end; buf++) {
5966		lba = scsi_8btou64(buf->lba);
5967		num_blocks = scsi_4btoul(buf->length);
5968		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5969		 || ((lba + num_blocks) < lba)) {
5970			ctl_set_lba_out_of_range(ctsio);
5971			ctl_done((union ctl_io *)ctsio);
5972			return (CTL_RETVAL_COMPLETE);
5973		}
5974	}
5975
5976	retval = lun->backend->config_write((union ctl_io *)ctsio);
5977
5978	return (retval);
5979}
5980
5981/*
5982 * Note that this function currently doesn't actually do anything inside
5983 * CTL to enforce things if the DQue bit is turned on.
5984 *
5985 * Also note that this function can't be used in the default case, because
5986 * the DQue bit isn't set in the changeable mask for the control mode page
5987 * anyway.  This is just here as an example for how to implement a page
5988 * handler, and a placeholder in case we want to allow the user to turn
5989 * tagged queueing on and off.
5990 *
5991 * The D_SENSE bit handling is functional, however, and will turn
5992 * descriptor sense on and off for a given LUN.
5993 */
5994int
5995ctl_control_page_handler(struct ctl_scsiio *ctsio,
5996			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5997{
5998	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5999	struct ctl_lun *lun;
6000	struct ctl_softc *softc;
6001	int set_ua;
6002	uint32_t initidx;
6003
6004	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6005	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6006	set_ua = 0;
6007
6008	user_cp = (struct scsi_control_page *)page_ptr;
6009	current_cp = (struct scsi_control_page *)
6010		(page_index->page_data + (page_index->page_len *
6011		CTL_PAGE_CURRENT));
6012	saved_cp = (struct scsi_control_page *)
6013		(page_index->page_data + (page_index->page_len *
6014		CTL_PAGE_SAVED));
6015
6016	softc = control_softc;
6017
6018	mtx_lock(&lun->lun_lock);
6019	if (((current_cp->rlec & SCP_DSENSE) == 0)
6020	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6021		/*
6022		 * Descriptor sense is currently turned off and the user
6023		 * wants to turn it on.
6024		 */
6025		current_cp->rlec |= SCP_DSENSE;
6026		saved_cp->rlec |= SCP_DSENSE;
6027		lun->flags |= CTL_LUN_SENSE_DESC;
6028		set_ua = 1;
6029	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6030		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6031		/*
6032		 * Descriptor sense is currently turned on, and the user
6033		 * wants to turn it off.
6034		 */
6035		current_cp->rlec &= ~SCP_DSENSE;
6036		saved_cp->rlec &= ~SCP_DSENSE;
6037		lun->flags &= ~CTL_LUN_SENSE_DESC;
6038		set_ua = 1;
6039	}
6040	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
6041		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6042#ifdef NEEDTOPORT
6043			csevent_log(CSC_CTL | CSC_SHELF_SW |
6044				    CTL_UNTAG_TO_UNTAG,
6045				    csevent_LogType_Trace,
6046				    csevent_Severity_Information,
6047				    csevent_AlertLevel_Green,
6048				    csevent_FRU_Firmware,
6049				    csevent_FRU_Unknown,
6050				    "Received untagged to untagged transition");
6051#endif /* NEEDTOPORT */
6052		} else {
6053#ifdef NEEDTOPORT
6054			csevent_log(CSC_CTL | CSC_SHELF_SW |
6055				    CTL_UNTAG_TO_TAG,
6056				    csevent_LogType_ConfigChange,
6057				    csevent_Severity_Information,
6058				    csevent_AlertLevel_Green,
6059				    csevent_FRU_Firmware,
6060				    csevent_FRU_Unknown,
6061				    "Received untagged to tagged "
6062				    "queueing transition");
6063#endif /* NEEDTOPORT */
6064
6065			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6066			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6067			set_ua = 1;
6068		}
6069	} else {
6070		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6071#ifdef NEEDTOPORT
6072			csevent_log(CSC_CTL | CSC_SHELF_SW |
6073				    CTL_TAG_TO_UNTAG,
6074				    csevent_LogType_ConfigChange,
6075				    csevent_Severity_Warning,
6076				    csevent_AlertLevel_Yellow,
6077				    csevent_FRU_Firmware,
6078				    csevent_FRU_Unknown,
6079				    "Received tagged queueing to untagged "
6080				    "transition");
6081#endif /* NEEDTOPORT */
6082
6083			current_cp->queue_flags |= SCP_QUEUE_DQUE;
6084			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
6085			set_ua = 1;
6086		} else {
6087#ifdef NEEDTOPORT
6088			csevent_log(CSC_CTL | CSC_SHELF_SW |
6089				    CTL_TAG_TO_TAG,
6090				    csevent_LogType_Trace,
6091				    csevent_Severity_Information,
6092				    csevent_AlertLevel_Green,
6093				    csevent_FRU_Firmware,
6094				    csevent_FRU_Unknown,
6095				    "Received tagged queueing to tagged "
6096				    "queueing transition");
6097#endif /* NEEDTOPORT */
6098		}
6099	}
6100	if (set_ua != 0) {
6101		int i;
6102		/*
6103		 * Let other initiators know that the mode
6104		 * parameters for this LUN have changed.
6105		 */
6106		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6107			if (i == initidx)
6108				continue;
6109
6110			lun->pending_sense[i].ua_pending |=
6111				CTL_UA_MODE_CHANGE;
6112		}
6113	}
6114	mtx_unlock(&lun->lun_lock);
6115
6116	return (0);
6117}
6118
6119int
6120ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6121		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6122{
6123	return (0);
6124}
6125
6126int
6127ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6128			   struct ctl_page_index *page_index, int pc)
6129{
6130	struct copan_power_subpage *page;
6131
6132	page = (struct copan_power_subpage *)page_index->page_data +
6133		(page_index->page_len * pc);
6134
6135	switch (pc) {
6136	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6137		/*
6138		 * We don't update the changable bits for this page.
6139		 */
6140		break;
6141	case SMS_PAGE_CTRL_CURRENT >> 6:
6142	case SMS_PAGE_CTRL_DEFAULT >> 6:
6143	case SMS_PAGE_CTRL_SAVED >> 6:
6144#ifdef NEEDTOPORT
6145		ctl_update_power_subpage(page);
6146#endif
6147		break;
6148	default:
6149#ifdef NEEDTOPORT
6150		EPRINT(0, "Invalid PC %d!!", pc);
6151#endif
6152		break;
6153	}
6154	return (0);
6155}
6156
6157
6158int
6159ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6160		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6161{
6162	struct copan_aps_subpage *user_sp;
6163	struct copan_aps_subpage *current_sp;
6164	union ctl_modepage_info *modepage_info;
6165	struct ctl_softc *softc;
6166	struct ctl_lun *lun;
6167	int retval;
6168
6169	retval = CTL_RETVAL_COMPLETE;
6170	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6171		     (page_index->page_len * CTL_PAGE_CURRENT));
6172	softc = control_softc;
6173	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6174
6175	user_sp = (struct copan_aps_subpage *)page_ptr;
6176
6177	modepage_info = (union ctl_modepage_info *)
6178		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6179
6180	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6181	modepage_info->header.subpage = page_index->subpage;
6182	modepage_info->aps.lock_active = user_sp->lock_active;
6183
6184	mtx_lock(&softc->ctl_lock);
6185
6186	/*
6187	 * If there is a request to lock the LUN and another LUN is locked
6188	 * this is an error. If the requested LUN is already locked ignore
6189	 * the request. If no LUN is locked attempt to lock it.
6190	 * if there is a request to unlock the LUN and the LUN is currently
6191	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6192	 * if another LUN is locked or no LUN is locked.
6193	 */
6194	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6195		if (softc->aps_locked_lun == lun->lun) {
6196			/*
6197			 * This LUN is already locked, so we're done.
6198			 */
6199			retval = CTL_RETVAL_COMPLETE;
6200		} else if (softc->aps_locked_lun == 0) {
6201			/*
6202			 * No one has the lock, pass the request to the
6203			 * backend.
6204			 */
6205			retval = lun->backend->config_write(
6206				(union ctl_io *)ctsio);
6207		} else {
6208			/*
6209			 * Someone else has the lock, throw out the request.
6210			 */
6211			ctl_set_already_locked(ctsio);
6212			free(ctsio->kern_data_ptr, M_CTL);
6213			ctl_done((union ctl_io *)ctsio);
6214
6215			/*
6216			 * Set the return value so that ctl_do_mode_select()
6217			 * won't try to complete the command.  We already
6218			 * completed it here.
6219			 */
6220			retval = CTL_RETVAL_ERROR;
6221		}
6222	} else if (softc->aps_locked_lun == lun->lun) {
6223		/*
6224		 * This LUN is locked, so pass the unlock request to the
6225		 * backend.
6226		 */
6227		retval = lun->backend->config_write((union ctl_io *)ctsio);
6228	}
6229	mtx_unlock(&softc->ctl_lock);
6230
6231	return (retval);
6232}
6233
6234int
6235ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6236				struct ctl_page_index *page_index,
6237				uint8_t *page_ptr)
6238{
6239	uint8_t *c;
6240	int i;
6241
6242	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6243	ctl_time_io_secs =
6244		(c[0] << 8) |
6245		(c[1] << 0) |
6246		0;
6247	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6248	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6249	printf("page data:");
6250	for (i=0; i<8; i++)
6251		printf(" %.2x",page_ptr[i]);
6252	printf("\n");
6253	return (0);
6254}
6255
6256int
6257ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6258			       struct ctl_page_index *page_index,
6259			       int pc)
6260{
6261	struct copan_debugconf_subpage *page;
6262
6263	page = (struct copan_debugconf_subpage *)page_index->page_data +
6264		(page_index->page_len * pc);
6265
6266	switch (pc) {
6267	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6268	case SMS_PAGE_CTRL_DEFAULT >> 6:
6269	case SMS_PAGE_CTRL_SAVED >> 6:
6270		/*
6271		 * We don't update the changable or default bits for this page.
6272		 */
6273		break;
6274	case SMS_PAGE_CTRL_CURRENT >> 6:
6275		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6276		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6277		break;
6278	default:
6279#ifdef NEEDTOPORT
6280		EPRINT(0, "Invalid PC %d!!", pc);
6281#endif /* NEEDTOPORT */
6282		break;
6283	}
6284	return (0);
6285}
6286
6287
6288static int
6289ctl_do_mode_select(union ctl_io *io)
6290{
6291	struct scsi_mode_page_header *page_header;
6292	struct ctl_page_index *page_index;
6293	struct ctl_scsiio *ctsio;
6294	int control_dev, page_len;
6295	int page_len_offset, page_len_size;
6296	union ctl_modepage_info *modepage_info;
6297	struct ctl_lun *lun;
6298	int *len_left, *len_used;
6299	int retval, i;
6300
6301	ctsio = &io->scsiio;
6302	page_index = NULL;
6303	page_len = 0;
6304	retval = CTL_RETVAL_COMPLETE;
6305
6306	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6307
6308	if (lun->be_lun->lun_type != T_DIRECT)
6309		control_dev = 1;
6310	else
6311		control_dev = 0;
6312
6313	modepage_info = (union ctl_modepage_info *)
6314		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6315	len_left = &modepage_info->header.len_left;
6316	len_used = &modepage_info->header.len_used;
6317
6318do_next_page:
6319
6320	page_header = (struct scsi_mode_page_header *)
6321		(ctsio->kern_data_ptr + *len_used);
6322
6323	if (*len_left == 0) {
6324		free(ctsio->kern_data_ptr, M_CTL);
6325		ctl_set_success(ctsio);
6326		ctl_done((union ctl_io *)ctsio);
6327		return (CTL_RETVAL_COMPLETE);
6328	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6329
6330		free(ctsio->kern_data_ptr, M_CTL);
6331		ctl_set_param_len_error(ctsio);
6332		ctl_done((union ctl_io *)ctsio);
6333		return (CTL_RETVAL_COMPLETE);
6334
6335	} else if ((page_header->page_code & SMPH_SPF)
6336		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6337
6338		free(ctsio->kern_data_ptr, M_CTL);
6339		ctl_set_param_len_error(ctsio);
6340		ctl_done((union ctl_io *)ctsio);
6341		return (CTL_RETVAL_COMPLETE);
6342	}
6343
6344
6345	/*
6346	 * XXX KDM should we do something with the block descriptor?
6347	 */
6348	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6349
6350		if ((control_dev != 0)
6351		 && (lun->mode_pages.index[i].page_flags &
6352		     CTL_PAGE_FLAG_DISK_ONLY))
6353			continue;
6354
6355		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6356		    (page_header->page_code & SMPH_PC_MASK))
6357			continue;
6358
6359		/*
6360		 * If neither page has a subpage code, then we've got a
6361		 * match.
6362		 */
6363		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6364		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6365			page_index = &lun->mode_pages.index[i];
6366			page_len = page_header->page_length;
6367			break;
6368		}
6369
6370		/*
6371		 * If both pages have subpages, then the subpage numbers
6372		 * have to match.
6373		 */
6374		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6375		  && (page_header->page_code & SMPH_SPF)) {
6376			struct scsi_mode_page_header_sp *sph;
6377
6378			sph = (struct scsi_mode_page_header_sp *)page_header;
6379
6380			if (lun->mode_pages.index[i].subpage ==
6381			    sph->subpage) {
6382				page_index = &lun->mode_pages.index[i];
6383				page_len = scsi_2btoul(sph->page_length);
6384				break;
6385			}
6386		}
6387	}
6388
6389	/*
6390	 * If we couldn't find the page, or if we don't have a mode select
6391	 * handler for it, send back an error to the user.
6392	 */
6393	if ((page_index == NULL)
6394	 || (page_index->select_handler == NULL)) {
6395		ctl_set_invalid_field(ctsio,
6396				      /*sks_valid*/ 1,
6397				      /*command*/ 0,
6398				      /*field*/ *len_used,
6399				      /*bit_valid*/ 0,
6400				      /*bit*/ 0);
6401		free(ctsio->kern_data_ptr, M_CTL);
6402		ctl_done((union ctl_io *)ctsio);
6403		return (CTL_RETVAL_COMPLETE);
6404	}
6405
6406	if (page_index->page_code & SMPH_SPF) {
6407		page_len_offset = 2;
6408		page_len_size = 2;
6409	} else {
6410		page_len_size = 1;
6411		page_len_offset = 1;
6412	}
6413
6414	/*
6415	 * If the length the initiator gives us isn't the one we specify in
6416	 * the mode page header, or if they didn't specify enough data in
6417	 * the CDB to avoid truncating this page, kick out the request.
6418	 */
6419	if ((page_len != (page_index->page_len - page_len_offset -
6420			  page_len_size))
6421	 || (*len_left < page_index->page_len)) {
6422
6423
6424		ctl_set_invalid_field(ctsio,
6425				      /*sks_valid*/ 1,
6426				      /*command*/ 0,
6427				      /*field*/ *len_used + page_len_offset,
6428				      /*bit_valid*/ 0,
6429				      /*bit*/ 0);
6430		free(ctsio->kern_data_ptr, M_CTL);
6431		ctl_done((union ctl_io *)ctsio);
6432		return (CTL_RETVAL_COMPLETE);
6433	}
6434
6435	/*
6436	 * Run through the mode page, checking to make sure that the bits
6437	 * the user changed are actually legal for him to change.
6438	 */
6439	for (i = 0; i < page_index->page_len; i++) {
6440		uint8_t *user_byte, *change_mask, *current_byte;
6441		int bad_bit;
6442		int j;
6443
6444		user_byte = (uint8_t *)page_header + i;
6445		change_mask = page_index->page_data +
6446			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6447		current_byte = page_index->page_data +
6448			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6449
6450		/*
6451		 * Check to see whether the user set any bits in this byte
6452		 * that he is not allowed to set.
6453		 */
6454		if ((*user_byte & ~(*change_mask)) ==
6455		    (*current_byte & ~(*change_mask)))
6456			continue;
6457
6458		/*
6459		 * Go through bit by bit to determine which one is illegal.
6460		 */
6461		bad_bit = 0;
6462		for (j = 7; j >= 0; j--) {
6463			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6464			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6465				bad_bit = i;
6466				break;
6467			}
6468		}
6469		ctl_set_invalid_field(ctsio,
6470				      /*sks_valid*/ 1,
6471				      /*command*/ 0,
6472				      /*field*/ *len_used + i,
6473				      /*bit_valid*/ 1,
6474				      /*bit*/ bad_bit);
6475		free(ctsio->kern_data_ptr, M_CTL);
6476		ctl_done((union ctl_io *)ctsio);
6477		return (CTL_RETVAL_COMPLETE);
6478	}
6479
6480	/*
6481	 * Decrement these before we call the page handler, since we may
6482	 * end up getting called back one way or another before the handler
6483	 * returns to this context.
6484	 */
6485	*len_left -= page_index->page_len;
6486	*len_used += page_index->page_len;
6487
6488	retval = page_index->select_handler(ctsio, page_index,
6489					    (uint8_t *)page_header);
6490
6491	/*
6492	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6493	 * wait until this queued command completes to finish processing
6494	 * the mode page.  If it returns anything other than
6495	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6496	 * already set the sense information, freed the data pointer, and
6497	 * completed the io for us.
6498	 */
6499	if (retval != CTL_RETVAL_COMPLETE)
6500		goto bailout_no_done;
6501
6502	/*
6503	 * If the initiator sent us more than one page, parse the next one.
6504	 */
6505	if (*len_left > 0)
6506		goto do_next_page;
6507
6508	ctl_set_success(ctsio);
6509	free(ctsio->kern_data_ptr, M_CTL);
6510	ctl_done((union ctl_io *)ctsio);
6511
6512bailout_no_done:
6513
6514	return (CTL_RETVAL_COMPLETE);
6515
6516}
6517
6518int
6519ctl_mode_select(struct ctl_scsiio *ctsio)
6520{
6521	int param_len, pf, sp;
6522	int header_size, bd_len;
6523	int len_left, len_used;
6524	struct ctl_page_index *page_index;
6525	struct ctl_lun *lun;
6526	int control_dev, page_len;
6527	union ctl_modepage_info *modepage_info;
6528	int retval;
6529
6530	pf = 0;
6531	sp = 0;
6532	page_len = 0;
6533	len_used = 0;
6534	len_left = 0;
6535	retval = 0;
6536	bd_len = 0;
6537	page_index = NULL;
6538
6539	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6540
6541	if (lun->be_lun->lun_type != T_DIRECT)
6542		control_dev = 1;
6543	else
6544		control_dev = 0;
6545
6546	switch (ctsio->cdb[0]) {
6547	case MODE_SELECT_6: {
6548		struct scsi_mode_select_6 *cdb;
6549
6550		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6551
6552		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6553		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6554
6555		param_len = cdb->length;
6556		header_size = sizeof(struct scsi_mode_header_6);
6557		break;
6558	}
6559	case MODE_SELECT_10: {
6560		struct scsi_mode_select_10 *cdb;
6561
6562		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6563
6564		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6565		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6566
6567		param_len = scsi_2btoul(cdb->length);
6568		header_size = sizeof(struct scsi_mode_header_10);
6569		break;
6570	}
6571	default:
6572		ctl_set_invalid_opcode(ctsio);
6573		ctl_done((union ctl_io *)ctsio);
6574		return (CTL_RETVAL_COMPLETE);
6575		break; /* NOTREACHED */
6576	}
6577
6578	/*
6579	 * From SPC-3:
6580	 * "A parameter list length of zero indicates that the Data-Out Buffer
6581	 * shall be empty. This condition shall not be considered as an error."
6582	 */
6583	if (param_len == 0) {
6584		ctl_set_success(ctsio);
6585		ctl_done((union ctl_io *)ctsio);
6586		return (CTL_RETVAL_COMPLETE);
6587	}
6588
6589	/*
6590	 * Since we'll hit this the first time through, prior to
6591	 * allocation, we don't need to free a data buffer here.
6592	 */
6593	if (param_len < header_size) {
6594		ctl_set_param_len_error(ctsio);
6595		ctl_done((union ctl_io *)ctsio);
6596		return (CTL_RETVAL_COMPLETE);
6597	}
6598
6599	/*
6600	 * Allocate the data buffer and grab the user's data.  In theory,
6601	 * we shouldn't have to sanity check the parameter list length here
6602	 * because the maximum size is 64K.  We should be able to malloc
6603	 * that much without too many problems.
6604	 */
6605	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6606		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6607		ctsio->kern_data_len = param_len;
6608		ctsio->kern_total_len = param_len;
6609		ctsio->kern_data_resid = 0;
6610		ctsio->kern_rel_offset = 0;
6611		ctsio->kern_sg_entries = 0;
6612		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6613		ctsio->be_move_done = ctl_config_move_done;
6614		ctl_datamove((union ctl_io *)ctsio);
6615
6616		return (CTL_RETVAL_COMPLETE);
6617	}
6618
6619	switch (ctsio->cdb[0]) {
6620	case MODE_SELECT_6: {
6621		struct scsi_mode_header_6 *mh6;
6622
6623		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6624		bd_len = mh6->blk_desc_len;
6625		break;
6626	}
6627	case MODE_SELECT_10: {
6628		struct scsi_mode_header_10 *mh10;
6629
6630		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6631		bd_len = scsi_2btoul(mh10->blk_desc_len);
6632		break;
6633	}
6634	default:
6635		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6636		break;
6637	}
6638
6639	if (param_len < (header_size + bd_len)) {
6640		free(ctsio->kern_data_ptr, M_CTL);
6641		ctl_set_param_len_error(ctsio);
6642		ctl_done((union ctl_io *)ctsio);
6643		return (CTL_RETVAL_COMPLETE);
6644	}
6645
6646	/*
6647	 * Set the IO_CONT flag, so that if this I/O gets passed to
6648	 * ctl_config_write_done(), it'll get passed back to
6649	 * ctl_do_mode_select() for further processing, or completion if
6650	 * we're all done.
6651	 */
6652	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6653	ctsio->io_cont = ctl_do_mode_select;
6654
6655	modepage_info = (union ctl_modepage_info *)
6656		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6657
6658	memset(modepage_info, 0, sizeof(*modepage_info));
6659
6660	len_left = param_len - header_size - bd_len;
6661	len_used = header_size + bd_len;
6662
6663	modepage_info->header.len_left = len_left;
6664	modepage_info->header.len_used = len_used;
6665
6666	return (ctl_do_mode_select((union ctl_io *)ctsio));
6667}
6668
6669int
6670ctl_mode_sense(struct ctl_scsiio *ctsio)
6671{
6672	struct ctl_lun *lun;
6673	int pc, page_code, dbd, llba, subpage;
6674	int alloc_len, page_len, header_len, total_len;
6675	struct scsi_mode_block_descr *block_desc;
6676	struct ctl_page_index *page_index;
6677	int control_dev;
6678
6679	dbd = 0;
6680	llba = 0;
6681	block_desc = NULL;
6682	page_index = NULL;
6683
6684	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6685
6686	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6687
6688	if (lun->be_lun->lun_type != T_DIRECT)
6689		control_dev = 1;
6690	else
6691		control_dev = 0;
6692
6693	if (lun->flags & CTL_LUN_PR_RESERVED) {
6694		uint32_t residx;
6695
6696		/*
6697		 * XXX KDM need a lock here.
6698		 */
6699		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6700		if ((lun->res_type == SPR_TYPE_EX_AC
6701		  && residx != lun->pr_res_idx)
6702		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6703		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6704		  && !lun->per_res[residx].registered)) {
6705			ctl_set_reservation_conflict(ctsio);
6706			ctl_done((union ctl_io *)ctsio);
6707			return (CTL_RETVAL_COMPLETE);
6708		}
6709	}
6710
6711	switch (ctsio->cdb[0]) {
6712	case MODE_SENSE_6: {
6713		struct scsi_mode_sense_6 *cdb;
6714
6715		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6716
6717		header_len = sizeof(struct scsi_mode_hdr_6);
6718		if (cdb->byte2 & SMS_DBD)
6719			dbd = 1;
6720		else
6721			header_len += sizeof(struct scsi_mode_block_descr);
6722
6723		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6724		page_code = cdb->page & SMS_PAGE_CODE;
6725		subpage = cdb->subpage;
6726		alloc_len = cdb->length;
6727		break;
6728	}
6729	case MODE_SENSE_10: {
6730		struct scsi_mode_sense_10 *cdb;
6731
6732		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6733
6734		header_len = sizeof(struct scsi_mode_hdr_10);
6735
6736		if (cdb->byte2 & SMS_DBD)
6737			dbd = 1;
6738		else
6739			header_len += sizeof(struct scsi_mode_block_descr);
6740		if (cdb->byte2 & SMS10_LLBAA)
6741			llba = 1;
6742		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6743		page_code = cdb->page & SMS_PAGE_CODE;
6744		subpage = cdb->subpage;
6745		alloc_len = scsi_2btoul(cdb->length);
6746		break;
6747	}
6748	default:
6749		ctl_set_invalid_opcode(ctsio);
6750		ctl_done((union ctl_io *)ctsio);
6751		return (CTL_RETVAL_COMPLETE);
6752		break; /* NOTREACHED */
6753	}
6754
6755	/*
6756	 * We have to make a first pass through to calculate the size of
6757	 * the pages that match the user's query.  Then we allocate enough
6758	 * memory to hold it, and actually copy the data into the buffer.
6759	 */
6760	switch (page_code) {
6761	case SMS_ALL_PAGES_PAGE: {
6762		int i;
6763
6764		page_len = 0;
6765
6766		/*
6767		 * At the moment, values other than 0 and 0xff here are
6768		 * reserved according to SPC-3.
6769		 */
6770		if ((subpage != SMS_SUBPAGE_PAGE_0)
6771		 && (subpage != SMS_SUBPAGE_ALL)) {
6772			ctl_set_invalid_field(ctsio,
6773					      /*sks_valid*/ 1,
6774					      /*command*/ 1,
6775					      /*field*/ 3,
6776					      /*bit_valid*/ 0,
6777					      /*bit*/ 0);
6778			ctl_done((union ctl_io *)ctsio);
6779			return (CTL_RETVAL_COMPLETE);
6780		}
6781
6782		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6783			if ((control_dev != 0)
6784			 && (lun->mode_pages.index[i].page_flags &
6785			     CTL_PAGE_FLAG_DISK_ONLY))
6786				continue;
6787
6788			/*
6789			 * We don't use this subpage if the user didn't
6790			 * request all subpages.
6791			 */
6792			if ((lun->mode_pages.index[i].subpage != 0)
6793			 && (subpage == SMS_SUBPAGE_PAGE_0))
6794				continue;
6795
6796#if 0
6797			printf("found page %#x len %d\n",
6798			       lun->mode_pages.index[i].page_code &
6799			       SMPH_PC_MASK,
6800			       lun->mode_pages.index[i].page_len);
6801#endif
6802			page_len += lun->mode_pages.index[i].page_len;
6803		}
6804		break;
6805	}
6806	default: {
6807		int i;
6808
6809		page_len = 0;
6810
6811		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6812			/* Look for the right page code */
6813			if ((lun->mode_pages.index[i].page_code &
6814			     SMPH_PC_MASK) != page_code)
6815				continue;
6816
6817			/* Look for the right subpage or the subpage wildcard*/
6818			if ((lun->mode_pages.index[i].subpage != subpage)
6819			 && (subpage != SMS_SUBPAGE_ALL))
6820				continue;
6821
6822			/* Make sure the page is supported for this dev type */
6823			if ((control_dev != 0)
6824			 && (lun->mode_pages.index[i].page_flags &
6825			     CTL_PAGE_FLAG_DISK_ONLY))
6826				continue;
6827
6828#if 0
6829			printf("found page %#x len %d\n",
6830			       lun->mode_pages.index[i].page_code &
6831			       SMPH_PC_MASK,
6832			       lun->mode_pages.index[i].page_len);
6833#endif
6834
6835			page_len += lun->mode_pages.index[i].page_len;
6836		}
6837
6838		if (page_len == 0) {
6839			ctl_set_invalid_field(ctsio,
6840					      /*sks_valid*/ 1,
6841					      /*command*/ 1,
6842					      /*field*/ 2,
6843					      /*bit_valid*/ 1,
6844					      /*bit*/ 5);
6845			ctl_done((union ctl_io *)ctsio);
6846			return (CTL_RETVAL_COMPLETE);
6847		}
6848		break;
6849	}
6850	}
6851
6852	total_len = header_len + page_len;
6853#if 0
6854	printf("header_len = %d, page_len = %d, total_len = %d\n",
6855	       header_len, page_len, total_len);
6856#endif
6857
6858	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6859	ctsio->kern_sg_entries = 0;
6860	ctsio->kern_data_resid = 0;
6861	ctsio->kern_rel_offset = 0;
6862	if (total_len < alloc_len) {
6863		ctsio->residual = alloc_len - total_len;
6864		ctsio->kern_data_len = total_len;
6865		ctsio->kern_total_len = total_len;
6866	} else {
6867		ctsio->residual = 0;
6868		ctsio->kern_data_len = alloc_len;
6869		ctsio->kern_total_len = alloc_len;
6870	}
6871
6872	switch (ctsio->cdb[0]) {
6873	case MODE_SENSE_6: {
6874		struct scsi_mode_hdr_6 *header;
6875
6876		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6877
6878		header->datalen = ctl_min(total_len - 1, 254);
6879
6880		if (dbd)
6881			header->block_descr_len = 0;
6882		else
6883			header->block_descr_len =
6884				sizeof(struct scsi_mode_block_descr);
6885		block_desc = (struct scsi_mode_block_descr *)&header[1];
6886		break;
6887	}
6888	case MODE_SENSE_10: {
6889		struct scsi_mode_hdr_10 *header;
6890		int datalen;
6891
6892		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6893
6894		datalen = ctl_min(total_len - 2, 65533);
6895		scsi_ulto2b(datalen, header->datalen);
6896		if (dbd)
6897			scsi_ulto2b(0, header->block_descr_len);
6898		else
6899			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6900				    header->block_descr_len);
6901		block_desc = (struct scsi_mode_block_descr *)&header[1];
6902		break;
6903	}
6904	default:
6905		panic("invalid CDB type %#x", ctsio->cdb[0]);
6906		break; /* NOTREACHED */
6907	}
6908
6909	/*
6910	 * If we've got a disk, use its blocksize in the block
6911	 * descriptor.  Otherwise, just set it to 0.
6912	 */
6913	if (dbd == 0) {
6914		if (control_dev != 0)
6915			scsi_ulto3b(lun->be_lun->blocksize,
6916				    block_desc->block_len);
6917		else
6918			scsi_ulto3b(0, block_desc->block_len);
6919	}
6920
6921	switch (page_code) {
6922	case SMS_ALL_PAGES_PAGE: {
6923		int i, data_used;
6924
6925		data_used = header_len;
6926		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6927			struct ctl_page_index *page_index;
6928
6929			page_index = &lun->mode_pages.index[i];
6930
6931			if ((control_dev != 0)
6932			 && (page_index->page_flags &
6933			    CTL_PAGE_FLAG_DISK_ONLY))
6934				continue;
6935
6936			/*
6937			 * We don't use this subpage if the user didn't
6938			 * request all subpages.  We already checked (above)
6939			 * to make sure the user only specified a subpage
6940			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6941			 */
6942			if ((page_index->subpage != 0)
6943			 && (subpage == SMS_SUBPAGE_PAGE_0))
6944				continue;
6945
6946			/*
6947			 * Call the handler, if it exists, to update the
6948			 * page to the latest values.
6949			 */
6950			if (page_index->sense_handler != NULL)
6951				page_index->sense_handler(ctsio, page_index,pc);
6952
6953			memcpy(ctsio->kern_data_ptr + data_used,
6954			       page_index->page_data +
6955			       (page_index->page_len * pc),
6956			       page_index->page_len);
6957			data_used += page_index->page_len;
6958		}
6959		break;
6960	}
6961	default: {
6962		int i, data_used;
6963
6964		data_used = header_len;
6965
6966		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6967			struct ctl_page_index *page_index;
6968
6969			page_index = &lun->mode_pages.index[i];
6970
6971			/* Look for the right page code */
6972			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6973				continue;
6974
6975			/* Look for the right subpage or the subpage wildcard*/
6976			if ((page_index->subpage != subpage)
6977			 && (subpage != SMS_SUBPAGE_ALL))
6978				continue;
6979
6980			/* Make sure the page is supported for this dev type */
6981			if ((control_dev != 0)
6982			 && (page_index->page_flags &
6983			     CTL_PAGE_FLAG_DISK_ONLY))
6984				continue;
6985
6986			/*
6987			 * Call the handler, if it exists, to update the
6988			 * page to the latest values.
6989			 */
6990			if (page_index->sense_handler != NULL)
6991				page_index->sense_handler(ctsio, page_index,pc);
6992
6993			memcpy(ctsio->kern_data_ptr + data_used,
6994			       page_index->page_data +
6995			       (page_index->page_len * pc),
6996			       page_index->page_len);
6997			data_used += page_index->page_len;
6998		}
6999		break;
7000	}
7001	}
7002
7003	ctsio->scsi_status = SCSI_STATUS_OK;
7004
7005	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7006	ctsio->be_move_done = ctl_config_move_done;
7007	ctl_datamove((union ctl_io *)ctsio);
7008
7009	return (CTL_RETVAL_COMPLETE);
7010}
7011
7012int
7013ctl_read_capacity(struct ctl_scsiio *ctsio)
7014{
7015	struct scsi_read_capacity *cdb;
7016	struct scsi_read_capacity_data *data;
7017	struct ctl_lun *lun;
7018	uint32_t lba;
7019
7020	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7021
7022	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7023
7024	lba = scsi_4btoul(cdb->addr);
7025	if (((cdb->pmi & SRC_PMI) == 0)
7026	 && (lba != 0)) {
7027		ctl_set_invalid_field(/*ctsio*/ ctsio,
7028				      /*sks_valid*/ 1,
7029				      /*command*/ 1,
7030				      /*field*/ 2,
7031				      /*bit_valid*/ 0,
7032				      /*bit*/ 0);
7033		ctl_done((union ctl_io *)ctsio);
7034		return (CTL_RETVAL_COMPLETE);
7035	}
7036
7037	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7038
7039	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7040	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7041	ctsio->residual = 0;
7042	ctsio->kern_data_len = sizeof(*data);
7043	ctsio->kern_total_len = sizeof(*data);
7044	ctsio->kern_data_resid = 0;
7045	ctsio->kern_rel_offset = 0;
7046	ctsio->kern_sg_entries = 0;
7047
7048	/*
7049	 * If the maximum LBA is greater than 0xfffffffe, the user must
7050	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7051	 * serivce action set.
7052	 */
7053	if (lun->be_lun->maxlba > 0xfffffffe)
7054		scsi_ulto4b(0xffffffff, data->addr);
7055	else
7056		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7057
7058	/*
7059	 * XXX KDM this may not be 512 bytes...
7060	 */
7061	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7062
7063	ctsio->scsi_status = SCSI_STATUS_OK;
7064
7065	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7066	ctsio->be_move_done = ctl_config_move_done;
7067	ctl_datamove((union ctl_io *)ctsio);
7068
7069	return (CTL_RETVAL_COMPLETE);
7070}
7071
7072int
7073ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7074{
7075	struct scsi_read_capacity_16 *cdb;
7076	struct scsi_read_capacity_data_long *data;
7077	struct ctl_lun *lun;
7078	uint64_t lba;
7079	uint32_t alloc_len;
7080
7081	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7082
7083	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7084
7085	alloc_len = scsi_4btoul(cdb->alloc_len);
7086	lba = scsi_8btou64(cdb->addr);
7087
7088	if ((cdb->reladr & SRC16_PMI)
7089	 && (lba != 0)) {
7090		ctl_set_invalid_field(/*ctsio*/ ctsio,
7091				      /*sks_valid*/ 1,
7092				      /*command*/ 1,
7093				      /*field*/ 2,
7094				      /*bit_valid*/ 0,
7095				      /*bit*/ 0);
7096		ctl_done((union ctl_io *)ctsio);
7097		return (CTL_RETVAL_COMPLETE);
7098	}
7099
7100	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7101
7102	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7103	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7104
7105	if (sizeof(*data) < alloc_len) {
7106		ctsio->residual = alloc_len - sizeof(*data);
7107		ctsio->kern_data_len = sizeof(*data);
7108		ctsio->kern_total_len = sizeof(*data);
7109	} else {
7110		ctsio->residual = 0;
7111		ctsio->kern_data_len = alloc_len;
7112		ctsio->kern_total_len = alloc_len;
7113	}
7114	ctsio->kern_data_resid = 0;
7115	ctsio->kern_rel_offset = 0;
7116	ctsio->kern_sg_entries = 0;
7117
7118	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7119	/* XXX KDM this may not be 512 bytes... */
7120	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7121	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7122	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7123	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7124		data->lalba_lbp[0] |= SRC16_LBPME;
7125
7126	ctsio->scsi_status = SCSI_STATUS_OK;
7127
7128	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7129	ctsio->be_move_done = ctl_config_move_done;
7130	ctl_datamove((union ctl_io *)ctsio);
7131
7132	return (CTL_RETVAL_COMPLETE);
7133}
7134
7135int
7136ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7137{
7138	struct scsi_maintenance_in *cdb;
7139	int retval;
7140	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7141	int num_target_port_groups, num_target_ports, single;
7142	struct ctl_lun *lun;
7143	struct ctl_softc *softc;
7144	struct ctl_port *port;
7145	struct scsi_target_group_data *rtg_ptr;
7146	struct scsi_target_group_data_extended *rtg_ext_ptr;
7147	struct scsi_target_port_group_descriptor *tpg_desc;
7148
7149	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7150
7151	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7152	softc = control_softc;
7153	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7154
7155	retval = CTL_RETVAL_COMPLETE;
7156
7157	switch (cdb->byte2 & STG_PDF_MASK) {
7158	case STG_PDF_LENGTH:
7159		ext = 0;
7160		break;
7161	case STG_PDF_EXTENDED:
7162		ext = 1;
7163		break;
7164	default:
7165		ctl_set_invalid_field(/*ctsio*/ ctsio,
7166				      /*sks_valid*/ 1,
7167				      /*command*/ 1,
7168				      /*field*/ 2,
7169				      /*bit_valid*/ 1,
7170				      /*bit*/ 5);
7171		ctl_done((union ctl_io *)ctsio);
7172		return(retval);
7173	}
7174
7175	single = ctl_is_single;
7176	if (single)
7177		num_target_port_groups = 1;
7178	else
7179		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7180	num_target_ports = 0;
7181	mtx_lock(&softc->ctl_lock);
7182	STAILQ_FOREACH(port, &softc->port_list, links) {
7183		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7184			continue;
7185		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7186			continue;
7187		num_target_ports++;
7188	}
7189	mtx_unlock(&softc->ctl_lock);
7190
7191	if (ext)
7192		total_len = sizeof(struct scsi_target_group_data_extended);
7193	else
7194		total_len = sizeof(struct scsi_target_group_data);
7195	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7196		num_target_port_groups +
7197	    sizeof(struct scsi_target_port_descriptor) *
7198		num_target_ports * num_target_port_groups;
7199
7200	alloc_len = scsi_4btoul(cdb->length);
7201
7202	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7203
7204	ctsio->kern_sg_entries = 0;
7205
7206	if (total_len < alloc_len) {
7207		ctsio->residual = alloc_len - total_len;
7208		ctsio->kern_data_len = total_len;
7209		ctsio->kern_total_len = total_len;
7210	} else {
7211		ctsio->residual = 0;
7212		ctsio->kern_data_len = alloc_len;
7213		ctsio->kern_total_len = alloc_len;
7214	}
7215	ctsio->kern_data_resid = 0;
7216	ctsio->kern_rel_offset = 0;
7217
7218	if (ext) {
7219		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7220		    ctsio->kern_data_ptr;
7221		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7222		rtg_ext_ptr->format_type = 0x10;
7223		rtg_ext_ptr->implicit_transition_time = 0;
7224		tpg_desc = &rtg_ext_ptr->groups[0];
7225	} else {
7226		rtg_ptr = (struct scsi_target_group_data *)
7227		    ctsio->kern_data_ptr;
7228		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7229		tpg_desc = &rtg_ptr->groups[0];
7230	}
7231
7232	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7233	mtx_lock(&softc->ctl_lock);
7234	for (g = 0; g < num_target_port_groups; g++) {
7235		if (g == pg)
7236			tpg_desc->pref_state = TPG_PRIMARY |
7237			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7238		else
7239			tpg_desc->pref_state =
7240			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7241		tpg_desc->support = TPG_AO_SUP;
7242		if (!single)
7243			tpg_desc->support = TPG_AN_SUP;
7244		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7245		tpg_desc->status = TPG_IMPLICIT;
7246		pc = 0;
7247		STAILQ_FOREACH(port, &softc->port_list, links) {
7248			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7249				continue;
7250			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7251			    CTL_MAX_LUNS)
7252				continue;
7253			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7254			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7255			    relative_target_port_identifier);
7256			pc++;
7257		}
7258		tpg_desc->target_port_count = pc;
7259		tpg_desc = (struct scsi_target_port_group_descriptor *)
7260		    &tpg_desc->descriptors[pc];
7261	}
7262	mtx_unlock(&softc->ctl_lock);
7263
7264	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7265	ctsio->be_move_done = ctl_config_move_done;
7266
7267	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7268			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7269			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7270			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7271			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7272
7273	ctl_datamove((union ctl_io *)ctsio);
7274	return(retval);
7275}
7276
7277int
7278ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7279{
7280	struct ctl_lun *lun;
7281	struct scsi_report_supported_opcodes *cdb;
7282	const struct ctl_cmd_entry *entry, *sentry;
7283	struct scsi_report_supported_opcodes_all *all;
7284	struct scsi_report_supported_opcodes_descr *descr;
7285	struct scsi_report_supported_opcodes_one *one;
7286	int retval;
7287	int alloc_len, total_len;
7288	int opcode, service_action, i, j, num;
7289
7290	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7291
7292	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7293	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7294
7295	retval = CTL_RETVAL_COMPLETE;
7296
7297	opcode = cdb->requested_opcode;
7298	service_action = scsi_2btoul(cdb->requested_service_action);
7299	switch (cdb->options & RSO_OPTIONS_MASK) {
7300	case RSO_OPTIONS_ALL:
7301		num = 0;
7302		for (i = 0; i < 256; i++) {
7303			entry = &ctl_cmd_table[i];
7304			if (entry->flags & CTL_CMD_FLAG_SA5) {
7305				for (j = 0; j < 32; j++) {
7306					sentry = &((const struct ctl_cmd_entry *)
7307					    entry->execute)[j];
7308					if (ctl_cmd_applicable(
7309					    lun->be_lun->lun_type, sentry))
7310						num++;
7311				}
7312			} else {
7313				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7314				    entry))
7315					num++;
7316			}
7317		}
7318		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7319		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7320		break;
7321	case RSO_OPTIONS_OC:
7322		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7323			ctl_set_invalid_field(/*ctsio*/ ctsio,
7324					      /*sks_valid*/ 1,
7325					      /*command*/ 1,
7326					      /*field*/ 2,
7327					      /*bit_valid*/ 1,
7328					      /*bit*/ 2);
7329			ctl_done((union ctl_io *)ctsio);
7330			return (CTL_RETVAL_COMPLETE);
7331		}
7332		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7333		break;
7334	case RSO_OPTIONS_OC_SA:
7335		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7336		    service_action >= 32) {
7337			ctl_set_invalid_field(/*ctsio*/ ctsio,
7338					      /*sks_valid*/ 1,
7339					      /*command*/ 1,
7340					      /*field*/ 2,
7341					      /*bit_valid*/ 1,
7342					      /*bit*/ 2);
7343			ctl_done((union ctl_io *)ctsio);
7344			return (CTL_RETVAL_COMPLETE);
7345		}
7346		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7347		break;
7348	default:
7349		ctl_set_invalid_field(/*ctsio*/ ctsio,
7350				      /*sks_valid*/ 1,
7351				      /*command*/ 1,
7352				      /*field*/ 2,
7353				      /*bit_valid*/ 1,
7354				      /*bit*/ 2);
7355		ctl_done((union ctl_io *)ctsio);
7356		return (CTL_RETVAL_COMPLETE);
7357	}
7358
7359	alloc_len = scsi_4btoul(cdb->length);
7360
7361	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7362
7363	ctsio->kern_sg_entries = 0;
7364
7365	if (total_len < alloc_len) {
7366		ctsio->residual = alloc_len - total_len;
7367		ctsio->kern_data_len = total_len;
7368		ctsio->kern_total_len = total_len;
7369	} else {
7370		ctsio->residual = 0;
7371		ctsio->kern_data_len = alloc_len;
7372		ctsio->kern_total_len = alloc_len;
7373	}
7374	ctsio->kern_data_resid = 0;
7375	ctsio->kern_rel_offset = 0;
7376
7377	switch (cdb->options & RSO_OPTIONS_MASK) {
7378	case RSO_OPTIONS_ALL:
7379		all = (struct scsi_report_supported_opcodes_all *)
7380		    ctsio->kern_data_ptr;
7381		num = 0;
7382		for (i = 0; i < 256; i++) {
7383			entry = &ctl_cmd_table[i];
7384			if (entry->flags & CTL_CMD_FLAG_SA5) {
7385				for (j = 0; j < 32; j++) {
7386					sentry = &((const struct ctl_cmd_entry *)
7387					    entry->execute)[j];
7388					if (!ctl_cmd_applicable(
7389					    lun->be_lun->lun_type, sentry))
7390						continue;
7391					descr = &all->descr[num++];
7392					descr->opcode = i;
7393					scsi_ulto2b(j, descr->service_action);
7394					descr->flags = RSO_SERVACTV;
7395					scsi_ulto2b(sentry->length,
7396					    descr->cdb_length);
7397				}
7398			} else {
7399				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7400				    entry))
7401					continue;
7402				descr = &all->descr[num++];
7403				descr->opcode = i;
7404				scsi_ulto2b(0, descr->service_action);
7405				descr->flags = 0;
7406				scsi_ulto2b(entry->length, descr->cdb_length);
7407			}
7408		}
7409		scsi_ulto4b(
7410		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7411		    all->length);
7412		break;
7413	case RSO_OPTIONS_OC:
7414		one = (struct scsi_report_supported_opcodes_one *)
7415		    ctsio->kern_data_ptr;
7416		entry = &ctl_cmd_table[opcode];
7417		goto fill_one;
7418	case RSO_OPTIONS_OC_SA:
7419		one = (struct scsi_report_supported_opcodes_one *)
7420		    ctsio->kern_data_ptr;
7421		entry = &ctl_cmd_table[opcode];
7422		entry = &((const struct ctl_cmd_entry *)
7423		    entry->execute)[service_action];
7424fill_one:
7425		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7426			one->support = 3;
7427			scsi_ulto2b(entry->length, one->cdb_length);
7428			one->cdb_usage[0] = opcode;
7429			memcpy(&one->cdb_usage[1], entry->usage,
7430			    entry->length - 1);
7431		} else
7432			one->support = 1;
7433		break;
7434	}
7435
7436	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7437	ctsio->be_move_done = ctl_config_move_done;
7438
7439	ctl_datamove((union ctl_io *)ctsio);
7440	return(retval);
7441}
7442
7443int
7444ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7445{
7446	struct ctl_lun *lun;
7447	struct scsi_report_supported_tmf *cdb;
7448	struct scsi_report_supported_tmf_data *data;
7449	int retval;
7450	int alloc_len, total_len;
7451
7452	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7453
7454	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7455	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7456
7457	retval = CTL_RETVAL_COMPLETE;
7458
7459	total_len = sizeof(struct scsi_report_supported_tmf_data);
7460	alloc_len = scsi_4btoul(cdb->length);
7461
7462	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7463
7464	ctsio->kern_sg_entries = 0;
7465
7466	if (total_len < alloc_len) {
7467		ctsio->residual = alloc_len - total_len;
7468		ctsio->kern_data_len = total_len;
7469		ctsio->kern_total_len = total_len;
7470	} else {
7471		ctsio->residual = 0;
7472		ctsio->kern_data_len = alloc_len;
7473		ctsio->kern_total_len = alloc_len;
7474	}
7475	ctsio->kern_data_resid = 0;
7476	ctsio->kern_rel_offset = 0;
7477
7478	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7479	data->byte1 |= RST_ATS | RST_LURS | RST_TRS;
7480
7481	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7482	ctsio->be_move_done = ctl_config_move_done;
7483
7484	ctl_datamove((union ctl_io *)ctsio);
7485	return (retval);
7486}
7487
7488int
7489ctl_report_timestamp(struct ctl_scsiio *ctsio)
7490{
7491	struct ctl_lun *lun;
7492	struct scsi_report_timestamp *cdb;
7493	struct scsi_report_timestamp_data *data;
7494	struct timeval tv;
7495	int64_t timestamp;
7496	int retval;
7497	int alloc_len, total_len;
7498
7499	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7500
7501	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7502	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7503
7504	retval = CTL_RETVAL_COMPLETE;
7505
7506	total_len = sizeof(struct scsi_report_timestamp_data);
7507	alloc_len = scsi_4btoul(cdb->length);
7508
7509	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7510
7511	ctsio->kern_sg_entries = 0;
7512
7513	if (total_len < alloc_len) {
7514		ctsio->residual = alloc_len - total_len;
7515		ctsio->kern_data_len = total_len;
7516		ctsio->kern_total_len = total_len;
7517	} else {
7518		ctsio->residual = 0;
7519		ctsio->kern_data_len = alloc_len;
7520		ctsio->kern_total_len = alloc_len;
7521	}
7522	ctsio->kern_data_resid = 0;
7523	ctsio->kern_rel_offset = 0;
7524
7525	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7526	scsi_ulto2b(sizeof(*data) - 2, data->length);
7527	data->origin = RTS_ORIG_OUTSIDE;
7528	getmicrotime(&tv);
7529	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7530	scsi_ulto4b(timestamp >> 16, data->timestamp);
7531	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7532
7533	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7534	ctsio->be_move_done = ctl_config_move_done;
7535
7536	ctl_datamove((union ctl_io *)ctsio);
7537	return (retval);
7538}
7539
7540int
7541ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7542{
7543	struct scsi_per_res_in *cdb;
7544	int alloc_len, total_len = 0;
7545	/* struct scsi_per_res_in_rsrv in_data; */
7546	struct ctl_lun *lun;
7547	struct ctl_softc *softc;
7548
7549	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7550
7551	softc = control_softc;
7552
7553	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7554
7555	alloc_len = scsi_2btoul(cdb->length);
7556
7557	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7558
7559retry:
7560	mtx_lock(&lun->lun_lock);
7561	switch (cdb->action) {
7562	case SPRI_RK: /* read keys */
7563		total_len = sizeof(struct scsi_per_res_in_keys) +
7564			lun->pr_key_count *
7565			sizeof(struct scsi_per_res_key);
7566		break;
7567	case SPRI_RR: /* read reservation */
7568		if (lun->flags & CTL_LUN_PR_RESERVED)
7569			total_len = sizeof(struct scsi_per_res_in_rsrv);
7570		else
7571			total_len = sizeof(struct scsi_per_res_in_header);
7572		break;
7573	case SPRI_RC: /* report capabilities */
7574		total_len = sizeof(struct scsi_per_res_cap);
7575		break;
7576	default:
7577		panic("Invalid PR type %x", cdb->action);
7578	}
7579	mtx_unlock(&lun->lun_lock);
7580
7581	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7582
7583	if (total_len < alloc_len) {
7584		ctsio->residual = alloc_len - total_len;
7585		ctsio->kern_data_len = total_len;
7586		ctsio->kern_total_len = total_len;
7587	} else {
7588		ctsio->residual = 0;
7589		ctsio->kern_data_len = alloc_len;
7590		ctsio->kern_total_len = alloc_len;
7591	}
7592
7593	ctsio->kern_data_resid = 0;
7594	ctsio->kern_rel_offset = 0;
7595	ctsio->kern_sg_entries = 0;
7596
7597	mtx_lock(&lun->lun_lock);
7598	switch (cdb->action) {
7599	case SPRI_RK: { // read keys
7600        struct scsi_per_res_in_keys *res_keys;
7601		int i, key_count;
7602
7603		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7604
7605		/*
7606		 * We had to drop the lock to allocate our buffer, which
7607		 * leaves time for someone to come in with another
7608		 * persistent reservation.  (That is unlikely, though,
7609		 * since this should be the only persistent reservation
7610		 * command active right now.)
7611		 */
7612		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7613		    (lun->pr_key_count *
7614		     sizeof(struct scsi_per_res_key)))){
7615			mtx_unlock(&lun->lun_lock);
7616			free(ctsio->kern_data_ptr, M_CTL);
7617			printf("%s: reservation length changed, retrying\n",
7618			       __func__);
7619			goto retry;
7620		}
7621
7622		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7623
7624		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7625			     lun->pr_key_count, res_keys->header.length);
7626
7627		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7628			if (!lun->per_res[i].registered)
7629				continue;
7630
7631			/*
7632			 * We used lun->pr_key_count to calculate the
7633			 * size to allocate.  If it turns out the number of
7634			 * initiators with the registered flag set is
7635			 * larger than that (i.e. they haven't been kept in
7636			 * sync), we've got a problem.
7637			 */
7638			if (key_count >= lun->pr_key_count) {
7639#ifdef NEEDTOPORT
7640				csevent_log(CSC_CTL | CSC_SHELF_SW |
7641					    CTL_PR_ERROR,
7642					    csevent_LogType_Fault,
7643					    csevent_AlertLevel_Yellow,
7644					    csevent_FRU_ShelfController,
7645					    csevent_FRU_Firmware,
7646				        csevent_FRU_Unknown,
7647					    "registered keys %d >= key "
7648					    "count %d", key_count,
7649					    lun->pr_key_count);
7650#endif
7651				key_count++;
7652				continue;
7653			}
7654			memcpy(res_keys->keys[key_count].key,
7655			       lun->per_res[i].res_key.key,
7656			       ctl_min(sizeof(res_keys->keys[key_count].key),
7657			       sizeof(lun->per_res[i].res_key)));
7658			key_count++;
7659		}
7660		break;
7661	}
7662	case SPRI_RR: { // read reservation
7663		struct scsi_per_res_in_rsrv *res;
7664		int tmp_len, header_only;
7665
7666		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7667
7668		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7669
7670		if (lun->flags & CTL_LUN_PR_RESERVED)
7671		{
7672			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7673			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7674				    res->header.length);
7675			header_only = 0;
7676		} else {
7677			tmp_len = sizeof(struct scsi_per_res_in_header);
7678			scsi_ulto4b(0, res->header.length);
7679			header_only = 1;
7680		}
7681
7682		/*
7683		 * We had to drop the lock to allocate our buffer, which
7684		 * leaves time for someone to come in with another
7685		 * persistent reservation.  (That is unlikely, though,
7686		 * since this should be the only persistent reservation
7687		 * command active right now.)
7688		 */
7689		if (tmp_len != total_len) {
7690			mtx_unlock(&lun->lun_lock);
7691			free(ctsio->kern_data_ptr, M_CTL);
7692			printf("%s: reservation status changed, retrying\n",
7693			       __func__);
7694			goto retry;
7695		}
7696
7697		/*
7698		 * No reservation held, so we're done.
7699		 */
7700		if (header_only != 0)
7701			break;
7702
7703		/*
7704		 * If the registration is an All Registrants type, the key
7705		 * is 0, since it doesn't really matter.
7706		 */
7707		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7708			memcpy(res->data.reservation,
7709			       &lun->per_res[lun->pr_res_idx].res_key,
7710			       sizeof(struct scsi_per_res_key));
7711		}
7712		res->data.scopetype = lun->res_type;
7713		break;
7714	}
7715	case SPRI_RC:     //report capabilities
7716	{
7717		struct scsi_per_res_cap *res_cap;
7718		uint16_t type_mask;
7719
7720		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7721		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7722		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_3;
7723		type_mask = SPRI_TM_WR_EX_AR |
7724			    SPRI_TM_EX_AC_RO |
7725			    SPRI_TM_WR_EX_RO |
7726			    SPRI_TM_EX_AC |
7727			    SPRI_TM_WR_EX |
7728			    SPRI_TM_EX_AC_AR;
7729		scsi_ulto2b(type_mask, res_cap->type_mask);
7730		break;
7731	}
7732	case SPRI_RS: //read full status
7733	default:
7734		/*
7735		 * This is a bug, because we just checked for this above,
7736		 * and should have returned an error.
7737		 */
7738		panic("Invalid PR type %x", cdb->action);
7739		break; /* NOTREACHED */
7740	}
7741	mtx_unlock(&lun->lun_lock);
7742
7743	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7744	ctsio->be_move_done = ctl_config_move_done;
7745
7746	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7747			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7748			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7749			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7750			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7751
7752	ctl_datamove((union ctl_io *)ctsio);
7753
7754	return (CTL_RETVAL_COMPLETE);
7755}
7756
7757/*
7758 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7759 * it should return.
7760 */
7761static int
7762ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7763		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7764		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7765		struct scsi_per_res_out_parms* param)
7766{
7767	union ctl_ha_msg persis_io;
7768	int retval, i;
7769	int isc_retval;
7770
7771	retval = 0;
7772
7773	mtx_lock(&lun->lun_lock);
7774	if (sa_res_key == 0) {
7775		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7776			/* validate scope and type */
7777			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7778			     SPR_LU_SCOPE) {
7779				mtx_unlock(&lun->lun_lock);
7780				ctl_set_invalid_field(/*ctsio*/ ctsio,
7781						      /*sks_valid*/ 1,
7782						      /*command*/ 1,
7783						      /*field*/ 2,
7784						      /*bit_valid*/ 1,
7785						      /*bit*/ 4);
7786				ctl_done((union ctl_io *)ctsio);
7787				return (1);
7788			}
7789
7790		        if (type>8 || type==2 || type==4 || type==0) {
7791				mtx_unlock(&lun->lun_lock);
7792				ctl_set_invalid_field(/*ctsio*/ ctsio,
7793       	           				      /*sks_valid*/ 1,
7794						      /*command*/ 1,
7795						      /*field*/ 2,
7796						      /*bit_valid*/ 1,
7797						      /*bit*/ 0);
7798				ctl_done((union ctl_io *)ctsio);
7799				return (1);
7800		        }
7801
7802			/* temporarily unregister this nexus */
7803			lun->per_res[residx].registered = 0;
7804
7805			/*
7806			 * Unregister everybody else and build UA for
7807			 * them
7808			 */
7809			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7810				if (lun->per_res[i].registered == 0)
7811					continue;
7812
7813				if (!persis_offset
7814				 && i <CTL_MAX_INITIATORS)
7815					lun->pending_sense[i].ua_pending |=
7816						CTL_UA_REG_PREEMPT;
7817				else if (persis_offset
7818				      && i >= persis_offset)
7819					lun->pending_sense[i-persis_offset
7820						].ua_pending |=
7821						CTL_UA_REG_PREEMPT;
7822				lun->per_res[i].registered = 0;
7823				memset(&lun->per_res[i].res_key, 0,
7824				       sizeof(struct scsi_per_res_key));
7825			}
7826			lun->per_res[residx].registered = 1;
7827			lun->pr_key_count = 1;
7828			lun->res_type = type;
7829			if (lun->res_type != SPR_TYPE_WR_EX_AR
7830			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7831				lun->pr_res_idx = residx;
7832
7833			/* send msg to other side */
7834			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7835			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7836			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7837			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7838			persis_io.pr.pr_info.res_type = type;
7839			memcpy(persis_io.pr.pr_info.sa_res_key,
7840			       param->serv_act_res_key,
7841			       sizeof(param->serv_act_res_key));
7842			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7843			     &persis_io, sizeof(persis_io), 0)) >
7844			     CTL_HA_STATUS_SUCCESS) {
7845				printf("CTL:Persis Out error returned "
7846				       "from ctl_ha_msg_send %d\n",
7847				       isc_retval);
7848			}
7849		} else {
7850			/* not all registrants */
7851			mtx_unlock(&lun->lun_lock);
7852			free(ctsio->kern_data_ptr, M_CTL);
7853			ctl_set_invalid_field(ctsio,
7854					      /*sks_valid*/ 1,
7855					      /*command*/ 0,
7856					      /*field*/ 8,
7857					      /*bit_valid*/ 0,
7858					      /*bit*/ 0);
7859			ctl_done((union ctl_io *)ctsio);
7860			return (1);
7861		}
7862	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7863		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7864		int found = 0;
7865
7866		if (res_key == sa_res_key) {
7867			/* special case */
7868			/*
7869			 * The spec implies this is not good but doesn't
7870			 * say what to do. There are two choices either
7871			 * generate a res conflict or check condition
7872			 * with illegal field in parameter data. Since
7873			 * that is what is done when the sa_res_key is
7874			 * zero I'll take that approach since this has
7875			 * to do with the sa_res_key.
7876			 */
7877			mtx_unlock(&lun->lun_lock);
7878			free(ctsio->kern_data_ptr, M_CTL);
7879			ctl_set_invalid_field(ctsio,
7880					      /*sks_valid*/ 1,
7881					      /*command*/ 0,
7882					      /*field*/ 8,
7883					      /*bit_valid*/ 0,
7884					      /*bit*/ 0);
7885			ctl_done((union ctl_io *)ctsio);
7886			return (1);
7887		}
7888
7889		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7890			if (lun->per_res[i].registered
7891			 && memcmp(param->serv_act_res_key,
7892			    lun->per_res[i].res_key.key,
7893			    sizeof(struct scsi_per_res_key)) != 0)
7894				continue;
7895
7896			found = 1;
7897			lun->per_res[i].registered = 0;
7898			memset(&lun->per_res[i].res_key, 0,
7899			       sizeof(struct scsi_per_res_key));
7900			lun->pr_key_count--;
7901
7902			if (!persis_offset
7903			 && i < CTL_MAX_INITIATORS)
7904				lun->pending_sense[i].ua_pending |=
7905					CTL_UA_REG_PREEMPT;
7906			else if (persis_offset
7907			      && i >= persis_offset)
7908				lun->pending_sense[i-persis_offset].ua_pending|=
7909					CTL_UA_REG_PREEMPT;
7910		}
7911		if (!found) {
7912			mtx_unlock(&lun->lun_lock);
7913			free(ctsio->kern_data_ptr, M_CTL);
7914			ctl_set_reservation_conflict(ctsio);
7915			ctl_done((union ctl_io *)ctsio);
7916			return (CTL_RETVAL_COMPLETE);
7917		}
7918		/* send msg to other side */
7919		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7920		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7921		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7922		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7923		persis_io.pr.pr_info.res_type = type;
7924		memcpy(persis_io.pr.pr_info.sa_res_key,
7925		       param->serv_act_res_key,
7926		       sizeof(param->serv_act_res_key));
7927		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7928		     &persis_io, sizeof(persis_io), 0)) >
7929		     CTL_HA_STATUS_SUCCESS) {
7930			printf("CTL:Persis Out error returned from "
7931			       "ctl_ha_msg_send %d\n", isc_retval);
7932		}
7933	} else {
7934		/* Reserved but not all registrants */
7935		/* sa_res_key is res holder */
7936		if (memcmp(param->serv_act_res_key,
7937                   lun->per_res[lun->pr_res_idx].res_key.key,
7938                   sizeof(struct scsi_per_res_key)) == 0) {
7939			/* validate scope and type */
7940			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7941			     SPR_LU_SCOPE) {
7942				mtx_unlock(&lun->lun_lock);
7943				ctl_set_invalid_field(/*ctsio*/ ctsio,
7944						      /*sks_valid*/ 1,
7945						      /*command*/ 1,
7946						      /*field*/ 2,
7947						      /*bit_valid*/ 1,
7948						      /*bit*/ 4);
7949				ctl_done((union ctl_io *)ctsio);
7950				return (1);
7951			}
7952
7953			if (type>8 || type==2 || type==4 || type==0) {
7954				mtx_unlock(&lun->lun_lock);
7955				ctl_set_invalid_field(/*ctsio*/ ctsio,
7956						      /*sks_valid*/ 1,
7957						      /*command*/ 1,
7958						      /*field*/ 2,
7959						      /*bit_valid*/ 1,
7960						      /*bit*/ 0);
7961				ctl_done((union ctl_io *)ctsio);
7962				return (1);
7963			}
7964
7965			/*
7966			 * Do the following:
7967			 * if sa_res_key != res_key remove all
7968			 * registrants w/sa_res_key and generate UA
7969			 * for these registrants(Registrations
7970			 * Preempted) if it wasn't an exclusive
7971			 * reservation generate UA(Reservations
7972			 * Preempted) for all other registered nexuses
7973			 * if the type has changed. Establish the new
7974			 * reservation and holder. If res_key and
7975			 * sa_res_key are the same do the above
7976			 * except don't unregister the res holder.
7977			 */
7978
7979			/*
7980			 * Temporarily unregister so it won't get
7981			 * removed or UA generated
7982			 */
7983			lun->per_res[residx].registered = 0;
7984			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7985				if (lun->per_res[i].registered == 0)
7986					continue;
7987
7988				if (memcmp(param->serv_act_res_key,
7989				    lun->per_res[i].res_key.key,
7990				    sizeof(struct scsi_per_res_key)) == 0) {
7991					lun->per_res[i].registered = 0;
7992					memset(&lun->per_res[i].res_key,
7993					       0,
7994					       sizeof(struct scsi_per_res_key));
7995					lun->pr_key_count--;
7996
7997					if (!persis_offset
7998					 && i < CTL_MAX_INITIATORS)
7999						lun->pending_sense[i
8000							].ua_pending |=
8001							CTL_UA_REG_PREEMPT;
8002					else if (persis_offset
8003					      && i >= persis_offset)
8004						lun->pending_sense[
8005						  i-persis_offset].ua_pending |=
8006						  CTL_UA_REG_PREEMPT;
8007				} else if (type != lun->res_type
8008					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8009					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8010						if (!persis_offset
8011						 && i < CTL_MAX_INITIATORS)
8012							lun->pending_sense[i
8013							].ua_pending |=
8014							CTL_UA_RES_RELEASE;
8015						else if (persis_offset
8016						      && i >= persis_offset)
8017							lun->pending_sense[
8018							i-persis_offset
8019							].ua_pending |=
8020							CTL_UA_RES_RELEASE;
8021				}
8022			}
8023			lun->per_res[residx].registered = 1;
8024			lun->res_type = type;
8025			if (lun->res_type != SPR_TYPE_WR_EX_AR
8026			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8027				lun->pr_res_idx = residx;
8028			else
8029				lun->pr_res_idx =
8030					CTL_PR_ALL_REGISTRANTS;
8031
8032			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8033			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8034			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8035			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8036			persis_io.pr.pr_info.res_type = type;
8037			memcpy(persis_io.pr.pr_info.sa_res_key,
8038			       param->serv_act_res_key,
8039			       sizeof(param->serv_act_res_key));
8040			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8041			     &persis_io, sizeof(persis_io), 0)) >
8042			     CTL_HA_STATUS_SUCCESS) {
8043				printf("CTL:Persis Out error returned "
8044				       "from ctl_ha_msg_send %d\n",
8045				       isc_retval);
8046			}
8047		} else {
8048			/*
8049			 * sa_res_key is not the res holder just
8050			 * remove registrants
8051			 */
8052			int found=0;
8053
8054			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8055				if (memcmp(param->serv_act_res_key,
8056				    lun->per_res[i].res_key.key,
8057				    sizeof(struct scsi_per_res_key)) != 0)
8058					continue;
8059
8060				found = 1;
8061				lun->per_res[i].registered = 0;
8062				memset(&lun->per_res[i].res_key, 0,
8063				       sizeof(struct scsi_per_res_key));
8064				lun->pr_key_count--;
8065
8066				if (!persis_offset
8067				 && i < CTL_MAX_INITIATORS)
8068					lun->pending_sense[i].ua_pending |=
8069						CTL_UA_REG_PREEMPT;
8070				else if (persis_offset
8071				      && i >= persis_offset)
8072					lun->pending_sense[
8073						i-persis_offset].ua_pending |=
8074						CTL_UA_REG_PREEMPT;
8075			}
8076
8077			if (!found) {
8078				mtx_unlock(&lun->lun_lock);
8079				free(ctsio->kern_data_ptr, M_CTL);
8080				ctl_set_reservation_conflict(ctsio);
8081				ctl_done((union ctl_io *)ctsio);
8082		        	return (1);
8083			}
8084			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8085			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8086			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8087			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8088			persis_io.pr.pr_info.res_type = type;
8089			memcpy(persis_io.pr.pr_info.sa_res_key,
8090			       param->serv_act_res_key,
8091			       sizeof(param->serv_act_res_key));
8092			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8093			     &persis_io, sizeof(persis_io), 0)) >
8094			     CTL_HA_STATUS_SUCCESS) {
8095				printf("CTL:Persis Out error returned "
8096				       "from ctl_ha_msg_send %d\n",
8097				isc_retval);
8098			}
8099		}
8100	}
8101
8102	lun->PRGeneration++;
8103	mtx_unlock(&lun->lun_lock);
8104
8105	return (retval);
8106}
8107
8108static void
8109ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8110{
8111	int i;
8112
8113	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8114	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8115	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8116		   msg->pr.pr_info.sa_res_key,
8117		   sizeof(struct scsi_per_res_key)) != 0) {
8118		uint64_t sa_res_key;
8119		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8120
8121		if (sa_res_key == 0) {
8122			/* temporarily unregister this nexus */
8123			lun->per_res[msg->pr.pr_info.residx].registered = 0;
8124
8125			/*
8126			 * Unregister everybody else and build UA for
8127			 * them
8128			 */
8129			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8130				if (lun->per_res[i].registered == 0)
8131					continue;
8132
8133				if (!persis_offset
8134				 && i < CTL_MAX_INITIATORS)
8135					lun->pending_sense[i].ua_pending |=
8136						CTL_UA_REG_PREEMPT;
8137				else if (persis_offset && i >= persis_offset)
8138					lun->pending_sense[i -
8139						persis_offset].ua_pending |=
8140						CTL_UA_REG_PREEMPT;
8141				lun->per_res[i].registered = 0;
8142				memset(&lun->per_res[i].res_key, 0,
8143				       sizeof(struct scsi_per_res_key));
8144			}
8145
8146			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8147			lun->pr_key_count = 1;
8148			lun->res_type = msg->pr.pr_info.res_type;
8149			if (lun->res_type != SPR_TYPE_WR_EX_AR
8150			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8151				lun->pr_res_idx = msg->pr.pr_info.residx;
8152		} else {
8153		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8154				if (memcmp(msg->pr.pr_info.sa_res_key,
8155		                   lun->per_res[i].res_key.key,
8156		                   sizeof(struct scsi_per_res_key)) != 0)
8157					continue;
8158
8159				lun->per_res[i].registered = 0;
8160				memset(&lun->per_res[i].res_key, 0,
8161				       sizeof(struct scsi_per_res_key));
8162				lun->pr_key_count--;
8163
8164				if (!persis_offset
8165				 && i < persis_offset)
8166					lun->pending_sense[i].ua_pending |=
8167						CTL_UA_REG_PREEMPT;
8168				else if (persis_offset
8169				      && i >= persis_offset)
8170					lun->pending_sense[i -
8171						persis_offset].ua_pending |=
8172						CTL_UA_REG_PREEMPT;
8173			}
8174		}
8175	} else {
8176		/*
8177		 * Temporarily unregister so it won't get removed
8178		 * or UA generated
8179		 */
8180		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8181		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8182			if (lun->per_res[i].registered == 0)
8183				continue;
8184
8185			if (memcmp(msg->pr.pr_info.sa_res_key,
8186	                   lun->per_res[i].res_key.key,
8187	                   sizeof(struct scsi_per_res_key)) == 0) {
8188				lun->per_res[i].registered = 0;
8189				memset(&lun->per_res[i].res_key, 0,
8190				       sizeof(struct scsi_per_res_key));
8191				lun->pr_key_count--;
8192				if (!persis_offset
8193				 && i < CTL_MAX_INITIATORS)
8194					lun->pending_sense[i].ua_pending |=
8195						CTL_UA_REG_PREEMPT;
8196				else if (persis_offset
8197				      && i >= persis_offset)
8198					lun->pending_sense[i -
8199						persis_offset].ua_pending |=
8200						CTL_UA_REG_PREEMPT;
8201			} else if (msg->pr.pr_info.res_type != lun->res_type
8202				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8203				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8204					if (!persis_offset
8205					 && i < persis_offset)
8206						lun->pending_sense[i
8207							].ua_pending |=
8208							CTL_UA_RES_RELEASE;
8209					else if (persis_offset
8210					      && i >= persis_offset)
8211					lun->pending_sense[i -
8212						persis_offset].ua_pending |=
8213						CTL_UA_RES_RELEASE;
8214			}
8215		}
8216		lun->per_res[msg->pr.pr_info.residx].registered = 1;
8217		lun->res_type = msg->pr.pr_info.res_type;
8218		if (lun->res_type != SPR_TYPE_WR_EX_AR
8219		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8220			lun->pr_res_idx = msg->pr.pr_info.residx;
8221		else
8222			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8223	}
8224	lun->PRGeneration++;
8225
8226}
8227
8228
8229int
8230ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8231{
8232	int retval;
8233	int isc_retval;
8234	u_int32_t param_len;
8235	struct scsi_per_res_out *cdb;
8236	struct ctl_lun *lun;
8237	struct scsi_per_res_out_parms* param;
8238	struct ctl_softc *softc;
8239	uint32_t residx;
8240	uint64_t res_key, sa_res_key;
8241	uint8_t type;
8242	union ctl_ha_msg persis_io;
8243	int    i;
8244
8245	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8246
8247	retval = CTL_RETVAL_COMPLETE;
8248
8249	softc = control_softc;
8250
8251	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8252	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8253
8254	/*
8255	 * We only support whole-LUN scope.  The scope & type are ignored for
8256	 * register, register and ignore existing key and clear.
8257	 * We sometimes ignore scope and type on preempts too!!
8258	 * Verify reservation type here as well.
8259	 */
8260	type = cdb->scope_type & SPR_TYPE_MASK;
8261	if ((cdb->action == SPRO_RESERVE)
8262	 || (cdb->action == SPRO_RELEASE)) {
8263		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8264			ctl_set_invalid_field(/*ctsio*/ ctsio,
8265					      /*sks_valid*/ 1,
8266					      /*command*/ 1,
8267					      /*field*/ 2,
8268					      /*bit_valid*/ 1,
8269					      /*bit*/ 4);
8270			ctl_done((union ctl_io *)ctsio);
8271			return (CTL_RETVAL_COMPLETE);
8272		}
8273
8274		if (type>8 || type==2 || type==4 || type==0) {
8275			ctl_set_invalid_field(/*ctsio*/ ctsio,
8276					      /*sks_valid*/ 1,
8277					      /*command*/ 1,
8278					      /*field*/ 2,
8279					      /*bit_valid*/ 1,
8280					      /*bit*/ 0);
8281			ctl_done((union ctl_io *)ctsio);
8282			return (CTL_RETVAL_COMPLETE);
8283		}
8284	}
8285
8286	param_len = scsi_4btoul(cdb->length);
8287
8288	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8289		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8290		ctsio->kern_data_len = param_len;
8291		ctsio->kern_total_len = param_len;
8292		ctsio->kern_data_resid = 0;
8293		ctsio->kern_rel_offset = 0;
8294		ctsio->kern_sg_entries = 0;
8295		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8296		ctsio->be_move_done = ctl_config_move_done;
8297		ctl_datamove((union ctl_io *)ctsio);
8298
8299		return (CTL_RETVAL_COMPLETE);
8300	}
8301
8302	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8303
8304	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8305	res_key = scsi_8btou64(param->res_key.key);
8306	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8307
8308	/*
8309	 * Validate the reservation key here except for SPRO_REG_IGNO
8310	 * This must be done for all other service actions
8311	 */
8312	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8313		mtx_lock(&lun->lun_lock);
8314		if (lun->per_res[residx].registered) {
8315		    if (memcmp(param->res_key.key,
8316			       lun->per_res[residx].res_key.key,
8317			       ctl_min(sizeof(param->res_key),
8318			       sizeof(lun->per_res[residx].res_key))) != 0) {
8319				/*
8320				 * The current key passed in doesn't match
8321				 * the one the initiator previously
8322				 * registered.
8323				 */
8324				mtx_unlock(&lun->lun_lock);
8325				free(ctsio->kern_data_ptr, M_CTL);
8326				ctl_set_reservation_conflict(ctsio);
8327				ctl_done((union ctl_io *)ctsio);
8328				return (CTL_RETVAL_COMPLETE);
8329			}
8330		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8331			/*
8332			 * We are not registered
8333			 */
8334			mtx_unlock(&lun->lun_lock);
8335			free(ctsio->kern_data_ptr, M_CTL);
8336			ctl_set_reservation_conflict(ctsio);
8337			ctl_done((union ctl_io *)ctsio);
8338			return (CTL_RETVAL_COMPLETE);
8339		} else if (res_key != 0) {
8340			/*
8341			 * We are not registered and trying to register but
8342			 * the register key isn't zero.
8343			 */
8344			mtx_unlock(&lun->lun_lock);
8345			free(ctsio->kern_data_ptr, M_CTL);
8346			ctl_set_reservation_conflict(ctsio);
8347			ctl_done((union ctl_io *)ctsio);
8348			return (CTL_RETVAL_COMPLETE);
8349		}
8350		mtx_unlock(&lun->lun_lock);
8351	}
8352
8353	switch (cdb->action & SPRO_ACTION_MASK) {
8354	case SPRO_REGISTER:
8355	case SPRO_REG_IGNO: {
8356
8357#if 0
8358		printf("Registration received\n");
8359#endif
8360
8361		/*
8362		 * We don't support any of these options, as we report in
8363		 * the read capabilities request (see
8364		 * ctl_persistent_reserve_in(), above).
8365		 */
8366		if ((param->flags & SPR_SPEC_I_PT)
8367		 || (param->flags & SPR_ALL_TG_PT)
8368		 || (param->flags & SPR_APTPL)) {
8369			int bit_ptr;
8370
8371			if (param->flags & SPR_APTPL)
8372				bit_ptr = 0;
8373			else if (param->flags & SPR_ALL_TG_PT)
8374				bit_ptr = 2;
8375			else /* SPR_SPEC_I_PT */
8376				bit_ptr = 3;
8377
8378			free(ctsio->kern_data_ptr, M_CTL);
8379			ctl_set_invalid_field(ctsio,
8380					      /*sks_valid*/ 1,
8381					      /*command*/ 0,
8382					      /*field*/ 20,
8383					      /*bit_valid*/ 1,
8384					      /*bit*/ bit_ptr);
8385			ctl_done((union ctl_io *)ctsio);
8386			return (CTL_RETVAL_COMPLETE);
8387		}
8388
8389		mtx_lock(&lun->lun_lock);
8390
8391		/*
8392		 * The initiator wants to clear the
8393		 * key/unregister.
8394		 */
8395		if (sa_res_key == 0) {
8396			if ((res_key == 0
8397			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8398			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8399			  && !lun->per_res[residx].registered)) {
8400				mtx_unlock(&lun->lun_lock);
8401				goto done;
8402			}
8403
8404			lun->per_res[residx].registered = 0;
8405			memset(&lun->per_res[residx].res_key,
8406			       0, sizeof(lun->per_res[residx].res_key));
8407			lun->pr_key_count--;
8408
8409			if (residx == lun->pr_res_idx) {
8410				lun->flags &= ~CTL_LUN_PR_RESERVED;
8411				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8412
8413				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8414				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8415				 && lun->pr_key_count) {
8416					/*
8417					 * If the reservation is a registrants
8418					 * only type we need to generate a UA
8419					 * for other registered inits.  The
8420					 * sense code should be RESERVATIONS
8421					 * RELEASED
8422					 */
8423
8424					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8425						if (lun->per_res[
8426						    i+persis_offset].registered
8427						    == 0)
8428							continue;
8429						lun->pending_sense[i
8430							].ua_pending |=
8431							CTL_UA_RES_RELEASE;
8432					}
8433				}
8434				lun->res_type = 0;
8435			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8436				if (lun->pr_key_count==0) {
8437					lun->flags &= ~CTL_LUN_PR_RESERVED;
8438					lun->res_type = 0;
8439					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8440				}
8441			}
8442			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8443			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8444			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8445			persis_io.pr.pr_info.residx = residx;
8446			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8447			     &persis_io, sizeof(persis_io), 0 )) >
8448			     CTL_HA_STATUS_SUCCESS) {
8449				printf("CTL:Persis Out error returned from "
8450				       "ctl_ha_msg_send %d\n", isc_retval);
8451			}
8452		} else /* sa_res_key != 0 */ {
8453
8454			/*
8455			 * If we aren't registered currently then increment
8456			 * the key count and set the registered flag.
8457			 */
8458			if (!lun->per_res[residx].registered) {
8459				lun->pr_key_count++;
8460				lun->per_res[residx].registered = 1;
8461			}
8462
8463			memcpy(&lun->per_res[residx].res_key,
8464			       param->serv_act_res_key,
8465			       ctl_min(sizeof(param->serv_act_res_key),
8466			       sizeof(lun->per_res[residx].res_key)));
8467
8468			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8469			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8470			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8471			persis_io.pr.pr_info.residx = residx;
8472			memcpy(persis_io.pr.pr_info.sa_res_key,
8473			       param->serv_act_res_key,
8474			       sizeof(param->serv_act_res_key));
8475			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8476			     &persis_io, sizeof(persis_io), 0)) >
8477			     CTL_HA_STATUS_SUCCESS) {
8478				printf("CTL:Persis Out error returned from "
8479				       "ctl_ha_msg_send %d\n", isc_retval);
8480			}
8481		}
8482		lun->PRGeneration++;
8483		mtx_unlock(&lun->lun_lock);
8484
8485		break;
8486	}
8487	case SPRO_RESERVE:
8488#if 0
8489                printf("Reserve executed type %d\n", type);
8490#endif
8491		mtx_lock(&lun->lun_lock);
8492		if (lun->flags & CTL_LUN_PR_RESERVED) {
8493			/*
8494			 * if this isn't the reservation holder and it's
8495			 * not a "all registrants" type or if the type is
8496			 * different then we have a conflict
8497			 */
8498			if ((lun->pr_res_idx != residx
8499			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8500			 || lun->res_type != type) {
8501				mtx_unlock(&lun->lun_lock);
8502				free(ctsio->kern_data_ptr, M_CTL);
8503				ctl_set_reservation_conflict(ctsio);
8504				ctl_done((union ctl_io *)ctsio);
8505				return (CTL_RETVAL_COMPLETE);
8506			}
8507			mtx_unlock(&lun->lun_lock);
8508		} else /* create a reservation */ {
8509			/*
8510			 * If it's not an "all registrants" type record
8511			 * reservation holder
8512			 */
8513			if (type != SPR_TYPE_WR_EX_AR
8514			 && type != SPR_TYPE_EX_AC_AR)
8515				lun->pr_res_idx = residx; /* Res holder */
8516			else
8517				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8518
8519			lun->flags |= CTL_LUN_PR_RESERVED;
8520			lun->res_type = type;
8521
8522			mtx_unlock(&lun->lun_lock);
8523
8524			/* send msg to other side */
8525			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8526			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8527			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8528			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8529			persis_io.pr.pr_info.res_type = type;
8530			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8531			     &persis_io, sizeof(persis_io), 0)) >
8532			     CTL_HA_STATUS_SUCCESS) {
8533				printf("CTL:Persis Out error returned from "
8534				       "ctl_ha_msg_send %d\n", isc_retval);
8535			}
8536		}
8537		break;
8538
8539	case SPRO_RELEASE:
8540		mtx_lock(&lun->lun_lock);
8541		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8542			/* No reservation exists return good status */
8543			mtx_unlock(&lun->lun_lock);
8544			goto done;
8545		}
8546		/*
8547		 * Is this nexus a reservation holder?
8548		 */
8549		if (lun->pr_res_idx != residx
8550		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8551			/*
8552			 * not a res holder return good status but
8553			 * do nothing
8554			 */
8555			mtx_unlock(&lun->lun_lock);
8556			goto done;
8557		}
8558
8559		if (lun->res_type != type) {
8560			mtx_unlock(&lun->lun_lock);
8561			free(ctsio->kern_data_ptr, M_CTL);
8562			ctl_set_illegal_pr_release(ctsio);
8563			ctl_done((union ctl_io *)ctsio);
8564			return (CTL_RETVAL_COMPLETE);
8565		}
8566
8567		/* okay to release */
8568		lun->flags &= ~CTL_LUN_PR_RESERVED;
8569		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8570		lun->res_type = 0;
8571
8572		/*
8573		 * if this isn't an exclusive access
8574		 * res generate UA for all other
8575		 * registrants.
8576		 */
8577		if (type != SPR_TYPE_EX_AC
8578		 && type != SPR_TYPE_WR_EX) {
8579			/*
8580			 * temporarily unregister so we don't generate UA
8581			 */
8582			lun->per_res[residx].registered = 0;
8583
8584			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8585				if (lun->per_res[i+persis_offset].registered
8586				    == 0)
8587					continue;
8588				lun->pending_sense[i].ua_pending |=
8589					CTL_UA_RES_RELEASE;
8590			}
8591
8592			lun->per_res[residx].registered = 1;
8593		}
8594		mtx_unlock(&lun->lun_lock);
8595		/* Send msg to other side */
8596		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8597		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8598		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8599		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8600		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8601			printf("CTL:Persis Out error returned from "
8602			       "ctl_ha_msg_send %d\n", isc_retval);
8603		}
8604		break;
8605
8606	case SPRO_CLEAR:
8607		/* send msg to other side */
8608
8609		mtx_lock(&lun->lun_lock);
8610		lun->flags &= ~CTL_LUN_PR_RESERVED;
8611		lun->res_type = 0;
8612		lun->pr_key_count = 0;
8613		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8614
8615
8616		memset(&lun->per_res[residx].res_key,
8617		       0, sizeof(lun->per_res[residx].res_key));
8618		lun->per_res[residx].registered = 0;
8619
8620		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8621			if (lun->per_res[i].registered) {
8622				if (!persis_offset && i < CTL_MAX_INITIATORS)
8623					lun->pending_sense[i].ua_pending |=
8624						CTL_UA_RES_PREEMPT;
8625				else if (persis_offset && i >= persis_offset)
8626					lun->pending_sense[i-persis_offset
8627					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8628
8629				memset(&lun->per_res[i].res_key,
8630				       0, sizeof(struct scsi_per_res_key));
8631				lun->per_res[i].registered = 0;
8632			}
8633		lun->PRGeneration++;
8634		mtx_unlock(&lun->lun_lock);
8635		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8636		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8637		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8638		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8639		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8640			printf("CTL:Persis Out error returned from "
8641			       "ctl_ha_msg_send %d\n", isc_retval);
8642		}
8643		break;
8644
8645	case SPRO_PREEMPT: {
8646		int nretval;
8647
8648		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8649					  residx, ctsio, cdb, param);
8650		if (nretval != 0)
8651			return (CTL_RETVAL_COMPLETE);
8652		break;
8653	}
8654	default:
8655		panic("Invalid PR type %x", cdb->action);
8656	}
8657
8658done:
8659	free(ctsio->kern_data_ptr, M_CTL);
8660	ctl_set_success(ctsio);
8661	ctl_done((union ctl_io *)ctsio);
8662
8663	return (retval);
8664}
8665
8666/*
8667 * This routine is for handling a message from the other SC pertaining to
8668 * persistent reserve out. All the error checking will have been done
8669 * so only perorming the action need be done here to keep the two
8670 * in sync.
8671 */
8672static void
8673ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8674{
8675	struct ctl_lun *lun;
8676	struct ctl_softc *softc;
8677	int i;
8678	uint32_t targ_lun;
8679
8680	softc = control_softc;
8681
8682	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8683	lun = softc->ctl_luns[targ_lun];
8684	mtx_lock(&lun->lun_lock);
8685	switch(msg->pr.pr_info.action) {
8686	case CTL_PR_REG_KEY:
8687		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8688			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8689			lun->pr_key_count++;
8690		}
8691		lun->PRGeneration++;
8692		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8693		       msg->pr.pr_info.sa_res_key,
8694		       sizeof(struct scsi_per_res_key));
8695		break;
8696
8697	case CTL_PR_UNREG_KEY:
8698		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8699		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8700		       0, sizeof(struct scsi_per_res_key));
8701		lun->pr_key_count--;
8702
8703		/* XXX Need to see if the reservation has been released */
8704		/* if so do we need to generate UA? */
8705		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8706			lun->flags &= ~CTL_LUN_PR_RESERVED;
8707			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8708
8709			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8710			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8711			 && lun->pr_key_count) {
8712				/*
8713				 * If the reservation is a registrants
8714				 * only type we need to generate a UA
8715				 * for other registered inits.  The
8716				 * sense code should be RESERVATIONS
8717				 * RELEASED
8718				 */
8719
8720				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8721					if (lun->per_res[i+
8722					    persis_offset].registered == 0)
8723						continue;
8724
8725					lun->pending_sense[i
8726						].ua_pending |=
8727						CTL_UA_RES_RELEASE;
8728				}
8729			}
8730			lun->res_type = 0;
8731		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8732			if (lun->pr_key_count==0) {
8733				lun->flags &= ~CTL_LUN_PR_RESERVED;
8734				lun->res_type = 0;
8735				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8736			}
8737		}
8738		lun->PRGeneration++;
8739		break;
8740
8741	case CTL_PR_RESERVE:
8742		lun->flags |= CTL_LUN_PR_RESERVED;
8743		lun->res_type = msg->pr.pr_info.res_type;
8744		lun->pr_res_idx = msg->pr.pr_info.residx;
8745
8746		break;
8747
8748	case CTL_PR_RELEASE:
8749		/*
8750		 * if this isn't an exclusive access res generate UA for all
8751		 * other registrants.
8752		 */
8753		if (lun->res_type != SPR_TYPE_EX_AC
8754		 && lun->res_type != SPR_TYPE_WR_EX) {
8755			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8756				if (lun->per_res[i+persis_offset].registered)
8757					lun->pending_sense[i].ua_pending |=
8758						CTL_UA_RES_RELEASE;
8759		}
8760
8761		lun->flags &= ~CTL_LUN_PR_RESERVED;
8762		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8763		lun->res_type = 0;
8764		break;
8765
8766	case CTL_PR_PREEMPT:
8767		ctl_pro_preempt_other(lun, msg);
8768		break;
8769	case CTL_PR_CLEAR:
8770		lun->flags &= ~CTL_LUN_PR_RESERVED;
8771		lun->res_type = 0;
8772		lun->pr_key_count = 0;
8773		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8774
8775		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8776			if (lun->per_res[i].registered == 0)
8777				continue;
8778			if (!persis_offset
8779			 && i < CTL_MAX_INITIATORS)
8780				lun->pending_sense[i].ua_pending |=
8781					CTL_UA_RES_PREEMPT;
8782			else if (persis_offset
8783			      && i >= persis_offset)
8784   				lun->pending_sense[i-persis_offset].ua_pending|=
8785					CTL_UA_RES_PREEMPT;
8786			memset(&lun->per_res[i].res_key, 0,
8787			       sizeof(struct scsi_per_res_key));
8788			lun->per_res[i].registered = 0;
8789		}
8790		lun->PRGeneration++;
8791		break;
8792	}
8793
8794	mtx_unlock(&lun->lun_lock);
8795}
8796
8797int
8798ctl_read_write(struct ctl_scsiio *ctsio)
8799{
8800	struct ctl_lun *lun;
8801	struct ctl_lba_len_flags *lbalen;
8802	uint64_t lba;
8803	uint32_t num_blocks;
8804	int fua, dpo;
8805	int retval;
8806	int isread;
8807
8808	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8809
8810	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8811
8812	fua = 0;
8813	dpo = 0;
8814
8815	retval = CTL_RETVAL_COMPLETE;
8816
8817	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8818	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8819	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8820		uint32_t residx;
8821
8822		/*
8823		 * XXX KDM need a lock here.
8824		 */
8825		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8826		if ((lun->res_type == SPR_TYPE_EX_AC
8827		  && residx != lun->pr_res_idx)
8828		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8829		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8830		  && !lun->per_res[residx].registered)) {
8831			ctl_set_reservation_conflict(ctsio);
8832			ctl_done((union ctl_io *)ctsio);
8833			return (CTL_RETVAL_COMPLETE);
8834	        }
8835	}
8836
8837	switch (ctsio->cdb[0]) {
8838	case READ_6:
8839	case WRITE_6: {
8840		struct scsi_rw_6 *cdb;
8841
8842		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8843
8844		lba = scsi_3btoul(cdb->addr);
8845		/* only 5 bits are valid in the most significant address byte */
8846		lba &= 0x1fffff;
8847		num_blocks = cdb->length;
8848		/*
8849		 * This is correct according to SBC-2.
8850		 */
8851		if (num_blocks == 0)
8852			num_blocks = 256;
8853		break;
8854	}
8855	case READ_10:
8856	case WRITE_10: {
8857		struct scsi_rw_10 *cdb;
8858
8859		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8860
8861		if (cdb->byte2 & SRW10_FUA)
8862			fua = 1;
8863		if (cdb->byte2 & SRW10_DPO)
8864			dpo = 1;
8865
8866		lba = scsi_4btoul(cdb->addr);
8867		num_blocks = scsi_2btoul(cdb->length);
8868		break;
8869	}
8870	case WRITE_VERIFY_10: {
8871		struct scsi_write_verify_10 *cdb;
8872
8873		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8874
8875		/*
8876		 * XXX KDM we should do actual write verify support at some
8877		 * point.  This is obviously fake, we're just translating
8878		 * things to a write.  So we don't even bother checking the
8879		 * BYTCHK field, since we don't do any verification.  If
8880		 * the user asks for it, we'll just pretend we did it.
8881		 */
8882		if (cdb->byte2 & SWV_DPO)
8883			dpo = 1;
8884
8885		lba = scsi_4btoul(cdb->addr);
8886		num_blocks = scsi_2btoul(cdb->length);
8887		break;
8888	}
8889	case READ_12:
8890	case WRITE_12: {
8891		struct scsi_rw_12 *cdb;
8892
8893		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8894
8895		if (cdb->byte2 & SRW12_FUA)
8896			fua = 1;
8897		if (cdb->byte2 & SRW12_DPO)
8898			dpo = 1;
8899		lba = scsi_4btoul(cdb->addr);
8900		num_blocks = scsi_4btoul(cdb->length);
8901		break;
8902	}
8903	case WRITE_VERIFY_12: {
8904		struct scsi_write_verify_12 *cdb;
8905
8906		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8907
8908		if (cdb->byte2 & SWV_DPO)
8909			dpo = 1;
8910
8911		lba = scsi_4btoul(cdb->addr);
8912		num_blocks = scsi_4btoul(cdb->length);
8913
8914		break;
8915	}
8916	case READ_16:
8917	case WRITE_16: {
8918		struct scsi_rw_16 *cdb;
8919
8920		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8921
8922		if (cdb->byte2 & SRW12_FUA)
8923			fua = 1;
8924		if (cdb->byte2 & SRW12_DPO)
8925			dpo = 1;
8926
8927		lba = scsi_8btou64(cdb->addr);
8928		num_blocks = scsi_4btoul(cdb->length);
8929		break;
8930	}
8931	case WRITE_VERIFY_16: {
8932		struct scsi_write_verify_16 *cdb;
8933
8934		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8935
8936		if (cdb->byte2 & SWV_DPO)
8937			dpo = 1;
8938
8939		lba = scsi_8btou64(cdb->addr);
8940		num_blocks = scsi_4btoul(cdb->length);
8941		break;
8942	}
8943	default:
8944		/*
8945		 * We got a command we don't support.  This shouldn't
8946		 * happen, commands should be filtered out above us.
8947		 */
8948		ctl_set_invalid_opcode(ctsio);
8949		ctl_done((union ctl_io *)ctsio);
8950
8951		return (CTL_RETVAL_COMPLETE);
8952		break; /* NOTREACHED */
8953	}
8954
8955	/*
8956	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8957	 * interesting for us, but if RAIDCore is in write-back mode,
8958	 * getting it to do write-through for a particular transaction may
8959	 * not be possible.
8960	 */
8961
8962	/*
8963	 * The first check is to make sure we're in bounds, the second
8964	 * check is to catch wrap-around problems.  If the lba + num blocks
8965	 * is less than the lba, then we've wrapped around and the block
8966	 * range is invalid anyway.
8967	 */
8968	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8969	 || ((lba + num_blocks) < lba)) {
8970		ctl_set_lba_out_of_range(ctsio);
8971		ctl_done((union ctl_io *)ctsio);
8972		return (CTL_RETVAL_COMPLETE);
8973	}
8974
8975	/*
8976	 * According to SBC-3, a transfer length of 0 is not an error.
8977	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8978	 * translates to 256 blocks for those commands.
8979	 */
8980	if (num_blocks == 0) {
8981		ctl_set_success(ctsio);
8982		ctl_done((union ctl_io *)ctsio);
8983		return (CTL_RETVAL_COMPLETE);
8984	}
8985
8986	lbalen = (struct ctl_lba_len_flags *)
8987	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8988	lbalen->lba = lba;
8989	lbalen->len = num_blocks;
8990	lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
8991
8992	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8993	ctsio->kern_rel_offset = 0;
8994
8995	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8996
8997	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8998
8999	return (retval);
9000}
9001
9002static int
9003ctl_cnw_cont(union ctl_io *io)
9004{
9005	struct ctl_scsiio *ctsio;
9006	struct ctl_lun *lun;
9007	struct ctl_lba_len_flags *lbalen;
9008	int retval;
9009
9010	ctsio = &io->scsiio;
9011	ctsio->io_hdr.status = CTL_STATUS_NONE;
9012	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9013	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9014	lbalen = (struct ctl_lba_len_flags *)
9015	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9016	lbalen->flags = CTL_LLF_WRITE;
9017
9018	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9019	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9020	return (retval);
9021}
9022
9023int
9024ctl_cnw(struct ctl_scsiio *ctsio)
9025{
9026	struct ctl_lun *lun;
9027	struct ctl_lba_len_flags *lbalen;
9028	uint64_t lba;
9029	uint32_t num_blocks;
9030	int fua, dpo;
9031	int retval;
9032
9033	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9034
9035	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9036
9037	fua = 0;
9038	dpo = 0;
9039
9040	retval = CTL_RETVAL_COMPLETE;
9041
9042	switch (ctsio->cdb[0]) {
9043	case COMPARE_AND_WRITE: {
9044		struct scsi_compare_and_write *cdb;
9045
9046		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9047
9048		if (cdb->byte2 & SRW10_FUA)
9049			fua = 1;
9050		if (cdb->byte2 & SRW10_DPO)
9051			dpo = 1;
9052		lba = scsi_8btou64(cdb->addr);
9053		num_blocks = cdb->length;
9054		break;
9055	}
9056	default:
9057		/*
9058		 * We got a command we don't support.  This shouldn't
9059		 * happen, commands should be filtered out above us.
9060		 */
9061		ctl_set_invalid_opcode(ctsio);
9062		ctl_done((union ctl_io *)ctsio);
9063
9064		return (CTL_RETVAL_COMPLETE);
9065		break; /* NOTREACHED */
9066	}
9067
9068	/*
9069	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
9070	 * interesting for us, but if RAIDCore is in write-back mode,
9071	 * getting it to do write-through for a particular transaction may
9072	 * not be possible.
9073	 */
9074
9075	/*
9076	 * The first check is to make sure we're in bounds, the second
9077	 * check is to catch wrap-around problems.  If the lba + num blocks
9078	 * is less than the lba, then we've wrapped around and the block
9079	 * range is invalid anyway.
9080	 */
9081	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9082	 || ((lba + num_blocks) < lba)) {
9083		ctl_set_lba_out_of_range(ctsio);
9084		ctl_done((union ctl_io *)ctsio);
9085		return (CTL_RETVAL_COMPLETE);
9086	}
9087
9088	/*
9089	 * According to SBC-3, a transfer length of 0 is not an error.
9090	 */
9091	if (num_blocks == 0) {
9092		ctl_set_success(ctsio);
9093		ctl_done((union ctl_io *)ctsio);
9094		return (CTL_RETVAL_COMPLETE);
9095	}
9096
9097	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9098	ctsio->kern_rel_offset = 0;
9099
9100	/*
9101	 * Set the IO_CONT flag, so that if this I/O gets passed to
9102	 * ctl_data_submit_done(), it'll get passed back to
9103	 * ctl_ctl_cnw_cont() for further processing.
9104	 */
9105	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9106	ctsio->io_cont = ctl_cnw_cont;
9107
9108	lbalen = (struct ctl_lba_len_flags *)
9109	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9110	lbalen->lba = lba;
9111	lbalen->len = num_blocks;
9112	lbalen->flags = CTL_LLF_COMPARE;
9113
9114	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9115	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9116	return (retval);
9117}
9118
9119int
9120ctl_verify(struct ctl_scsiio *ctsio)
9121{
9122	struct ctl_lun *lun;
9123	struct ctl_lba_len_flags *lbalen;
9124	uint64_t lba;
9125	uint32_t num_blocks;
9126	int bytchk, dpo;
9127	int retval;
9128
9129	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9130
9131	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9132
9133	bytchk = 0;
9134	dpo = 0;
9135	retval = CTL_RETVAL_COMPLETE;
9136
9137	switch (ctsio->cdb[0]) {
9138	case VERIFY_10: {
9139		struct scsi_verify_10 *cdb;
9140
9141		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9142		if (cdb->byte2 & SVFY_BYTCHK)
9143			bytchk = 1;
9144		if (cdb->byte2 & SVFY_DPO)
9145			dpo = 1;
9146		lba = scsi_4btoul(cdb->addr);
9147		num_blocks = scsi_2btoul(cdb->length);
9148		break;
9149	}
9150	case VERIFY_12: {
9151		struct scsi_verify_12 *cdb;
9152
9153		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9154		if (cdb->byte2 & SVFY_BYTCHK)
9155			bytchk = 1;
9156		if (cdb->byte2 & SVFY_DPO)
9157			dpo = 1;
9158		lba = scsi_4btoul(cdb->addr);
9159		num_blocks = scsi_4btoul(cdb->length);
9160		break;
9161	}
9162	case VERIFY_16: {
9163		struct scsi_rw_16 *cdb;
9164
9165		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9166		if (cdb->byte2 & SVFY_BYTCHK)
9167			bytchk = 1;
9168		if (cdb->byte2 & SVFY_DPO)
9169			dpo = 1;
9170		lba = scsi_8btou64(cdb->addr);
9171		num_blocks = scsi_4btoul(cdb->length);
9172		break;
9173	}
9174	default:
9175		/*
9176		 * We got a command we don't support.  This shouldn't
9177		 * happen, commands should be filtered out above us.
9178		 */
9179		ctl_set_invalid_opcode(ctsio);
9180		ctl_done((union ctl_io *)ctsio);
9181		return (CTL_RETVAL_COMPLETE);
9182	}
9183
9184	/*
9185	 * The first check is to make sure we're in bounds, the second
9186	 * check is to catch wrap-around problems.  If the lba + num blocks
9187	 * is less than the lba, then we've wrapped around and the block
9188	 * range is invalid anyway.
9189	 */
9190	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9191	 || ((lba + num_blocks) < lba)) {
9192		ctl_set_lba_out_of_range(ctsio);
9193		ctl_done((union ctl_io *)ctsio);
9194		return (CTL_RETVAL_COMPLETE);
9195	}
9196
9197	/*
9198	 * According to SBC-3, a transfer length of 0 is not an error.
9199	 */
9200	if (num_blocks == 0) {
9201		ctl_set_success(ctsio);
9202		ctl_done((union ctl_io *)ctsio);
9203		return (CTL_RETVAL_COMPLETE);
9204	}
9205
9206	lbalen = (struct ctl_lba_len_flags *)
9207	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9208	lbalen->lba = lba;
9209	lbalen->len = num_blocks;
9210	if (bytchk) {
9211		lbalen->flags = CTL_LLF_COMPARE;
9212		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9213	} else {
9214		lbalen->flags = CTL_LLF_VERIFY;
9215		ctsio->kern_total_len = 0;
9216	}
9217	ctsio->kern_rel_offset = 0;
9218
9219	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9220	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9221	return (retval);
9222}
9223
9224int
9225ctl_report_luns(struct ctl_scsiio *ctsio)
9226{
9227	struct scsi_report_luns *cdb;
9228	struct scsi_report_luns_data *lun_data;
9229	struct ctl_lun *lun, *request_lun;
9230	int num_luns, retval;
9231	uint32_t alloc_len, lun_datalen;
9232	int num_filled, well_known;
9233	uint32_t initidx, targ_lun_id, lun_id;
9234
9235	retval = CTL_RETVAL_COMPLETE;
9236	well_known = 0;
9237
9238	cdb = (struct scsi_report_luns *)ctsio->cdb;
9239
9240	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9241
9242	mtx_lock(&control_softc->ctl_lock);
9243	num_luns = control_softc->num_luns;
9244	mtx_unlock(&control_softc->ctl_lock);
9245
9246	switch (cdb->select_report) {
9247	case RPL_REPORT_DEFAULT:
9248	case RPL_REPORT_ALL:
9249		break;
9250	case RPL_REPORT_WELLKNOWN:
9251		well_known = 1;
9252		num_luns = 0;
9253		break;
9254	default:
9255		ctl_set_invalid_field(ctsio,
9256				      /*sks_valid*/ 1,
9257				      /*command*/ 1,
9258				      /*field*/ 2,
9259				      /*bit_valid*/ 0,
9260				      /*bit*/ 0);
9261		ctl_done((union ctl_io *)ctsio);
9262		return (retval);
9263		break; /* NOTREACHED */
9264	}
9265
9266	alloc_len = scsi_4btoul(cdb->length);
9267	/*
9268	 * The initiator has to allocate at least 16 bytes for this request,
9269	 * so he can at least get the header and the first LUN.  Otherwise
9270	 * we reject the request (per SPC-3 rev 14, section 6.21).
9271	 */
9272	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9273	    sizeof(struct scsi_report_luns_lundata))) {
9274		ctl_set_invalid_field(ctsio,
9275				      /*sks_valid*/ 1,
9276				      /*command*/ 1,
9277				      /*field*/ 6,
9278				      /*bit_valid*/ 0,
9279				      /*bit*/ 0);
9280		ctl_done((union ctl_io *)ctsio);
9281		return (retval);
9282	}
9283
9284	request_lun = (struct ctl_lun *)
9285		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9286
9287	lun_datalen = sizeof(*lun_data) +
9288		(num_luns * sizeof(struct scsi_report_luns_lundata));
9289
9290	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9291	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9292	ctsio->kern_sg_entries = 0;
9293
9294	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9295
9296	mtx_lock(&control_softc->ctl_lock);
9297	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9298		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9299		if (lun_id >= CTL_MAX_LUNS)
9300			continue;
9301		lun = control_softc->ctl_luns[lun_id];
9302		if (lun == NULL)
9303			continue;
9304
9305		if (targ_lun_id <= 0xff) {
9306			/*
9307			 * Peripheral addressing method, bus number 0.
9308			 */
9309			lun_data->luns[num_filled].lundata[0] =
9310				RPL_LUNDATA_ATYP_PERIPH;
9311			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9312			num_filled++;
9313		} else if (targ_lun_id <= 0x3fff) {
9314			/*
9315			 * Flat addressing method.
9316			 */
9317			lun_data->luns[num_filled].lundata[0] =
9318				RPL_LUNDATA_ATYP_FLAT |
9319				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9320#ifdef OLDCTLHEADERS
9321				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9322				(targ_lun_id & SRLD_BUS_LUN_MASK);
9323#endif
9324			lun_data->luns[num_filled].lundata[1] =
9325#ifdef OLDCTLHEADERS
9326				targ_lun_id >> SRLD_BUS_LUN_BITS;
9327#endif
9328				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9329			num_filled++;
9330		} else {
9331			printf("ctl_report_luns: bogus LUN number %jd, "
9332			       "skipping\n", (intmax_t)targ_lun_id);
9333		}
9334		/*
9335		 * According to SPC-3, rev 14 section 6.21:
9336		 *
9337		 * "The execution of a REPORT LUNS command to any valid and
9338		 * installed logical unit shall clear the REPORTED LUNS DATA
9339		 * HAS CHANGED unit attention condition for all logical
9340		 * units of that target with respect to the requesting
9341		 * initiator. A valid and installed logical unit is one
9342		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9343		 * INQUIRY data (see 6.4.2)."
9344		 *
9345		 * If request_lun is NULL, the LUN this report luns command
9346		 * was issued to is either disabled or doesn't exist. In that
9347		 * case, we shouldn't clear any pending lun change unit
9348		 * attention.
9349		 */
9350		if (request_lun != NULL) {
9351			mtx_lock(&lun->lun_lock);
9352			lun->pending_sense[initidx].ua_pending &=
9353				~CTL_UA_LUN_CHANGE;
9354			mtx_unlock(&lun->lun_lock);
9355		}
9356	}
9357	mtx_unlock(&control_softc->ctl_lock);
9358
9359	/*
9360	 * It's quite possible that we've returned fewer LUNs than we allocated
9361	 * space for.  Trim it.
9362	 */
9363	lun_datalen = sizeof(*lun_data) +
9364		(num_filled * sizeof(struct scsi_report_luns_lundata));
9365
9366	if (lun_datalen < alloc_len) {
9367		ctsio->residual = alloc_len - lun_datalen;
9368		ctsio->kern_data_len = lun_datalen;
9369		ctsio->kern_total_len = lun_datalen;
9370	} else {
9371		ctsio->residual = 0;
9372		ctsio->kern_data_len = alloc_len;
9373		ctsio->kern_total_len = alloc_len;
9374	}
9375	ctsio->kern_data_resid = 0;
9376	ctsio->kern_rel_offset = 0;
9377	ctsio->kern_sg_entries = 0;
9378
9379	/*
9380	 * We set this to the actual data length, regardless of how much
9381	 * space we actually have to return results.  If the user looks at
9382	 * this value, he'll know whether or not he allocated enough space
9383	 * and reissue the command if necessary.  We don't support well
9384	 * known logical units, so if the user asks for that, return none.
9385	 */
9386	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9387
9388	/*
9389	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9390	 * this request.
9391	 */
9392	ctsio->scsi_status = SCSI_STATUS_OK;
9393
9394	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9395	ctsio->be_move_done = ctl_config_move_done;
9396	ctl_datamove((union ctl_io *)ctsio);
9397
9398	return (retval);
9399}
9400
9401int
9402ctl_request_sense(struct ctl_scsiio *ctsio)
9403{
9404	struct scsi_request_sense *cdb;
9405	struct scsi_sense_data *sense_ptr;
9406	struct ctl_lun *lun;
9407	uint32_t initidx;
9408	int have_error;
9409	scsi_sense_data_type sense_format;
9410
9411	cdb = (struct scsi_request_sense *)ctsio->cdb;
9412
9413	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9414
9415	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9416
9417	/*
9418	 * Determine which sense format the user wants.
9419	 */
9420	if (cdb->byte2 & SRS_DESC)
9421		sense_format = SSD_TYPE_DESC;
9422	else
9423		sense_format = SSD_TYPE_FIXED;
9424
9425	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9426	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9427	ctsio->kern_sg_entries = 0;
9428
9429	/*
9430	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9431	 * larger than the largest allowed value for the length field in the
9432	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9433	 */
9434	ctsio->residual = 0;
9435	ctsio->kern_data_len = cdb->length;
9436	ctsio->kern_total_len = cdb->length;
9437
9438	ctsio->kern_data_resid = 0;
9439	ctsio->kern_rel_offset = 0;
9440	ctsio->kern_sg_entries = 0;
9441
9442	/*
9443	 * If we don't have a LUN, we don't have any pending sense.
9444	 */
9445	if (lun == NULL)
9446		goto no_sense;
9447
9448	have_error = 0;
9449	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9450	/*
9451	 * Check for pending sense, and then for pending unit attentions.
9452	 * Pending sense gets returned first, then pending unit attentions.
9453	 */
9454	mtx_lock(&lun->lun_lock);
9455	if (ctl_is_set(lun->have_ca, initidx)) {
9456		scsi_sense_data_type stored_format;
9457
9458		/*
9459		 * Check to see which sense format was used for the stored
9460		 * sense data.
9461		 */
9462		stored_format = scsi_sense_type(
9463		    &lun->pending_sense[initidx].sense);
9464
9465		/*
9466		 * If the user requested a different sense format than the
9467		 * one we stored, then we need to convert it to the other
9468		 * format.  If we're going from descriptor to fixed format
9469		 * sense data, we may lose things in translation, depending
9470		 * on what options were used.
9471		 *
9472		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9473		 * for some reason we'll just copy it out as-is.
9474		 */
9475		if ((stored_format == SSD_TYPE_FIXED)
9476		 && (sense_format == SSD_TYPE_DESC))
9477			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9478			    &lun->pending_sense[initidx].sense,
9479			    (struct scsi_sense_data_desc *)sense_ptr);
9480		else if ((stored_format == SSD_TYPE_DESC)
9481		      && (sense_format == SSD_TYPE_FIXED))
9482			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9483			    &lun->pending_sense[initidx].sense,
9484			    (struct scsi_sense_data_fixed *)sense_ptr);
9485		else
9486			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
9487			       ctl_min(sizeof(*sense_ptr),
9488			       sizeof(lun->pending_sense[initidx].sense)));
9489
9490		ctl_clear_mask(lun->have_ca, initidx);
9491		have_error = 1;
9492	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
9493		ctl_ua_type ua_type;
9494
9495		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
9496				       sense_ptr, sense_format);
9497		if (ua_type != CTL_UA_NONE) {
9498			have_error = 1;
9499			/* We're reporting this UA, so clear it */
9500			lun->pending_sense[initidx].ua_pending &= ~ua_type;
9501		}
9502	}
9503	mtx_unlock(&lun->lun_lock);
9504
9505	/*
9506	 * We already have a pending error, return it.
9507	 */
9508	if (have_error != 0) {
9509		/*
9510		 * We report the SCSI status as OK, since the status of the
9511		 * request sense command itself is OK.
9512		 */
9513		ctsio->scsi_status = SCSI_STATUS_OK;
9514
9515		/*
9516		 * We report 0 for the sense length, because we aren't doing
9517		 * autosense in this case.  We're reporting sense as
9518		 * parameter data.
9519		 */
9520		ctsio->sense_len = 0;
9521		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9522		ctsio->be_move_done = ctl_config_move_done;
9523		ctl_datamove((union ctl_io *)ctsio);
9524
9525		return (CTL_RETVAL_COMPLETE);
9526	}
9527
9528no_sense:
9529
9530	/*
9531	 * No sense information to report, so we report that everything is
9532	 * okay.
9533	 */
9534	ctl_set_sense_data(sense_ptr,
9535			   lun,
9536			   sense_format,
9537			   /*current_error*/ 1,
9538			   /*sense_key*/ SSD_KEY_NO_SENSE,
9539			   /*asc*/ 0x00,
9540			   /*ascq*/ 0x00,
9541			   SSD_ELEM_NONE);
9542
9543	ctsio->scsi_status = SCSI_STATUS_OK;
9544
9545	/*
9546	 * We report 0 for the sense length, because we aren't doing
9547	 * autosense in this case.  We're reporting sense as parameter data.
9548	 */
9549	ctsio->sense_len = 0;
9550	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9551	ctsio->be_move_done = ctl_config_move_done;
9552	ctl_datamove((union ctl_io *)ctsio);
9553
9554	return (CTL_RETVAL_COMPLETE);
9555}
9556
9557int
9558ctl_tur(struct ctl_scsiio *ctsio)
9559{
9560	struct ctl_lun *lun;
9561
9562	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9563
9564	CTL_DEBUG_PRINT(("ctl_tur\n"));
9565
9566	if (lun == NULL)
9567		return (EINVAL);
9568
9569	ctsio->scsi_status = SCSI_STATUS_OK;
9570	ctsio->io_hdr.status = CTL_SUCCESS;
9571
9572	ctl_done((union ctl_io *)ctsio);
9573
9574	return (CTL_RETVAL_COMPLETE);
9575}
9576
9577#ifdef notyet
9578static int
9579ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9580{
9581
9582}
9583#endif
9584
9585static int
9586ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9587{
9588	struct scsi_vpd_supported_pages *pages;
9589	int sup_page_size;
9590	struct ctl_lun *lun;
9591
9592	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9593
9594	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9595	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9596	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9597	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9598	ctsio->kern_sg_entries = 0;
9599
9600	if (sup_page_size < alloc_len) {
9601		ctsio->residual = alloc_len - sup_page_size;
9602		ctsio->kern_data_len = sup_page_size;
9603		ctsio->kern_total_len = sup_page_size;
9604	} else {
9605		ctsio->residual = 0;
9606		ctsio->kern_data_len = alloc_len;
9607		ctsio->kern_total_len = alloc_len;
9608	}
9609	ctsio->kern_data_resid = 0;
9610	ctsio->kern_rel_offset = 0;
9611	ctsio->kern_sg_entries = 0;
9612
9613	/*
9614	 * The control device is always connected.  The disk device, on the
9615	 * other hand, may not be online all the time.  Need to change this
9616	 * to figure out whether the disk device is actually online or not.
9617	 */
9618	if (lun != NULL)
9619		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9620				lun->be_lun->lun_type;
9621	else
9622		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9623
9624	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9625	/* Supported VPD pages */
9626	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9627	/* Serial Number */
9628	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9629	/* Device Identification */
9630	pages->page_list[2] = SVPD_DEVICE_ID;
9631	/* Block limits */
9632	pages->page_list[3] = SVPD_BLOCK_LIMITS;
9633	/* Logical Block Provisioning */
9634	pages->page_list[4] = SVPD_LBP;
9635
9636	ctsio->scsi_status = SCSI_STATUS_OK;
9637
9638	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9639	ctsio->be_move_done = ctl_config_move_done;
9640	ctl_datamove((union ctl_io *)ctsio);
9641
9642	return (CTL_RETVAL_COMPLETE);
9643}
9644
9645static int
9646ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9647{
9648	struct scsi_vpd_unit_serial_number *sn_ptr;
9649	struct ctl_lun *lun;
9650
9651	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9652
9653	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9654	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9655	ctsio->kern_sg_entries = 0;
9656
9657	if (sizeof(*sn_ptr) < alloc_len) {
9658		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9659		ctsio->kern_data_len = sizeof(*sn_ptr);
9660		ctsio->kern_total_len = sizeof(*sn_ptr);
9661	} else {
9662		ctsio->residual = 0;
9663		ctsio->kern_data_len = alloc_len;
9664		ctsio->kern_total_len = alloc_len;
9665	}
9666	ctsio->kern_data_resid = 0;
9667	ctsio->kern_rel_offset = 0;
9668	ctsio->kern_sg_entries = 0;
9669
9670	/*
9671	 * The control device is always connected.  The disk device, on the
9672	 * other hand, may not be online all the time.  Need to change this
9673	 * to figure out whether the disk device is actually online or not.
9674	 */
9675	if (lun != NULL)
9676		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9677				  lun->be_lun->lun_type;
9678	else
9679		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9680
9681	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9682	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9683	/*
9684	 * If we don't have a LUN, we just leave the serial number as
9685	 * all spaces.
9686	 */
9687	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9688	if (lun != NULL) {
9689		strncpy((char *)sn_ptr->serial_num,
9690			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9691	}
9692	ctsio->scsi_status = SCSI_STATUS_OK;
9693
9694	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9695	ctsio->be_move_done = ctl_config_move_done;
9696	ctl_datamove((union ctl_io *)ctsio);
9697
9698	return (CTL_RETVAL_COMPLETE);
9699}
9700
9701
9702static int
9703ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9704{
9705	struct scsi_vpd_device_id *devid_ptr;
9706	struct scsi_vpd_id_descriptor *desc;
9707	struct ctl_softc *ctl_softc;
9708	struct ctl_lun *lun;
9709	struct ctl_port *port;
9710	int data_len;
9711	uint8_t proto;
9712
9713	ctl_softc = control_softc;
9714
9715	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9716	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9717
9718	data_len = sizeof(struct scsi_vpd_device_id) +
9719	    sizeof(struct scsi_vpd_id_descriptor) +
9720		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9721	    sizeof(struct scsi_vpd_id_descriptor) +
9722		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9723	if (lun && lun->lun_devid)
9724		data_len += lun->lun_devid->len;
9725	if (port->port_devid)
9726		data_len += port->port_devid->len;
9727	if (port->target_devid)
9728		data_len += port->target_devid->len;
9729
9730	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9731	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9732	ctsio->kern_sg_entries = 0;
9733
9734	if (data_len < alloc_len) {
9735		ctsio->residual = alloc_len - data_len;
9736		ctsio->kern_data_len = data_len;
9737		ctsio->kern_total_len = data_len;
9738	} else {
9739		ctsio->residual = 0;
9740		ctsio->kern_data_len = alloc_len;
9741		ctsio->kern_total_len = alloc_len;
9742	}
9743	ctsio->kern_data_resid = 0;
9744	ctsio->kern_rel_offset = 0;
9745	ctsio->kern_sg_entries = 0;
9746
9747	/*
9748	 * The control device is always connected.  The disk device, on the
9749	 * other hand, may not be online all the time.
9750	 */
9751	if (lun != NULL)
9752		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9753				     lun->be_lun->lun_type;
9754	else
9755		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9756	devid_ptr->page_code = SVPD_DEVICE_ID;
9757	scsi_ulto2b(data_len - 4, devid_ptr->length);
9758
9759	if (port->port_type == CTL_PORT_FC)
9760		proto = SCSI_PROTO_FC << 4;
9761	else if (port->port_type == CTL_PORT_ISCSI)
9762		proto = SCSI_PROTO_ISCSI << 4;
9763	else
9764		proto = SCSI_PROTO_SPI << 4;
9765	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9766
9767	/*
9768	 * We're using a LUN association here.  i.e., this device ID is a
9769	 * per-LUN identifier.
9770	 */
9771	if (lun && lun->lun_devid) {
9772		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9773		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9774		    lun->lun_devid->len);
9775	}
9776
9777	/*
9778	 * This is for the WWPN which is a port association.
9779	 */
9780	if (port->port_devid) {
9781		memcpy(desc, port->port_devid->data, port->port_devid->len);
9782		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9783		    port->port_devid->len);
9784	}
9785
9786	/*
9787	 * This is for the Relative Target Port(type 4h) identifier
9788	 */
9789	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9790	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9791	    SVPD_ID_TYPE_RELTARG;
9792	desc->length = 4;
9793	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9794	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9795	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9796
9797	/*
9798	 * This is for the Target Port Group(type 5h) identifier
9799	 */
9800	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9801	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9802	    SVPD_ID_TYPE_TPORTGRP;
9803	desc->length = 4;
9804	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9805	    &desc->identifier[2]);
9806	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9807	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9808
9809	/*
9810	 * This is for the Target identifier
9811	 */
9812	if (port->target_devid) {
9813		memcpy(desc, port->target_devid->data, port->target_devid->len);
9814	}
9815
9816	ctsio->scsi_status = SCSI_STATUS_OK;
9817	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9818	ctsio->be_move_done = ctl_config_move_done;
9819	ctl_datamove((union ctl_io *)ctsio);
9820
9821	return (CTL_RETVAL_COMPLETE);
9822}
9823
9824static int
9825ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9826{
9827	struct scsi_vpd_block_limits *bl_ptr;
9828	struct ctl_lun *lun;
9829	int bs;
9830
9831	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9832	bs = lun->be_lun->blocksize;
9833
9834	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9835	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9836	ctsio->kern_sg_entries = 0;
9837
9838	if (sizeof(*bl_ptr) < alloc_len) {
9839		ctsio->residual = alloc_len - sizeof(*bl_ptr);
9840		ctsio->kern_data_len = sizeof(*bl_ptr);
9841		ctsio->kern_total_len = sizeof(*bl_ptr);
9842	} else {
9843		ctsio->residual = 0;
9844		ctsio->kern_data_len = alloc_len;
9845		ctsio->kern_total_len = alloc_len;
9846	}
9847	ctsio->kern_data_resid = 0;
9848	ctsio->kern_rel_offset = 0;
9849	ctsio->kern_sg_entries = 0;
9850
9851	/*
9852	 * The control device is always connected.  The disk device, on the
9853	 * other hand, may not be online all the time.  Need to change this
9854	 * to figure out whether the disk device is actually online or not.
9855	 */
9856	if (lun != NULL)
9857		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9858				  lun->be_lun->lun_type;
9859	else
9860		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9861
9862	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9863	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
9864	bl_ptr->max_cmp_write_len = 0xff;
9865	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9866	scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
9867	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9868		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9869		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9870	}
9871	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9872
9873	ctsio->scsi_status = SCSI_STATUS_OK;
9874	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9875	ctsio->be_move_done = ctl_config_move_done;
9876	ctl_datamove((union ctl_io *)ctsio);
9877
9878	return (CTL_RETVAL_COMPLETE);
9879}
9880
9881static int
9882ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9883{
9884	struct scsi_vpd_logical_block_prov *lbp_ptr;
9885	struct ctl_lun *lun;
9886	int bs;
9887
9888	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9889	bs = lun->be_lun->blocksize;
9890
9891	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9892	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9893	ctsio->kern_sg_entries = 0;
9894
9895	if (sizeof(*lbp_ptr) < alloc_len) {
9896		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9897		ctsio->kern_data_len = sizeof(*lbp_ptr);
9898		ctsio->kern_total_len = sizeof(*lbp_ptr);
9899	} else {
9900		ctsio->residual = 0;
9901		ctsio->kern_data_len = alloc_len;
9902		ctsio->kern_total_len = alloc_len;
9903	}
9904	ctsio->kern_data_resid = 0;
9905	ctsio->kern_rel_offset = 0;
9906	ctsio->kern_sg_entries = 0;
9907
9908	/*
9909	 * The control device is always connected.  The disk device, on the
9910	 * other hand, may not be online all the time.  Need to change this
9911	 * to figure out whether the disk device is actually online or not.
9912	 */
9913	if (lun != NULL)
9914		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9915				  lun->be_lun->lun_type;
9916	else
9917		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9918
9919	lbp_ptr->page_code = SVPD_LBP;
9920	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
9921		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10;
9922
9923	ctsio->scsi_status = SCSI_STATUS_OK;
9924	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9925	ctsio->be_move_done = ctl_config_move_done;
9926	ctl_datamove((union ctl_io *)ctsio);
9927
9928	return (CTL_RETVAL_COMPLETE);
9929}
9930
9931static int
9932ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9933{
9934	struct scsi_inquiry *cdb;
9935	struct ctl_lun *lun;
9936	int alloc_len, retval;
9937
9938	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9939	cdb = (struct scsi_inquiry *)ctsio->cdb;
9940
9941	retval = CTL_RETVAL_COMPLETE;
9942
9943	alloc_len = scsi_2btoul(cdb->length);
9944
9945	switch (cdb->page_code) {
9946	case SVPD_SUPPORTED_PAGES:
9947		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9948		break;
9949	case SVPD_UNIT_SERIAL_NUMBER:
9950		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9951		break;
9952	case SVPD_DEVICE_ID:
9953		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9954		break;
9955	case SVPD_BLOCK_LIMITS:
9956		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9957		break;
9958	case SVPD_LBP:
9959		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9960		break;
9961	default:
9962		ctl_set_invalid_field(ctsio,
9963				      /*sks_valid*/ 1,
9964				      /*command*/ 1,
9965				      /*field*/ 2,
9966				      /*bit_valid*/ 0,
9967				      /*bit*/ 0);
9968		ctl_done((union ctl_io *)ctsio);
9969		retval = CTL_RETVAL_COMPLETE;
9970		break;
9971	}
9972
9973	return (retval);
9974}
9975
9976static int
9977ctl_inquiry_std(struct ctl_scsiio *ctsio)
9978{
9979	struct scsi_inquiry_data *inq_ptr;
9980	struct scsi_inquiry *cdb;
9981	struct ctl_softc *ctl_softc;
9982	struct ctl_lun *lun;
9983	char *val;
9984	uint32_t alloc_len;
9985	int is_fc;
9986
9987	ctl_softc = control_softc;
9988
9989	/*
9990	 * Figure out whether we're talking to a Fibre Channel port or not.
9991	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9992	 * SCSI front ends.
9993	 */
9994	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9995	    CTL_PORT_FC)
9996		is_fc = 0;
9997	else
9998		is_fc = 1;
9999
10000	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10001	cdb = (struct scsi_inquiry *)ctsio->cdb;
10002	alloc_len = scsi_2btoul(cdb->length);
10003
10004	/*
10005	 * We malloc the full inquiry data size here and fill it
10006	 * in.  If the user only asks for less, we'll give him
10007	 * that much.
10008	 */
10009	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
10010	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10011	ctsio->kern_sg_entries = 0;
10012	ctsio->kern_data_resid = 0;
10013	ctsio->kern_rel_offset = 0;
10014
10015	if (sizeof(*inq_ptr) < alloc_len) {
10016		ctsio->residual = alloc_len - sizeof(*inq_ptr);
10017		ctsio->kern_data_len = sizeof(*inq_ptr);
10018		ctsio->kern_total_len = sizeof(*inq_ptr);
10019	} else {
10020		ctsio->residual = 0;
10021		ctsio->kern_data_len = alloc_len;
10022		ctsio->kern_total_len = alloc_len;
10023	}
10024
10025	/*
10026	 * If we have a LUN configured, report it as connected.  Otherwise,
10027	 * report that it is offline or no device is supported, depending
10028	 * on the value of inquiry_pq_no_lun.
10029	 *
10030	 * According to the spec (SPC-4 r34), the peripheral qualifier
10031	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10032	 *
10033	 * "A peripheral device having the specified peripheral device type
10034	 * is not connected to this logical unit. However, the device
10035	 * server is capable of supporting the specified peripheral device
10036	 * type on this logical unit."
10037	 *
10038	 * According to the same spec, the peripheral qualifier
10039	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10040	 *
10041	 * "The device server is not capable of supporting a peripheral
10042	 * device on this logical unit. For this peripheral qualifier the
10043	 * peripheral device type shall be set to 1Fh. All other peripheral
10044	 * device type values are reserved for this peripheral qualifier."
10045	 *
10046	 * Given the text, it would seem that we probably want to report that
10047	 * the LUN is offline here.  There is no LUN connected, but we can
10048	 * support a LUN at the given LUN number.
10049	 *
10050	 * In the real world, though, it sounds like things are a little
10051	 * different:
10052	 *
10053	 * - Linux, when presented with a LUN with the offline peripheral
10054	 *   qualifier, will create an sg driver instance for it.  So when
10055	 *   you attach it to CTL, you wind up with a ton of sg driver
10056	 *   instances.  (One for every LUN that Linux bothered to probe.)
10057	 *   Linux does this despite the fact that it issues a REPORT LUNs
10058	 *   to LUN 0 to get the inventory of supported LUNs.
10059	 *
10060	 * - There is other anecdotal evidence (from Emulex folks) about
10061	 *   arrays that use the offline peripheral qualifier for LUNs that
10062	 *   are on the "passive" path in an active/passive array.
10063	 *
10064	 * So the solution is provide a hopefully reasonable default
10065	 * (return bad/no LUN) and allow the user to change the behavior
10066	 * with a tunable/sysctl variable.
10067	 */
10068	if (lun != NULL)
10069		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10070				  lun->be_lun->lun_type;
10071	else if (ctl_softc->inquiry_pq_no_lun == 0)
10072		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10073	else
10074		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10075
10076	/* RMB in byte 2 is 0 */
10077	inq_ptr->version = SCSI_REV_SPC3;
10078
10079	/*
10080	 * According to SAM-3, even if a device only supports a single
10081	 * level of LUN addressing, it should still set the HISUP bit:
10082	 *
10083	 * 4.9.1 Logical unit numbers overview
10084	 *
10085	 * All logical unit number formats described in this standard are
10086	 * hierarchical in structure even when only a single level in that
10087	 * hierarchy is used. The HISUP bit shall be set to one in the
10088	 * standard INQUIRY data (see SPC-2) when any logical unit number
10089	 * format described in this standard is used.  Non-hierarchical
10090	 * formats are outside the scope of this standard.
10091	 *
10092	 * Therefore we set the HiSup bit here.
10093	 *
10094	 * The reponse format is 2, per SPC-3.
10095	 */
10096	inq_ptr->response_format = SID_HiSup | 2;
10097
10098	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
10099	CTL_DEBUG_PRINT(("additional_length = %d\n",
10100			 inq_ptr->additional_length));
10101
10102	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
10103	/* 16 bit addressing */
10104	if (is_fc == 0)
10105		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10106	/* XXX set the SID_MultiP bit here if we're actually going to
10107	   respond on multiple ports */
10108	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10109
10110	/* 16 bit data bus, synchronous transfers */
10111	/* XXX these flags don't apply for FC */
10112	if (is_fc == 0)
10113		inq_ptr->flags = SID_WBus16 | SID_Sync;
10114	/*
10115	 * XXX KDM do we want to support tagged queueing on the control
10116	 * device at all?
10117	 */
10118	if ((lun == NULL)
10119	 || (lun->be_lun->lun_type != T_PROCESSOR))
10120		inq_ptr->flags |= SID_CmdQue;
10121	/*
10122	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10123	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10124	 * name and 4 bytes for the revision.
10125	 */
10126	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10127	    "vendor")) == NULL) {
10128		strcpy(inq_ptr->vendor, CTL_VENDOR);
10129	} else {
10130		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10131		strncpy(inq_ptr->vendor, val,
10132		    min(sizeof(inq_ptr->vendor), strlen(val)));
10133	}
10134	if (lun == NULL) {
10135		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10136	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10137		switch (lun->be_lun->lun_type) {
10138		case T_DIRECT:
10139			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10140			break;
10141		case T_PROCESSOR:
10142			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
10143			break;
10144		default:
10145			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
10146			break;
10147		}
10148	} else {
10149		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10150		strncpy(inq_ptr->product, val,
10151		    min(sizeof(inq_ptr->product), strlen(val)));
10152	}
10153
10154	/*
10155	 * XXX make this a macro somewhere so it automatically gets
10156	 * incremented when we make changes.
10157	 */
10158	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10159	    "revision")) == NULL) {
10160		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10161	} else {
10162		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10163		strncpy(inq_ptr->revision, val,
10164		    min(sizeof(inq_ptr->revision), strlen(val)));
10165	}
10166
10167	/*
10168	 * For parallel SCSI, we support double transition and single
10169	 * transition clocking.  We also support QAS (Quick Arbitration
10170	 * and Selection) and Information Unit transfers on both the
10171	 * control and array devices.
10172	 */
10173	if (is_fc == 0)
10174		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10175				    SID_SPI_IUS;
10176
10177	/* SAM-3 */
10178	scsi_ulto2b(0x0060, inq_ptr->version1);
10179	/* SPC-3 (no version claimed) XXX should we claim a version? */
10180	scsi_ulto2b(0x0300, inq_ptr->version2);
10181	if (is_fc) {
10182		/* FCP-2 ANSI INCITS.350:2003 */
10183		scsi_ulto2b(0x0917, inq_ptr->version3);
10184	} else {
10185		/* SPI-4 ANSI INCITS.362:200x */
10186		scsi_ulto2b(0x0B56, inq_ptr->version3);
10187	}
10188
10189	if (lun == NULL) {
10190		/* SBC-2 (no version claimed) XXX should we claim a version? */
10191		scsi_ulto2b(0x0320, inq_ptr->version4);
10192	} else {
10193		switch (lun->be_lun->lun_type) {
10194		case T_DIRECT:
10195			/*
10196			 * SBC-2 (no version claimed) XXX should we claim a
10197			 * version?
10198			 */
10199			scsi_ulto2b(0x0320, inq_ptr->version4);
10200			break;
10201		case T_PROCESSOR:
10202		default:
10203			break;
10204		}
10205	}
10206
10207	ctsio->scsi_status = SCSI_STATUS_OK;
10208	if (ctsio->kern_data_len > 0) {
10209		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10210		ctsio->be_move_done = ctl_config_move_done;
10211		ctl_datamove((union ctl_io *)ctsio);
10212	} else {
10213		ctsio->io_hdr.status = CTL_SUCCESS;
10214		ctl_done((union ctl_io *)ctsio);
10215	}
10216
10217	return (CTL_RETVAL_COMPLETE);
10218}
10219
10220int
10221ctl_inquiry(struct ctl_scsiio *ctsio)
10222{
10223	struct scsi_inquiry *cdb;
10224	int retval;
10225
10226	cdb = (struct scsi_inquiry *)ctsio->cdb;
10227
10228	retval = 0;
10229
10230	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10231
10232	/*
10233	 * Right now, we don't support the CmdDt inquiry information.
10234	 * This would be nice to support in the future.  When we do
10235	 * support it, we should change this test so that it checks to make
10236	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10237	 */
10238#ifdef notyet
10239	if (((cdb->byte2 & SI_EVPD)
10240	 && (cdb->byte2 & SI_CMDDT)))
10241#endif
10242	if (cdb->byte2 & SI_CMDDT) {
10243		/*
10244		 * Point to the SI_CMDDT bit.  We might change this
10245		 * when we support SI_CMDDT, but since both bits would be
10246		 * "wrong", this should probably just stay as-is then.
10247		 */
10248		ctl_set_invalid_field(ctsio,
10249				      /*sks_valid*/ 1,
10250				      /*command*/ 1,
10251				      /*field*/ 1,
10252				      /*bit_valid*/ 1,
10253				      /*bit*/ 1);
10254		ctl_done((union ctl_io *)ctsio);
10255		return (CTL_RETVAL_COMPLETE);
10256	}
10257	if (cdb->byte2 & SI_EVPD)
10258		retval = ctl_inquiry_evpd(ctsio);
10259#ifdef notyet
10260	else if (cdb->byte2 & SI_CMDDT)
10261		retval = ctl_inquiry_cmddt(ctsio);
10262#endif
10263	else
10264		retval = ctl_inquiry_std(ctsio);
10265
10266	return (retval);
10267}
10268
10269/*
10270 * For known CDB types, parse the LBA and length.
10271 */
10272static int
10273ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10274{
10275	if (io->io_hdr.io_type != CTL_IO_SCSI)
10276		return (1);
10277
10278	switch (io->scsiio.cdb[0]) {
10279	case COMPARE_AND_WRITE: {
10280		struct scsi_compare_and_write *cdb;
10281
10282		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10283
10284		*lba = scsi_8btou64(cdb->addr);
10285		*len = cdb->length;
10286		break;
10287	}
10288	case READ_6:
10289	case WRITE_6: {
10290		struct scsi_rw_6 *cdb;
10291
10292		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10293
10294		*lba = scsi_3btoul(cdb->addr);
10295		/* only 5 bits are valid in the most significant address byte */
10296		*lba &= 0x1fffff;
10297		*len = cdb->length;
10298		break;
10299	}
10300	case READ_10:
10301	case WRITE_10: {
10302		struct scsi_rw_10 *cdb;
10303
10304		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10305
10306		*lba = scsi_4btoul(cdb->addr);
10307		*len = scsi_2btoul(cdb->length);
10308		break;
10309	}
10310	case WRITE_VERIFY_10: {
10311		struct scsi_write_verify_10 *cdb;
10312
10313		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10314
10315		*lba = scsi_4btoul(cdb->addr);
10316		*len = scsi_2btoul(cdb->length);
10317		break;
10318	}
10319	case READ_12:
10320	case WRITE_12: {
10321		struct scsi_rw_12 *cdb;
10322
10323		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10324
10325		*lba = scsi_4btoul(cdb->addr);
10326		*len = scsi_4btoul(cdb->length);
10327		break;
10328	}
10329	case WRITE_VERIFY_12: {
10330		struct scsi_write_verify_12 *cdb;
10331
10332		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10333
10334		*lba = scsi_4btoul(cdb->addr);
10335		*len = scsi_4btoul(cdb->length);
10336		break;
10337	}
10338	case READ_16:
10339	case WRITE_16: {
10340		struct scsi_rw_16 *cdb;
10341
10342		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10343
10344		*lba = scsi_8btou64(cdb->addr);
10345		*len = scsi_4btoul(cdb->length);
10346		break;
10347	}
10348	case WRITE_VERIFY_16: {
10349		struct scsi_write_verify_16 *cdb;
10350
10351		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10352
10353
10354		*lba = scsi_8btou64(cdb->addr);
10355		*len = scsi_4btoul(cdb->length);
10356		break;
10357	}
10358	case WRITE_SAME_10: {
10359		struct scsi_write_same_10 *cdb;
10360
10361		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10362
10363		*lba = scsi_4btoul(cdb->addr);
10364		*len = scsi_2btoul(cdb->length);
10365		break;
10366	}
10367	case WRITE_SAME_16: {
10368		struct scsi_write_same_16 *cdb;
10369
10370		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10371
10372		*lba = scsi_8btou64(cdb->addr);
10373		*len = scsi_4btoul(cdb->length);
10374		break;
10375	}
10376	case VERIFY_10: {
10377		struct scsi_verify_10 *cdb;
10378
10379		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10380
10381		*lba = scsi_4btoul(cdb->addr);
10382		*len = scsi_2btoul(cdb->length);
10383		break;
10384	}
10385	case VERIFY_12: {
10386		struct scsi_verify_12 *cdb;
10387
10388		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10389
10390		*lba = scsi_4btoul(cdb->addr);
10391		*len = scsi_4btoul(cdb->length);
10392		break;
10393	}
10394	case VERIFY_16: {
10395		struct scsi_verify_16 *cdb;
10396
10397		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10398
10399		*lba = scsi_8btou64(cdb->addr);
10400		*len = scsi_4btoul(cdb->length);
10401		break;
10402	}
10403	default:
10404		return (1);
10405		break; /* NOTREACHED */
10406	}
10407
10408	return (0);
10409}
10410
10411static ctl_action
10412ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10413{
10414	uint64_t endlba1, endlba2;
10415
10416	endlba1 = lba1 + len1 - 1;
10417	endlba2 = lba2 + len2 - 1;
10418
10419	if ((endlba1 < lba2)
10420	 || (endlba2 < lba1))
10421		return (CTL_ACTION_PASS);
10422	else
10423		return (CTL_ACTION_BLOCK);
10424}
10425
10426static ctl_action
10427ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10428{
10429	uint64_t lba1, lba2;
10430	uint32_t len1, len2;
10431	int retval;
10432
10433	retval = ctl_get_lba_len(io1, &lba1, &len1);
10434	if (retval != 0)
10435		return (CTL_ACTION_ERROR);
10436
10437	retval = ctl_get_lba_len(io2, &lba2, &len2);
10438	if (retval != 0)
10439		return (CTL_ACTION_ERROR);
10440
10441	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10442}
10443
10444static ctl_action
10445ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10446{
10447	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10448	ctl_serialize_action *serialize_row;
10449
10450	/*
10451	 * The initiator attempted multiple untagged commands at the same
10452	 * time.  Can't do that.
10453	 */
10454	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10455	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10456	 && ((pending_io->io_hdr.nexus.targ_port ==
10457	      ooa_io->io_hdr.nexus.targ_port)
10458	  && (pending_io->io_hdr.nexus.initid.id ==
10459	      ooa_io->io_hdr.nexus.initid.id))
10460	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10461		return (CTL_ACTION_OVERLAP);
10462
10463	/*
10464	 * The initiator attempted to send multiple tagged commands with
10465	 * the same ID.  (It's fine if different initiators have the same
10466	 * tag ID.)
10467	 *
10468	 * Even if all of those conditions are true, we don't kill the I/O
10469	 * if the command ahead of us has been aborted.  We won't end up
10470	 * sending it to the FETD, and it's perfectly legal to resend a
10471	 * command with the same tag number as long as the previous
10472	 * instance of this tag number has been aborted somehow.
10473	 */
10474	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10475	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10476	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10477	 && ((pending_io->io_hdr.nexus.targ_port ==
10478	      ooa_io->io_hdr.nexus.targ_port)
10479	  && (pending_io->io_hdr.nexus.initid.id ==
10480	      ooa_io->io_hdr.nexus.initid.id))
10481	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10482		return (CTL_ACTION_OVERLAP_TAG);
10483
10484	/*
10485	 * If we get a head of queue tag, SAM-3 says that we should
10486	 * immediately execute it.
10487	 *
10488	 * What happens if this command would normally block for some other
10489	 * reason?  e.g. a request sense with a head of queue tag
10490	 * immediately after a write.  Normally that would block, but this
10491	 * will result in its getting executed immediately...
10492	 *
10493	 * We currently return "pass" instead of "skip", so we'll end up
10494	 * going through the rest of the queue to check for overlapped tags.
10495	 *
10496	 * XXX KDM check for other types of blockage first??
10497	 */
10498	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10499		return (CTL_ACTION_PASS);
10500
10501	/*
10502	 * Ordered tags have to block until all items ahead of them
10503	 * have completed.  If we get called with an ordered tag, we always
10504	 * block, if something else is ahead of us in the queue.
10505	 */
10506	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10507		return (CTL_ACTION_BLOCK);
10508
10509	/*
10510	 * Simple tags get blocked until all head of queue and ordered tags
10511	 * ahead of them have completed.  I'm lumping untagged commands in
10512	 * with simple tags here.  XXX KDM is that the right thing to do?
10513	 */
10514	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10515	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10516	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10517	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10518		return (CTL_ACTION_BLOCK);
10519
10520	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio);
10521	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio);
10522
10523	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10524
10525	switch (serialize_row[pending_entry->seridx]) {
10526	case CTL_SER_BLOCK:
10527		return (CTL_ACTION_BLOCK);
10528		break; /* NOTREACHED */
10529	case CTL_SER_EXTENT:
10530		return (ctl_extent_check(pending_io, ooa_io));
10531		break; /* NOTREACHED */
10532	case CTL_SER_PASS:
10533		return (CTL_ACTION_PASS);
10534		break; /* NOTREACHED */
10535	case CTL_SER_SKIP:
10536		return (CTL_ACTION_SKIP);
10537		break;
10538	default:
10539		panic("invalid serialization value %d",
10540		      serialize_row[pending_entry->seridx]);
10541		break; /* NOTREACHED */
10542	}
10543
10544	return (CTL_ACTION_ERROR);
10545}
10546
10547/*
10548 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10549 * Assumptions:
10550 * - pending_io is generally either incoming, or on the blocked queue
10551 * - starting I/O is the I/O we want to start the check with.
10552 */
10553static ctl_action
10554ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10555	      union ctl_io *starting_io)
10556{
10557	union ctl_io *ooa_io;
10558	ctl_action action;
10559
10560	mtx_assert(&lun->lun_lock, MA_OWNED);
10561
10562	/*
10563	 * Run back along the OOA queue, starting with the current
10564	 * blocked I/O and going through every I/O before it on the
10565	 * queue.  If starting_io is NULL, we'll just end up returning
10566	 * CTL_ACTION_PASS.
10567	 */
10568	for (ooa_io = starting_io; ooa_io != NULL;
10569	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10570	     ooa_links)){
10571
10572		/*
10573		 * This routine just checks to see whether
10574		 * cur_blocked is blocked by ooa_io, which is ahead
10575		 * of it in the queue.  It doesn't queue/dequeue
10576		 * cur_blocked.
10577		 */
10578		action = ctl_check_for_blockage(pending_io, ooa_io);
10579		switch (action) {
10580		case CTL_ACTION_BLOCK:
10581		case CTL_ACTION_OVERLAP:
10582		case CTL_ACTION_OVERLAP_TAG:
10583		case CTL_ACTION_SKIP:
10584		case CTL_ACTION_ERROR:
10585			return (action);
10586			break; /* NOTREACHED */
10587		case CTL_ACTION_PASS:
10588			break;
10589		default:
10590			panic("invalid action %d", action);
10591			break;  /* NOTREACHED */
10592		}
10593	}
10594
10595	return (CTL_ACTION_PASS);
10596}
10597
10598/*
10599 * Assumptions:
10600 * - An I/O has just completed, and has been removed from the per-LUN OOA
10601 *   queue, so some items on the blocked queue may now be unblocked.
10602 */
10603static int
10604ctl_check_blocked(struct ctl_lun *lun)
10605{
10606	union ctl_io *cur_blocked, *next_blocked;
10607
10608	mtx_assert(&lun->lun_lock, MA_OWNED);
10609
10610	/*
10611	 * Run forward from the head of the blocked queue, checking each
10612	 * entry against the I/Os prior to it on the OOA queue to see if
10613	 * there is still any blockage.
10614	 *
10615	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10616	 * with our removing a variable on it while it is traversing the
10617	 * list.
10618	 */
10619	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10620	     cur_blocked != NULL; cur_blocked = next_blocked) {
10621		union ctl_io *prev_ooa;
10622		ctl_action action;
10623
10624		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10625							  blocked_links);
10626
10627		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10628						      ctl_ooaq, ooa_links);
10629
10630		/*
10631		 * If cur_blocked happens to be the first item in the OOA
10632		 * queue now, prev_ooa will be NULL, and the action
10633		 * returned will just be CTL_ACTION_PASS.
10634		 */
10635		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10636
10637		switch (action) {
10638		case CTL_ACTION_BLOCK:
10639			/* Nothing to do here, still blocked */
10640			break;
10641		case CTL_ACTION_OVERLAP:
10642		case CTL_ACTION_OVERLAP_TAG:
10643			/*
10644			 * This shouldn't happen!  In theory we've already
10645			 * checked this command for overlap...
10646			 */
10647			break;
10648		case CTL_ACTION_PASS:
10649		case CTL_ACTION_SKIP: {
10650			struct ctl_softc *softc;
10651			const struct ctl_cmd_entry *entry;
10652			uint32_t initidx;
10653			int isc_retval;
10654
10655			/*
10656			 * The skip case shouldn't happen, this transaction
10657			 * should have never made it onto the blocked queue.
10658			 */
10659			/*
10660			 * This I/O is no longer blocked, we can remove it
10661			 * from the blocked queue.  Since this is a TAILQ
10662			 * (doubly linked list), we can do O(1) removals
10663			 * from any place on the list.
10664			 */
10665			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10666				     blocked_links);
10667			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10668
10669			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10670				/*
10671				 * Need to send IO back to original side to
10672				 * run
10673				 */
10674				union ctl_ha_msg msg_info;
10675
10676				msg_info.hdr.original_sc =
10677					cur_blocked->io_hdr.original_sc;
10678				msg_info.hdr.serializing_sc = cur_blocked;
10679				msg_info.hdr.msg_type = CTL_MSG_R2R;
10680				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10681				     &msg_info, sizeof(msg_info), 0)) >
10682				     CTL_HA_STATUS_SUCCESS) {
10683					printf("CTL:Check Blocked error from "
10684					       "ctl_ha_msg_send %d\n",
10685					       isc_retval);
10686				}
10687				break;
10688			}
10689			entry = ctl_get_cmd_entry(&cur_blocked->scsiio);
10690			softc = control_softc;
10691
10692			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
10693
10694			/*
10695			 * Check this I/O for LUN state changes that may
10696			 * have happened while this command was blocked.
10697			 * The LUN state may have been changed by a command
10698			 * ahead of us in the queue, so we need to re-check
10699			 * for any states that can be caused by SCSI
10700			 * commands.
10701			 */
10702			if (ctl_scsiio_lun_check(softc, lun, entry,
10703						 &cur_blocked->scsiio) == 0) {
10704				cur_blocked->io_hdr.flags |=
10705				                      CTL_FLAG_IS_WAS_ON_RTR;
10706				ctl_enqueue_rtr(cur_blocked);
10707			} else
10708				ctl_done(cur_blocked);
10709			break;
10710		}
10711		default:
10712			/*
10713			 * This probably shouldn't happen -- we shouldn't
10714			 * get CTL_ACTION_ERROR, or anything else.
10715			 */
10716			break;
10717		}
10718	}
10719
10720	return (CTL_RETVAL_COMPLETE);
10721}
10722
10723/*
10724 * This routine (with one exception) checks LUN flags that can be set by
10725 * commands ahead of us in the OOA queue.  These flags have to be checked
10726 * when a command initially comes in, and when we pull a command off the
10727 * blocked queue and are preparing to execute it.  The reason we have to
10728 * check these flags for commands on the blocked queue is that the LUN
10729 * state may have been changed by a command ahead of us while we're on the
10730 * blocked queue.
10731 *
10732 * Ordering is somewhat important with these checks, so please pay
10733 * careful attention to the placement of any new checks.
10734 */
10735static int
10736ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
10737    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10738{
10739	int retval;
10740
10741	retval = 0;
10742
10743	mtx_assert(&lun->lun_lock, MA_OWNED);
10744
10745	/*
10746	 * If this shelf is a secondary shelf controller, we have to reject
10747	 * any media access commands.
10748	 */
10749#if 0
10750	/* No longer needed for HA */
10751	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10752	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10753		ctl_set_lun_standby(ctsio);
10754		retval = 1;
10755		goto bailout;
10756	}
10757#endif
10758
10759	/*
10760	 * Check for a reservation conflict.  If this command isn't allowed
10761	 * even on reserved LUNs, and if this initiator isn't the one who
10762	 * reserved us, reject the command with a reservation conflict.
10763	 */
10764	if ((lun->flags & CTL_LUN_RESERVED)
10765	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10766		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10767		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10768		 || (ctsio->io_hdr.nexus.targ_target.id !=
10769		     lun->rsv_nexus.targ_target.id)) {
10770			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10771			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10772			retval = 1;
10773			goto bailout;
10774		}
10775	}
10776
10777	if ( (lun->flags & CTL_LUN_PR_RESERVED)
10778	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10779		uint32_t residx;
10780
10781		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10782		/*
10783		 * if we aren't registered or it's a res holder type
10784		 * reservation and this isn't the res holder then set a
10785		 * conflict.
10786		 * NOTE: Commands which might be allowed on write exclusive
10787		 * type reservations are checked in the particular command
10788		 * for a conflict. Read and SSU are the only ones.
10789		 */
10790		if (!lun->per_res[residx].registered
10791		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10792			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10793			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10794			retval = 1;
10795			goto bailout;
10796		}
10797
10798	}
10799
10800	if ((lun->flags & CTL_LUN_OFFLINE)
10801	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10802		ctl_set_lun_not_ready(ctsio);
10803		retval = 1;
10804		goto bailout;
10805	}
10806
10807	/*
10808	 * If the LUN is stopped, see if this particular command is allowed
10809	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10810	 */
10811	if ((lun->flags & CTL_LUN_STOPPED)
10812	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10813		/* "Logical unit not ready, initializing cmd. required" */
10814		ctl_set_lun_stopped(ctsio);
10815		retval = 1;
10816		goto bailout;
10817	}
10818
10819	if ((lun->flags & CTL_LUN_INOPERABLE)
10820	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10821		/* "Medium format corrupted" */
10822		ctl_set_medium_format_corrupted(ctsio);
10823		retval = 1;
10824		goto bailout;
10825	}
10826
10827bailout:
10828	return (retval);
10829
10830}
10831
10832static void
10833ctl_failover_io(union ctl_io *io, int have_lock)
10834{
10835	ctl_set_busy(&io->scsiio);
10836	ctl_done(io);
10837}
10838
10839static void
10840ctl_failover(void)
10841{
10842	struct ctl_lun *lun;
10843	struct ctl_softc *ctl_softc;
10844	union ctl_io *next_io, *pending_io;
10845	union ctl_io *io;
10846	int lun_idx;
10847	int i;
10848
10849	ctl_softc = control_softc;
10850
10851	mtx_lock(&ctl_softc->ctl_lock);
10852	/*
10853	 * Remove any cmds from the other SC from the rtr queue.  These
10854	 * will obviously only be for LUNs for which we're the primary.
10855	 * We can't send status or get/send data for these commands.
10856	 * Since they haven't been executed yet, we can just remove them.
10857	 * We'll either abort them or delete them below, depending on
10858	 * which HA mode we're in.
10859	 */
10860#ifdef notyet
10861	mtx_lock(&ctl_softc->queue_lock);
10862	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10863	     io != NULL; io = next_io) {
10864		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10865		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10866			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10867				      ctl_io_hdr, links);
10868	}
10869	mtx_unlock(&ctl_softc->queue_lock);
10870#endif
10871
10872	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10873		lun = ctl_softc->ctl_luns[lun_idx];
10874		if (lun==NULL)
10875			continue;
10876
10877		/*
10878		 * Processor LUNs are primary on both sides.
10879		 * XXX will this always be true?
10880		 */
10881		if (lun->be_lun->lun_type == T_PROCESSOR)
10882			continue;
10883
10884		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10885		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10886			printf("FAILOVER: primary lun %d\n", lun_idx);
10887		        /*
10888			 * Remove all commands from the other SC. First from the
10889			 * blocked queue then from the ooa queue. Once we have
10890			 * removed them. Call ctl_check_blocked to see if there
10891			 * is anything that can run.
10892			 */
10893			for (io = (union ctl_io *)TAILQ_FIRST(
10894			     &lun->blocked_queue); io != NULL; io = next_io) {
10895
10896		        	next_io = (union ctl_io *)TAILQ_NEXT(
10897				    &io->io_hdr, blocked_links);
10898
10899				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10900					TAILQ_REMOVE(&lun->blocked_queue,
10901						     &io->io_hdr,blocked_links);
10902					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10903					TAILQ_REMOVE(&lun->ooa_queue,
10904						     &io->io_hdr, ooa_links);
10905
10906					ctl_free_io(io);
10907				}
10908			}
10909
10910			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10911	     		     io != NULL; io = next_io) {
10912
10913		        	next_io = (union ctl_io *)TAILQ_NEXT(
10914				    &io->io_hdr, ooa_links);
10915
10916				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10917
10918					TAILQ_REMOVE(&lun->ooa_queue,
10919						&io->io_hdr,
10920					     	ooa_links);
10921
10922					ctl_free_io(io);
10923				}
10924			}
10925			ctl_check_blocked(lun);
10926		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10927			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10928
10929			printf("FAILOVER: primary lun %d\n", lun_idx);
10930			/*
10931			 * Abort all commands from the other SC.  We can't
10932			 * send status back for them now.  These should get
10933			 * cleaned up when they are completed or come out
10934			 * for a datamove operation.
10935			 */
10936			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10937	     		     io != NULL; io = next_io) {
10938		        	next_io = (union ctl_io *)TAILQ_NEXT(
10939					&io->io_hdr, ooa_links);
10940
10941				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10942					io->io_hdr.flags |= CTL_FLAG_ABORT;
10943			}
10944		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10945			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10946
10947			printf("FAILOVER: secondary lun %d\n", lun_idx);
10948
10949			lun->flags |= CTL_LUN_PRIMARY_SC;
10950
10951			/*
10952			 * We send all I/O that was sent to this controller
10953			 * and redirected to the other side back with
10954			 * busy status, and have the initiator retry it.
10955			 * Figuring out how much data has been transferred,
10956			 * etc. and picking up where we left off would be
10957			 * very tricky.
10958			 *
10959			 * XXX KDM need to remove I/O from the blocked
10960			 * queue as well!
10961			 */
10962			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10963			     &lun->ooa_queue); pending_io != NULL;
10964			     pending_io = next_io) {
10965
10966				next_io =  (union ctl_io *)TAILQ_NEXT(
10967					&pending_io->io_hdr, ooa_links);
10968
10969				pending_io->io_hdr.flags &=
10970					~CTL_FLAG_SENT_2OTHER_SC;
10971
10972				if (pending_io->io_hdr.flags &
10973				    CTL_FLAG_IO_ACTIVE) {
10974					pending_io->io_hdr.flags |=
10975						CTL_FLAG_FAILOVER;
10976				} else {
10977					ctl_set_busy(&pending_io->scsiio);
10978					ctl_done(pending_io);
10979				}
10980			}
10981
10982			/*
10983			 * Build Unit Attention
10984			 */
10985			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10986				lun->pending_sense[i].ua_pending |=
10987				                     CTL_UA_ASYM_ACC_CHANGE;
10988			}
10989		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10990			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10991			printf("FAILOVER: secondary lun %d\n", lun_idx);
10992			/*
10993			 * if the first io on the OOA is not on the RtR queue
10994			 * add it.
10995			 */
10996			lun->flags |= CTL_LUN_PRIMARY_SC;
10997
10998			pending_io = (union ctl_io *)TAILQ_FIRST(
10999			    &lun->ooa_queue);
11000			if (pending_io==NULL) {
11001				printf("Nothing on OOA queue\n");
11002				continue;
11003			}
11004
11005			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11006			if ((pending_io->io_hdr.flags &
11007			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11008				pending_io->io_hdr.flags |=
11009				    CTL_FLAG_IS_WAS_ON_RTR;
11010				ctl_enqueue_rtr(pending_io);
11011			}
11012#if 0
11013			else
11014			{
11015				printf("Tag 0x%04x is running\n",
11016				      pending_io->scsiio.tag_num);
11017			}
11018#endif
11019
11020			next_io = (union ctl_io *)TAILQ_NEXT(
11021			    &pending_io->io_hdr, ooa_links);
11022			for (pending_io=next_io; pending_io != NULL;
11023			     pending_io = next_io) {
11024				pending_io->io_hdr.flags &=
11025				    ~CTL_FLAG_SENT_2OTHER_SC;
11026				next_io = (union ctl_io *)TAILQ_NEXT(
11027					&pending_io->io_hdr, ooa_links);
11028				if (pending_io->io_hdr.flags &
11029				    CTL_FLAG_IS_WAS_ON_RTR) {
11030#if 0
11031				        printf("Tag 0x%04x is running\n",
11032				      		pending_io->scsiio.tag_num);
11033#endif
11034					continue;
11035				}
11036
11037				switch (ctl_check_ooa(lun, pending_io,
11038			            (union ctl_io *)TAILQ_PREV(
11039				    &pending_io->io_hdr, ctl_ooaq,
11040				    ooa_links))) {
11041
11042				case CTL_ACTION_BLOCK:
11043					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11044							  &pending_io->io_hdr,
11045							  blocked_links);
11046					pending_io->io_hdr.flags |=
11047					    CTL_FLAG_BLOCKED;
11048					break;
11049				case CTL_ACTION_PASS:
11050				case CTL_ACTION_SKIP:
11051					pending_io->io_hdr.flags |=
11052					    CTL_FLAG_IS_WAS_ON_RTR;
11053					ctl_enqueue_rtr(pending_io);
11054					break;
11055				case CTL_ACTION_OVERLAP:
11056					ctl_set_overlapped_cmd(
11057					    (struct ctl_scsiio *)pending_io);
11058					ctl_done(pending_io);
11059					break;
11060				case CTL_ACTION_OVERLAP_TAG:
11061					ctl_set_overlapped_tag(
11062					    (struct ctl_scsiio *)pending_io,
11063					    pending_io->scsiio.tag_num & 0xff);
11064					ctl_done(pending_io);
11065					break;
11066				case CTL_ACTION_ERROR:
11067				default:
11068					ctl_set_internal_failure(
11069						(struct ctl_scsiio *)pending_io,
11070						0,  // sks_valid
11071						0); //retry count
11072					ctl_done(pending_io);
11073					break;
11074				}
11075			}
11076
11077			/*
11078			 * Build Unit Attention
11079			 */
11080			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11081				lun->pending_sense[i].ua_pending |=
11082				                     CTL_UA_ASYM_ACC_CHANGE;
11083			}
11084		} else {
11085			panic("Unhandled HA mode failover, LUN flags = %#x, "
11086			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11087		}
11088	}
11089	ctl_pause_rtr = 0;
11090	mtx_unlock(&ctl_softc->ctl_lock);
11091}
11092
11093static int
11094ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11095{
11096	struct ctl_lun *lun;
11097	const struct ctl_cmd_entry *entry;
11098	uint32_t initidx, targ_lun;
11099	int retval;
11100
11101	retval = 0;
11102
11103	lun = NULL;
11104
11105	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11106	if ((targ_lun < CTL_MAX_LUNS)
11107	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11108		lun = ctl_softc->ctl_luns[targ_lun];
11109		/*
11110		 * If the LUN is invalid, pretend that it doesn't exist.
11111		 * It will go away as soon as all pending I/O has been
11112		 * completed.
11113		 */
11114		if (lun->flags & CTL_LUN_DISABLED) {
11115			lun = NULL;
11116		} else {
11117			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11118			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11119				lun->be_lun;
11120			if (lun->be_lun->lun_type == T_PROCESSOR) {
11121				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11122			}
11123
11124			/*
11125			 * Every I/O goes into the OOA queue for a
11126			 * particular LUN, and stays there until completion.
11127			 */
11128			mtx_lock(&lun->lun_lock);
11129			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11130			    ooa_links);
11131		}
11132	} else {
11133		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11134		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11135	}
11136
11137	/* Get command entry and return error if it is unsuppotyed. */
11138	entry = ctl_validate_command(ctsio);
11139	if (entry == NULL) {
11140		if (lun)
11141			mtx_unlock(&lun->lun_lock);
11142		return (retval);
11143	}
11144
11145	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11146	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11147
11148	/*
11149	 * Check to see whether we can send this command to LUNs that don't
11150	 * exist.  This should pretty much only be the case for inquiry
11151	 * and request sense.  Further checks, below, really require having
11152	 * a LUN, so we can't really check the command anymore.  Just put
11153	 * it on the rtr queue.
11154	 */
11155	if (lun == NULL) {
11156		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11157			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11158			ctl_enqueue_rtr((union ctl_io *)ctsio);
11159			return (retval);
11160		}
11161
11162		ctl_set_unsupported_lun(ctsio);
11163		ctl_done((union ctl_io *)ctsio);
11164		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11165		return (retval);
11166	} else {
11167		/*
11168		 * Make sure we support this particular command on this LUN.
11169		 * e.g., we don't support writes to the control LUN.
11170		 */
11171		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11172			mtx_unlock(&lun->lun_lock);
11173			ctl_set_invalid_opcode(ctsio);
11174			ctl_done((union ctl_io *)ctsio);
11175			return (retval);
11176		}
11177	}
11178
11179	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11180
11181	/*
11182	 * If we've got a request sense, it'll clear the contingent
11183	 * allegiance condition.  Otherwise, if we have a CA condition for
11184	 * this initiator, clear it, because it sent down a command other
11185	 * than request sense.
11186	 */
11187	if ((ctsio->cdb[0] != REQUEST_SENSE)
11188	 && (ctl_is_set(lun->have_ca, initidx)))
11189		ctl_clear_mask(lun->have_ca, initidx);
11190
11191	/*
11192	 * If the command has this flag set, it handles its own unit
11193	 * attention reporting, we shouldn't do anything.  Otherwise we
11194	 * check for any pending unit attentions, and send them back to the
11195	 * initiator.  We only do this when a command initially comes in,
11196	 * not when we pull it off the blocked queue.
11197	 *
11198	 * According to SAM-3, section 5.3.2, the order that things get
11199	 * presented back to the host is basically unit attentions caused
11200	 * by some sort of reset event, busy status, reservation conflicts
11201	 * or task set full, and finally any other status.
11202	 *
11203	 * One issue here is that some of the unit attentions we report
11204	 * don't fall into the "reset" category (e.g. "reported luns data
11205	 * has changed").  So reporting it here, before the reservation
11206	 * check, may be technically wrong.  I guess the only thing to do
11207	 * would be to check for and report the reset events here, and then
11208	 * check for the other unit attention types after we check for a
11209	 * reservation conflict.
11210	 *
11211	 * XXX KDM need to fix this
11212	 */
11213	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11214		ctl_ua_type ua_type;
11215
11216		ua_type = lun->pending_sense[initidx].ua_pending;
11217		if (ua_type != CTL_UA_NONE) {
11218			scsi_sense_data_type sense_format;
11219
11220			if (lun != NULL)
11221				sense_format = (lun->flags &
11222				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11223				    SSD_TYPE_FIXED;
11224			else
11225				sense_format = SSD_TYPE_FIXED;
11226
11227			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11228					       sense_format);
11229			if (ua_type != CTL_UA_NONE) {
11230				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11231				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11232						       CTL_AUTOSENSE;
11233				ctsio->sense_len = SSD_FULL_SIZE;
11234				lun->pending_sense[initidx].ua_pending &=
11235					~ua_type;
11236				mtx_unlock(&lun->lun_lock);
11237				ctl_done((union ctl_io *)ctsio);
11238				return (retval);
11239			}
11240		}
11241	}
11242
11243
11244	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11245		mtx_unlock(&lun->lun_lock);
11246		ctl_done((union ctl_io *)ctsio);
11247		return (retval);
11248	}
11249
11250	/*
11251	 * XXX CHD this is where we want to send IO to other side if
11252	 * this LUN is secondary on this SC. We will need to make a copy
11253	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11254	 * the copy we send as FROM_OTHER.
11255	 * We also need to stuff the address of the original IO so we can
11256	 * find it easily. Something similar will need be done on the other
11257	 * side so when we are done we can find the copy.
11258	 */
11259	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11260		union ctl_ha_msg msg_info;
11261		int isc_retval;
11262
11263		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11264
11265		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11266		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11267#if 0
11268		printf("1. ctsio %p\n", ctsio);
11269#endif
11270		msg_info.hdr.serializing_sc = NULL;
11271		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11272		msg_info.scsi.tag_num = ctsio->tag_num;
11273		msg_info.scsi.tag_type = ctsio->tag_type;
11274		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11275
11276		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11277
11278		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11279		    (void *)&msg_info, sizeof(msg_info), 0)) >
11280		    CTL_HA_STATUS_SUCCESS) {
11281			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11282			       isc_retval);
11283			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11284		} else {
11285#if 0
11286			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11287#endif
11288		}
11289
11290		/*
11291		 * XXX KDM this I/O is off the incoming queue, but hasn't
11292		 * been inserted on any other queue.  We may need to come
11293		 * up with a holding queue while we wait for serialization
11294		 * so that we have an idea of what we're waiting for from
11295		 * the other side.
11296		 */
11297		mtx_unlock(&lun->lun_lock);
11298		return (retval);
11299	}
11300
11301	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11302			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11303			      ctl_ooaq, ooa_links))) {
11304	case CTL_ACTION_BLOCK:
11305		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11306		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11307				  blocked_links);
11308		mtx_unlock(&lun->lun_lock);
11309		return (retval);
11310	case CTL_ACTION_PASS:
11311	case CTL_ACTION_SKIP:
11312		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11313		mtx_unlock(&lun->lun_lock);
11314		ctl_enqueue_rtr((union ctl_io *)ctsio);
11315		break;
11316	case CTL_ACTION_OVERLAP:
11317		mtx_unlock(&lun->lun_lock);
11318		ctl_set_overlapped_cmd(ctsio);
11319		ctl_done((union ctl_io *)ctsio);
11320		break;
11321	case CTL_ACTION_OVERLAP_TAG:
11322		mtx_unlock(&lun->lun_lock);
11323		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11324		ctl_done((union ctl_io *)ctsio);
11325		break;
11326	case CTL_ACTION_ERROR:
11327	default:
11328		mtx_unlock(&lun->lun_lock);
11329		ctl_set_internal_failure(ctsio,
11330					 /*sks_valid*/ 0,
11331					 /*retry_count*/ 0);
11332		ctl_done((union ctl_io *)ctsio);
11333		break;
11334	}
11335	return (retval);
11336}
11337
11338const struct ctl_cmd_entry *
11339ctl_get_cmd_entry(struct ctl_scsiio *ctsio)
11340{
11341	const struct ctl_cmd_entry *entry;
11342	int service_action;
11343
11344	entry = &ctl_cmd_table[ctsio->cdb[0]];
11345	if (entry->flags & CTL_CMD_FLAG_SA5) {
11346		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11347		entry = &((const struct ctl_cmd_entry *)
11348		    entry->execute)[service_action];
11349	}
11350	return (entry);
11351}
11352
11353const struct ctl_cmd_entry *
11354ctl_validate_command(struct ctl_scsiio *ctsio)
11355{
11356	const struct ctl_cmd_entry *entry;
11357	int i;
11358	uint8_t diff;
11359
11360	entry = ctl_get_cmd_entry(ctsio);
11361	if (entry->execute == NULL) {
11362		ctl_set_invalid_opcode(ctsio);
11363		ctl_done((union ctl_io *)ctsio);
11364		return (NULL);
11365	}
11366	KASSERT(entry->length > 0,
11367	    ("Not defined length for command 0x%02x/0x%02x",
11368	     ctsio->cdb[0], ctsio->cdb[1]));
11369	for (i = 1; i < entry->length; i++) {
11370		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11371		if (diff == 0)
11372			continue;
11373		ctl_set_invalid_field(ctsio,
11374				      /*sks_valid*/ 1,
11375				      /*command*/ 1,
11376				      /*field*/ i,
11377				      /*bit_valid*/ 1,
11378				      /*bit*/ fls(diff) - 1);
11379		ctl_done((union ctl_io *)ctsio);
11380		return (NULL);
11381	}
11382	return (entry);
11383}
11384
11385static int
11386ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11387{
11388
11389	switch (lun_type) {
11390	case T_PROCESSOR:
11391		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11392		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11393			return (0);
11394		break;
11395	case T_DIRECT:
11396		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11397		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11398			return (0);
11399		break;
11400	default:
11401		return (0);
11402	}
11403	return (1);
11404}
11405
11406static int
11407ctl_scsiio(struct ctl_scsiio *ctsio)
11408{
11409	int retval;
11410	const struct ctl_cmd_entry *entry;
11411
11412	retval = CTL_RETVAL_COMPLETE;
11413
11414	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11415
11416	entry = ctl_get_cmd_entry(ctsio);
11417
11418	/*
11419	 * If this I/O has been aborted, just send it straight to
11420	 * ctl_done() without executing it.
11421	 */
11422	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11423		ctl_done((union ctl_io *)ctsio);
11424		goto bailout;
11425	}
11426
11427	/*
11428	 * All the checks should have been handled by ctl_scsiio_precheck().
11429	 * We should be clear now to just execute the I/O.
11430	 */
11431	retval = entry->execute(ctsio);
11432
11433bailout:
11434	return (retval);
11435}
11436
11437/*
11438 * Since we only implement one target right now, a bus reset simply resets
11439 * our single target.
11440 */
11441static int
11442ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11443{
11444	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11445}
11446
11447static int
11448ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11449		 ctl_ua_type ua_type)
11450{
11451	struct ctl_lun *lun;
11452	int retval;
11453
11454	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11455		union ctl_ha_msg msg_info;
11456
11457		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11458		msg_info.hdr.nexus = io->io_hdr.nexus;
11459		if (ua_type==CTL_UA_TARG_RESET)
11460			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11461		else
11462			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11463		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11464		msg_info.hdr.original_sc = NULL;
11465		msg_info.hdr.serializing_sc = NULL;
11466		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11467		    (void *)&msg_info, sizeof(msg_info), 0)) {
11468		}
11469	}
11470	retval = 0;
11471
11472	mtx_lock(&ctl_softc->ctl_lock);
11473	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11474		retval += ctl_lun_reset(lun, io, ua_type);
11475	mtx_unlock(&ctl_softc->ctl_lock);
11476
11477	return (retval);
11478}
11479
11480/*
11481 * The LUN should always be set.  The I/O is optional, and is used to
11482 * distinguish between I/Os sent by this initiator, and by other
11483 * initiators.  We set unit attention for initiators other than this one.
11484 * SAM-3 is vague on this point.  It does say that a unit attention should
11485 * be established for other initiators when a LUN is reset (see section
11486 * 5.7.3), but it doesn't specifically say that the unit attention should
11487 * be established for this particular initiator when a LUN is reset.  Here
11488 * is the relevant text, from SAM-3 rev 8:
11489 *
11490 * 5.7.2 When a SCSI initiator port aborts its own tasks
11491 *
11492 * When a SCSI initiator port causes its own task(s) to be aborted, no
11493 * notification that the task(s) have been aborted shall be returned to
11494 * the SCSI initiator port other than the completion response for the
11495 * command or task management function action that caused the task(s) to
11496 * be aborted and notification(s) associated with related effects of the
11497 * action (e.g., a reset unit attention condition).
11498 *
11499 * XXX KDM for now, we're setting unit attention for all initiators.
11500 */
11501static int
11502ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11503{
11504	union ctl_io *xio;
11505#if 0
11506	uint32_t initindex;
11507#endif
11508	int i;
11509
11510	mtx_lock(&lun->lun_lock);
11511	/*
11512	 * Run through the OOA queue and abort each I/O.
11513	 */
11514#if 0
11515	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11516#endif
11517	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11518	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11519		xio->io_hdr.flags |= CTL_FLAG_ABORT;
11520	}
11521
11522	/*
11523	 * This version sets unit attention for every
11524	 */
11525#if 0
11526	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11527	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11528		if (initindex == i)
11529			continue;
11530		lun->pending_sense[i].ua_pending |= ua_type;
11531	}
11532#endif
11533
11534	/*
11535	 * A reset (any kind, really) clears reservations established with
11536	 * RESERVE/RELEASE.  It does not clear reservations established
11537	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11538	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11539	 * reservations made with the RESERVE/RELEASE commands, because
11540	 * those commands are obsolete in SPC-3.
11541	 */
11542	lun->flags &= ~CTL_LUN_RESERVED;
11543
11544	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11545		ctl_clear_mask(lun->have_ca, i);
11546		lun->pending_sense[i].ua_pending |= ua_type;
11547	}
11548	mtx_unlock(&lun->lun_lock);
11549
11550	return (0);
11551}
11552
11553static int
11554ctl_abort_task(union ctl_io *io)
11555{
11556	union ctl_io *xio;
11557	struct ctl_lun *lun;
11558	struct ctl_softc *ctl_softc;
11559#if 0
11560	struct sbuf sb;
11561	char printbuf[128];
11562#endif
11563	int found;
11564	uint32_t targ_lun;
11565
11566	ctl_softc = control_softc;
11567	found = 0;
11568
11569	/*
11570	 * Look up the LUN.
11571	 */
11572	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11573	mtx_lock(&ctl_softc->ctl_lock);
11574	if ((targ_lun < CTL_MAX_LUNS)
11575	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11576		lun = ctl_softc->ctl_luns[targ_lun];
11577	else {
11578		mtx_unlock(&ctl_softc->ctl_lock);
11579		goto bailout;
11580	}
11581
11582#if 0
11583	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11584	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11585#endif
11586
11587	mtx_lock(&lun->lun_lock);
11588	mtx_unlock(&ctl_softc->ctl_lock);
11589	/*
11590	 * Run through the OOA queue and attempt to find the given I/O.
11591	 * The target port, initiator ID, tag type and tag number have to
11592	 * match the values that we got from the initiator.  If we have an
11593	 * untagged command to abort, simply abort the first untagged command
11594	 * we come to.  We only allow one untagged command at a time of course.
11595	 */
11596#if 0
11597	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11598#endif
11599	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11600	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11601#if 0
11602		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11603
11604		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11605			    lun->lun, xio->scsiio.tag_num,
11606			    xio->scsiio.tag_type,
11607			    (xio->io_hdr.blocked_links.tqe_prev
11608			    == NULL) ? "" : " BLOCKED",
11609			    (xio->io_hdr.flags &
11610			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11611			    (xio->io_hdr.flags &
11612			    CTL_FLAG_ABORT) ? " ABORT" : "",
11613			    (xio->io_hdr.flags &
11614			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11615		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11616		sbuf_finish(&sb);
11617		printf("%s\n", sbuf_data(&sb));
11618#endif
11619
11620		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
11621		 && (xio->io_hdr.nexus.initid.id ==
11622		     io->io_hdr.nexus.initid.id)) {
11623			/*
11624			 * If the abort says that the task is untagged, the
11625			 * task in the queue must be untagged.  Otherwise,
11626			 * we just check to see whether the tag numbers
11627			 * match.  This is because the QLogic firmware
11628			 * doesn't pass back the tag type in an abort
11629			 * request.
11630			 */
11631#if 0
11632			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11633			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11634			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
11635#endif
11636			/*
11637			 * XXX KDM we've got problems with FC, because it
11638			 * doesn't send down a tag type with aborts.  So we
11639			 * can only really go by the tag number...
11640			 * This may cause problems with parallel SCSI.
11641			 * Need to figure that out!!
11642			 */
11643			if (xio->scsiio.tag_num == io->taskio.tag_num) {
11644				xio->io_hdr.flags |= CTL_FLAG_ABORT;
11645				found = 1;
11646				if ((io->io_hdr.flags &
11647				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
11648				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11649					union ctl_ha_msg msg_info;
11650
11651					io->io_hdr.flags |=
11652					                CTL_FLAG_SENT_2OTHER_SC;
11653					msg_info.hdr.nexus = io->io_hdr.nexus;
11654					msg_info.task.task_action =
11655						CTL_TASK_ABORT_TASK;
11656					msg_info.task.tag_num =
11657						io->taskio.tag_num;
11658					msg_info.task.tag_type =
11659						io->taskio.tag_type;
11660					msg_info.hdr.msg_type =
11661						CTL_MSG_MANAGE_TASKS;
11662					msg_info.hdr.original_sc = NULL;
11663					msg_info.hdr.serializing_sc = NULL;
11664#if 0
11665					printf("Sent Abort to other side\n");
11666#endif
11667					if (CTL_HA_STATUS_SUCCESS !=
11668					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11669		    				(void *)&msg_info,
11670						sizeof(msg_info), 0)) {
11671					}
11672				}
11673#if 0
11674				printf("ctl_abort_task: found I/O to abort\n");
11675#endif
11676				break;
11677			}
11678		}
11679	}
11680	mtx_unlock(&lun->lun_lock);
11681
11682bailout:
11683
11684	if (found == 0) {
11685		/*
11686		 * This isn't really an error.  It's entirely possible for
11687		 * the abort and command completion to cross on the wire.
11688		 * This is more of an informative/diagnostic error.
11689		 */
11690#if 0
11691		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11692		       "%d:%d:%d:%d tag %d type %d\n",
11693		       io->io_hdr.nexus.initid.id,
11694		       io->io_hdr.nexus.targ_port,
11695		       io->io_hdr.nexus.targ_target.id,
11696		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11697		       io->taskio.tag_type);
11698#endif
11699		return (1);
11700	} else
11701		return (0);
11702}
11703
11704static void
11705ctl_run_task(union ctl_io *io)
11706{
11707	struct ctl_softc *ctl_softc;
11708	int retval;
11709	const char *task_desc;
11710
11711	CTL_DEBUG_PRINT(("ctl_run_task\n"));
11712
11713	ctl_softc = control_softc;
11714	retval = 0;
11715
11716	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
11717	    ("ctl_run_task: Unextected io_type %d\n",
11718	     io->io_hdr.io_type));
11719
11720	task_desc = ctl_scsi_task_string(&io->taskio);
11721	if (task_desc != NULL) {
11722#ifdef NEEDTOPORT
11723		csevent_log(CSC_CTL | CSC_SHELF_SW |
11724			    CTL_TASK_REPORT,
11725			    csevent_LogType_Trace,
11726			    csevent_Severity_Information,
11727			    csevent_AlertLevel_Green,
11728			    csevent_FRU_Firmware,
11729			    csevent_FRU_Unknown,
11730			    "CTL: received task: %s",task_desc);
11731#endif
11732	} else {
11733#ifdef NEEDTOPORT
11734		csevent_log(CSC_CTL | CSC_SHELF_SW |
11735			    CTL_TASK_REPORT,
11736			    csevent_LogType_Trace,
11737			    csevent_Severity_Information,
11738			    csevent_AlertLevel_Green,
11739			    csevent_FRU_Firmware,
11740			    csevent_FRU_Unknown,
11741			    "CTL: received unknown task "
11742			    "type: %d (%#x)",
11743			    io->taskio.task_action,
11744			    io->taskio.task_action);
11745#endif
11746	}
11747	switch (io->taskio.task_action) {
11748	case CTL_TASK_ABORT_TASK:
11749		retval = ctl_abort_task(io);
11750		break;
11751	case CTL_TASK_ABORT_TASK_SET:
11752		break;
11753	case CTL_TASK_CLEAR_ACA:
11754		break;
11755	case CTL_TASK_CLEAR_TASK_SET:
11756		break;
11757	case CTL_TASK_LUN_RESET: {
11758		struct ctl_lun *lun;
11759		uint32_t targ_lun;
11760		int retval;
11761
11762		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11763		mtx_lock(&ctl_softc->ctl_lock);
11764		if ((targ_lun < CTL_MAX_LUNS)
11765		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11766			lun = ctl_softc->ctl_luns[targ_lun];
11767		else {
11768			mtx_unlock(&ctl_softc->ctl_lock);
11769			retval = 1;
11770			break;
11771		}
11772
11773		if (!(io->io_hdr.flags &
11774		    CTL_FLAG_FROM_OTHER_SC)) {
11775			union ctl_ha_msg msg_info;
11776
11777			io->io_hdr.flags |=
11778				CTL_FLAG_SENT_2OTHER_SC;
11779			msg_info.hdr.msg_type =
11780				CTL_MSG_MANAGE_TASKS;
11781			msg_info.hdr.nexus = io->io_hdr.nexus;
11782			msg_info.task.task_action =
11783				CTL_TASK_LUN_RESET;
11784			msg_info.hdr.original_sc = NULL;
11785			msg_info.hdr.serializing_sc = NULL;
11786			if (CTL_HA_STATUS_SUCCESS !=
11787			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11788			    (void *)&msg_info,
11789			    sizeof(msg_info), 0)) {
11790			}
11791		}
11792
11793		retval = ctl_lun_reset(lun, io,
11794				       CTL_UA_LUN_RESET);
11795		mtx_unlock(&ctl_softc->ctl_lock);
11796		break;
11797	}
11798	case CTL_TASK_TARGET_RESET:
11799		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
11800		break;
11801	case CTL_TASK_BUS_RESET:
11802		retval = ctl_bus_reset(ctl_softc, io);
11803		break;
11804	case CTL_TASK_PORT_LOGIN:
11805		break;
11806	case CTL_TASK_PORT_LOGOUT:
11807		break;
11808	default:
11809		printf("ctl_run_task: got unknown task management event %d\n",
11810		       io->taskio.task_action);
11811		break;
11812	}
11813	if (retval == 0)
11814		io->io_hdr.status = CTL_SUCCESS;
11815	else
11816		io->io_hdr.status = CTL_ERROR;
11817
11818	/*
11819	 * This will queue this I/O to the done queue, but the
11820	 * work thread won't be able to process it until we
11821	 * return and the lock is released.
11822	 */
11823	ctl_done(io);
11824}
11825
11826/*
11827 * For HA operation.  Handle commands that come in from the other
11828 * controller.
11829 */
11830static void
11831ctl_handle_isc(union ctl_io *io)
11832{
11833	int free_io;
11834	struct ctl_lun *lun;
11835	struct ctl_softc *ctl_softc;
11836	uint32_t targ_lun;
11837
11838	ctl_softc = control_softc;
11839
11840	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11841	lun = ctl_softc->ctl_luns[targ_lun];
11842
11843	switch (io->io_hdr.msg_type) {
11844	case CTL_MSG_SERIALIZE:
11845		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
11846		break;
11847	case CTL_MSG_R2R: {
11848		const struct ctl_cmd_entry *entry;
11849
11850		/*
11851		 * This is only used in SER_ONLY mode.
11852		 */
11853		free_io = 0;
11854		entry = ctl_get_cmd_entry(&io->scsiio);
11855		mtx_lock(&lun->lun_lock);
11856		if (ctl_scsiio_lun_check(ctl_softc, lun,
11857		    entry, (struct ctl_scsiio *)io) != 0) {
11858			mtx_unlock(&lun->lun_lock);
11859			ctl_done(io);
11860			break;
11861		}
11862		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11863		mtx_unlock(&lun->lun_lock);
11864		ctl_enqueue_rtr(io);
11865		break;
11866	}
11867	case CTL_MSG_FINISH_IO:
11868		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11869			free_io = 0;
11870			ctl_done(io);
11871		} else {
11872			free_io = 1;
11873			mtx_lock(&lun->lun_lock);
11874			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11875				     ooa_links);
11876			ctl_check_blocked(lun);
11877			mtx_unlock(&lun->lun_lock);
11878		}
11879		break;
11880	case CTL_MSG_PERS_ACTION:
11881		ctl_hndl_per_res_out_on_other_sc(
11882			(union ctl_ha_msg *)&io->presio.pr_msg);
11883		free_io = 1;
11884		break;
11885	case CTL_MSG_BAD_JUJU:
11886		free_io = 0;
11887		ctl_done(io);
11888		break;
11889	case CTL_MSG_DATAMOVE:
11890		/* Only used in XFER mode */
11891		free_io = 0;
11892		ctl_datamove_remote(io);
11893		break;
11894	case CTL_MSG_DATAMOVE_DONE:
11895		/* Only used in XFER mode */
11896		free_io = 0;
11897		io->scsiio.be_move_done(io);
11898		break;
11899	default:
11900		free_io = 1;
11901		printf("%s: Invalid message type %d\n",
11902		       __func__, io->io_hdr.msg_type);
11903		break;
11904	}
11905	if (free_io)
11906		ctl_free_io(io);
11907
11908}
11909
11910
11911/*
11912 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11913 * there is no match.
11914 */
11915static ctl_lun_error_pattern
11916ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11917{
11918	const struct ctl_cmd_entry *entry;
11919	ctl_lun_error_pattern filtered_pattern, pattern;
11920
11921	pattern = desc->error_pattern;
11922
11923	/*
11924	 * XXX KDM we need more data passed into this function to match a
11925	 * custom pattern, and we actually need to implement custom pattern
11926	 * matching.
11927	 */
11928	if (pattern & CTL_LUN_PAT_CMD)
11929		return (CTL_LUN_PAT_CMD);
11930
11931	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11932		return (CTL_LUN_PAT_ANY);
11933
11934	entry = ctl_get_cmd_entry(ctsio);
11935
11936	filtered_pattern = entry->pattern & pattern;
11937
11938	/*
11939	 * If the user requested specific flags in the pattern (e.g.
11940	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11941	 * flags.
11942	 *
11943	 * If the user did not specify any flags, it doesn't matter whether
11944	 * or not the command supports the flags.
11945	 */
11946	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11947	     (pattern & ~CTL_LUN_PAT_MASK))
11948		return (CTL_LUN_PAT_NONE);
11949
11950	/*
11951	 * If the user asked for a range check, see if the requested LBA
11952	 * range overlaps with this command's LBA range.
11953	 */
11954	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11955		uint64_t lba1;
11956		uint32_t len1;
11957		ctl_action action;
11958		int retval;
11959
11960		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11961		if (retval != 0)
11962			return (CTL_LUN_PAT_NONE);
11963
11964		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11965					      desc->lba_range.len);
11966		/*
11967		 * A "pass" means that the LBA ranges don't overlap, so
11968		 * this doesn't match the user's range criteria.
11969		 */
11970		if (action == CTL_ACTION_PASS)
11971			return (CTL_LUN_PAT_NONE);
11972	}
11973
11974	return (filtered_pattern);
11975}
11976
11977static void
11978ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11979{
11980	struct ctl_error_desc *desc, *desc2;
11981
11982	mtx_assert(&lun->lun_lock, MA_OWNED);
11983
11984	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11985		ctl_lun_error_pattern pattern;
11986		/*
11987		 * Check to see whether this particular command matches
11988		 * the pattern in the descriptor.
11989		 */
11990		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11991		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11992			continue;
11993
11994		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11995		case CTL_LUN_INJ_ABORTED:
11996			ctl_set_aborted(&io->scsiio);
11997			break;
11998		case CTL_LUN_INJ_MEDIUM_ERR:
11999			ctl_set_medium_error(&io->scsiio);
12000			break;
12001		case CTL_LUN_INJ_UA:
12002			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12003			 * OCCURRED */
12004			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12005			break;
12006		case CTL_LUN_INJ_CUSTOM:
12007			/*
12008			 * We're assuming the user knows what he is doing.
12009			 * Just copy the sense information without doing
12010			 * checks.
12011			 */
12012			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12013			      ctl_min(sizeof(desc->custom_sense),
12014				      sizeof(io->scsiio.sense_data)));
12015			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12016			io->scsiio.sense_len = SSD_FULL_SIZE;
12017			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12018			break;
12019		case CTL_LUN_INJ_NONE:
12020		default:
12021			/*
12022			 * If this is an error injection type we don't know
12023			 * about, clear the continuous flag (if it is set)
12024			 * so it will get deleted below.
12025			 */
12026			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12027			break;
12028		}
12029		/*
12030		 * By default, each error injection action is a one-shot
12031		 */
12032		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12033			continue;
12034
12035		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12036
12037		free(desc, M_CTL);
12038	}
12039}
12040
12041#ifdef CTL_IO_DELAY
12042static void
12043ctl_datamove_timer_wakeup(void *arg)
12044{
12045	union ctl_io *io;
12046
12047	io = (union ctl_io *)arg;
12048
12049	ctl_datamove(io);
12050}
12051#endif /* CTL_IO_DELAY */
12052
12053void
12054ctl_datamove(union ctl_io *io)
12055{
12056	void (*fe_datamove)(union ctl_io *io);
12057
12058	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12059
12060	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12061
12062#ifdef CTL_TIME_IO
12063	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12064		char str[256];
12065		char path_str[64];
12066		struct sbuf sb;
12067
12068		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12069		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12070
12071		sbuf_cat(&sb, path_str);
12072		switch (io->io_hdr.io_type) {
12073		case CTL_IO_SCSI:
12074			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12075			sbuf_printf(&sb, "\n");
12076			sbuf_cat(&sb, path_str);
12077			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12078				    io->scsiio.tag_num, io->scsiio.tag_type);
12079			break;
12080		case CTL_IO_TASK:
12081			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12082				    "Tag Type: %d\n", io->taskio.task_action,
12083				    io->taskio.tag_num, io->taskio.tag_type);
12084			break;
12085		default:
12086			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12087			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12088			break;
12089		}
12090		sbuf_cat(&sb, path_str);
12091		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12092			    (intmax_t)time_uptime - io->io_hdr.start_time);
12093		sbuf_finish(&sb);
12094		printf("%s", sbuf_data(&sb));
12095	}
12096#endif /* CTL_TIME_IO */
12097
12098#ifdef CTL_IO_DELAY
12099	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12100		struct ctl_lun *lun;
12101
12102		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12103
12104		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12105	} else {
12106		struct ctl_lun *lun;
12107
12108		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12109		if ((lun != NULL)
12110		 && (lun->delay_info.datamove_delay > 0)) {
12111			struct callout *callout;
12112
12113			callout = (struct callout *)&io->io_hdr.timer_bytes;
12114			callout_init(callout, /*mpsafe*/ 1);
12115			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12116			callout_reset(callout,
12117				      lun->delay_info.datamove_delay * hz,
12118				      ctl_datamove_timer_wakeup, io);
12119			if (lun->delay_info.datamove_type ==
12120			    CTL_DELAY_TYPE_ONESHOT)
12121				lun->delay_info.datamove_delay = 0;
12122			return;
12123		}
12124	}
12125#endif
12126
12127	/*
12128	 * This command has been aborted.  Set the port status, so we fail
12129	 * the data move.
12130	 */
12131	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12132		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12133		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12134		       io->io_hdr.nexus.targ_port,
12135		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12136		       io->io_hdr.nexus.targ_lun);
12137		io->io_hdr.status = CTL_CMD_ABORTED;
12138		io->io_hdr.port_status = 31337;
12139		/*
12140		 * Note that the backend, in this case, will get the
12141		 * callback in its context.  In other cases it may get
12142		 * called in the frontend's interrupt thread context.
12143		 */
12144		io->scsiio.be_move_done(io);
12145		return;
12146	}
12147
12148	/*
12149	 * If we're in XFER mode and this I/O is from the other shelf
12150	 * controller, we need to send the DMA to the other side to
12151	 * actually transfer the data to/from the host.  In serialize only
12152	 * mode the transfer happens below CTL and ctl_datamove() is only
12153	 * called on the machine that originally received the I/O.
12154	 */
12155	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12156	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12157		union ctl_ha_msg msg;
12158		uint32_t sg_entries_sent;
12159		int do_sg_copy;
12160		int i;
12161
12162		memset(&msg, 0, sizeof(msg));
12163		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12164		msg.hdr.original_sc = io->io_hdr.original_sc;
12165		msg.hdr.serializing_sc = io;
12166		msg.hdr.nexus = io->io_hdr.nexus;
12167		msg.dt.flags = io->io_hdr.flags;
12168		/*
12169		 * We convert everything into a S/G list here.  We can't
12170		 * pass by reference, only by value between controllers.
12171		 * So we can't pass a pointer to the S/G list, only as many
12172		 * S/G entries as we can fit in here.  If it's possible for
12173		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12174		 * then we need to break this up into multiple transfers.
12175		 */
12176		if (io->scsiio.kern_sg_entries == 0) {
12177			msg.dt.kern_sg_entries = 1;
12178			/*
12179			 * If this is in cached memory, flush the cache
12180			 * before we send the DMA request to the other
12181			 * controller.  We want to do this in either the
12182			 * read or the write case.  The read case is
12183			 * straightforward.  In the write case, we want to
12184			 * make sure nothing is in the local cache that
12185			 * could overwrite the DMAed data.
12186			 */
12187			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12188				/*
12189				 * XXX KDM use bus_dmamap_sync() here.
12190				 */
12191			}
12192
12193			/*
12194			 * Convert to a physical address if this is a
12195			 * virtual address.
12196			 */
12197			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12198				msg.dt.sg_list[0].addr =
12199					io->scsiio.kern_data_ptr;
12200			} else {
12201				/*
12202				 * XXX KDM use busdma here!
12203				 */
12204#if 0
12205				msg.dt.sg_list[0].addr = (void *)
12206					vtophys(io->scsiio.kern_data_ptr);
12207#endif
12208			}
12209
12210			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12211			do_sg_copy = 0;
12212		} else {
12213			struct ctl_sg_entry *sgl;
12214
12215			do_sg_copy = 1;
12216			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12217			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12218			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12219				/*
12220				 * XXX KDM use bus_dmamap_sync() here.
12221				 */
12222			}
12223		}
12224
12225		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12226		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12227		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12228		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12229		msg.dt.sg_sequence = 0;
12230
12231		/*
12232		 * Loop until we've sent all of the S/G entries.  On the
12233		 * other end, we'll recompose these S/G entries into one
12234		 * contiguous list before passing it to the
12235		 */
12236		for (sg_entries_sent = 0; sg_entries_sent <
12237		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12238			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12239				sizeof(msg.dt.sg_list[0])),
12240				msg.dt.kern_sg_entries - sg_entries_sent);
12241
12242			if (do_sg_copy != 0) {
12243				struct ctl_sg_entry *sgl;
12244				int j;
12245
12246				sgl = (struct ctl_sg_entry *)
12247					io->scsiio.kern_data_ptr;
12248				/*
12249				 * If this is in cached memory, flush the cache
12250				 * before we send the DMA request to the other
12251				 * controller.  We want to do this in either
12252				 * the * read or the write case.  The read
12253				 * case is straightforward.  In the write
12254				 * case, we want to make sure nothing is
12255				 * in the local cache that could overwrite
12256				 * the DMAed data.
12257				 */
12258
12259				for (i = sg_entries_sent, j = 0;
12260				     i < msg.dt.cur_sg_entries; i++, j++) {
12261					if ((io->io_hdr.flags &
12262					     CTL_FLAG_NO_DATASYNC) == 0) {
12263						/*
12264						 * XXX KDM use bus_dmamap_sync()
12265						 */
12266					}
12267					if ((io->io_hdr.flags &
12268					     CTL_FLAG_BUS_ADDR) == 0) {
12269						/*
12270						 * XXX KDM use busdma.
12271						 */
12272#if 0
12273						msg.dt.sg_list[j].addr =(void *)
12274						       vtophys(sgl[i].addr);
12275#endif
12276					} else {
12277						msg.dt.sg_list[j].addr =
12278							sgl[i].addr;
12279					}
12280					msg.dt.sg_list[j].len = sgl[i].len;
12281				}
12282			}
12283
12284			sg_entries_sent += msg.dt.cur_sg_entries;
12285			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12286				msg.dt.sg_last = 1;
12287			else
12288				msg.dt.sg_last = 0;
12289
12290			/*
12291			 * XXX KDM drop and reacquire the lock here?
12292			 */
12293			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12294			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12295				/*
12296				 * XXX do something here.
12297				 */
12298			}
12299
12300			msg.dt.sent_sg_entries = sg_entries_sent;
12301		}
12302		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12303		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12304			ctl_failover_io(io, /*have_lock*/ 0);
12305
12306	} else {
12307
12308		/*
12309		 * Lookup the fe_datamove() function for this particular
12310		 * front end.
12311		 */
12312		fe_datamove =
12313		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12314
12315		fe_datamove(io);
12316	}
12317}
12318
12319static void
12320ctl_send_datamove_done(union ctl_io *io, int have_lock)
12321{
12322	union ctl_ha_msg msg;
12323	int isc_status;
12324
12325	memset(&msg, 0, sizeof(msg));
12326
12327	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12328	msg.hdr.original_sc = io;
12329	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12330	msg.hdr.nexus = io->io_hdr.nexus;
12331	msg.hdr.status = io->io_hdr.status;
12332	msg.scsi.tag_num = io->scsiio.tag_num;
12333	msg.scsi.tag_type = io->scsiio.tag_type;
12334	msg.scsi.scsi_status = io->scsiio.scsi_status;
12335	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12336	       sizeof(io->scsiio.sense_data));
12337	msg.scsi.sense_len = io->scsiio.sense_len;
12338	msg.scsi.sense_residual = io->scsiio.sense_residual;
12339	msg.scsi.fetd_status = io->io_hdr.port_status;
12340	msg.scsi.residual = io->scsiio.residual;
12341	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12342
12343	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12344		ctl_failover_io(io, /*have_lock*/ have_lock);
12345		return;
12346	}
12347
12348	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12349	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12350		/* XXX do something if this fails */
12351	}
12352
12353}
12354
12355/*
12356 * The DMA to the remote side is done, now we need to tell the other side
12357 * we're done so it can continue with its data movement.
12358 */
12359static void
12360ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12361{
12362	union ctl_io *io;
12363
12364	io = rq->context;
12365
12366	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12367		printf("%s: ISC DMA write failed with error %d", __func__,
12368		       rq->ret);
12369		ctl_set_internal_failure(&io->scsiio,
12370					 /*sks_valid*/ 1,
12371					 /*retry_count*/ rq->ret);
12372	}
12373
12374	ctl_dt_req_free(rq);
12375
12376	/*
12377	 * In this case, we had to malloc the memory locally.  Free it.
12378	 */
12379	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12380		int i;
12381		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12382			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12383	}
12384	/*
12385	 * The data is in local and remote memory, so now we need to send
12386	 * status (good or back) back to the other side.
12387	 */
12388	ctl_send_datamove_done(io, /*have_lock*/ 0);
12389}
12390
12391/*
12392 * We've moved the data from the host/controller into local memory.  Now we
12393 * need to push it over to the remote controller's memory.
12394 */
12395static int
12396ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12397{
12398	int retval;
12399
12400	retval = 0;
12401
12402	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12403					  ctl_datamove_remote_write_cb);
12404
12405	return (retval);
12406}
12407
12408static void
12409ctl_datamove_remote_write(union ctl_io *io)
12410{
12411	int retval;
12412	void (*fe_datamove)(union ctl_io *io);
12413
12414	/*
12415	 * - Get the data from the host/HBA into local memory.
12416	 * - DMA memory from the local controller to the remote controller.
12417	 * - Send status back to the remote controller.
12418	 */
12419
12420	retval = ctl_datamove_remote_sgl_setup(io);
12421	if (retval != 0)
12422		return;
12423
12424	/* Switch the pointer over so the FETD knows what to do */
12425	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12426
12427	/*
12428	 * Use a custom move done callback, since we need to send completion
12429	 * back to the other controller, not to the backend on this side.
12430	 */
12431	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12432
12433	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12434
12435	fe_datamove(io);
12436
12437	return;
12438
12439}
12440
12441static int
12442ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12443{
12444#if 0
12445	char str[256];
12446	char path_str[64];
12447	struct sbuf sb;
12448#endif
12449
12450	/*
12451	 * In this case, we had to malloc the memory locally.  Free it.
12452	 */
12453	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12454		int i;
12455		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12456			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12457	}
12458
12459#if 0
12460	scsi_path_string(io, path_str, sizeof(path_str));
12461	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12462	sbuf_cat(&sb, path_str);
12463	scsi_command_string(&io->scsiio, NULL, &sb);
12464	sbuf_printf(&sb, "\n");
12465	sbuf_cat(&sb, path_str);
12466	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12467		    io->scsiio.tag_num, io->scsiio.tag_type);
12468	sbuf_cat(&sb, path_str);
12469	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12470		    io->io_hdr.flags, io->io_hdr.status);
12471	sbuf_finish(&sb);
12472	printk("%s", sbuf_data(&sb));
12473#endif
12474
12475
12476	/*
12477	 * The read is done, now we need to send status (good or bad) back
12478	 * to the other side.
12479	 */
12480	ctl_send_datamove_done(io, /*have_lock*/ 0);
12481
12482	return (0);
12483}
12484
12485static void
12486ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12487{
12488	union ctl_io *io;
12489	void (*fe_datamove)(union ctl_io *io);
12490
12491	io = rq->context;
12492
12493	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12494		printf("%s: ISC DMA read failed with error %d", __func__,
12495		       rq->ret);
12496		ctl_set_internal_failure(&io->scsiio,
12497					 /*sks_valid*/ 1,
12498					 /*retry_count*/ rq->ret);
12499	}
12500
12501	ctl_dt_req_free(rq);
12502
12503	/* Switch the pointer over so the FETD knows what to do */
12504	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12505
12506	/*
12507	 * Use a custom move done callback, since we need to send completion
12508	 * back to the other controller, not to the backend on this side.
12509	 */
12510	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12511
12512	/* XXX KDM add checks like the ones in ctl_datamove? */
12513
12514	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12515
12516	fe_datamove(io);
12517}
12518
12519static int
12520ctl_datamove_remote_sgl_setup(union ctl_io *io)
12521{
12522	struct ctl_sg_entry *local_sglist, *remote_sglist;
12523	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12524	struct ctl_softc *softc;
12525	int retval;
12526	int i;
12527
12528	retval = 0;
12529	softc = control_softc;
12530
12531	local_sglist = io->io_hdr.local_sglist;
12532	local_dma_sglist = io->io_hdr.local_dma_sglist;
12533	remote_sglist = io->io_hdr.remote_sglist;
12534	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12535
12536	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12537		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12538			local_sglist[i].len = remote_sglist[i].len;
12539
12540			/*
12541			 * XXX Detect the situation where the RS-level I/O
12542			 * redirector on the other side has already read the
12543			 * data off of the AOR RS on this side, and
12544			 * transferred it to remote (mirror) memory on the
12545			 * other side.  Since we already have the data in
12546			 * memory here, we just need to use it.
12547			 *
12548			 * XXX KDM this can probably be removed once we
12549			 * get the cache device code in and take the
12550			 * current AOR implementation out.
12551			 */
12552#ifdef NEEDTOPORT
12553			if ((remote_sglist[i].addr >=
12554			     (void *)vtophys(softc->mirr->addr))
12555			 && (remote_sglist[i].addr <
12556			     ((void *)vtophys(softc->mirr->addr) +
12557			     CacheMirrorOffset))) {
12558				local_sglist[i].addr = remote_sglist[i].addr -
12559					CacheMirrorOffset;
12560				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12561				     CTL_FLAG_DATA_IN)
12562					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12563			} else {
12564				local_sglist[i].addr = remote_sglist[i].addr +
12565					CacheMirrorOffset;
12566			}
12567#endif
12568#if 0
12569			printf("%s: local %p, remote %p, len %d\n",
12570			       __func__, local_sglist[i].addr,
12571			       remote_sglist[i].addr, local_sglist[i].len);
12572#endif
12573		}
12574	} else {
12575		uint32_t len_to_go;
12576
12577		/*
12578		 * In this case, we don't have automatically allocated
12579		 * memory for this I/O on this controller.  This typically
12580		 * happens with internal CTL I/O -- e.g. inquiry, mode
12581		 * sense, etc.  Anything coming from RAIDCore will have
12582		 * a mirror area available.
12583		 */
12584		len_to_go = io->scsiio.kern_data_len;
12585
12586		/*
12587		 * Clear the no datasync flag, we have to use malloced
12588		 * buffers.
12589		 */
12590		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12591
12592		/*
12593		 * The difficult thing here is that the size of the various
12594		 * S/G segments may be different than the size from the
12595		 * remote controller.  That'll make it harder when DMAing
12596		 * the data back to the other side.
12597		 */
12598		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12599		     sizeof(io->io_hdr.remote_sglist[0])) &&
12600		     (len_to_go > 0); i++) {
12601			local_sglist[i].len = ctl_min(len_to_go, 131072);
12602			CTL_SIZE_8B(local_dma_sglist[i].len,
12603				    local_sglist[i].len);
12604			local_sglist[i].addr =
12605				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
12606
12607			local_dma_sglist[i].addr = local_sglist[i].addr;
12608
12609			if (local_sglist[i].addr == NULL) {
12610				int j;
12611
12612				printf("malloc failed for %zd bytes!",
12613				       local_dma_sglist[i].len);
12614				for (j = 0; j < i; j++) {
12615					free(local_sglist[j].addr, M_CTL);
12616				}
12617				ctl_set_internal_failure(&io->scsiio,
12618							 /*sks_valid*/ 1,
12619							 /*retry_count*/ 4857);
12620				retval = 1;
12621				goto bailout_error;
12622
12623			}
12624			/* XXX KDM do we need a sync here? */
12625
12626			len_to_go -= local_sglist[i].len;
12627		}
12628		/*
12629		 * Reset the number of S/G entries accordingly.  The
12630		 * original number of S/G entries is available in
12631		 * rem_sg_entries.
12632		 */
12633		io->scsiio.kern_sg_entries = i;
12634
12635#if 0
12636		printf("%s: kern_sg_entries = %d\n", __func__,
12637		       io->scsiio.kern_sg_entries);
12638		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12639			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
12640			       local_sglist[i].addr, local_sglist[i].len,
12641			       local_dma_sglist[i].len);
12642#endif
12643	}
12644
12645
12646	return (retval);
12647
12648bailout_error:
12649
12650	ctl_send_datamove_done(io, /*have_lock*/ 0);
12651
12652	return (retval);
12653}
12654
12655static int
12656ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12657			 ctl_ha_dt_cb callback)
12658{
12659	struct ctl_ha_dt_req *rq;
12660	struct ctl_sg_entry *remote_sglist, *local_sglist;
12661	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
12662	uint32_t local_used, remote_used, total_used;
12663	int retval;
12664	int i, j;
12665
12666	retval = 0;
12667
12668	rq = ctl_dt_req_alloc();
12669
12670	/*
12671	 * If we failed to allocate the request, and if the DMA didn't fail
12672	 * anyway, set busy status.  This is just a resource allocation
12673	 * failure.
12674	 */
12675	if ((rq == NULL)
12676	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12677		ctl_set_busy(&io->scsiio);
12678
12679	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12680
12681		if (rq != NULL)
12682			ctl_dt_req_free(rq);
12683
12684		/*
12685		 * The data move failed.  We need to return status back
12686		 * to the other controller.  No point in trying to DMA
12687		 * data to the remote controller.
12688		 */
12689
12690		ctl_send_datamove_done(io, /*have_lock*/ 0);
12691
12692		retval = 1;
12693
12694		goto bailout;
12695	}
12696
12697	local_sglist = io->io_hdr.local_sglist;
12698	local_dma_sglist = io->io_hdr.local_dma_sglist;
12699	remote_sglist = io->io_hdr.remote_sglist;
12700	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12701	local_used = 0;
12702	remote_used = 0;
12703	total_used = 0;
12704
12705	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12706		rq->ret = CTL_HA_STATUS_SUCCESS;
12707		rq->context = io;
12708		callback(rq);
12709		goto bailout;
12710	}
12711
12712	/*
12713	 * Pull/push the data over the wire from/to the other controller.
12714	 * This takes into account the possibility that the local and
12715	 * remote sglists may not be identical in terms of the size of
12716	 * the elements and the number of elements.
12717	 *
12718	 * One fundamental assumption here is that the length allocated for
12719	 * both the local and remote sglists is identical.  Otherwise, we've
12720	 * essentially got a coding error of some sort.
12721	 */
12722	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12723		int isc_ret;
12724		uint32_t cur_len, dma_length;
12725		uint8_t *tmp_ptr;
12726
12727		rq->id = CTL_HA_DATA_CTL;
12728		rq->command = command;
12729		rq->context = io;
12730
12731		/*
12732		 * Both pointers should be aligned.  But it is possible
12733		 * that the allocation length is not.  They should both
12734		 * also have enough slack left over at the end, though,
12735		 * to round up to the next 8 byte boundary.
12736		 */
12737		cur_len = ctl_min(local_sglist[i].len - local_used,
12738				  remote_sglist[j].len - remote_used);
12739
12740		/*
12741		 * In this case, we have a size issue and need to decrease
12742		 * the size, except in the case where we actually have less
12743		 * than 8 bytes left.  In that case, we need to increase
12744		 * the DMA length to get the last bit.
12745		 */
12746		if ((cur_len & 0x7) != 0) {
12747			if (cur_len > 0x7) {
12748				cur_len = cur_len - (cur_len & 0x7);
12749				dma_length = cur_len;
12750			} else {
12751				CTL_SIZE_8B(dma_length, cur_len);
12752			}
12753
12754		} else
12755			dma_length = cur_len;
12756
12757		/*
12758		 * If we had to allocate memory for this I/O, instead of using
12759		 * the non-cached mirror memory, we'll need to flush the cache
12760		 * before trying to DMA to the other controller.
12761		 *
12762		 * We could end up doing this multiple times for the same
12763		 * segment if we have a larger local segment than remote
12764		 * segment.  That shouldn't be an issue.
12765		 */
12766		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12767			/*
12768			 * XXX KDM use bus_dmamap_sync() here.
12769			 */
12770		}
12771
12772		rq->size = dma_length;
12773
12774		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12775		tmp_ptr += local_used;
12776
12777		/* Use physical addresses when talking to ISC hardware */
12778		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12779			/* XXX KDM use busdma */
12780#if 0
12781			rq->local = vtophys(tmp_ptr);
12782#endif
12783		} else
12784			rq->local = tmp_ptr;
12785
12786		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12787		tmp_ptr += remote_used;
12788		rq->remote = tmp_ptr;
12789
12790		rq->callback = NULL;
12791
12792		local_used += cur_len;
12793		if (local_used >= local_sglist[i].len) {
12794			i++;
12795			local_used = 0;
12796		}
12797
12798		remote_used += cur_len;
12799		if (remote_used >= remote_sglist[j].len) {
12800			j++;
12801			remote_used = 0;
12802		}
12803		total_used += cur_len;
12804
12805		if (total_used >= io->scsiio.kern_data_len)
12806			rq->callback = callback;
12807
12808		if ((rq->size & 0x7) != 0) {
12809			printf("%s: warning: size %d is not on 8b boundary\n",
12810			       __func__, rq->size);
12811		}
12812		if (((uintptr_t)rq->local & 0x7) != 0) {
12813			printf("%s: warning: local %p not on 8b boundary\n",
12814			       __func__, rq->local);
12815		}
12816		if (((uintptr_t)rq->remote & 0x7) != 0) {
12817			printf("%s: warning: remote %p not on 8b boundary\n",
12818			       __func__, rq->local);
12819		}
12820#if 0
12821		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12822		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12823		       rq->local, rq->remote, rq->size);
12824#endif
12825
12826		isc_ret = ctl_dt_single(rq);
12827		if (isc_ret == CTL_HA_STATUS_WAIT)
12828			continue;
12829
12830		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12831			rq->ret = CTL_HA_STATUS_SUCCESS;
12832		} else {
12833			rq->ret = isc_ret;
12834		}
12835		callback(rq);
12836		goto bailout;
12837	}
12838
12839bailout:
12840	return (retval);
12841
12842}
12843
12844static void
12845ctl_datamove_remote_read(union ctl_io *io)
12846{
12847	int retval;
12848	int i;
12849
12850	/*
12851	 * This will send an error to the other controller in the case of a
12852	 * failure.
12853	 */
12854	retval = ctl_datamove_remote_sgl_setup(io);
12855	if (retval != 0)
12856		return;
12857
12858	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12859					  ctl_datamove_remote_read_cb);
12860	if ((retval != 0)
12861	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12862		/*
12863		 * Make sure we free memory if there was an error..  The
12864		 * ctl_datamove_remote_xfer() function will send the
12865		 * datamove done message, or call the callback with an
12866		 * error if there is a problem.
12867		 */
12868		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12869			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12870	}
12871
12872	return;
12873}
12874
12875/*
12876 * Process a datamove request from the other controller.  This is used for
12877 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12878 * first.  Once that is complete, the data gets DMAed into the remote
12879 * controller's memory.  For reads, we DMA from the remote controller's
12880 * memory into our memory first, and then move it out to the FETD.
12881 */
12882static void
12883ctl_datamove_remote(union ctl_io *io)
12884{
12885	struct ctl_softc *softc;
12886
12887	softc = control_softc;
12888
12889	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
12890
12891	/*
12892	 * Note that we look for an aborted I/O here, but don't do some of
12893	 * the other checks that ctl_datamove() normally does.  We don't
12894	 * need to run the task queue, because this I/O is on the ISC
12895	 * queue, which is executed by the work thread after the task queue.
12896	 * We don't need to run the datamove delay code, since that should
12897	 * have been done if need be on the other controller.
12898	 */
12899	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12900
12901		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12902		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12903		       io->io_hdr.nexus.targ_port,
12904		       io->io_hdr.nexus.targ_target.id,
12905		       io->io_hdr.nexus.targ_lun);
12906		io->io_hdr.status = CTL_CMD_ABORTED;
12907		io->io_hdr.port_status = 31338;
12908
12909		ctl_send_datamove_done(io, /*have_lock*/ 0);
12910
12911		return;
12912	}
12913
12914	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12915		ctl_datamove_remote_write(io);
12916	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12917		ctl_datamove_remote_read(io);
12918	} else {
12919		union ctl_ha_msg msg;
12920		struct scsi_sense_data *sense;
12921		uint8_t sks[3];
12922		int retry_count;
12923
12924		memset(&msg, 0, sizeof(msg));
12925
12926		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12927		msg.hdr.status = CTL_SCSI_ERROR;
12928		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12929
12930		retry_count = 4243;
12931
12932		sense = &msg.scsi.sense_data;
12933		sks[0] = SSD_SCS_VALID;
12934		sks[1] = (retry_count >> 8) & 0xff;
12935		sks[2] = retry_count & 0xff;
12936
12937		/* "Internal target failure" */
12938		scsi_set_sense_data(sense,
12939				    /*sense_format*/ SSD_TYPE_NONE,
12940				    /*current_error*/ 1,
12941				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12942				    /*asc*/ 0x44,
12943				    /*ascq*/ 0x00,
12944				    /*type*/ SSD_ELEM_SKS,
12945				    /*size*/ sizeof(sks),
12946				    /*data*/ sks,
12947				    SSD_ELEM_NONE);
12948
12949		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12950		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12951			ctl_failover_io(io, /*have_lock*/ 1);
12952			return;
12953		}
12954
12955		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
12956		    CTL_HA_STATUS_SUCCESS) {
12957			/* XXX KDM what to do if this fails? */
12958		}
12959		return;
12960	}
12961
12962}
12963
12964static int
12965ctl_process_done(union ctl_io *io)
12966{
12967	struct ctl_lun *lun;
12968	struct ctl_softc *ctl_softc;
12969	void (*fe_done)(union ctl_io *io);
12970	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
12971
12972	CTL_DEBUG_PRINT(("ctl_process_done\n"));
12973
12974	fe_done =
12975	    control_softc->ctl_ports[targ_port]->fe_done;
12976
12977#ifdef CTL_TIME_IO
12978	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12979		char str[256];
12980		char path_str[64];
12981		struct sbuf sb;
12982
12983		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12984		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12985
12986		sbuf_cat(&sb, path_str);
12987		switch (io->io_hdr.io_type) {
12988		case CTL_IO_SCSI:
12989			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12990			sbuf_printf(&sb, "\n");
12991			sbuf_cat(&sb, path_str);
12992			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12993				    io->scsiio.tag_num, io->scsiio.tag_type);
12994			break;
12995		case CTL_IO_TASK:
12996			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12997				    "Tag Type: %d\n", io->taskio.task_action,
12998				    io->taskio.tag_num, io->taskio.tag_type);
12999			break;
13000		default:
13001			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13002			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13003			break;
13004		}
13005		sbuf_cat(&sb, path_str);
13006		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13007			    (intmax_t)time_uptime - io->io_hdr.start_time);
13008		sbuf_finish(&sb);
13009		printf("%s", sbuf_data(&sb));
13010	}
13011#endif /* CTL_TIME_IO */
13012
13013	switch (io->io_hdr.io_type) {
13014	case CTL_IO_SCSI:
13015		break;
13016	case CTL_IO_TASK:
13017		if (bootverbose || verbose > 0)
13018			ctl_io_error_print(io, NULL);
13019		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13020			ctl_free_io(io);
13021		else
13022			fe_done(io);
13023		return (CTL_RETVAL_COMPLETE);
13024		break;
13025	default:
13026		printf("ctl_process_done: invalid io type %d\n",
13027		       io->io_hdr.io_type);
13028		panic("ctl_process_done: invalid io type %d\n",
13029		      io->io_hdr.io_type);
13030		break; /* NOTREACHED */
13031	}
13032
13033	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13034	if (lun == NULL) {
13035		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13036				 io->io_hdr.nexus.targ_mapped_lun));
13037		fe_done(io);
13038		goto bailout;
13039	}
13040	ctl_softc = lun->ctl_softc;
13041
13042	mtx_lock(&lun->lun_lock);
13043
13044	/*
13045	 * Check to see if we have any errors to inject here.  We only
13046	 * inject errors for commands that don't already have errors set.
13047	 */
13048	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13049	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13050		ctl_inject_error(lun, io);
13051
13052	/*
13053	 * XXX KDM how do we treat commands that aren't completed
13054	 * successfully?
13055	 *
13056	 * XXX KDM should we also track I/O latency?
13057	 */
13058	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13059	    io->io_hdr.io_type == CTL_IO_SCSI) {
13060#ifdef CTL_TIME_IO
13061		struct bintime cur_bt;
13062#endif
13063		int type;
13064
13065		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13066		    CTL_FLAG_DATA_IN)
13067			type = CTL_STATS_READ;
13068		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13069		    CTL_FLAG_DATA_OUT)
13070			type = CTL_STATS_WRITE;
13071		else
13072			type = CTL_STATS_NO_IO;
13073
13074		lun->stats.ports[targ_port].bytes[type] +=
13075		    io->scsiio.kern_total_len;
13076		lun->stats.ports[targ_port].operations[type]++;
13077#ifdef CTL_TIME_IO
13078		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13079		   &io->io_hdr.dma_bt);
13080		lun->stats.ports[targ_port].num_dmas[type] +=
13081		    io->io_hdr.num_dmas;
13082		getbintime(&cur_bt);
13083		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13084		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13085#endif
13086	}
13087
13088	/*
13089	 * Remove this from the OOA queue.
13090	 */
13091	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13092
13093	/*
13094	 * Run through the blocked queue on this LUN and see if anything
13095	 * has become unblocked, now that this transaction is done.
13096	 */
13097	ctl_check_blocked(lun);
13098
13099	/*
13100	 * If the LUN has been invalidated, free it if there is nothing
13101	 * left on its OOA queue.
13102	 */
13103	if ((lun->flags & CTL_LUN_INVALID)
13104	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13105		mtx_unlock(&lun->lun_lock);
13106		mtx_lock(&ctl_softc->ctl_lock);
13107		ctl_free_lun(lun);
13108		mtx_unlock(&ctl_softc->ctl_lock);
13109	} else
13110		mtx_unlock(&lun->lun_lock);
13111
13112	/*
13113	 * If this command has been aborted, make sure we set the status
13114	 * properly.  The FETD is responsible for freeing the I/O and doing
13115	 * whatever it needs to do to clean up its state.
13116	 */
13117	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13118		io->io_hdr.status = CTL_CMD_ABORTED;
13119
13120	/*
13121	 * We print out status for every task management command.  For SCSI
13122	 * commands, we filter out any unit attention errors; they happen
13123	 * on every boot, and would clutter up the log.  Note:  task
13124	 * management commands aren't printed here, they are printed above,
13125	 * since they should never even make it down here.
13126	 */
13127	switch (io->io_hdr.io_type) {
13128	case CTL_IO_SCSI: {
13129		int error_code, sense_key, asc, ascq;
13130
13131		sense_key = 0;
13132
13133		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13134		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13135			/*
13136			 * Since this is just for printing, no need to
13137			 * show errors here.
13138			 */
13139			scsi_extract_sense_len(&io->scsiio.sense_data,
13140					       io->scsiio.sense_len,
13141					       &error_code,
13142					       &sense_key,
13143					       &asc,
13144					       &ascq,
13145					       /*show_errors*/ 0);
13146		}
13147
13148		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13149		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13150		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13151		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13152
13153			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13154				ctl_softc->skipped_prints++;
13155			} else {
13156				uint32_t skipped_prints;
13157
13158				skipped_prints = ctl_softc->skipped_prints;
13159
13160				ctl_softc->skipped_prints = 0;
13161				ctl_softc->last_print_jiffies = time_uptime;
13162
13163				if (skipped_prints > 0) {
13164#ifdef NEEDTOPORT
13165					csevent_log(CSC_CTL | CSC_SHELF_SW |
13166					    CTL_ERROR_REPORT,
13167					    csevent_LogType_Trace,
13168					    csevent_Severity_Information,
13169					    csevent_AlertLevel_Green,
13170					    csevent_FRU_Firmware,
13171					    csevent_FRU_Unknown,
13172					    "High CTL error volume, %d prints "
13173					    "skipped", skipped_prints);
13174#endif
13175				}
13176				if (bootverbose || verbose > 0)
13177					ctl_io_error_print(io, NULL);
13178			}
13179		}
13180		break;
13181	}
13182	case CTL_IO_TASK:
13183		if (bootverbose || verbose > 0)
13184			ctl_io_error_print(io, NULL);
13185		break;
13186	default:
13187		break;
13188	}
13189
13190	/*
13191	 * Tell the FETD or the other shelf controller we're done with this
13192	 * command.  Note that only SCSI commands get to this point.  Task
13193	 * management commands are completed above.
13194	 *
13195	 * We only send status to the other controller if we're in XFER
13196	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13197	 * received the I/O (from CTL's perspective), and so the status is
13198	 * generated there.
13199	 *
13200	 * XXX KDM if we hold the lock here, we could cause a deadlock
13201	 * if the frontend comes back in in this context to queue
13202	 * something.
13203	 */
13204	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13205	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13206		union ctl_ha_msg msg;
13207
13208		memset(&msg, 0, sizeof(msg));
13209		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13210		msg.hdr.original_sc = io->io_hdr.original_sc;
13211		msg.hdr.nexus = io->io_hdr.nexus;
13212		msg.hdr.status = io->io_hdr.status;
13213		msg.scsi.scsi_status = io->scsiio.scsi_status;
13214		msg.scsi.tag_num = io->scsiio.tag_num;
13215		msg.scsi.tag_type = io->scsiio.tag_type;
13216		msg.scsi.sense_len = io->scsiio.sense_len;
13217		msg.scsi.sense_residual = io->scsiio.sense_residual;
13218		msg.scsi.residual = io->scsiio.residual;
13219		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13220		       sizeof(io->scsiio.sense_data));
13221		/*
13222		 * We copy this whether or not this is an I/O-related
13223		 * command.  Otherwise, we'd have to go and check to see
13224		 * whether it's a read/write command, and it really isn't
13225		 * worth it.
13226		 */
13227		memcpy(&msg.scsi.lbalen,
13228		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13229		       sizeof(msg.scsi.lbalen));
13230
13231		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13232				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13233			/* XXX do something here */
13234		}
13235
13236		ctl_free_io(io);
13237	} else
13238		fe_done(io);
13239
13240bailout:
13241
13242	return (CTL_RETVAL_COMPLETE);
13243}
13244
13245/*
13246 * Front end should call this if it doesn't do autosense.  When the request
13247 * sense comes back in from the initiator, we'll dequeue this and send it.
13248 */
13249int
13250ctl_queue_sense(union ctl_io *io)
13251{
13252	struct ctl_lun *lun;
13253	struct ctl_softc *ctl_softc;
13254	uint32_t initidx, targ_lun;
13255
13256	ctl_softc = control_softc;
13257
13258	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13259
13260	/*
13261	 * LUN lookup will likely move to the ctl_work_thread() once we
13262	 * have our new queueing infrastructure (that doesn't put things on
13263	 * a per-LUN queue initially).  That is so that we can handle
13264	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13265	 * can't deal with that right now.
13266	 */
13267	mtx_lock(&ctl_softc->ctl_lock);
13268
13269	/*
13270	 * If we don't have a LUN for this, just toss the sense
13271	 * information.
13272	 */
13273	targ_lun = io->io_hdr.nexus.targ_lun;
13274	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13275	if ((targ_lun < CTL_MAX_LUNS)
13276	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13277		lun = ctl_softc->ctl_luns[targ_lun];
13278	else
13279		goto bailout;
13280
13281	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13282
13283	mtx_lock(&lun->lun_lock);
13284	/*
13285	 * Already have CA set for this LUN...toss the sense information.
13286	 */
13287	if (ctl_is_set(lun->have_ca, initidx)) {
13288		mtx_unlock(&lun->lun_lock);
13289		goto bailout;
13290	}
13291
13292	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
13293	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
13294	       sizeof(io->scsiio.sense_data)));
13295	ctl_set_mask(lun->have_ca, initidx);
13296	mtx_unlock(&lun->lun_lock);
13297
13298bailout:
13299	mtx_unlock(&ctl_softc->ctl_lock);
13300
13301	ctl_free_io(io);
13302
13303	return (CTL_RETVAL_COMPLETE);
13304}
13305
13306/*
13307 * Primary command inlet from frontend ports.  All SCSI and task I/O
13308 * requests must go through this function.
13309 */
13310int
13311ctl_queue(union ctl_io *io)
13312{
13313	struct ctl_softc *ctl_softc;
13314
13315	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13316
13317	ctl_softc = control_softc;
13318
13319#ifdef CTL_TIME_IO
13320	io->io_hdr.start_time = time_uptime;
13321	getbintime(&io->io_hdr.start_bt);
13322#endif /* CTL_TIME_IO */
13323
13324	/* Map FE-specific LUN ID into global one. */
13325	io->io_hdr.nexus.targ_mapped_lun =
13326	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13327
13328	switch (io->io_hdr.io_type) {
13329	case CTL_IO_SCSI:
13330	case CTL_IO_TASK:
13331		ctl_enqueue_incoming(io);
13332		break;
13333	default:
13334		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13335		return (EINVAL);
13336	}
13337
13338	return (CTL_RETVAL_COMPLETE);
13339}
13340
13341#ifdef CTL_IO_DELAY
13342static void
13343ctl_done_timer_wakeup(void *arg)
13344{
13345	union ctl_io *io;
13346
13347	io = (union ctl_io *)arg;
13348	ctl_done(io);
13349}
13350#endif /* CTL_IO_DELAY */
13351
13352void
13353ctl_done(union ctl_io *io)
13354{
13355	struct ctl_softc *ctl_softc;
13356
13357	ctl_softc = control_softc;
13358
13359	/*
13360	 * Enable this to catch duplicate completion issues.
13361	 */
13362#if 0
13363	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13364		printf("%s: type %d msg %d cdb %x iptl: "
13365		       "%d:%d:%d:%d tag 0x%04x "
13366		       "flag %#x status %x\n",
13367			__func__,
13368			io->io_hdr.io_type,
13369			io->io_hdr.msg_type,
13370			io->scsiio.cdb[0],
13371			io->io_hdr.nexus.initid.id,
13372			io->io_hdr.nexus.targ_port,
13373			io->io_hdr.nexus.targ_target.id,
13374			io->io_hdr.nexus.targ_lun,
13375			(io->io_hdr.io_type ==
13376			CTL_IO_TASK) ?
13377			io->taskio.tag_num :
13378			io->scsiio.tag_num,
13379		        io->io_hdr.flags,
13380			io->io_hdr.status);
13381	} else
13382		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13383#endif
13384
13385	/*
13386	 * This is an internal copy of an I/O, and should not go through
13387	 * the normal done processing logic.
13388	 */
13389	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13390		return;
13391
13392	/*
13393	 * We need to send a msg to the serializing shelf to finish the IO
13394	 * as well.  We don't send a finish message to the other shelf if
13395	 * this is a task management command.  Task management commands
13396	 * aren't serialized in the OOA queue, but rather just executed on
13397	 * both shelf controllers for commands that originated on that
13398	 * controller.
13399	 */
13400	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13401	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13402		union ctl_ha_msg msg_io;
13403
13404		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13405		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13406		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13407		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13408		}
13409		/* continue on to finish IO */
13410	}
13411#ifdef CTL_IO_DELAY
13412	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13413		struct ctl_lun *lun;
13414
13415		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13416
13417		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13418	} else {
13419		struct ctl_lun *lun;
13420
13421		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13422
13423		if ((lun != NULL)
13424		 && (lun->delay_info.done_delay > 0)) {
13425			struct callout *callout;
13426
13427			callout = (struct callout *)&io->io_hdr.timer_bytes;
13428			callout_init(callout, /*mpsafe*/ 1);
13429			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13430			callout_reset(callout,
13431				      lun->delay_info.done_delay * hz,
13432				      ctl_done_timer_wakeup, io);
13433			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13434				lun->delay_info.done_delay = 0;
13435			return;
13436		}
13437	}
13438#endif /* CTL_IO_DELAY */
13439
13440	ctl_enqueue_done(io);
13441}
13442
13443int
13444ctl_isc(struct ctl_scsiio *ctsio)
13445{
13446	struct ctl_lun *lun;
13447	int retval;
13448
13449	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13450
13451	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13452
13453	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13454
13455	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13456
13457	return (retval);
13458}
13459
13460
13461static void
13462ctl_work_thread(void *arg)
13463{
13464	struct ctl_thread *thr = (struct ctl_thread *)arg;
13465	struct ctl_softc *softc = thr->ctl_softc;
13466	union ctl_io *io;
13467	int retval;
13468
13469	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13470
13471	for (;;) {
13472		retval = 0;
13473
13474		/*
13475		 * We handle the queues in this order:
13476		 * - ISC
13477		 * - done queue (to free up resources, unblock other commands)
13478		 * - RtR queue
13479		 * - incoming queue
13480		 *
13481		 * If those queues are empty, we break out of the loop and
13482		 * go to sleep.
13483		 */
13484		mtx_lock(&thr->queue_lock);
13485		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13486		if (io != NULL) {
13487			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13488			mtx_unlock(&thr->queue_lock);
13489			ctl_handle_isc(io);
13490			continue;
13491		}
13492		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13493		if (io != NULL) {
13494			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13495			/* clear any blocked commands, call fe_done */
13496			mtx_unlock(&thr->queue_lock);
13497			retval = ctl_process_done(io);
13498			continue;
13499		}
13500		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13501		if (io != NULL) {
13502			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13503			mtx_unlock(&thr->queue_lock);
13504			if (io->io_hdr.io_type == CTL_IO_TASK)
13505				ctl_run_task(io);
13506			else
13507				ctl_scsiio_precheck(softc, &io->scsiio);
13508			continue;
13509		}
13510		if (!ctl_pause_rtr) {
13511			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13512			if (io != NULL) {
13513				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13514				mtx_unlock(&thr->queue_lock);
13515				retval = ctl_scsiio(&io->scsiio);
13516				if (retval != CTL_RETVAL_COMPLETE)
13517					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13518				continue;
13519			}
13520		}
13521
13522		/* Sleep until we have something to do. */
13523		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13524	}
13525}
13526
13527static void
13528ctl_lun_thread(void *arg)
13529{
13530	struct ctl_softc *softc = (struct ctl_softc *)arg;
13531	struct ctl_be_lun *be_lun;
13532	int retval;
13533
13534	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13535
13536	for (;;) {
13537		retval = 0;
13538		mtx_lock(&softc->ctl_lock);
13539		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13540		if (be_lun != NULL) {
13541			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13542			mtx_unlock(&softc->ctl_lock);
13543			ctl_create_lun(be_lun);
13544			continue;
13545		}
13546
13547		/* Sleep until we have something to do. */
13548		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13549		    PDROP | PRIBIO, "-", 0);
13550	}
13551}
13552
13553static void
13554ctl_enqueue_incoming(union ctl_io *io)
13555{
13556	struct ctl_softc *softc = control_softc;
13557	struct ctl_thread *thr;
13558	u_int idx;
13559
13560	idx = (io->io_hdr.nexus.targ_port * 127 +
13561	       io->io_hdr.nexus.initid.id) % worker_threads;
13562	thr = &softc->threads[idx];
13563	mtx_lock(&thr->queue_lock);
13564	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13565	mtx_unlock(&thr->queue_lock);
13566	wakeup(thr);
13567}
13568
13569static void
13570ctl_enqueue_rtr(union ctl_io *io)
13571{
13572	struct ctl_softc *softc = control_softc;
13573	struct ctl_thread *thr;
13574
13575	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13576	mtx_lock(&thr->queue_lock);
13577	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13578	mtx_unlock(&thr->queue_lock);
13579	wakeup(thr);
13580}
13581
13582static void
13583ctl_enqueue_done(union ctl_io *io)
13584{
13585	struct ctl_softc *softc = control_softc;
13586	struct ctl_thread *thr;
13587
13588	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13589	mtx_lock(&thr->queue_lock);
13590	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13591	mtx_unlock(&thr->queue_lock);
13592	wakeup(thr);
13593}
13594
13595static void
13596ctl_enqueue_isc(union ctl_io *io)
13597{
13598	struct ctl_softc *softc = control_softc;
13599	struct ctl_thread *thr;
13600
13601	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13602	mtx_lock(&thr->queue_lock);
13603	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13604	mtx_unlock(&thr->queue_lock);
13605	wakeup(thr);
13606}
13607
13608/* Initialization and failover */
13609
13610void
13611ctl_init_isc_msg(void)
13612{
13613	printf("CTL: Still calling this thing\n");
13614}
13615
13616/*
13617 * Init component
13618 * 	Initializes component into configuration defined by bootMode
13619 *	(see hasc-sv.c)
13620 *  	returns hasc_Status:
13621 * 		OK
13622 *		ERROR - fatal error
13623 */
13624static ctl_ha_comp_status
13625ctl_isc_init(struct ctl_ha_component *c)
13626{
13627	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13628
13629	c->status = ret;
13630	return ret;
13631}
13632
13633/* Start component
13634 * 	Starts component in state requested. If component starts successfully,
13635 *	it must set its own state to the requestrd state
13636 *	When requested state is HASC_STATE_HA, the component may refine it
13637 * 	by adding _SLAVE or _MASTER flags.
13638 *	Currently allowed state transitions are:
13639 *	UNKNOWN->HA		- initial startup
13640 *	UNKNOWN->SINGLE - initial startup when no parter detected
13641 *	HA->SINGLE		- failover
13642 * returns ctl_ha_comp_status:
13643 * 		OK	- component successfully started in requested state
13644 *		FAILED  - could not start the requested state, failover may
13645 * 			  be possible
13646 *		ERROR	- fatal error detected, no future startup possible
13647 */
13648static ctl_ha_comp_status
13649ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13650{
13651	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13652
13653	printf("%s: go\n", __func__);
13654
13655	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13656	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13657		ctl_is_single = 0;
13658		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13659		    != CTL_HA_STATUS_SUCCESS) {
13660			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13661			ret = CTL_HA_COMP_STATUS_ERROR;
13662		}
13663	} else if (CTL_HA_STATE_IS_HA(c->state)
13664		&& CTL_HA_STATE_IS_SINGLE(state)){
13665		// HA->SINGLE transition
13666	        ctl_failover();
13667		ctl_is_single = 1;
13668	} else {
13669		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13670		       c->state, state);
13671		ret = CTL_HA_COMP_STATUS_ERROR;
13672	}
13673	if (CTL_HA_STATE_IS_SINGLE(state))
13674		ctl_is_single = 1;
13675
13676	c->state = state;
13677	c->status = ret;
13678	return ret;
13679}
13680
13681/*
13682 * Quiesce component
13683 * The component must clear any error conditions (set status to OK) and
13684 * prepare itself to another Start call
13685 * returns ctl_ha_comp_status:
13686 * 	OK
13687 *	ERROR
13688 */
13689static ctl_ha_comp_status
13690ctl_isc_quiesce(struct ctl_ha_component *c)
13691{
13692	int ret = CTL_HA_COMP_STATUS_OK;
13693
13694	ctl_pause_rtr = 1;
13695	c->status = ret;
13696	return ret;
13697}
13698
13699struct ctl_ha_component ctl_ha_component_ctlisc =
13700{
13701	.name = "CTL ISC",
13702	.state = CTL_HA_STATE_UNKNOWN,
13703	.init = ctl_isc_init,
13704	.start = ctl_isc_start,
13705	.quiesce = ctl_isc_quiesce
13706};
13707
13708/*
13709 *  vim: ts=8
13710 */
13711