ctl.c revision 268447
1/*-
2 * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3 * Copyright (c) 2012 The FreeBSD Foundation
4 * All rights reserved.
5 *
6 * Portions of this software were developed by Edward Tomasz Napierala
7 * under sponsorship from the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions, and the following disclaimer,
14 *    without modification.
15 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16 *    substantially similar to the "NO WARRANTY" disclaimer below
17 *    ("Disclaimer") and any redistribution must be conditioned upon
18 *    including a substantially similar Disclaimer requirement for further
19 *    binary redistribution.
20 *
21 * NO WARRANTY
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGES.
33 *
34 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
35 */
36/*
37 * CAM Target Layer, a SCSI device emulation subsystem.
38 *
39 * Author: Ken Merry <ken@FreeBSD.org>
40 */
41
42#define _CTL_C
43
44#include <sys/cdefs.h>
45__FBSDID("$FreeBSD: head/sys/cam/ctl/ctl.c 268447 2014-07-09 04:37:50Z 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	/*flags4*/SCP_TAS,
286	/*aen_holdoff_period*/{0, 0},
287	/*busy_timeout_period*/{0, 0},
288	/*extended_selftest_completion_time*/{0, 0}
289};
290
291static struct scsi_control_page control_page_changeable = {
292	/*page_code*/SMS_CONTROL_MODE_PAGE,
293	/*page_length*/sizeof(struct scsi_control_page) - 2,
294	/*rlec*/SCP_DSENSE,
295	/*queue_flags*/0,
296	/*eca_and_aen*/0,
297	/*flags4*/0,
298	/*aen_holdoff_period*/{0, 0},
299	/*busy_timeout_period*/{0, 0},
300	/*extended_selftest_completion_time*/{0, 0}
301};
302
303
304/*
305 * XXX KDM move these into the softc.
306 */
307static int rcv_sync_msg;
308static int persis_offset;
309static uint8_t ctl_pause_rtr;
310static int     ctl_is_single = 1;
311static int     index_to_aps_page;
312
313SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
314static int worker_threads = -1;
315SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
316    &worker_threads, 1, "Number of worker threads");
317static int verbose = 0;
318SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
319    &verbose, 0, "Show SCSI errors returned to initiator");
320
321/*
322 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
323 * SCSI Ports (0x88), Block limits (0xB0) and
324 * Logical Block Provisioning (0xB2)
325 */
326#define SCSI_EVPD_NUM_SUPPORTED_PAGES	6
327
328static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
329				  int param);
330static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
331static int ctl_init(void);
332void ctl_shutdown(void);
333static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
334static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
335static void ctl_ioctl_online(void *arg);
336static void ctl_ioctl_offline(void *arg);
337static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
338static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
339static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
340static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
341static int ctl_ioctl_submit_wait(union ctl_io *io);
342static void ctl_ioctl_datamove(union ctl_io *io);
343static void ctl_ioctl_done(union ctl_io *io);
344static void ctl_ioctl_hard_startstop_callback(void *arg,
345					      struct cfi_metatask *metatask);
346static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
347static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
348			      struct ctl_ooa *ooa_hdr,
349			      struct ctl_ooa_entry *kern_entries);
350static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
351		     struct thread *td);
352uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
353uint32_t ctl_port_idx(int port_num);
354static uint32_t ctl_map_lun(int port_num, uint32_t lun);
355static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
356#ifdef unused
357static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
358				   uint32_t targ_target, uint32_t targ_lun,
359				   int can_wait);
360static void ctl_kfree_io(union ctl_io *io);
361#endif /* unused */
362static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
363			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
364static int ctl_free_lun(struct ctl_lun *lun);
365static void ctl_create_lun(struct ctl_be_lun *be_lun);
366/**
367static void ctl_failover_change_pages(struct ctl_softc *softc,
368				      struct ctl_scsiio *ctsio, int master);
369**/
370
371static int ctl_do_mode_select(union ctl_io *io);
372static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
373			   uint64_t res_key, uint64_t sa_res_key,
374			   uint8_t type, uint32_t residx,
375			   struct ctl_scsiio *ctsio,
376			   struct scsi_per_res_out *cdb,
377			   struct scsi_per_res_out_parms* param);
378static void ctl_pro_preempt_other(struct ctl_lun *lun,
379				  union ctl_ha_msg *msg);
380static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
381static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
382static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
383static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
384static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
385					 int alloc_len);
386static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
387					 int alloc_len);
388static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
389static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
390static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
391static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
392static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
393static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
394					 union ctl_io *ooa_io);
395static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
396				union ctl_io *starting_io);
397static int ctl_check_blocked(struct ctl_lun *lun);
398static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
399				struct ctl_lun *lun,
400				const struct ctl_cmd_entry *entry,
401				struct ctl_scsiio *ctsio);
402//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
403static void ctl_failover(void);
404static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
405			       struct ctl_scsiio *ctsio);
406static int ctl_scsiio(struct ctl_scsiio *ctsio);
407
408static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
409static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
410			    ctl_ua_type ua_type);
411static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
412			 ctl_ua_type ua_type);
413static int ctl_abort_task(union ctl_io *io);
414static int ctl_abort_task_set(union ctl_io *io);
415static int ctl_i_t_nexus_reset(union ctl_io *io);
416static void ctl_run_task(union ctl_io *io);
417#ifdef CTL_IO_DELAY
418static void ctl_datamove_timer_wakeup(void *arg);
419static void ctl_done_timer_wakeup(void *arg);
420#endif /* CTL_IO_DELAY */
421
422static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
423static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
424static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
425static void ctl_datamove_remote_write(union ctl_io *io);
426static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
427static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
428static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
429static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
430				    ctl_ha_dt_cb callback);
431static void ctl_datamove_remote_read(union ctl_io *io);
432static void ctl_datamove_remote(union ctl_io *io);
433static int ctl_process_done(union ctl_io *io);
434static void ctl_lun_thread(void *arg);
435static void ctl_work_thread(void *arg);
436static void ctl_enqueue_incoming(union ctl_io *io);
437static void ctl_enqueue_rtr(union ctl_io *io);
438static void ctl_enqueue_done(union ctl_io *io);
439static void ctl_enqueue_isc(union ctl_io *io);
440static const struct ctl_cmd_entry *
441    ctl_get_cmd_entry(struct ctl_scsiio *ctsio);
442static const struct ctl_cmd_entry *
443    ctl_validate_command(struct ctl_scsiio *ctsio);
444static int ctl_cmd_applicable(uint8_t lun_type,
445    const struct ctl_cmd_entry *entry);
446
447/*
448 * Load the serialization table.  This isn't very pretty, but is probably
449 * the easiest way to do it.
450 */
451#include "ctl_ser_table.c"
452
453/*
454 * We only need to define open, close and ioctl routines for this driver.
455 */
456static struct cdevsw ctl_cdevsw = {
457	.d_version =	D_VERSION,
458	.d_flags =	0,
459	.d_open =	ctl_open,
460	.d_close =	ctl_close,
461	.d_ioctl =	ctl_ioctl,
462	.d_name =	"ctl",
463};
464
465
466MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
467MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
468
469static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
470
471static moduledata_t ctl_moduledata = {
472	"ctl",
473	ctl_module_event_handler,
474	NULL
475};
476
477DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
478MODULE_VERSION(ctl, 1);
479
480static struct ctl_frontend ioctl_frontend =
481{
482	.name = "ioctl",
483};
484
485static void
486ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
487			    union ctl_ha_msg *msg_info)
488{
489	struct ctl_scsiio *ctsio;
490
491	if (msg_info->hdr.original_sc == NULL) {
492		printf("%s: original_sc == NULL!\n", __func__);
493		/* XXX KDM now what? */
494		return;
495	}
496
497	ctsio = &msg_info->hdr.original_sc->scsiio;
498	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
499	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
500	ctsio->io_hdr.status = msg_info->hdr.status;
501	ctsio->scsi_status = msg_info->scsi.scsi_status;
502	ctsio->sense_len = msg_info->scsi.sense_len;
503	ctsio->sense_residual = msg_info->scsi.sense_residual;
504	ctsio->residual = msg_info->scsi.residual;
505	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
506	       sizeof(ctsio->sense_data));
507	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
508	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
509	ctl_enqueue_isc((union ctl_io *)ctsio);
510}
511
512static void
513ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
514				union ctl_ha_msg *msg_info)
515{
516	struct ctl_scsiio *ctsio;
517
518	if (msg_info->hdr.serializing_sc == NULL) {
519		printf("%s: serializing_sc == NULL!\n", __func__);
520		/* XXX KDM now what? */
521		return;
522	}
523
524	ctsio = &msg_info->hdr.serializing_sc->scsiio;
525#if 0
526	/*
527	 * Attempt to catch the situation where an I/O has
528	 * been freed, and we're using it again.
529	 */
530	if (ctsio->io_hdr.io_type == 0xff) {
531		union ctl_io *tmp_io;
532		tmp_io = (union ctl_io *)ctsio;
533		printf("%s: %p use after free!\n", __func__,
534		       ctsio);
535		printf("%s: type %d msg %d cdb %x iptl: "
536		       "%d:%d:%d:%d tag 0x%04x "
537		       "flag %#x status %x\n",
538			__func__,
539			tmp_io->io_hdr.io_type,
540			tmp_io->io_hdr.msg_type,
541			tmp_io->scsiio.cdb[0],
542			tmp_io->io_hdr.nexus.initid.id,
543			tmp_io->io_hdr.nexus.targ_port,
544			tmp_io->io_hdr.nexus.targ_target.id,
545			tmp_io->io_hdr.nexus.targ_lun,
546			(tmp_io->io_hdr.io_type ==
547			CTL_IO_TASK) ?
548			tmp_io->taskio.tag_num :
549			tmp_io->scsiio.tag_num,
550		        tmp_io->io_hdr.flags,
551			tmp_io->io_hdr.status);
552	}
553#endif
554	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
555	ctl_enqueue_isc((union ctl_io *)ctsio);
556}
557
558/*
559 * ISC (Inter Shelf Communication) event handler.  Events from the HA
560 * subsystem come in here.
561 */
562static void
563ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
564{
565	struct ctl_softc *ctl_softc;
566	union ctl_io *io;
567	struct ctl_prio *presio;
568	ctl_ha_status isc_status;
569
570	ctl_softc = control_softc;
571	io = NULL;
572
573
574#if 0
575	printf("CTL: Isc Msg event %d\n", event);
576#endif
577	if (event == CTL_HA_EVT_MSG_RECV) {
578		union ctl_ha_msg msg_info;
579
580		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
581					     sizeof(msg_info), /*wait*/ 0);
582#if 0
583		printf("CTL: msg_type %d\n", msg_info.msg_type);
584#endif
585		if (isc_status != 0) {
586			printf("Error receiving message, status = %d\n",
587			       isc_status);
588			return;
589		}
590
591		switch (msg_info.hdr.msg_type) {
592		case CTL_MSG_SERIALIZE:
593#if 0
594			printf("Serialize\n");
595#endif
596			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
597			if (io == NULL) {
598				printf("ctl_isc_event_handler: can't allocate "
599				       "ctl_io!\n");
600				/* Bad Juju */
601				/* Need to set busy and send msg back */
602				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
603				msg_info.hdr.status = CTL_SCSI_ERROR;
604				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
605				msg_info.scsi.sense_len = 0;
606			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
607				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
608				}
609				goto bailout;
610			}
611			ctl_zero_io(io);
612			// populate ctsio from msg_info
613			io->io_hdr.io_type = CTL_IO_SCSI;
614			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
615			io->io_hdr.original_sc = msg_info.hdr.original_sc;
616#if 0
617			printf("pOrig %x\n", (int)msg_info.original_sc);
618#endif
619			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
620					    CTL_FLAG_IO_ACTIVE;
621			/*
622			 * If we're in serialization-only mode, we don't
623			 * want to go through full done processing.  Thus
624			 * the COPY flag.
625			 *
626			 * XXX KDM add another flag that is more specific.
627			 */
628			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
629				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
630			io->io_hdr.nexus = msg_info.hdr.nexus;
631#if 0
632			printf("targ %d, port %d, iid %d, lun %d\n",
633			       io->io_hdr.nexus.targ_target.id,
634			       io->io_hdr.nexus.targ_port,
635			       io->io_hdr.nexus.initid.id,
636			       io->io_hdr.nexus.targ_lun);
637#endif
638			io->scsiio.tag_num = msg_info.scsi.tag_num;
639			io->scsiio.tag_type = msg_info.scsi.tag_type;
640			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
641			       CTL_MAX_CDBLEN);
642			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
643				const struct ctl_cmd_entry *entry;
644
645				entry = ctl_get_cmd_entry(&io->scsiio);
646				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
647				io->io_hdr.flags |=
648					entry->flags & CTL_FLAG_DATA_MASK;
649			}
650			ctl_enqueue_isc(io);
651			break;
652
653		/* Performed on the Originating SC, XFER mode only */
654		case CTL_MSG_DATAMOVE: {
655			struct ctl_sg_entry *sgl;
656			int i, j;
657
658			io = msg_info.hdr.original_sc;
659			if (io == NULL) {
660				printf("%s: original_sc == NULL!\n", __func__);
661				/* XXX KDM do something here */
662				break;
663			}
664			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
665			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
666			/*
667			 * Keep track of this, we need to send it back over
668			 * when the datamove is complete.
669			 */
670			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
671
672			if (msg_info.dt.sg_sequence == 0) {
673				/*
674				 * XXX KDM we use the preallocated S/G list
675				 * here, but we'll need to change this to
676				 * dynamic allocation if we need larger S/G
677				 * lists.
678				 */
679				if (msg_info.dt.kern_sg_entries >
680				    sizeof(io->io_hdr.remote_sglist) /
681				    sizeof(io->io_hdr.remote_sglist[0])) {
682					printf("%s: number of S/G entries "
683					    "needed %u > allocated num %zd\n",
684					    __func__,
685					    msg_info.dt.kern_sg_entries,
686					    sizeof(io->io_hdr.remote_sglist)/
687					    sizeof(io->io_hdr.remote_sglist[0]));
688
689					/*
690					 * XXX KDM send a message back to
691					 * the other side to shut down the
692					 * DMA.  The error will come back
693					 * through via the normal channel.
694					 */
695					break;
696				}
697				sgl = io->io_hdr.remote_sglist;
698				memset(sgl, 0,
699				       sizeof(io->io_hdr.remote_sglist));
700
701				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
702
703				io->scsiio.kern_sg_entries =
704					msg_info.dt.kern_sg_entries;
705				io->scsiio.rem_sg_entries =
706					msg_info.dt.kern_sg_entries;
707				io->scsiio.kern_data_len =
708					msg_info.dt.kern_data_len;
709				io->scsiio.kern_total_len =
710					msg_info.dt.kern_total_len;
711				io->scsiio.kern_data_resid =
712					msg_info.dt.kern_data_resid;
713				io->scsiio.kern_rel_offset =
714					msg_info.dt.kern_rel_offset;
715				/*
716				 * Clear out per-DMA flags.
717				 */
718				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
719				/*
720				 * Add per-DMA flags that are set for this
721				 * particular DMA request.
722				 */
723				io->io_hdr.flags |= msg_info.dt.flags &
724						    CTL_FLAG_RDMA_MASK;
725			} else
726				sgl = (struct ctl_sg_entry *)
727					io->scsiio.kern_data_ptr;
728
729			for (i = msg_info.dt.sent_sg_entries, j = 0;
730			     i < (msg_info.dt.sent_sg_entries +
731			     msg_info.dt.cur_sg_entries); i++, j++) {
732				sgl[i].addr = msg_info.dt.sg_list[j].addr;
733				sgl[i].len = msg_info.dt.sg_list[j].len;
734
735#if 0
736				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
737				       __func__,
738				       msg_info.dt.sg_list[j].addr,
739				       msg_info.dt.sg_list[j].len,
740				       sgl[i].addr, sgl[i].len, j, i);
741#endif
742			}
743#if 0
744			memcpy(&sgl[msg_info.dt.sent_sg_entries],
745			       msg_info.dt.sg_list,
746			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
747#endif
748
749			/*
750			 * If this is the last piece of the I/O, we've got
751			 * the full S/G list.  Queue processing in the thread.
752			 * Otherwise wait for the next piece.
753			 */
754			if (msg_info.dt.sg_last != 0)
755				ctl_enqueue_isc(io);
756			break;
757		}
758		/* Performed on the Serializing (primary) SC, XFER mode only */
759		case CTL_MSG_DATAMOVE_DONE: {
760			if (msg_info.hdr.serializing_sc == NULL) {
761				printf("%s: serializing_sc == NULL!\n",
762				       __func__);
763				/* XXX KDM now what? */
764				break;
765			}
766			/*
767			 * We grab the sense information here in case
768			 * there was a failure, so we can return status
769			 * back to the initiator.
770			 */
771			io = msg_info.hdr.serializing_sc;
772			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
773			io->io_hdr.status = msg_info.hdr.status;
774			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
775			io->scsiio.sense_len = msg_info.scsi.sense_len;
776			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
777			io->io_hdr.port_status = msg_info.scsi.fetd_status;
778			io->scsiio.residual = msg_info.scsi.residual;
779			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
780			       sizeof(io->scsiio.sense_data));
781			ctl_enqueue_isc(io);
782			break;
783		}
784
785		/* Preformed on Originating SC, SER_ONLY mode */
786		case CTL_MSG_R2R:
787			io = msg_info.hdr.original_sc;
788			if (io == NULL) {
789				printf("%s: Major Bummer\n", __func__);
790				return;
791			} else {
792#if 0
793				printf("pOrig %x\n",(int) ctsio);
794#endif
795			}
796			io->io_hdr.msg_type = CTL_MSG_R2R;
797			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
798			ctl_enqueue_isc(io);
799			break;
800
801		/*
802		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
803		 * mode.
804		 * Performed on the Originating (i.e. secondary) SC in XFER
805		 * mode
806		 */
807		case CTL_MSG_FINISH_IO:
808			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
809				ctl_isc_handler_finish_xfer(ctl_softc,
810							    &msg_info);
811			else
812				ctl_isc_handler_finish_ser_only(ctl_softc,
813								&msg_info);
814			break;
815
816		/* Preformed on Originating SC */
817		case CTL_MSG_BAD_JUJU:
818			io = msg_info.hdr.original_sc;
819			if (io == NULL) {
820				printf("%s: Bad JUJU!, original_sc is NULL!\n",
821				       __func__);
822				break;
823			}
824			ctl_copy_sense_data(&msg_info, io);
825			/*
826			 * IO should have already been cleaned up on other
827			 * SC so clear this flag so we won't send a message
828			 * back to finish the IO there.
829			 */
830			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
831			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
832
833			/* io = msg_info.hdr.serializing_sc; */
834			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
835			ctl_enqueue_isc(io);
836			break;
837
838		/* Handle resets sent from the other side */
839		case CTL_MSG_MANAGE_TASKS: {
840			struct ctl_taskio *taskio;
841			taskio = (struct ctl_taskio *)ctl_alloc_io(
842				(void *)ctl_softc->othersc_pool);
843			if (taskio == NULL) {
844				printf("ctl_isc_event_handler: can't allocate "
845				       "ctl_io!\n");
846				/* Bad Juju */
847				/* should I just call the proper reset func
848				   here??? */
849				goto bailout;
850			}
851			ctl_zero_io((union ctl_io *)taskio);
852			taskio->io_hdr.io_type = CTL_IO_TASK;
853			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
854			taskio->io_hdr.nexus = msg_info.hdr.nexus;
855			taskio->task_action = msg_info.task.task_action;
856			taskio->tag_num = msg_info.task.tag_num;
857			taskio->tag_type = msg_info.task.tag_type;
858#ifdef CTL_TIME_IO
859			taskio->io_hdr.start_time = time_uptime;
860			getbintime(&taskio->io_hdr.start_bt);
861#if 0
862			cs_prof_gettime(&taskio->io_hdr.start_ticks);
863#endif
864#endif /* CTL_TIME_IO */
865			ctl_run_task((union ctl_io *)taskio);
866			break;
867		}
868		/* Persistent Reserve action which needs attention */
869		case CTL_MSG_PERS_ACTION:
870			presio = (struct ctl_prio *)ctl_alloc_io(
871				(void *)ctl_softc->othersc_pool);
872			if (presio == NULL) {
873				printf("ctl_isc_event_handler: can't allocate "
874				       "ctl_io!\n");
875				/* Bad Juju */
876				/* Need to set busy and send msg back */
877				goto bailout;
878			}
879			ctl_zero_io((union ctl_io *)presio);
880			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
881			presio->pr_msg = msg_info.pr;
882			ctl_enqueue_isc((union ctl_io *)presio);
883			break;
884		case CTL_MSG_SYNC_FE:
885			rcv_sync_msg = 1;
886			break;
887		case CTL_MSG_APS_LOCK: {
888			// It's quicker to execute this then to
889			// queue it.
890			struct ctl_lun *lun;
891			struct ctl_page_index *page_index;
892			struct copan_aps_subpage *current_sp;
893			uint32_t targ_lun;
894
895			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
896			lun = ctl_softc->ctl_luns[targ_lun];
897			mtx_lock(&lun->lun_lock);
898			page_index = &lun->mode_pages.index[index_to_aps_page];
899			current_sp = (struct copan_aps_subpage *)
900				     (page_index->page_data +
901				     (page_index->page_len * CTL_PAGE_CURRENT));
902
903			current_sp->lock_active = msg_info.aps.lock_flag;
904			mtx_unlock(&lun->lun_lock);
905		        break;
906		}
907		default:
908		        printf("How did I get here?\n");
909		}
910	} else if (event == CTL_HA_EVT_MSG_SENT) {
911		if (param != CTL_HA_STATUS_SUCCESS) {
912			printf("Bad status from ctl_ha_msg_send status %d\n",
913			       param);
914		}
915		return;
916	} else if (event == CTL_HA_EVT_DISCONNECT) {
917		printf("CTL: Got a disconnect from Isc\n");
918		return;
919	} else {
920		printf("ctl_isc_event_handler: Unknown event %d\n", event);
921		return;
922	}
923
924bailout:
925	return;
926}
927
928static void
929ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
930{
931	struct scsi_sense_data *sense;
932
933	sense = &dest->scsiio.sense_data;
934	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
935	dest->scsiio.scsi_status = src->scsi.scsi_status;
936	dest->scsiio.sense_len = src->scsi.sense_len;
937	dest->io_hdr.status = src->hdr.status;
938}
939
940static int
941ctl_init(void)
942{
943	struct ctl_softc *softc;
944	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
945	struct ctl_port *port;
946        uint8_t sc_id =0;
947	int i, error, retval;
948	//int isc_retval;
949
950	retval = 0;
951	ctl_pause_rtr = 0;
952        rcv_sync_msg = 0;
953
954	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
955			       M_WAITOK | M_ZERO);
956	softc = control_softc;
957
958	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
959			      "cam/ctl");
960
961	softc->dev->si_drv1 = softc;
962
963	/*
964	 * By default, return a "bad LUN" peripheral qualifier for unknown
965	 * LUNs.  The user can override this default using the tunable or
966	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
967	 */
968	softc->inquiry_pq_no_lun = 1;
969	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
970			  &softc->inquiry_pq_no_lun);
971	sysctl_ctx_init(&softc->sysctl_ctx);
972	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
973		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
974		CTLFLAG_RD, 0, "CAM Target Layer");
975
976	if (softc->sysctl_tree == NULL) {
977		printf("%s: unable to allocate sysctl tree\n", __func__);
978		destroy_dev(softc->dev);
979		free(control_softc, M_DEVBUF);
980		control_softc = NULL;
981		return (ENOMEM);
982	}
983
984	SYSCTL_ADD_INT(&softc->sysctl_ctx,
985		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
986		       "inquiry_pq_no_lun", CTLFLAG_RW,
987		       &softc->inquiry_pq_no_lun, 0,
988		       "Report no lun possible for invalid LUNs");
989
990	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
991	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
992	softc->open_count = 0;
993
994	/*
995	 * Default to actually sending a SYNCHRONIZE CACHE command down to
996	 * the drive.
997	 */
998	softc->flags = CTL_FLAG_REAL_SYNC;
999
1000	/*
1001	 * In Copan's HA scheme, the "master" and "slave" roles are
1002	 * figured out through the slot the controller is in.  Although it
1003	 * is an active/active system, someone has to be in charge.
1004 	 */
1005#ifdef NEEDTOPORT
1006        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1007#endif
1008
1009        if (sc_id == 0) {
1010		softc->flags |= CTL_FLAG_MASTER_SHELF;
1011		persis_offset = 0;
1012	} else
1013		persis_offset = CTL_MAX_INITIATORS;
1014
1015	/*
1016	 * XXX KDM need to figure out where we want to get our target ID
1017	 * and WWID.  Is it different on each port?
1018	 */
1019	softc->target.id = 0;
1020	softc->target.wwid[0] = 0x12345678;
1021	softc->target.wwid[1] = 0x87654321;
1022	STAILQ_INIT(&softc->lun_list);
1023	STAILQ_INIT(&softc->pending_lun_queue);
1024	STAILQ_INIT(&softc->fe_list);
1025	STAILQ_INIT(&softc->port_list);
1026	STAILQ_INIT(&softc->be_list);
1027	STAILQ_INIT(&softc->io_pools);
1028
1029	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1030			    &internal_pool)!= 0){
1031		printf("ctl: can't allocate %d entry internal pool, "
1032		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1033		return (ENOMEM);
1034	}
1035
1036	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1037			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1038		printf("ctl: can't allocate %d entry emergency pool, "
1039		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1040		ctl_pool_free(internal_pool);
1041		return (ENOMEM);
1042	}
1043
1044	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1045	                    &other_pool) != 0)
1046	{
1047		printf("ctl: can't allocate %d entry other SC pool, "
1048		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1049		ctl_pool_free(internal_pool);
1050		ctl_pool_free(emergency_pool);
1051		return (ENOMEM);
1052	}
1053
1054	softc->internal_pool = internal_pool;
1055	softc->emergency_pool = emergency_pool;
1056	softc->othersc_pool = other_pool;
1057
1058	if (worker_threads <= 0)
1059		worker_threads = max(1, mp_ncpus / 4);
1060	if (worker_threads > CTL_MAX_THREADS)
1061		worker_threads = CTL_MAX_THREADS;
1062
1063	for (i = 0; i < worker_threads; i++) {
1064		struct ctl_thread *thr = &softc->threads[i];
1065
1066		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1067		thr->ctl_softc = softc;
1068		STAILQ_INIT(&thr->incoming_queue);
1069		STAILQ_INIT(&thr->rtr_queue);
1070		STAILQ_INIT(&thr->done_queue);
1071		STAILQ_INIT(&thr->isc_queue);
1072
1073		error = kproc_kthread_add(ctl_work_thread, thr,
1074		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1075		if (error != 0) {
1076			printf("error creating CTL work thread!\n");
1077			ctl_pool_free(internal_pool);
1078			ctl_pool_free(emergency_pool);
1079			ctl_pool_free(other_pool);
1080			return (error);
1081		}
1082	}
1083	error = kproc_kthread_add(ctl_lun_thread, softc,
1084	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1085	if (error != 0) {
1086		printf("error creating CTL lun thread!\n");
1087		ctl_pool_free(internal_pool);
1088		ctl_pool_free(emergency_pool);
1089		ctl_pool_free(other_pool);
1090		return (error);
1091	}
1092	if (bootverbose)
1093		printf("ctl: CAM Target Layer loaded\n");
1094
1095	/*
1096	 * Initialize the ioctl front end.
1097	 */
1098	ctl_frontend_register(&ioctl_frontend);
1099	port = &softc->ioctl_info.port;
1100	port->frontend = &ioctl_frontend;
1101	sprintf(softc->ioctl_info.port_name, "ioctl");
1102	port->port_type = CTL_PORT_IOCTL;
1103	port->num_requested_ctl_io = 100;
1104	port->port_name = softc->ioctl_info.port_name;
1105	port->port_online = ctl_ioctl_online;
1106	port->port_offline = ctl_ioctl_offline;
1107	port->onoff_arg = &softc->ioctl_info;
1108	port->lun_enable = ctl_ioctl_lun_enable;
1109	port->lun_disable = ctl_ioctl_lun_disable;
1110	port->targ_lun_arg = &softc->ioctl_info;
1111	port->fe_datamove = ctl_ioctl_datamove;
1112	port->fe_done = ctl_ioctl_done;
1113	port->max_targets = 15;
1114	port->max_target_id = 15;
1115
1116	if (ctl_port_register(&softc->ioctl_info.port,
1117	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1118		printf("ctl: ioctl front end registration failed, will "
1119		       "continue anyway\n");
1120	}
1121
1122#ifdef CTL_IO_DELAY
1123	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1124		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1125		       sizeof(struct callout), CTL_TIMER_BYTES);
1126		return (EINVAL);
1127	}
1128#endif /* CTL_IO_DELAY */
1129
1130	return (0);
1131}
1132
1133void
1134ctl_shutdown(void)
1135{
1136	struct ctl_softc *softc;
1137	struct ctl_lun *lun, *next_lun;
1138	struct ctl_io_pool *pool;
1139
1140	softc = (struct ctl_softc *)control_softc;
1141
1142	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1143		printf("ctl: ioctl front end deregistration failed\n");
1144
1145	mtx_lock(&softc->ctl_lock);
1146
1147	/*
1148	 * Free up each LUN.
1149	 */
1150	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1151		next_lun = STAILQ_NEXT(lun, links);
1152		ctl_free_lun(lun);
1153	}
1154
1155	mtx_unlock(&softc->ctl_lock);
1156
1157	ctl_frontend_deregister(&ioctl_frontend);
1158
1159	/*
1160	 * This will rip the rug out from under any FETDs or anyone else
1161	 * that has a pool allocated.  Since we increment our module
1162	 * refcount any time someone outside the main CTL module allocates
1163	 * a pool, we shouldn't have any problems here.  The user won't be
1164	 * able to unload the CTL module until client modules have
1165	 * successfully unloaded.
1166	 */
1167	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1168		ctl_pool_free(pool);
1169
1170#if 0
1171	ctl_shutdown_thread(softc->work_thread);
1172	mtx_destroy(&softc->queue_lock);
1173#endif
1174
1175	mtx_destroy(&softc->pool_lock);
1176	mtx_destroy(&softc->ctl_lock);
1177
1178	destroy_dev(softc->dev);
1179
1180	sysctl_ctx_free(&softc->sysctl_ctx);
1181
1182	free(control_softc, M_DEVBUF);
1183	control_softc = NULL;
1184
1185	if (bootverbose)
1186		printf("ctl: CAM Target Layer unloaded\n");
1187}
1188
1189static int
1190ctl_module_event_handler(module_t mod, int what, void *arg)
1191{
1192
1193	switch (what) {
1194	case MOD_LOAD:
1195		return (ctl_init());
1196	case MOD_UNLOAD:
1197		return (EBUSY);
1198	default:
1199		return (EOPNOTSUPP);
1200	}
1201}
1202
1203/*
1204 * XXX KDM should we do some access checks here?  Bump a reference count to
1205 * prevent a CTL module from being unloaded while someone has it open?
1206 */
1207static int
1208ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1209{
1210	return (0);
1211}
1212
1213static int
1214ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1215{
1216	return (0);
1217}
1218
1219int
1220ctl_port_enable(ctl_port_type port_type)
1221{
1222	struct ctl_softc *softc;
1223	struct ctl_port *port;
1224
1225	if (ctl_is_single == 0) {
1226		union ctl_ha_msg msg_info;
1227		int isc_retval;
1228
1229#if 0
1230		printf("%s: HA mode, synchronizing frontend enable\n",
1231		        __func__);
1232#endif
1233		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1234	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1235		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1236			printf("Sync msg send error retval %d\n", isc_retval);
1237		}
1238		if (!rcv_sync_msg) {
1239			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1240			        sizeof(msg_info), 1);
1241		}
1242#if 0
1243        	printf("CTL:Frontend Enable\n");
1244	} else {
1245		printf("%s: single mode, skipping frontend synchronization\n",
1246		        __func__);
1247#endif
1248	}
1249
1250	softc = control_softc;
1251
1252	STAILQ_FOREACH(port, &softc->port_list, links) {
1253		if (port_type & port->port_type)
1254		{
1255#if 0
1256			printf("port %d\n", port->targ_port);
1257#endif
1258			ctl_port_online(port);
1259		}
1260	}
1261
1262	return (0);
1263}
1264
1265int
1266ctl_port_disable(ctl_port_type port_type)
1267{
1268	struct ctl_softc *softc;
1269	struct ctl_port *port;
1270
1271	softc = control_softc;
1272
1273	STAILQ_FOREACH(port, &softc->port_list, links) {
1274		if (port_type & port->port_type)
1275			ctl_port_offline(port);
1276	}
1277
1278	return (0);
1279}
1280
1281/*
1282 * Returns 0 for success, 1 for failure.
1283 * Currently the only failure mode is if there aren't enough entries
1284 * allocated.  So, in case of a failure, look at num_entries_dropped,
1285 * reallocate and try again.
1286 */
1287int
1288ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1289	      int *num_entries_filled, int *num_entries_dropped,
1290	      ctl_port_type port_type, int no_virtual)
1291{
1292	struct ctl_softc *softc;
1293	struct ctl_port *port;
1294	int entries_dropped, entries_filled;
1295	int retval;
1296	int i;
1297
1298	softc = control_softc;
1299
1300	retval = 0;
1301	entries_filled = 0;
1302	entries_dropped = 0;
1303
1304	i = 0;
1305	mtx_lock(&softc->ctl_lock);
1306	STAILQ_FOREACH(port, &softc->port_list, links) {
1307		struct ctl_port_entry *entry;
1308
1309		if ((port->port_type & port_type) == 0)
1310			continue;
1311
1312		if ((no_virtual != 0)
1313		 && (port->virtual_port != 0))
1314			continue;
1315
1316		if (entries_filled >= num_entries_alloced) {
1317			entries_dropped++;
1318			continue;
1319		}
1320		entry = &entries[i];
1321
1322		entry->port_type = port->port_type;
1323		strlcpy(entry->port_name, port->port_name,
1324			sizeof(entry->port_name));
1325		entry->physical_port = port->physical_port;
1326		entry->virtual_port = port->virtual_port;
1327		entry->wwnn = port->wwnn;
1328		entry->wwpn = port->wwpn;
1329
1330		i++;
1331		entries_filled++;
1332	}
1333
1334	mtx_unlock(&softc->ctl_lock);
1335
1336	if (entries_dropped > 0)
1337		retval = 1;
1338
1339	*num_entries_dropped = entries_dropped;
1340	*num_entries_filled = entries_filled;
1341
1342	return (retval);
1343}
1344
1345static void
1346ctl_ioctl_online(void *arg)
1347{
1348	struct ctl_ioctl_info *ioctl_info;
1349
1350	ioctl_info = (struct ctl_ioctl_info *)arg;
1351
1352	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1353}
1354
1355static void
1356ctl_ioctl_offline(void *arg)
1357{
1358	struct ctl_ioctl_info *ioctl_info;
1359
1360	ioctl_info = (struct ctl_ioctl_info *)arg;
1361
1362	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1363}
1364
1365/*
1366 * Remove an initiator by port number and initiator ID.
1367 * Returns 0 for success, -1 for failure.
1368 */
1369int
1370ctl_remove_initiator(struct ctl_port *port, int iid)
1371{
1372	struct ctl_softc *softc = control_softc;
1373
1374	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1375
1376	if (iid > CTL_MAX_INIT_PER_PORT) {
1377		printf("%s: initiator ID %u > maximun %u!\n",
1378		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1379		return (-1);
1380	}
1381
1382	mtx_lock(&softc->ctl_lock);
1383	port->wwpn_iid[iid].in_use--;
1384	port->wwpn_iid[iid].last_use = time_uptime;
1385	mtx_unlock(&softc->ctl_lock);
1386
1387	return (0);
1388}
1389
1390/*
1391 * Add an initiator to the initiator map.
1392 * Returns iid for success, < 0 for failure.
1393 */
1394int
1395ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1396{
1397	struct ctl_softc *softc = control_softc;
1398	time_t best_time;
1399	int i, best;
1400
1401	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1402
1403	if (iid >= CTL_MAX_INIT_PER_PORT) {
1404		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1405		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1406		free(name, M_CTL);
1407		return (-1);
1408	}
1409
1410	mtx_lock(&softc->ctl_lock);
1411
1412	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1413		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1414			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1415				iid = i;
1416				break;
1417			}
1418			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1419			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1420				iid = i;
1421				break;
1422			}
1423		}
1424	}
1425
1426	if (iid < 0) {
1427		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1428			if (port->wwpn_iid[i].in_use == 0 &&
1429			    port->wwpn_iid[i].wwpn == 0 &&
1430			    port->wwpn_iid[i].name == NULL) {
1431				iid = i;
1432				break;
1433			}
1434		}
1435	}
1436
1437	if (iid < 0) {
1438		best = -1;
1439		best_time = INT32_MAX;
1440		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1441			if (port->wwpn_iid[i].in_use == 0) {
1442				if (port->wwpn_iid[i].last_use < best_time) {
1443					best = i;
1444					best_time = port->wwpn_iid[i].last_use;
1445				}
1446			}
1447		}
1448		iid = best;
1449	}
1450
1451	if (iid < 0) {
1452		mtx_unlock(&softc->ctl_lock);
1453		free(name, M_CTL);
1454		return (-2);
1455	}
1456
1457	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1458		/*
1459		 * This is not an error yet.
1460		 */
1461		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1462#if 0
1463			printf("%s: port %d iid %u WWPN %#jx arrived"
1464			    " again\n", __func__, port->targ_port,
1465			    iid, (uintmax_t)wwpn);
1466#endif
1467			goto take;
1468		}
1469		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1470		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1471#if 0
1472			printf("%s: port %d iid %u name '%s' arrived"
1473			    " again\n", __func__, port->targ_port,
1474			    iid, name);
1475#endif
1476			goto take;
1477		}
1478
1479		/*
1480		 * This is an error, but what do we do about it?  The
1481		 * driver is telling us we have a new WWPN for this
1482		 * initiator ID, so we pretty much need to use it.
1483		 */
1484		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1485		    " but WWPN %#jx '%s' is still at that address\n",
1486		    __func__, port->targ_port, iid, wwpn, name,
1487		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1488		    port->wwpn_iid[iid].name);
1489
1490		/*
1491		 * XXX KDM clear have_ca and ua_pending on each LUN for
1492		 * this initiator.
1493		 */
1494	}
1495take:
1496	free(port->wwpn_iid[iid].name, M_CTL);
1497	port->wwpn_iid[iid].name = name;
1498	port->wwpn_iid[iid].wwpn = wwpn;
1499	port->wwpn_iid[iid].in_use++;
1500	mtx_unlock(&softc->ctl_lock);
1501
1502	return (iid);
1503}
1504
1505static int
1506ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1507{
1508	int len;
1509
1510	switch (port->port_type) {
1511	case CTL_PORT_FC:
1512	{
1513		struct scsi_transportid_fcp *id =
1514		    (struct scsi_transportid_fcp *)buf;
1515		if (port->wwpn_iid[iid].wwpn == 0)
1516			return (0);
1517		memset(id, 0, sizeof(*id));
1518		id->format_protocol = SCSI_PROTO_FC;
1519		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1520		return (sizeof(*id));
1521	}
1522	case CTL_PORT_ISCSI:
1523	{
1524		struct scsi_transportid_iscsi_port *id =
1525		    (struct scsi_transportid_iscsi_port *)buf;
1526		if (port->wwpn_iid[iid].name == NULL)
1527			return (0);
1528		memset(id, 0, 256);
1529		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1530		    SCSI_PROTO_ISCSI;
1531		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1532		len = roundup2(min(len, 252), 4);
1533		scsi_ulto2b(len, id->additional_length);
1534		return (sizeof(*id) + len);
1535	}
1536	case CTL_PORT_SAS:
1537	{
1538		struct scsi_transportid_sas *id =
1539		    (struct scsi_transportid_sas *)buf;
1540		if (port->wwpn_iid[iid].wwpn == 0)
1541			return (0);
1542		memset(id, 0, sizeof(*id));
1543		id->format_protocol = SCSI_PROTO_SAS;
1544		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1545		return (sizeof(*id));
1546	}
1547	default:
1548	{
1549		struct scsi_transportid_spi *id =
1550		    (struct scsi_transportid_spi *)buf;
1551		memset(id, 0, sizeof(*id));
1552		id->format_protocol = SCSI_PROTO_SPI;
1553		scsi_ulto2b(iid, id->scsi_addr);
1554		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1555		return (sizeof(*id));
1556	}
1557	}
1558}
1559
1560static int
1561ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1562{
1563	return (0);
1564}
1565
1566static int
1567ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1568{
1569	return (0);
1570}
1571
1572/*
1573 * Data movement routine for the CTL ioctl frontend port.
1574 */
1575static int
1576ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1577{
1578	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1579	struct ctl_sg_entry ext_entry, kern_entry;
1580	int ext_sglen, ext_sg_entries, kern_sg_entries;
1581	int ext_sg_start, ext_offset;
1582	int len_to_copy, len_copied;
1583	int kern_watermark, ext_watermark;
1584	int ext_sglist_malloced;
1585	int i, j;
1586
1587	ext_sglist_malloced = 0;
1588	ext_sg_start = 0;
1589	ext_offset = 0;
1590
1591	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1592
1593	/*
1594	 * If this flag is set, fake the data transfer.
1595	 */
1596	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1597		ctsio->ext_data_filled = ctsio->ext_data_len;
1598		goto bailout;
1599	}
1600
1601	/*
1602	 * To simplify things here, if we have a single buffer, stick it in
1603	 * a S/G entry and just make it a single entry S/G list.
1604	 */
1605	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1606		int len_seen;
1607
1608		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1609
1610		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1611							   M_WAITOK);
1612		ext_sglist_malloced = 1;
1613		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1614				   ext_sglen) != 0) {
1615			ctl_set_internal_failure(ctsio,
1616						 /*sks_valid*/ 0,
1617						 /*retry_count*/ 0);
1618			goto bailout;
1619		}
1620		ext_sg_entries = ctsio->ext_sg_entries;
1621		len_seen = 0;
1622		for (i = 0; i < ext_sg_entries; i++) {
1623			if ((len_seen + ext_sglist[i].len) >=
1624			     ctsio->ext_data_filled) {
1625				ext_sg_start = i;
1626				ext_offset = ctsio->ext_data_filled - len_seen;
1627				break;
1628			}
1629			len_seen += ext_sglist[i].len;
1630		}
1631	} else {
1632		ext_sglist = &ext_entry;
1633		ext_sglist->addr = ctsio->ext_data_ptr;
1634		ext_sglist->len = ctsio->ext_data_len;
1635		ext_sg_entries = 1;
1636		ext_sg_start = 0;
1637		ext_offset = ctsio->ext_data_filled;
1638	}
1639
1640	if (ctsio->kern_sg_entries > 0) {
1641		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1642		kern_sg_entries = ctsio->kern_sg_entries;
1643	} else {
1644		kern_sglist = &kern_entry;
1645		kern_sglist->addr = ctsio->kern_data_ptr;
1646		kern_sglist->len = ctsio->kern_data_len;
1647		kern_sg_entries = 1;
1648	}
1649
1650
1651	kern_watermark = 0;
1652	ext_watermark = ext_offset;
1653	len_copied = 0;
1654	for (i = ext_sg_start, j = 0;
1655	     i < ext_sg_entries && j < kern_sg_entries;) {
1656		uint8_t *ext_ptr, *kern_ptr;
1657
1658		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1659				      kern_sglist[j].len - kern_watermark);
1660
1661		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1662		ext_ptr = ext_ptr + ext_watermark;
1663		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1664			/*
1665			 * XXX KDM fix this!
1666			 */
1667			panic("need to implement bus address support");
1668#if 0
1669			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1670#endif
1671		} else
1672			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1673		kern_ptr = kern_ptr + kern_watermark;
1674
1675		kern_watermark += len_to_copy;
1676		ext_watermark += len_to_copy;
1677
1678		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1679		     CTL_FLAG_DATA_IN) {
1680			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1681					 "bytes to user\n", len_to_copy));
1682			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1683					 "to %p\n", kern_ptr, ext_ptr));
1684			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1685				ctl_set_internal_failure(ctsio,
1686							 /*sks_valid*/ 0,
1687							 /*retry_count*/ 0);
1688				goto bailout;
1689			}
1690		} else {
1691			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1692					 "bytes from user\n", len_to_copy));
1693			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1694					 "to %p\n", ext_ptr, kern_ptr));
1695			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1696				ctl_set_internal_failure(ctsio,
1697							 /*sks_valid*/ 0,
1698							 /*retry_count*/0);
1699				goto bailout;
1700			}
1701		}
1702
1703		len_copied += len_to_copy;
1704
1705		if (ext_sglist[i].len == ext_watermark) {
1706			i++;
1707			ext_watermark = 0;
1708		}
1709
1710		if (kern_sglist[j].len == kern_watermark) {
1711			j++;
1712			kern_watermark = 0;
1713		}
1714	}
1715
1716	ctsio->ext_data_filled += len_copied;
1717
1718	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1719			 "kern_sg_entries: %d\n", ext_sg_entries,
1720			 kern_sg_entries));
1721	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1722			 "kern_data_len = %d\n", ctsio->ext_data_len,
1723			 ctsio->kern_data_len));
1724
1725
1726	/* XXX KDM set residual?? */
1727bailout:
1728
1729	if (ext_sglist_malloced != 0)
1730		free(ext_sglist, M_CTL);
1731
1732	return (CTL_RETVAL_COMPLETE);
1733}
1734
1735/*
1736 * Serialize a command that went down the "wrong" side, and so was sent to
1737 * this controller for execution.  The logic is a little different than the
1738 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1739 * sent back to the other side, but in the success case, we execute the
1740 * command on this side (XFER mode) or tell the other side to execute it
1741 * (SER_ONLY mode).
1742 */
1743static int
1744ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1745{
1746	struct ctl_softc *ctl_softc;
1747	union ctl_ha_msg msg_info;
1748	struct ctl_lun *lun;
1749	int retval = 0;
1750	uint32_t targ_lun;
1751
1752	ctl_softc = control_softc;
1753
1754	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1755	lun = ctl_softc->ctl_luns[targ_lun];
1756	if (lun==NULL)
1757	{
1758		/*
1759		 * Why isn't LUN defined? The other side wouldn't
1760		 * send a cmd if the LUN is undefined.
1761		 */
1762		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1763
1764		/* "Logical unit not supported" */
1765		ctl_set_sense_data(&msg_info.scsi.sense_data,
1766				   lun,
1767				   /*sense_format*/SSD_TYPE_NONE,
1768				   /*current_error*/ 1,
1769				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1770				   /*asc*/ 0x25,
1771				   /*ascq*/ 0x00,
1772				   SSD_ELEM_NONE);
1773
1774		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1775		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1776		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1777		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1778		msg_info.hdr.serializing_sc = NULL;
1779		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1780	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1781				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1782		}
1783		return(1);
1784
1785	}
1786
1787	mtx_lock(&lun->lun_lock);
1788    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1789
1790	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1791		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1792		 ooa_links))) {
1793	case CTL_ACTION_BLOCK:
1794		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1795		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1796				  blocked_links);
1797		break;
1798	case CTL_ACTION_PASS:
1799	case CTL_ACTION_SKIP:
1800		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1801			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1802			ctl_enqueue_rtr((union ctl_io *)ctsio);
1803		} else {
1804
1805			/* send msg back to other side */
1806			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1807			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1808			msg_info.hdr.msg_type = CTL_MSG_R2R;
1809#if 0
1810			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1811#endif
1812		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1813			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1814			}
1815		}
1816		break;
1817	case CTL_ACTION_OVERLAP:
1818		/* OVERLAPPED COMMANDS ATTEMPTED */
1819		ctl_set_sense_data(&msg_info.scsi.sense_data,
1820				   lun,
1821				   /*sense_format*/SSD_TYPE_NONE,
1822				   /*current_error*/ 1,
1823				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1824				   /*asc*/ 0x4E,
1825				   /*ascq*/ 0x00,
1826				   SSD_ELEM_NONE);
1827
1828		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1829		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1830		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1831		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1832		msg_info.hdr.serializing_sc = NULL;
1833		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1834#if 0
1835		printf("BAD JUJU:Major Bummer Overlap\n");
1836#endif
1837		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1838		retval = 1;
1839		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1840		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1841		}
1842		break;
1843	case CTL_ACTION_OVERLAP_TAG:
1844		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1845		ctl_set_sense_data(&msg_info.scsi.sense_data,
1846				   lun,
1847				   /*sense_format*/SSD_TYPE_NONE,
1848				   /*current_error*/ 1,
1849				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1850				   /*asc*/ 0x4D,
1851				   /*ascq*/ ctsio->tag_num & 0xff,
1852				   SSD_ELEM_NONE);
1853
1854		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1855		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1856		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1857		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1858		msg_info.hdr.serializing_sc = NULL;
1859		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1860#if 0
1861		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1862#endif
1863		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1864		retval = 1;
1865		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1866		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1867		}
1868		break;
1869	case CTL_ACTION_ERROR:
1870	default:
1871		/* "Internal target failure" */
1872		ctl_set_sense_data(&msg_info.scsi.sense_data,
1873				   lun,
1874				   /*sense_format*/SSD_TYPE_NONE,
1875				   /*current_error*/ 1,
1876				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1877				   /*asc*/ 0x44,
1878				   /*ascq*/ 0x00,
1879				   SSD_ELEM_NONE);
1880
1881		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1882		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1883		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1884		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1885		msg_info.hdr.serializing_sc = NULL;
1886		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1887#if 0
1888		printf("BAD JUJU:Major Bummer HW Error\n");
1889#endif
1890		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1891		retval = 1;
1892		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1893		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1894		}
1895		break;
1896	}
1897	mtx_unlock(&lun->lun_lock);
1898	return (retval);
1899}
1900
1901static int
1902ctl_ioctl_submit_wait(union ctl_io *io)
1903{
1904	struct ctl_fe_ioctl_params params;
1905	ctl_fe_ioctl_state last_state;
1906	int done, retval;
1907
1908	retval = 0;
1909
1910	bzero(&params, sizeof(params));
1911
1912	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1913	cv_init(&params.sem, "ctlioccv");
1914	params.state = CTL_IOCTL_INPROG;
1915	last_state = params.state;
1916
1917	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1918
1919	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1920
1921	/* This shouldn't happen */
1922	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1923		return (retval);
1924
1925	done = 0;
1926
1927	do {
1928		mtx_lock(&params.ioctl_mtx);
1929		/*
1930		 * Check the state here, and don't sleep if the state has
1931		 * already changed (i.e. wakeup has already occured, but we
1932		 * weren't waiting yet).
1933		 */
1934		if (params.state == last_state) {
1935			/* XXX KDM cv_wait_sig instead? */
1936			cv_wait(&params.sem, &params.ioctl_mtx);
1937		}
1938		last_state = params.state;
1939
1940		switch (params.state) {
1941		case CTL_IOCTL_INPROG:
1942			/* Why did we wake up? */
1943			/* XXX KDM error here? */
1944			mtx_unlock(&params.ioctl_mtx);
1945			break;
1946		case CTL_IOCTL_DATAMOVE:
1947			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1948
1949			/*
1950			 * change last_state back to INPROG to avoid
1951			 * deadlock on subsequent data moves.
1952			 */
1953			params.state = last_state = CTL_IOCTL_INPROG;
1954
1955			mtx_unlock(&params.ioctl_mtx);
1956			ctl_ioctl_do_datamove(&io->scsiio);
1957			/*
1958			 * Note that in some cases, most notably writes,
1959			 * this will queue the I/O and call us back later.
1960			 * In other cases, generally reads, this routine
1961			 * will immediately call back and wake us up,
1962			 * probably using our own context.
1963			 */
1964			io->scsiio.be_move_done(io);
1965			break;
1966		case CTL_IOCTL_DONE:
1967			mtx_unlock(&params.ioctl_mtx);
1968			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1969			done = 1;
1970			break;
1971		default:
1972			mtx_unlock(&params.ioctl_mtx);
1973			/* XXX KDM error here? */
1974			break;
1975		}
1976	} while (done == 0);
1977
1978	mtx_destroy(&params.ioctl_mtx);
1979	cv_destroy(&params.sem);
1980
1981	return (CTL_RETVAL_COMPLETE);
1982}
1983
1984static void
1985ctl_ioctl_datamove(union ctl_io *io)
1986{
1987	struct ctl_fe_ioctl_params *params;
1988
1989	params = (struct ctl_fe_ioctl_params *)
1990		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1991
1992	mtx_lock(&params->ioctl_mtx);
1993	params->state = CTL_IOCTL_DATAMOVE;
1994	cv_broadcast(&params->sem);
1995	mtx_unlock(&params->ioctl_mtx);
1996}
1997
1998static void
1999ctl_ioctl_done(union ctl_io *io)
2000{
2001	struct ctl_fe_ioctl_params *params;
2002
2003	params = (struct ctl_fe_ioctl_params *)
2004		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2005
2006	mtx_lock(&params->ioctl_mtx);
2007	params->state = CTL_IOCTL_DONE;
2008	cv_broadcast(&params->sem);
2009	mtx_unlock(&params->ioctl_mtx);
2010}
2011
2012static void
2013ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2014{
2015	struct ctl_fe_ioctl_startstop_info *sd_info;
2016
2017	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2018
2019	sd_info->hs_info.status = metatask->status;
2020	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2021	sd_info->hs_info.luns_complete =
2022		metatask->taskinfo.startstop.luns_complete;
2023	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2024
2025	cv_broadcast(&sd_info->sem);
2026}
2027
2028static void
2029ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2030{
2031	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2032
2033	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2034
2035	mtx_lock(fe_bbr_info->lock);
2036	fe_bbr_info->bbr_info->status = metatask->status;
2037	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2038	fe_bbr_info->wakeup_done = 1;
2039	mtx_unlock(fe_bbr_info->lock);
2040
2041	cv_broadcast(&fe_bbr_info->sem);
2042}
2043
2044/*
2045 * Returns 0 for success, errno for failure.
2046 */
2047static int
2048ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2049		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2050{
2051	union ctl_io *io;
2052	int retval;
2053
2054	retval = 0;
2055
2056	mtx_lock(&lun->lun_lock);
2057	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2058	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2059	     ooa_links)) {
2060		struct ctl_ooa_entry *entry;
2061
2062		/*
2063		 * If we've got more than we can fit, just count the
2064		 * remaining entries.
2065		 */
2066		if (*cur_fill_num >= ooa_hdr->alloc_num)
2067			continue;
2068
2069		entry = &kern_entries[*cur_fill_num];
2070
2071		entry->tag_num = io->scsiio.tag_num;
2072		entry->lun_num = lun->lun;
2073#ifdef CTL_TIME_IO
2074		entry->start_bt = io->io_hdr.start_bt;
2075#endif
2076		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2077		entry->cdb_len = io->scsiio.cdb_len;
2078		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2079			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2080
2081		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2082			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2083
2084		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2085			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2086
2087		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2088			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2089
2090		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2091			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2092	}
2093	mtx_unlock(&lun->lun_lock);
2094
2095	return (retval);
2096}
2097
2098static void *
2099ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2100		 size_t error_str_len)
2101{
2102	void *kptr;
2103
2104	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2105
2106	if (copyin(user_addr, kptr, len) != 0) {
2107		snprintf(error_str, error_str_len, "Error copying %d bytes "
2108			 "from user address %p to kernel address %p", len,
2109			 user_addr, kptr);
2110		free(kptr, M_CTL);
2111		return (NULL);
2112	}
2113
2114	return (kptr);
2115}
2116
2117static void
2118ctl_free_args(int num_args, struct ctl_be_arg *args)
2119{
2120	int i;
2121
2122	if (args == NULL)
2123		return;
2124
2125	for (i = 0; i < num_args; i++) {
2126		free(args[i].kname, M_CTL);
2127		free(args[i].kvalue, M_CTL);
2128	}
2129
2130	free(args, M_CTL);
2131}
2132
2133static struct ctl_be_arg *
2134ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2135		char *error_str, size_t error_str_len)
2136{
2137	struct ctl_be_arg *args;
2138	int i;
2139
2140	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2141				error_str, error_str_len);
2142
2143	if (args == NULL)
2144		goto bailout;
2145
2146	for (i = 0; i < num_args; i++) {
2147		args[i].kname = NULL;
2148		args[i].kvalue = NULL;
2149	}
2150
2151	for (i = 0; i < num_args; i++) {
2152		uint8_t *tmpptr;
2153
2154		args[i].kname = ctl_copyin_alloc(args[i].name,
2155			args[i].namelen, error_str, error_str_len);
2156		if (args[i].kname == NULL)
2157			goto bailout;
2158
2159		if (args[i].kname[args[i].namelen - 1] != '\0') {
2160			snprintf(error_str, error_str_len, "Argument %d "
2161				 "name is not NUL-terminated", i);
2162			goto bailout;
2163		}
2164
2165		if (args[i].flags & CTL_BEARG_RD) {
2166			tmpptr = ctl_copyin_alloc(args[i].value,
2167				args[i].vallen, error_str, error_str_len);
2168			if (tmpptr == NULL)
2169				goto bailout;
2170			if ((args[i].flags & CTL_BEARG_ASCII)
2171			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2172				snprintf(error_str, error_str_len, "Argument "
2173				    "%d value is not NUL-terminated", i);
2174				goto bailout;
2175			}
2176			args[i].kvalue = tmpptr;
2177		} else {
2178			args[i].kvalue = malloc(args[i].vallen,
2179			    M_CTL, M_WAITOK | M_ZERO);
2180		}
2181	}
2182
2183	return (args);
2184bailout:
2185
2186	ctl_free_args(num_args, args);
2187
2188	return (NULL);
2189}
2190
2191static void
2192ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2193{
2194	int i;
2195
2196	for (i = 0; i < num_args; i++) {
2197		if (args[i].flags & CTL_BEARG_WR)
2198			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2199	}
2200}
2201
2202/*
2203 * Escape characters that are illegal or not recommended in XML.
2204 */
2205int
2206ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2207{
2208	int retval;
2209
2210	retval = 0;
2211
2212	for (; *str; str++) {
2213		switch (*str) {
2214		case '&':
2215			retval = sbuf_printf(sb, "&amp;");
2216			break;
2217		case '>':
2218			retval = sbuf_printf(sb, "&gt;");
2219			break;
2220		case '<':
2221			retval = sbuf_printf(sb, "&lt;");
2222			break;
2223		default:
2224			retval = sbuf_putc(sb, *str);
2225			break;
2226		}
2227
2228		if (retval != 0)
2229			break;
2230
2231	}
2232
2233	return (retval);
2234}
2235
2236static int
2237ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2238	  struct thread *td)
2239{
2240	struct ctl_softc *softc;
2241	int retval;
2242
2243	softc = control_softc;
2244
2245	retval = 0;
2246
2247	switch (cmd) {
2248	case CTL_IO: {
2249		union ctl_io *io;
2250		void *pool_tmp;
2251
2252		/*
2253		 * If we haven't been "enabled", don't allow any SCSI I/O
2254		 * to this FETD.
2255		 */
2256		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2257			retval = EPERM;
2258			break;
2259		}
2260
2261		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2262		if (io == NULL) {
2263			printf("ctl_ioctl: can't allocate ctl_io!\n");
2264			retval = ENOSPC;
2265			break;
2266		}
2267
2268		/*
2269		 * Need to save the pool reference so it doesn't get
2270		 * spammed by the user's ctl_io.
2271		 */
2272		pool_tmp = io->io_hdr.pool;
2273
2274		memcpy(io, (void *)addr, sizeof(*io));
2275
2276		io->io_hdr.pool = pool_tmp;
2277		/*
2278		 * No status yet, so make sure the status is set properly.
2279		 */
2280		io->io_hdr.status = CTL_STATUS_NONE;
2281
2282		/*
2283		 * The user sets the initiator ID, target and LUN IDs.
2284		 */
2285		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2286		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2287		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2288		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2289			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2290
2291		retval = ctl_ioctl_submit_wait(io);
2292
2293		if (retval != 0) {
2294			ctl_free_io(io);
2295			break;
2296		}
2297
2298		memcpy((void *)addr, io, sizeof(*io));
2299
2300		/* return this to our pool */
2301		ctl_free_io(io);
2302
2303		break;
2304	}
2305	case CTL_ENABLE_PORT:
2306	case CTL_DISABLE_PORT:
2307	case CTL_SET_PORT_WWNS: {
2308		struct ctl_port *port;
2309		struct ctl_port_entry *entry;
2310
2311		entry = (struct ctl_port_entry *)addr;
2312
2313		mtx_lock(&softc->ctl_lock);
2314		STAILQ_FOREACH(port, &softc->port_list, links) {
2315			int action, done;
2316
2317			action = 0;
2318			done = 0;
2319
2320			if ((entry->port_type == CTL_PORT_NONE)
2321			 && (entry->targ_port == port->targ_port)) {
2322				/*
2323				 * If the user only wants to enable or
2324				 * disable or set WWNs on a specific port,
2325				 * do the operation and we're done.
2326				 */
2327				action = 1;
2328				done = 1;
2329			} else if (entry->port_type & port->port_type) {
2330				/*
2331				 * Compare the user's type mask with the
2332				 * particular frontend type to see if we
2333				 * have a match.
2334				 */
2335				action = 1;
2336				done = 0;
2337
2338				/*
2339				 * Make sure the user isn't trying to set
2340				 * WWNs on multiple ports at the same time.
2341				 */
2342				if (cmd == CTL_SET_PORT_WWNS) {
2343					printf("%s: Can't set WWNs on "
2344					       "multiple ports\n", __func__);
2345					retval = EINVAL;
2346					break;
2347				}
2348			}
2349			if (action != 0) {
2350				/*
2351				 * XXX KDM we have to drop the lock here,
2352				 * because the online/offline operations
2353				 * can potentially block.  We need to
2354				 * reference count the frontends so they
2355				 * can't go away,
2356				 */
2357				mtx_unlock(&softc->ctl_lock);
2358
2359				if (cmd == CTL_ENABLE_PORT) {
2360					struct ctl_lun *lun;
2361
2362					STAILQ_FOREACH(lun, &softc->lun_list,
2363						       links) {
2364						port->lun_enable(port->targ_lun_arg,
2365						    lun->target,
2366						    lun->lun);
2367					}
2368
2369					ctl_port_online(port);
2370				} else if (cmd == CTL_DISABLE_PORT) {
2371					struct ctl_lun *lun;
2372
2373					ctl_port_offline(port);
2374
2375					STAILQ_FOREACH(lun, &softc->lun_list,
2376						       links) {
2377						port->lun_disable(
2378						    port->targ_lun_arg,
2379						    lun->target,
2380						    lun->lun);
2381					}
2382				}
2383
2384				mtx_lock(&softc->ctl_lock);
2385
2386				if (cmd == CTL_SET_PORT_WWNS)
2387					ctl_port_set_wwns(port,
2388					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2389					    1 : 0, entry->wwnn,
2390					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2391					    1 : 0, entry->wwpn);
2392			}
2393			if (done != 0)
2394				break;
2395		}
2396		mtx_unlock(&softc->ctl_lock);
2397		break;
2398	}
2399	case CTL_GET_PORT_LIST: {
2400		struct ctl_port *port;
2401		struct ctl_port_list *list;
2402		int i;
2403
2404		list = (struct ctl_port_list *)addr;
2405
2406		if (list->alloc_len != (list->alloc_num *
2407		    sizeof(struct ctl_port_entry))) {
2408			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2409			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2410			       "%zu\n", __func__, list->alloc_len,
2411			       list->alloc_num, sizeof(struct ctl_port_entry));
2412			retval = EINVAL;
2413			break;
2414		}
2415		list->fill_len = 0;
2416		list->fill_num = 0;
2417		list->dropped_num = 0;
2418		i = 0;
2419		mtx_lock(&softc->ctl_lock);
2420		STAILQ_FOREACH(port, &softc->port_list, links) {
2421			struct ctl_port_entry entry, *list_entry;
2422
2423			if (list->fill_num >= list->alloc_num) {
2424				list->dropped_num++;
2425				continue;
2426			}
2427
2428			entry.port_type = port->port_type;
2429			strlcpy(entry.port_name, port->port_name,
2430				sizeof(entry.port_name));
2431			entry.targ_port = port->targ_port;
2432			entry.physical_port = port->physical_port;
2433			entry.virtual_port = port->virtual_port;
2434			entry.wwnn = port->wwnn;
2435			entry.wwpn = port->wwpn;
2436			if (port->status & CTL_PORT_STATUS_ONLINE)
2437				entry.online = 1;
2438			else
2439				entry.online = 0;
2440
2441			list_entry = &list->entries[i];
2442
2443			retval = copyout(&entry, list_entry, sizeof(entry));
2444			if (retval != 0) {
2445				printf("%s: CTL_GET_PORT_LIST: copyout "
2446				       "returned %d\n", __func__, retval);
2447				break;
2448			}
2449			i++;
2450			list->fill_num++;
2451			list->fill_len += sizeof(entry);
2452		}
2453		mtx_unlock(&softc->ctl_lock);
2454
2455		/*
2456		 * If this is non-zero, we had a copyout fault, so there's
2457		 * probably no point in attempting to set the status inside
2458		 * the structure.
2459		 */
2460		if (retval != 0)
2461			break;
2462
2463		if (list->dropped_num > 0)
2464			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2465		else
2466			list->status = CTL_PORT_LIST_OK;
2467		break;
2468	}
2469	case CTL_DUMP_OOA: {
2470		struct ctl_lun *lun;
2471		union ctl_io *io;
2472		char printbuf[128];
2473		struct sbuf sb;
2474
2475		mtx_lock(&softc->ctl_lock);
2476		printf("Dumping OOA queues:\n");
2477		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2478			mtx_lock(&lun->lun_lock);
2479			for (io = (union ctl_io *)TAILQ_FIRST(
2480			     &lun->ooa_queue); io != NULL;
2481			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2482			     ooa_links)) {
2483				sbuf_new(&sb, printbuf, sizeof(printbuf),
2484					 SBUF_FIXEDLEN);
2485				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2486					    (intmax_t)lun->lun,
2487					    io->scsiio.tag_num,
2488					    (io->io_hdr.flags &
2489					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2490					    (io->io_hdr.flags &
2491					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2492					    (io->io_hdr.flags &
2493					    CTL_FLAG_ABORT) ? " ABORT" : "",
2494			                    (io->io_hdr.flags &
2495		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2496				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2497				sbuf_finish(&sb);
2498				printf("%s\n", sbuf_data(&sb));
2499			}
2500			mtx_unlock(&lun->lun_lock);
2501		}
2502		printf("OOA queues dump done\n");
2503		mtx_unlock(&softc->ctl_lock);
2504		break;
2505	}
2506	case CTL_GET_OOA: {
2507		struct ctl_lun *lun;
2508		struct ctl_ooa *ooa_hdr;
2509		struct ctl_ooa_entry *entries;
2510		uint32_t cur_fill_num;
2511
2512		ooa_hdr = (struct ctl_ooa *)addr;
2513
2514		if ((ooa_hdr->alloc_len == 0)
2515		 || (ooa_hdr->alloc_num == 0)) {
2516			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2517			       "must be non-zero\n", __func__,
2518			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2519			retval = EINVAL;
2520			break;
2521		}
2522
2523		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2524		    sizeof(struct ctl_ooa_entry))) {
2525			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2526			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2527			       __func__, ooa_hdr->alloc_len,
2528			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2529			retval = EINVAL;
2530			break;
2531		}
2532
2533		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2534		if (entries == NULL) {
2535			printf("%s: could not allocate %d bytes for OOA "
2536			       "dump\n", __func__, ooa_hdr->alloc_len);
2537			retval = ENOMEM;
2538			break;
2539		}
2540
2541		mtx_lock(&softc->ctl_lock);
2542		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2543		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2544		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2545			mtx_unlock(&softc->ctl_lock);
2546			free(entries, M_CTL);
2547			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2548			       __func__, (uintmax_t)ooa_hdr->lun_num);
2549			retval = EINVAL;
2550			break;
2551		}
2552
2553		cur_fill_num = 0;
2554
2555		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2556			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2557				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2558					ooa_hdr, entries);
2559				if (retval != 0)
2560					break;
2561			}
2562			if (retval != 0) {
2563				mtx_unlock(&softc->ctl_lock);
2564				free(entries, M_CTL);
2565				break;
2566			}
2567		} else {
2568			lun = softc->ctl_luns[ooa_hdr->lun_num];
2569
2570			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2571						    entries);
2572		}
2573		mtx_unlock(&softc->ctl_lock);
2574
2575		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2576		ooa_hdr->fill_len = ooa_hdr->fill_num *
2577			sizeof(struct ctl_ooa_entry);
2578		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2579		if (retval != 0) {
2580			printf("%s: error copying out %d bytes for OOA dump\n",
2581			       __func__, ooa_hdr->fill_len);
2582		}
2583
2584		getbintime(&ooa_hdr->cur_bt);
2585
2586		if (cur_fill_num > ooa_hdr->alloc_num) {
2587			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2588			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2589		} else {
2590			ooa_hdr->dropped_num = 0;
2591			ooa_hdr->status = CTL_OOA_OK;
2592		}
2593
2594		free(entries, M_CTL);
2595		break;
2596	}
2597	case CTL_CHECK_OOA: {
2598		union ctl_io *io;
2599		struct ctl_lun *lun;
2600		struct ctl_ooa_info *ooa_info;
2601
2602
2603		ooa_info = (struct ctl_ooa_info *)addr;
2604
2605		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2606			ooa_info->status = CTL_OOA_INVALID_LUN;
2607			break;
2608		}
2609		mtx_lock(&softc->ctl_lock);
2610		lun = softc->ctl_luns[ooa_info->lun_id];
2611		if (lun == NULL) {
2612			mtx_unlock(&softc->ctl_lock);
2613			ooa_info->status = CTL_OOA_INVALID_LUN;
2614			break;
2615		}
2616		mtx_lock(&lun->lun_lock);
2617		mtx_unlock(&softc->ctl_lock);
2618		ooa_info->num_entries = 0;
2619		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2620		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2621		     &io->io_hdr, ooa_links)) {
2622			ooa_info->num_entries++;
2623		}
2624		mtx_unlock(&lun->lun_lock);
2625
2626		ooa_info->status = CTL_OOA_SUCCESS;
2627
2628		break;
2629	}
2630	case CTL_HARD_START:
2631	case CTL_HARD_STOP: {
2632		struct ctl_fe_ioctl_startstop_info ss_info;
2633		struct cfi_metatask *metatask;
2634		struct mtx hs_mtx;
2635
2636		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2637
2638		cv_init(&ss_info.sem, "hard start/stop cv" );
2639
2640		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2641		if (metatask == NULL) {
2642			retval = ENOMEM;
2643			mtx_destroy(&hs_mtx);
2644			break;
2645		}
2646
2647		if (cmd == CTL_HARD_START)
2648			metatask->tasktype = CFI_TASK_STARTUP;
2649		else
2650			metatask->tasktype = CFI_TASK_SHUTDOWN;
2651
2652		metatask->callback = ctl_ioctl_hard_startstop_callback;
2653		metatask->callback_arg = &ss_info;
2654
2655		cfi_action(metatask);
2656
2657		/* Wait for the callback */
2658		mtx_lock(&hs_mtx);
2659		cv_wait_sig(&ss_info.sem, &hs_mtx);
2660		mtx_unlock(&hs_mtx);
2661
2662		/*
2663		 * All information has been copied from the metatask by the
2664		 * time cv_broadcast() is called, so we free the metatask here.
2665		 */
2666		cfi_free_metatask(metatask);
2667
2668		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2669
2670		mtx_destroy(&hs_mtx);
2671		break;
2672	}
2673	case CTL_BBRREAD: {
2674		struct ctl_bbrread_info *bbr_info;
2675		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2676		struct mtx bbr_mtx;
2677		struct cfi_metatask *metatask;
2678
2679		bbr_info = (struct ctl_bbrread_info *)addr;
2680
2681		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2682
2683		bzero(&bbr_mtx, sizeof(bbr_mtx));
2684		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2685
2686		fe_bbr_info.bbr_info = bbr_info;
2687		fe_bbr_info.lock = &bbr_mtx;
2688
2689		cv_init(&fe_bbr_info.sem, "BBR read cv");
2690		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2691
2692		if (metatask == NULL) {
2693			mtx_destroy(&bbr_mtx);
2694			cv_destroy(&fe_bbr_info.sem);
2695			retval = ENOMEM;
2696			break;
2697		}
2698		metatask->tasktype = CFI_TASK_BBRREAD;
2699		metatask->callback = ctl_ioctl_bbrread_callback;
2700		metatask->callback_arg = &fe_bbr_info;
2701		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2702		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2703		metatask->taskinfo.bbrread.len = bbr_info->len;
2704
2705		cfi_action(metatask);
2706
2707		mtx_lock(&bbr_mtx);
2708		while (fe_bbr_info.wakeup_done == 0)
2709			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2710		mtx_unlock(&bbr_mtx);
2711
2712		bbr_info->status = metatask->status;
2713		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2714		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2715		memcpy(&bbr_info->sense_data,
2716		       &metatask->taskinfo.bbrread.sense_data,
2717		       ctl_min(sizeof(bbr_info->sense_data),
2718			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2719
2720		cfi_free_metatask(metatask);
2721
2722		mtx_destroy(&bbr_mtx);
2723		cv_destroy(&fe_bbr_info.sem);
2724
2725		break;
2726	}
2727	case CTL_DELAY_IO: {
2728		struct ctl_io_delay_info *delay_info;
2729#ifdef CTL_IO_DELAY
2730		struct ctl_lun *lun;
2731#endif /* CTL_IO_DELAY */
2732
2733		delay_info = (struct ctl_io_delay_info *)addr;
2734
2735#ifdef CTL_IO_DELAY
2736		mtx_lock(&softc->ctl_lock);
2737
2738		if ((delay_info->lun_id > CTL_MAX_LUNS)
2739		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2740			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2741		} else {
2742			lun = softc->ctl_luns[delay_info->lun_id];
2743			mtx_lock(&lun->lun_lock);
2744
2745			delay_info->status = CTL_DELAY_STATUS_OK;
2746
2747			switch (delay_info->delay_type) {
2748			case CTL_DELAY_TYPE_CONT:
2749				break;
2750			case CTL_DELAY_TYPE_ONESHOT:
2751				break;
2752			default:
2753				delay_info->status =
2754					CTL_DELAY_STATUS_INVALID_TYPE;
2755				break;
2756			}
2757
2758			switch (delay_info->delay_loc) {
2759			case CTL_DELAY_LOC_DATAMOVE:
2760				lun->delay_info.datamove_type =
2761					delay_info->delay_type;
2762				lun->delay_info.datamove_delay =
2763					delay_info->delay_secs;
2764				break;
2765			case CTL_DELAY_LOC_DONE:
2766				lun->delay_info.done_type =
2767					delay_info->delay_type;
2768				lun->delay_info.done_delay =
2769					delay_info->delay_secs;
2770				break;
2771			default:
2772				delay_info->status =
2773					CTL_DELAY_STATUS_INVALID_LOC;
2774				break;
2775			}
2776			mtx_unlock(&lun->lun_lock);
2777		}
2778
2779		mtx_unlock(&softc->ctl_lock);
2780#else
2781		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2782#endif /* CTL_IO_DELAY */
2783		break;
2784	}
2785	case CTL_REALSYNC_SET: {
2786		int *syncstate;
2787
2788		syncstate = (int *)addr;
2789
2790		mtx_lock(&softc->ctl_lock);
2791		switch (*syncstate) {
2792		case 0:
2793			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2794			break;
2795		case 1:
2796			softc->flags |= CTL_FLAG_REAL_SYNC;
2797			break;
2798		default:
2799			retval = EINVAL;
2800			break;
2801		}
2802		mtx_unlock(&softc->ctl_lock);
2803		break;
2804	}
2805	case CTL_REALSYNC_GET: {
2806		int *syncstate;
2807
2808		syncstate = (int*)addr;
2809
2810		mtx_lock(&softc->ctl_lock);
2811		if (softc->flags & CTL_FLAG_REAL_SYNC)
2812			*syncstate = 1;
2813		else
2814			*syncstate = 0;
2815		mtx_unlock(&softc->ctl_lock);
2816
2817		break;
2818	}
2819	case CTL_SETSYNC:
2820	case CTL_GETSYNC: {
2821		struct ctl_sync_info *sync_info;
2822		struct ctl_lun *lun;
2823
2824		sync_info = (struct ctl_sync_info *)addr;
2825
2826		mtx_lock(&softc->ctl_lock);
2827		lun = softc->ctl_luns[sync_info->lun_id];
2828		if (lun == NULL) {
2829			mtx_unlock(&softc->ctl_lock);
2830			sync_info->status = CTL_GS_SYNC_NO_LUN;
2831		}
2832		/*
2833		 * Get or set the sync interval.  We're not bounds checking
2834		 * in the set case, hopefully the user won't do something
2835		 * silly.
2836		 */
2837		mtx_lock(&lun->lun_lock);
2838		mtx_unlock(&softc->ctl_lock);
2839		if (cmd == CTL_GETSYNC)
2840			sync_info->sync_interval = lun->sync_interval;
2841		else
2842			lun->sync_interval = sync_info->sync_interval;
2843		mtx_unlock(&lun->lun_lock);
2844
2845		sync_info->status = CTL_GS_SYNC_OK;
2846
2847		break;
2848	}
2849	case CTL_GETSTATS: {
2850		struct ctl_stats *stats;
2851		struct ctl_lun *lun;
2852		int i;
2853
2854		stats = (struct ctl_stats *)addr;
2855
2856		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2857		     stats->alloc_len) {
2858			stats->status = CTL_SS_NEED_MORE_SPACE;
2859			stats->num_luns = softc->num_luns;
2860			break;
2861		}
2862		/*
2863		 * XXX KDM no locking here.  If the LUN list changes,
2864		 * things can blow up.
2865		 */
2866		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2867		     i++, lun = STAILQ_NEXT(lun, links)) {
2868			retval = copyout(&lun->stats, &stats->lun_stats[i],
2869					 sizeof(lun->stats));
2870			if (retval != 0)
2871				break;
2872		}
2873		stats->num_luns = softc->num_luns;
2874		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2875				 softc->num_luns;
2876		stats->status = CTL_SS_OK;
2877#ifdef CTL_TIME_IO
2878		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2879#else
2880		stats->flags = CTL_STATS_FLAG_NONE;
2881#endif
2882		getnanouptime(&stats->timestamp);
2883		break;
2884	}
2885	case CTL_ERROR_INJECT: {
2886		struct ctl_error_desc *err_desc, *new_err_desc;
2887		struct ctl_lun *lun;
2888
2889		err_desc = (struct ctl_error_desc *)addr;
2890
2891		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2892				      M_WAITOK | M_ZERO);
2893		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2894
2895		mtx_lock(&softc->ctl_lock);
2896		lun = softc->ctl_luns[err_desc->lun_id];
2897		if (lun == NULL) {
2898			mtx_unlock(&softc->ctl_lock);
2899			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2900			       __func__, (uintmax_t)err_desc->lun_id);
2901			retval = EINVAL;
2902			break;
2903		}
2904		mtx_lock(&lun->lun_lock);
2905		mtx_unlock(&softc->ctl_lock);
2906
2907		/*
2908		 * We could do some checking here to verify the validity
2909		 * of the request, but given the complexity of error
2910		 * injection requests, the checking logic would be fairly
2911		 * complex.
2912		 *
2913		 * For now, if the request is invalid, it just won't get
2914		 * executed and might get deleted.
2915		 */
2916		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2917
2918		/*
2919		 * XXX KDM check to make sure the serial number is unique,
2920		 * in case we somehow manage to wrap.  That shouldn't
2921		 * happen for a very long time, but it's the right thing to
2922		 * do.
2923		 */
2924		new_err_desc->serial = lun->error_serial;
2925		err_desc->serial = lun->error_serial;
2926		lun->error_serial++;
2927
2928		mtx_unlock(&lun->lun_lock);
2929		break;
2930	}
2931	case CTL_ERROR_INJECT_DELETE: {
2932		struct ctl_error_desc *delete_desc, *desc, *desc2;
2933		struct ctl_lun *lun;
2934		int delete_done;
2935
2936		delete_desc = (struct ctl_error_desc *)addr;
2937		delete_done = 0;
2938
2939		mtx_lock(&softc->ctl_lock);
2940		lun = softc->ctl_luns[delete_desc->lun_id];
2941		if (lun == NULL) {
2942			mtx_unlock(&softc->ctl_lock);
2943			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2944			       __func__, (uintmax_t)delete_desc->lun_id);
2945			retval = EINVAL;
2946			break;
2947		}
2948		mtx_lock(&lun->lun_lock);
2949		mtx_unlock(&softc->ctl_lock);
2950		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2951			if (desc->serial != delete_desc->serial)
2952				continue;
2953
2954			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2955				      links);
2956			free(desc, M_CTL);
2957			delete_done = 1;
2958		}
2959		mtx_unlock(&lun->lun_lock);
2960		if (delete_done == 0) {
2961			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2962			       "error serial %ju on LUN %u\n", __func__,
2963			       delete_desc->serial, delete_desc->lun_id);
2964			retval = EINVAL;
2965			break;
2966		}
2967		break;
2968	}
2969	case CTL_DUMP_STRUCTS: {
2970		int i, j, k, idx;
2971		struct ctl_port *port;
2972		struct ctl_frontend *fe;
2973
2974		mtx_lock(&softc->ctl_lock);
2975		printf("CTL Persistent Reservation information start:\n");
2976		for (i = 0; i < CTL_MAX_LUNS; i++) {
2977			struct ctl_lun *lun;
2978
2979			lun = softc->ctl_luns[i];
2980
2981			if ((lun == NULL)
2982			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2983				continue;
2984
2985			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2986				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2987					idx = j * CTL_MAX_INIT_PER_PORT + k;
2988					if (lun->per_res[idx].registered == 0)
2989						continue;
2990					printf("  LUN %d port %d iid %d key "
2991					       "%#jx\n", i, j, k,
2992					       (uintmax_t)scsi_8btou64(
2993					       lun->per_res[idx].res_key.key));
2994				}
2995			}
2996		}
2997		printf("CTL Persistent Reservation information end\n");
2998		printf("CTL Ports:\n");
2999		STAILQ_FOREACH(port, &softc->port_list, links) {
3000			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3001			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3002			       port->frontend->name, port->port_type,
3003			       port->physical_port, port->virtual_port,
3004			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3005			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3006				if (port->wwpn_iid[j].in_use == 0 &&
3007				    port->wwpn_iid[j].wwpn == 0 &&
3008				    port->wwpn_iid[j].name == NULL)
3009					continue;
3010
3011				printf("    iid %u use %d WWPN %#jx '%s'\n",
3012				    j, port->wwpn_iid[j].in_use,
3013				    (uintmax_t)port->wwpn_iid[j].wwpn,
3014				    port->wwpn_iid[j].name);
3015			}
3016		}
3017		printf("CTL Port information end\n");
3018		mtx_unlock(&softc->ctl_lock);
3019		/*
3020		 * XXX KDM calling this without a lock.  We'd likely want
3021		 * to drop the lock before calling the frontend's dump
3022		 * routine anyway.
3023		 */
3024		printf("CTL Frontends:\n");
3025		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3026			printf("  Frontend '%s'\n", fe->name);
3027			if (fe->fe_dump != NULL)
3028				fe->fe_dump();
3029		}
3030		printf("CTL Frontend information end\n");
3031		break;
3032	}
3033	case CTL_LUN_REQ: {
3034		struct ctl_lun_req *lun_req;
3035		struct ctl_backend_driver *backend;
3036
3037		lun_req = (struct ctl_lun_req *)addr;
3038
3039		backend = ctl_backend_find(lun_req->backend);
3040		if (backend == NULL) {
3041			lun_req->status = CTL_LUN_ERROR;
3042			snprintf(lun_req->error_str,
3043				 sizeof(lun_req->error_str),
3044				 "Backend \"%s\" not found.",
3045				 lun_req->backend);
3046			break;
3047		}
3048		if (lun_req->num_be_args > 0) {
3049			lun_req->kern_be_args = ctl_copyin_args(
3050				lun_req->num_be_args,
3051				lun_req->be_args,
3052				lun_req->error_str,
3053				sizeof(lun_req->error_str));
3054			if (lun_req->kern_be_args == NULL) {
3055				lun_req->status = CTL_LUN_ERROR;
3056				break;
3057			}
3058		}
3059
3060		retval = backend->ioctl(dev, cmd, addr, flag, td);
3061
3062		if (lun_req->num_be_args > 0) {
3063			ctl_copyout_args(lun_req->num_be_args,
3064				      lun_req->kern_be_args);
3065			ctl_free_args(lun_req->num_be_args,
3066				      lun_req->kern_be_args);
3067		}
3068		break;
3069	}
3070	case CTL_LUN_LIST: {
3071		struct sbuf *sb;
3072		struct ctl_lun *lun;
3073		struct ctl_lun_list *list;
3074		struct ctl_option *opt;
3075
3076		list = (struct ctl_lun_list *)addr;
3077
3078		/*
3079		 * Allocate a fixed length sbuf here, based on the length
3080		 * of the user's buffer.  We could allocate an auto-extending
3081		 * buffer, and then tell the user how much larger our
3082		 * amount of data is than his buffer, but that presents
3083		 * some problems:
3084		 *
3085		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3086		 *     we can't hold a lock while calling them with an
3087		 *     auto-extending buffer.
3088 		 *
3089		 * 2.  There is not currently a LUN reference counting
3090		 *     mechanism, outside of outstanding transactions on
3091		 *     the LUN's OOA queue.  So a LUN could go away on us
3092		 *     while we're getting the LUN number, backend-specific
3093		 *     information, etc.  Thus, given the way things
3094		 *     currently work, we need to hold the CTL lock while
3095		 *     grabbing LUN information.
3096		 *
3097		 * So, from the user's standpoint, the best thing to do is
3098		 * allocate what he thinks is a reasonable buffer length,
3099		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3100		 * double the buffer length and try again.  (And repeat
3101		 * that until he succeeds.)
3102		 */
3103		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3104		if (sb == NULL) {
3105			list->status = CTL_LUN_LIST_ERROR;
3106			snprintf(list->error_str, sizeof(list->error_str),
3107				 "Unable to allocate %d bytes for LUN list",
3108				 list->alloc_len);
3109			break;
3110		}
3111
3112		sbuf_printf(sb, "<ctllunlist>\n");
3113
3114		mtx_lock(&softc->ctl_lock);
3115		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3116			mtx_lock(&lun->lun_lock);
3117			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3118					     (uintmax_t)lun->lun);
3119
3120			/*
3121			 * Bail out as soon as we see that we've overfilled
3122			 * the buffer.
3123			 */
3124			if (retval != 0)
3125				break;
3126
3127			retval = sbuf_printf(sb, "\t<backend_type>%s"
3128					     "</backend_type>\n",
3129					     (lun->backend == NULL) ?  "none" :
3130					     lun->backend->name);
3131
3132			if (retval != 0)
3133				break;
3134
3135			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3136					     lun->be_lun->lun_type);
3137
3138			if (retval != 0)
3139				break;
3140
3141			if (lun->backend == NULL) {
3142				retval = sbuf_printf(sb, "</lun>\n");
3143				if (retval != 0)
3144					break;
3145				continue;
3146			}
3147
3148			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3149					     (lun->be_lun->maxlba > 0) ?
3150					     lun->be_lun->maxlba + 1 : 0);
3151
3152			if (retval != 0)
3153				break;
3154
3155			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3156					     lun->be_lun->blocksize);
3157
3158			if (retval != 0)
3159				break;
3160
3161			retval = sbuf_printf(sb, "\t<serial_number>");
3162
3163			if (retval != 0)
3164				break;
3165
3166			retval = ctl_sbuf_printf_esc(sb,
3167						     lun->be_lun->serial_num);
3168
3169			if (retval != 0)
3170				break;
3171
3172			retval = sbuf_printf(sb, "</serial_number>\n");
3173
3174			if (retval != 0)
3175				break;
3176
3177			retval = sbuf_printf(sb, "\t<device_id>");
3178
3179			if (retval != 0)
3180				break;
3181
3182			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3183
3184			if (retval != 0)
3185				break;
3186
3187			retval = sbuf_printf(sb, "</device_id>\n");
3188
3189			if (retval != 0)
3190				break;
3191
3192			if (lun->backend->lun_info != NULL) {
3193				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3194				if (retval != 0)
3195					break;
3196			}
3197			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3198				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3199				    opt->name, opt->value, opt->name);
3200				if (retval != 0)
3201					break;
3202			}
3203
3204			retval = sbuf_printf(sb, "</lun>\n");
3205
3206			if (retval != 0)
3207				break;
3208			mtx_unlock(&lun->lun_lock);
3209		}
3210		if (lun != NULL)
3211			mtx_unlock(&lun->lun_lock);
3212		mtx_unlock(&softc->ctl_lock);
3213
3214		if ((retval != 0)
3215		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3216			retval = 0;
3217			sbuf_delete(sb);
3218			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3219			snprintf(list->error_str, sizeof(list->error_str),
3220				 "Out of space, %d bytes is too small",
3221				 list->alloc_len);
3222			break;
3223		}
3224
3225		sbuf_finish(sb);
3226
3227		retval = copyout(sbuf_data(sb), list->lun_xml,
3228				 sbuf_len(sb) + 1);
3229
3230		list->fill_len = sbuf_len(sb) + 1;
3231		list->status = CTL_LUN_LIST_OK;
3232		sbuf_delete(sb);
3233		break;
3234	}
3235	case CTL_ISCSI: {
3236		struct ctl_iscsi *ci;
3237		struct ctl_frontend *fe;
3238
3239		ci = (struct ctl_iscsi *)addr;
3240
3241		fe = ctl_frontend_find("iscsi");
3242		if (fe == NULL) {
3243			ci->status = CTL_ISCSI_ERROR;
3244			snprintf(ci->error_str, sizeof(ci->error_str),
3245			    "Frontend \"iscsi\" not found.");
3246			break;
3247		}
3248
3249		retval = fe->ioctl(dev, cmd, addr, flag, td);
3250		break;
3251	}
3252	case CTL_PORT_REQ: {
3253		struct ctl_req *req;
3254		struct ctl_frontend *fe;
3255
3256		req = (struct ctl_req *)addr;
3257
3258		fe = ctl_frontend_find(req->driver);
3259		if (fe == NULL) {
3260			req->status = CTL_LUN_ERROR;
3261			snprintf(req->error_str, sizeof(req->error_str),
3262			    "Frontend \"%s\" not found.", req->driver);
3263			break;
3264		}
3265		if (req->num_args > 0) {
3266			req->kern_args = ctl_copyin_args(req->num_args,
3267			    req->args, req->error_str, sizeof(req->error_str));
3268			if (req->kern_args == NULL) {
3269				req->status = CTL_LUN_ERROR;
3270				break;
3271			}
3272		}
3273
3274		retval = fe->ioctl(dev, cmd, addr, flag, td);
3275
3276		if (req->num_args > 0) {
3277			ctl_copyout_args(req->num_args, req->kern_args);
3278			ctl_free_args(req->num_args, req->kern_args);
3279		}
3280		break;
3281	}
3282	case CTL_PORT_LIST: {
3283		struct sbuf *sb;
3284		struct ctl_port *port;
3285		struct ctl_lun_list *list;
3286		struct ctl_option *opt;
3287
3288		list = (struct ctl_lun_list *)addr;
3289
3290		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3291		if (sb == NULL) {
3292			list->status = CTL_LUN_LIST_ERROR;
3293			snprintf(list->error_str, sizeof(list->error_str),
3294				 "Unable to allocate %d bytes for LUN list",
3295				 list->alloc_len);
3296			break;
3297		}
3298
3299		sbuf_printf(sb, "<ctlportlist>\n");
3300
3301		mtx_lock(&softc->ctl_lock);
3302		STAILQ_FOREACH(port, &softc->port_list, links) {
3303			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3304					     (uintmax_t)port->targ_port);
3305
3306			/*
3307			 * Bail out as soon as we see that we've overfilled
3308			 * the buffer.
3309			 */
3310			if (retval != 0)
3311				break;
3312
3313			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3314			    "</frontend_type>\n", port->frontend->name);
3315			if (retval != 0)
3316				break;
3317
3318			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3319					     port->port_type);
3320			if (retval != 0)
3321				break;
3322
3323			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3324			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3325			if (retval != 0)
3326				break;
3327
3328			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3329			    port->port_name);
3330			if (retval != 0)
3331				break;
3332
3333			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3334			    port->physical_port);
3335			if (retval != 0)
3336				break;
3337
3338			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3339			    port->virtual_port);
3340			if (retval != 0)
3341				break;
3342
3343			retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3344			    (uintmax_t)port->wwnn);
3345			if (retval != 0)
3346				break;
3347
3348			retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3349			    (uintmax_t)port->wwpn);
3350			if (retval != 0)
3351				break;
3352
3353			if (port->port_info != NULL) {
3354				retval = port->port_info(port->onoff_arg, sb);
3355				if (retval != 0)
3356					break;
3357			}
3358			STAILQ_FOREACH(opt, &port->options, links) {
3359				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3360				    opt->name, opt->value, opt->name);
3361				if (retval != 0)
3362					break;
3363			}
3364
3365			retval = sbuf_printf(sb, "</targ_port>\n");
3366			if (retval != 0)
3367				break;
3368		}
3369		mtx_unlock(&softc->ctl_lock);
3370
3371		if ((retval != 0)
3372		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3373			retval = 0;
3374			sbuf_delete(sb);
3375			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3376			snprintf(list->error_str, sizeof(list->error_str),
3377				 "Out of space, %d bytes is too small",
3378				 list->alloc_len);
3379			break;
3380		}
3381
3382		sbuf_finish(sb);
3383
3384		retval = copyout(sbuf_data(sb), list->lun_xml,
3385				 sbuf_len(sb) + 1);
3386
3387		list->fill_len = sbuf_len(sb) + 1;
3388		list->status = CTL_LUN_LIST_OK;
3389		sbuf_delete(sb);
3390		break;
3391	}
3392	default: {
3393		/* XXX KDM should we fix this? */
3394#if 0
3395		struct ctl_backend_driver *backend;
3396		unsigned int type;
3397		int found;
3398
3399		found = 0;
3400
3401		/*
3402		 * We encode the backend type as the ioctl type for backend
3403		 * ioctls.  So parse it out here, and then search for a
3404		 * backend of this type.
3405		 */
3406		type = _IOC_TYPE(cmd);
3407
3408		STAILQ_FOREACH(backend, &softc->be_list, links) {
3409			if (backend->type == type) {
3410				found = 1;
3411				break;
3412			}
3413		}
3414		if (found == 0) {
3415			printf("ctl: unknown ioctl command %#lx or backend "
3416			       "%d\n", cmd, type);
3417			retval = EINVAL;
3418			break;
3419		}
3420		retval = backend->ioctl(dev, cmd, addr, flag, td);
3421#endif
3422		retval = ENOTTY;
3423		break;
3424	}
3425	}
3426	return (retval);
3427}
3428
3429uint32_t
3430ctl_get_initindex(struct ctl_nexus *nexus)
3431{
3432	if (nexus->targ_port < CTL_MAX_PORTS)
3433		return (nexus->initid.id +
3434			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3435	else
3436		return (nexus->initid.id +
3437		       ((nexus->targ_port - CTL_MAX_PORTS) *
3438			CTL_MAX_INIT_PER_PORT));
3439}
3440
3441uint32_t
3442ctl_get_resindex(struct ctl_nexus *nexus)
3443{
3444	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3445}
3446
3447uint32_t
3448ctl_port_idx(int port_num)
3449{
3450	if (port_num < CTL_MAX_PORTS)
3451		return(port_num);
3452	else
3453		return(port_num - CTL_MAX_PORTS);
3454}
3455
3456static uint32_t
3457ctl_map_lun(int port_num, uint32_t lun_id)
3458{
3459	struct ctl_port *port;
3460
3461	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3462	if (port == NULL)
3463		return (UINT32_MAX);
3464	if (port->lun_map == NULL)
3465		return (lun_id);
3466	return (port->lun_map(port->targ_lun_arg, lun_id));
3467}
3468
3469static uint32_t
3470ctl_map_lun_back(int port_num, uint32_t lun_id)
3471{
3472	struct ctl_port *port;
3473	uint32_t i;
3474
3475	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3476	if (port->lun_map == NULL)
3477		return (lun_id);
3478	for (i = 0; i < CTL_MAX_LUNS; i++) {
3479		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3480			return (i);
3481	}
3482	return (UINT32_MAX);
3483}
3484
3485/*
3486 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3487 * that are a power of 2.
3488 */
3489int
3490ctl_ffz(uint32_t *mask, uint32_t size)
3491{
3492	uint32_t num_chunks, num_pieces;
3493	int i, j;
3494
3495	num_chunks = (size >> 5);
3496	if (num_chunks == 0)
3497		num_chunks++;
3498	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3499
3500	for (i = 0; i < num_chunks; i++) {
3501		for (j = 0; j < num_pieces; j++) {
3502			if ((mask[i] & (1 << j)) == 0)
3503				return ((i << 5) + j);
3504		}
3505	}
3506
3507	return (-1);
3508}
3509
3510int
3511ctl_set_mask(uint32_t *mask, uint32_t bit)
3512{
3513	uint32_t chunk, piece;
3514
3515	chunk = bit >> 5;
3516	piece = bit % (sizeof(uint32_t) * 8);
3517
3518	if ((mask[chunk] & (1 << piece)) != 0)
3519		return (-1);
3520	else
3521		mask[chunk] |= (1 << piece);
3522
3523	return (0);
3524}
3525
3526int
3527ctl_clear_mask(uint32_t *mask, uint32_t bit)
3528{
3529	uint32_t chunk, piece;
3530
3531	chunk = bit >> 5;
3532	piece = bit % (sizeof(uint32_t) * 8);
3533
3534	if ((mask[chunk] & (1 << piece)) == 0)
3535		return (-1);
3536	else
3537		mask[chunk] &= ~(1 << piece);
3538
3539	return (0);
3540}
3541
3542int
3543ctl_is_set(uint32_t *mask, uint32_t bit)
3544{
3545	uint32_t chunk, piece;
3546
3547	chunk = bit >> 5;
3548	piece = bit % (sizeof(uint32_t) * 8);
3549
3550	if ((mask[chunk] & (1 << piece)) == 0)
3551		return (0);
3552	else
3553		return (1);
3554}
3555
3556#ifdef unused
3557/*
3558 * The bus, target and lun are optional, they can be filled in later.
3559 * can_wait is used to determine whether we can wait on the malloc or not.
3560 */
3561union ctl_io*
3562ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3563	      uint32_t targ_lun, int can_wait)
3564{
3565	union ctl_io *io;
3566
3567	if (can_wait)
3568		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3569	else
3570		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3571
3572	if (io != NULL) {
3573		io->io_hdr.io_type = io_type;
3574		io->io_hdr.targ_port = targ_port;
3575		/*
3576		 * XXX KDM this needs to change/go away.  We need to move
3577		 * to a preallocated pool of ctl_scsiio structures.
3578		 */
3579		io->io_hdr.nexus.targ_target.id = targ_target;
3580		io->io_hdr.nexus.targ_lun = targ_lun;
3581	}
3582
3583	return (io);
3584}
3585
3586void
3587ctl_kfree_io(union ctl_io *io)
3588{
3589	free(io, M_CTL);
3590}
3591#endif /* unused */
3592
3593/*
3594 * ctl_softc, pool_type, total_ctl_io are passed in.
3595 * npool is passed out.
3596 */
3597int
3598ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3599		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3600{
3601	uint32_t i;
3602	union ctl_io *cur_io, *next_io;
3603	struct ctl_io_pool *pool;
3604	int retval;
3605
3606	retval = 0;
3607
3608	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3609					    M_NOWAIT | M_ZERO);
3610	if (pool == NULL) {
3611		retval = ENOMEM;
3612		goto bailout;
3613	}
3614
3615	pool->type = pool_type;
3616	pool->ctl_softc = ctl_softc;
3617
3618	mtx_lock(&ctl_softc->pool_lock);
3619	pool->id = ctl_softc->cur_pool_id++;
3620	mtx_unlock(&ctl_softc->pool_lock);
3621
3622	pool->flags = CTL_POOL_FLAG_NONE;
3623	pool->refcount = 1;		/* Reference for validity. */
3624	STAILQ_INIT(&pool->free_queue);
3625
3626	/*
3627	 * XXX KDM other options here:
3628	 * - allocate a page at a time
3629	 * - allocate one big chunk of memory.
3630	 * Page allocation might work well, but would take a little more
3631	 * tracking.
3632	 */
3633	for (i = 0; i < total_ctl_io; i++) {
3634		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3635						M_NOWAIT);
3636		if (cur_io == NULL) {
3637			retval = ENOMEM;
3638			break;
3639		}
3640		cur_io->io_hdr.pool = pool;
3641		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3642		pool->total_ctl_io++;
3643		pool->free_ctl_io++;
3644	}
3645
3646	if (retval != 0) {
3647		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3648		     cur_io != NULL; cur_io = next_io) {
3649			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3650							      links);
3651			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3652				      ctl_io_hdr, links);
3653			free(cur_io, M_CTLIO);
3654		}
3655
3656		free(pool, M_CTL);
3657		goto bailout;
3658	}
3659	mtx_lock(&ctl_softc->pool_lock);
3660	ctl_softc->num_pools++;
3661	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3662	/*
3663	 * Increment our usage count if this is an external consumer, so we
3664	 * can't get unloaded until the external consumer (most likely a
3665	 * FETD) unloads and frees his pool.
3666	 *
3667	 * XXX KDM will this increment the caller's module use count, or
3668	 * mine?
3669	 */
3670#if 0
3671	if ((pool_type != CTL_POOL_EMERGENCY)
3672	 && (pool_type != CTL_POOL_INTERNAL)
3673	 && (pool_type != CTL_POOL_4OTHERSC))
3674		MOD_INC_USE_COUNT;
3675#endif
3676
3677	mtx_unlock(&ctl_softc->pool_lock);
3678
3679	*npool = pool;
3680
3681bailout:
3682
3683	return (retval);
3684}
3685
3686static int
3687ctl_pool_acquire(struct ctl_io_pool *pool)
3688{
3689
3690	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3691
3692	if (pool->flags & CTL_POOL_FLAG_INVALID)
3693		return (EINVAL);
3694
3695	pool->refcount++;
3696
3697	return (0);
3698}
3699
3700static void
3701ctl_pool_release(struct ctl_io_pool *pool)
3702{
3703	struct ctl_softc *ctl_softc = pool->ctl_softc;
3704	union ctl_io *io;
3705
3706	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3707
3708	if (--pool->refcount != 0)
3709		return;
3710
3711	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3712		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3713			      links);
3714		free(io, M_CTLIO);
3715	}
3716
3717	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3718	ctl_softc->num_pools--;
3719
3720	/*
3721	 * XXX KDM will this decrement the caller's usage count or mine?
3722	 */
3723#if 0
3724	if ((pool->type != CTL_POOL_EMERGENCY)
3725	 && (pool->type != CTL_POOL_INTERNAL)
3726	 && (pool->type != CTL_POOL_4OTHERSC))
3727		MOD_DEC_USE_COUNT;
3728#endif
3729
3730	free(pool, M_CTL);
3731}
3732
3733void
3734ctl_pool_free(struct ctl_io_pool *pool)
3735{
3736	struct ctl_softc *ctl_softc;
3737
3738	if (pool == NULL)
3739		return;
3740
3741	ctl_softc = pool->ctl_softc;
3742	mtx_lock(&ctl_softc->pool_lock);
3743	pool->flags |= CTL_POOL_FLAG_INVALID;
3744	ctl_pool_release(pool);
3745	mtx_unlock(&ctl_softc->pool_lock);
3746}
3747
3748/*
3749 * This routine does not block (except for spinlocks of course).
3750 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3751 * possible.
3752 */
3753union ctl_io *
3754ctl_alloc_io(void *pool_ref)
3755{
3756	union ctl_io *io;
3757	struct ctl_softc *ctl_softc;
3758	struct ctl_io_pool *pool, *npool;
3759	struct ctl_io_pool *emergency_pool;
3760
3761	pool = (struct ctl_io_pool *)pool_ref;
3762
3763	if (pool == NULL) {
3764		printf("%s: pool is NULL\n", __func__);
3765		return (NULL);
3766	}
3767
3768	emergency_pool = NULL;
3769
3770	ctl_softc = pool->ctl_softc;
3771
3772	mtx_lock(&ctl_softc->pool_lock);
3773	/*
3774	 * First, try to get the io structure from the user's pool.
3775	 */
3776	if (ctl_pool_acquire(pool) == 0) {
3777		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3778		if (io != NULL) {
3779			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3780			pool->total_allocated++;
3781			pool->free_ctl_io--;
3782			mtx_unlock(&ctl_softc->pool_lock);
3783			return (io);
3784		} else
3785			ctl_pool_release(pool);
3786	}
3787	/*
3788	 * If he doesn't have any io structures left, search for an
3789	 * emergency pool and grab one from there.
3790	 */
3791	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3792		if (npool->type != CTL_POOL_EMERGENCY)
3793			continue;
3794
3795		if (ctl_pool_acquire(npool) != 0)
3796			continue;
3797
3798		emergency_pool = npool;
3799
3800		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3801		if (io != NULL) {
3802			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3803			npool->total_allocated++;
3804			npool->free_ctl_io--;
3805			mtx_unlock(&ctl_softc->pool_lock);
3806			return (io);
3807		} else
3808			ctl_pool_release(npool);
3809	}
3810
3811	/* Drop the spinlock before we malloc */
3812	mtx_unlock(&ctl_softc->pool_lock);
3813
3814	/*
3815	 * The emergency pool (if it exists) didn't have one, so try an
3816	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3817	 */
3818	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3819	if (io != NULL) {
3820		/*
3821		 * If the emergency pool exists but is empty, add this
3822		 * ctl_io to its list when it gets freed.
3823		 */
3824		if (emergency_pool != NULL) {
3825			mtx_lock(&ctl_softc->pool_lock);
3826			if (ctl_pool_acquire(emergency_pool) == 0) {
3827				io->io_hdr.pool = emergency_pool;
3828				emergency_pool->total_ctl_io++;
3829				/*
3830				 * Need to bump this, otherwise
3831				 * total_allocated and total_freed won't
3832				 * match when we no longer have anything
3833				 * outstanding.
3834				 */
3835				emergency_pool->total_allocated++;
3836			}
3837			mtx_unlock(&ctl_softc->pool_lock);
3838		} else
3839			io->io_hdr.pool = NULL;
3840	}
3841
3842	return (io);
3843}
3844
3845void
3846ctl_free_io(union ctl_io *io)
3847{
3848	if (io == NULL)
3849		return;
3850
3851	/*
3852	 * If this ctl_io has a pool, return it to that pool.
3853	 */
3854	if (io->io_hdr.pool != NULL) {
3855		struct ctl_io_pool *pool;
3856
3857		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3858		mtx_lock(&pool->ctl_softc->pool_lock);
3859		io->io_hdr.io_type = 0xff;
3860		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3861		pool->total_freed++;
3862		pool->free_ctl_io++;
3863		ctl_pool_release(pool);
3864		mtx_unlock(&pool->ctl_softc->pool_lock);
3865	} else {
3866		/*
3867		 * Otherwise, just free it.  We probably malloced it and
3868		 * the emergency pool wasn't available.
3869		 */
3870		free(io, M_CTLIO);
3871	}
3872
3873}
3874
3875void
3876ctl_zero_io(union ctl_io *io)
3877{
3878	void *pool_ref;
3879
3880	if (io == NULL)
3881		return;
3882
3883	/*
3884	 * May need to preserve linked list pointers at some point too.
3885	 */
3886	pool_ref = io->io_hdr.pool;
3887
3888	memset(io, 0, sizeof(*io));
3889
3890	io->io_hdr.pool = pool_ref;
3891}
3892
3893/*
3894 * This routine is currently used for internal copies of ctl_ios that need
3895 * to persist for some reason after we've already returned status to the
3896 * FETD.  (Thus the flag set.)
3897 *
3898 * XXX XXX
3899 * Note that this makes a blind copy of all fields in the ctl_io, except
3900 * for the pool reference.  This includes any memory that has been
3901 * allocated!  That memory will no longer be valid after done has been
3902 * called, so this would be VERY DANGEROUS for command that actually does
3903 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3904 * start and stop commands, which don't transfer any data, so this is not a
3905 * problem.  If it is used for anything else, the caller would also need to
3906 * allocate data buffer space and this routine would need to be modified to
3907 * copy the data buffer(s) as well.
3908 */
3909void
3910ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3911{
3912	void *pool_ref;
3913
3914	if ((src == NULL)
3915	 || (dest == NULL))
3916		return;
3917
3918	/*
3919	 * May need to preserve linked list pointers at some point too.
3920	 */
3921	pool_ref = dest->io_hdr.pool;
3922
3923	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3924
3925	dest->io_hdr.pool = pool_ref;
3926	/*
3927	 * We need to know that this is an internal copy, and doesn't need
3928	 * to get passed back to the FETD that allocated it.
3929	 */
3930	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3931}
3932
3933#ifdef NEEDTOPORT
3934static void
3935ctl_update_power_subpage(struct copan_power_subpage *page)
3936{
3937	int num_luns, num_partitions, config_type;
3938	struct ctl_softc *softc;
3939	cs_BOOL_t aor_present, shelf_50pct_power;
3940	cs_raidset_personality_t rs_type;
3941	int max_active_luns;
3942
3943	softc = control_softc;
3944
3945	/* subtract out the processor LUN */
3946	num_luns = softc->num_luns - 1;
3947	/*
3948	 * Default to 7 LUNs active, which was the only number we allowed
3949	 * in the past.
3950	 */
3951	max_active_luns = 7;
3952
3953	num_partitions = config_GetRsPartitionInfo();
3954	config_type = config_GetConfigType();
3955	shelf_50pct_power = config_GetShelfPowerMode();
3956	aor_present = config_IsAorRsPresent();
3957
3958	rs_type = ddb_GetRsRaidType(1);
3959	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3960	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3961		EPRINT(0, "Unsupported RS type %d!", rs_type);
3962	}
3963
3964
3965	page->total_luns = num_luns;
3966
3967	switch (config_type) {
3968	case 40:
3969		/*
3970		 * In a 40 drive configuration, it doesn't matter what DC
3971		 * cards we have, whether we have AOR enabled or not,
3972		 * partitioning or not, or what type of RAIDset we have.
3973		 * In that scenario, we can power up every LUN we present
3974		 * to the user.
3975		 */
3976		max_active_luns = num_luns;
3977
3978		break;
3979	case 64:
3980		if (shelf_50pct_power == CS_FALSE) {
3981			/* 25% power */
3982			if (aor_present == CS_TRUE) {
3983				if (rs_type ==
3984				     CS_RAIDSET_PERSONALITY_RAID5) {
3985					max_active_luns = 7;
3986				} else if (rs_type ==
3987					 CS_RAIDSET_PERSONALITY_RAID1){
3988					max_active_luns = 14;
3989				} else {
3990					/* XXX KDM now what?? */
3991				}
3992			} else {
3993				if (rs_type ==
3994				     CS_RAIDSET_PERSONALITY_RAID5) {
3995					max_active_luns = 8;
3996				} else if (rs_type ==
3997					 CS_RAIDSET_PERSONALITY_RAID1){
3998					max_active_luns = 16;
3999				} else {
4000					/* XXX KDM now what?? */
4001				}
4002			}
4003		} else {
4004			/* 50% power */
4005			/*
4006			 * With 50% power in a 64 drive configuration, we
4007			 * can power all LUNs we present.
4008			 */
4009			max_active_luns = num_luns;
4010		}
4011		break;
4012	case 112:
4013		if (shelf_50pct_power == CS_FALSE) {
4014			/* 25% power */
4015			if (aor_present == CS_TRUE) {
4016				if (rs_type ==
4017				     CS_RAIDSET_PERSONALITY_RAID5) {
4018					max_active_luns = 7;
4019				} else if (rs_type ==
4020					 CS_RAIDSET_PERSONALITY_RAID1){
4021					max_active_luns = 14;
4022				} else {
4023					/* XXX KDM now what?? */
4024				}
4025			} else {
4026				if (rs_type ==
4027				     CS_RAIDSET_PERSONALITY_RAID5) {
4028					max_active_luns = 8;
4029				} else if (rs_type ==
4030					 CS_RAIDSET_PERSONALITY_RAID1){
4031					max_active_luns = 16;
4032				} else {
4033					/* XXX KDM now what?? */
4034				}
4035			}
4036		} else {
4037			/* 50% power */
4038			if (aor_present == CS_TRUE) {
4039				if (rs_type ==
4040				     CS_RAIDSET_PERSONALITY_RAID5) {
4041					max_active_luns = 14;
4042				} else if (rs_type ==
4043					 CS_RAIDSET_PERSONALITY_RAID1){
4044					/*
4045					 * We're assuming here that disk
4046					 * caching is enabled, and so we're
4047					 * able to power up half of each
4048					 * LUN, and cache all writes.
4049					 */
4050					max_active_luns = num_luns;
4051				} else {
4052					/* XXX KDM now what?? */
4053				}
4054			} else {
4055				if (rs_type ==
4056				     CS_RAIDSET_PERSONALITY_RAID5) {
4057					max_active_luns = 15;
4058				} else if (rs_type ==
4059					 CS_RAIDSET_PERSONALITY_RAID1){
4060					max_active_luns = 30;
4061				} else {
4062					/* XXX KDM now what?? */
4063				}
4064			}
4065		}
4066		break;
4067	default:
4068		/*
4069		 * In this case, we have an unknown configuration, so we
4070		 * just use the default from above.
4071		 */
4072		break;
4073	}
4074
4075	page->max_active_luns = max_active_luns;
4076#if 0
4077	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4078	       page->total_luns, page->max_active_luns);
4079#endif
4080}
4081#endif /* NEEDTOPORT */
4082
4083/*
4084 * This routine could be used in the future to load default and/or saved
4085 * mode page parameters for a particuar lun.
4086 */
4087static int
4088ctl_init_page_index(struct ctl_lun *lun)
4089{
4090	int i;
4091	struct ctl_page_index *page_index;
4092	struct ctl_softc *softc;
4093
4094	memcpy(&lun->mode_pages.index, page_index_template,
4095	       sizeof(page_index_template));
4096
4097	softc = lun->ctl_softc;
4098
4099	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4100
4101		page_index = &lun->mode_pages.index[i];
4102		/*
4103		 * If this is a disk-only mode page, there's no point in
4104		 * setting it up.  For some pages, we have to have some
4105		 * basic information about the disk in order to calculate the
4106		 * mode page data.
4107		 */
4108		if ((lun->be_lun->lun_type != T_DIRECT)
4109		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4110			continue;
4111
4112		switch (page_index->page_code & SMPH_PC_MASK) {
4113		case SMS_FORMAT_DEVICE_PAGE: {
4114			struct scsi_format_page *format_page;
4115
4116			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4117				panic("subpage is incorrect!");
4118
4119			/*
4120			 * Sectors per track are set above.  Bytes per
4121			 * sector need to be set here on a per-LUN basis.
4122			 */
4123			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4124			       &format_page_default,
4125			       sizeof(format_page_default));
4126			memcpy(&lun->mode_pages.format_page[
4127			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4128			       sizeof(format_page_changeable));
4129			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4130			       &format_page_default,
4131			       sizeof(format_page_default));
4132			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4133			       &format_page_default,
4134			       sizeof(format_page_default));
4135
4136			format_page = &lun->mode_pages.format_page[
4137				CTL_PAGE_CURRENT];
4138			scsi_ulto2b(lun->be_lun->blocksize,
4139				    format_page->bytes_per_sector);
4140
4141			format_page = &lun->mode_pages.format_page[
4142				CTL_PAGE_DEFAULT];
4143			scsi_ulto2b(lun->be_lun->blocksize,
4144				    format_page->bytes_per_sector);
4145
4146			format_page = &lun->mode_pages.format_page[
4147				CTL_PAGE_SAVED];
4148			scsi_ulto2b(lun->be_lun->blocksize,
4149				    format_page->bytes_per_sector);
4150
4151			page_index->page_data =
4152				(uint8_t *)lun->mode_pages.format_page;
4153			break;
4154		}
4155		case SMS_RIGID_DISK_PAGE: {
4156			struct scsi_rigid_disk_page *rigid_disk_page;
4157			uint32_t sectors_per_cylinder;
4158			uint64_t cylinders;
4159#ifndef	__XSCALE__
4160			int shift;
4161#endif /* !__XSCALE__ */
4162
4163			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4164				panic("invalid subpage value %d",
4165				      page_index->subpage);
4166
4167			/*
4168			 * Rotation rate and sectors per track are set
4169			 * above.  We calculate the cylinders here based on
4170			 * capacity.  Due to the number of heads and
4171			 * sectors per track we're using, smaller arrays
4172			 * may turn out to have 0 cylinders.  Linux and
4173			 * FreeBSD don't pay attention to these mode pages
4174			 * to figure out capacity, but Solaris does.  It
4175			 * seems to deal with 0 cylinders just fine, and
4176			 * works out a fake geometry based on the capacity.
4177			 */
4178			memcpy(&lun->mode_pages.rigid_disk_page[
4179			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4180			       sizeof(rigid_disk_page_default));
4181			memcpy(&lun->mode_pages.rigid_disk_page[
4182			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4183			       sizeof(rigid_disk_page_changeable));
4184			memcpy(&lun->mode_pages.rigid_disk_page[
4185			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4186			       sizeof(rigid_disk_page_default));
4187			memcpy(&lun->mode_pages.rigid_disk_page[
4188			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4189			       sizeof(rigid_disk_page_default));
4190
4191			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4192				CTL_DEFAULT_HEADS;
4193
4194			/*
4195			 * The divide method here will be more accurate,
4196			 * probably, but results in floating point being
4197			 * used in the kernel on i386 (__udivdi3()).  On the
4198			 * XScale, though, __udivdi3() is implemented in
4199			 * software.
4200			 *
4201			 * The shift method for cylinder calculation is
4202			 * accurate if sectors_per_cylinder is a power of
4203			 * 2.  Otherwise it might be slightly off -- you
4204			 * might have a bit of a truncation problem.
4205			 */
4206#ifdef	__XSCALE__
4207			cylinders = (lun->be_lun->maxlba + 1) /
4208				sectors_per_cylinder;
4209#else
4210			for (shift = 31; shift > 0; shift--) {
4211				if (sectors_per_cylinder & (1 << shift))
4212					break;
4213			}
4214			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4215#endif
4216
4217			/*
4218			 * We've basically got 3 bytes, or 24 bits for the
4219			 * cylinder size in the mode page.  If we're over,
4220			 * just round down to 2^24.
4221			 */
4222			if (cylinders > 0xffffff)
4223				cylinders = 0xffffff;
4224
4225			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4226				CTL_PAGE_CURRENT];
4227			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4228
4229			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4230				CTL_PAGE_DEFAULT];
4231			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4232
4233			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4234				CTL_PAGE_SAVED];
4235			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4236
4237			page_index->page_data =
4238				(uint8_t *)lun->mode_pages.rigid_disk_page;
4239			break;
4240		}
4241		case SMS_CACHING_PAGE: {
4242
4243			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4244				panic("invalid subpage value %d",
4245				      page_index->subpage);
4246			/*
4247			 * Defaults should be okay here, no calculations
4248			 * needed.
4249			 */
4250			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4251			       &caching_page_default,
4252			       sizeof(caching_page_default));
4253			memcpy(&lun->mode_pages.caching_page[
4254			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4255			       sizeof(caching_page_changeable));
4256			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4257			       &caching_page_default,
4258			       sizeof(caching_page_default));
4259			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4260			       &caching_page_default,
4261			       sizeof(caching_page_default));
4262			page_index->page_data =
4263				(uint8_t *)lun->mode_pages.caching_page;
4264			break;
4265		}
4266		case SMS_CONTROL_MODE_PAGE: {
4267
4268			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4269				panic("invalid subpage value %d",
4270				      page_index->subpage);
4271
4272			/*
4273			 * Defaults should be okay here, no calculations
4274			 * needed.
4275			 */
4276			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4277			       &control_page_default,
4278			       sizeof(control_page_default));
4279			memcpy(&lun->mode_pages.control_page[
4280			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4281			       sizeof(control_page_changeable));
4282			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4283			       &control_page_default,
4284			       sizeof(control_page_default));
4285			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4286			       &control_page_default,
4287			       sizeof(control_page_default));
4288			page_index->page_data =
4289				(uint8_t *)lun->mode_pages.control_page;
4290			break;
4291
4292		}
4293		case SMS_VENDOR_SPECIFIC_PAGE:{
4294			switch (page_index->subpage) {
4295			case PWR_SUBPAGE_CODE: {
4296				struct copan_power_subpage *current_page,
4297							   *saved_page;
4298
4299				memcpy(&lun->mode_pages.power_subpage[
4300				       CTL_PAGE_CURRENT],
4301				       &power_page_default,
4302				       sizeof(power_page_default));
4303				memcpy(&lun->mode_pages.power_subpage[
4304				       CTL_PAGE_CHANGEABLE],
4305				       &power_page_changeable,
4306				       sizeof(power_page_changeable));
4307				memcpy(&lun->mode_pages.power_subpage[
4308				       CTL_PAGE_DEFAULT],
4309				       &power_page_default,
4310				       sizeof(power_page_default));
4311				memcpy(&lun->mode_pages.power_subpage[
4312				       CTL_PAGE_SAVED],
4313				       &power_page_default,
4314				       sizeof(power_page_default));
4315				page_index->page_data =
4316				    (uint8_t *)lun->mode_pages.power_subpage;
4317
4318				current_page = (struct copan_power_subpage *)
4319					(page_index->page_data +
4320					 (page_index->page_len *
4321					  CTL_PAGE_CURRENT));
4322			        saved_page = (struct copan_power_subpage *)
4323				        (page_index->page_data +
4324					 (page_index->page_len *
4325					  CTL_PAGE_SAVED));
4326				break;
4327			}
4328			case APS_SUBPAGE_CODE: {
4329				struct copan_aps_subpage *current_page,
4330							 *saved_page;
4331
4332				// This gets set multiple times but
4333				// it should always be the same. It's
4334				// only done during init so who cares.
4335				index_to_aps_page = i;
4336
4337				memcpy(&lun->mode_pages.aps_subpage[
4338				       CTL_PAGE_CURRENT],
4339				       &aps_page_default,
4340				       sizeof(aps_page_default));
4341				memcpy(&lun->mode_pages.aps_subpage[
4342				       CTL_PAGE_CHANGEABLE],
4343				       &aps_page_changeable,
4344				       sizeof(aps_page_changeable));
4345				memcpy(&lun->mode_pages.aps_subpage[
4346				       CTL_PAGE_DEFAULT],
4347				       &aps_page_default,
4348				       sizeof(aps_page_default));
4349				memcpy(&lun->mode_pages.aps_subpage[
4350				       CTL_PAGE_SAVED],
4351				       &aps_page_default,
4352				       sizeof(aps_page_default));
4353				page_index->page_data =
4354					(uint8_t *)lun->mode_pages.aps_subpage;
4355
4356				current_page = (struct copan_aps_subpage *)
4357					(page_index->page_data +
4358					 (page_index->page_len *
4359					  CTL_PAGE_CURRENT));
4360				saved_page = (struct copan_aps_subpage *)
4361					(page_index->page_data +
4362					 (page_index->page_len *
4363					  CTL_PAGE_SAVED));
4364				break;
4365			}
4366			case DBGCNF_SUBPAGE_CODE: {
4367				struct copan_debugconf_subpage *current_page,
4368							       *saved_page;
4369
4370				memcpy(&lun->mode_pages.debugconf_subpage[
4371				       CTL_PAGE_CURRENT],
4372				       &debugconf_page_default,
4373				       sizeof(debugconf_page_default));
4374				memcpy(&lun->mode_pages.debugconf_subpage[
4375				       CTL_PAGE_CHANGEABLE],
4376				       &debugconf_page_changeable,
4377				       sizeof(debugconf_page_changeable));
4378				memcpy(&lun->mode_pages.debugconf_subpage[
4379				       CTL_PAGE_DEFAULT],
4380				       &debugconf_page_default,
4381				       sizeof(debugconf_page_default));
4382				memcpy(&lun->mode_pages.debugconf_subpage[
4383				       CTL_PAGE_SAVED],
4384				       &debugconf_page_default,
4385				       sizeof(debugconf_page_default));
4386				page_index->page_data =
4387					(uint8_t *)lun->mode_pages.debugconf_subpage;
4388
4389				current_page = (struct copan_debugconf_subpage *)
4390					(page_index->page_data +
4391					 (page_index->page_len *
4392					  CTL_PAGE_CURRENT));
4393				saved_page = (struct copan_debugconf_subpage *)
4394					(page_index->page_data +
4395					 (page_index->page_len *
4396					  CTL_PAGE_SAVED));
4397				break;
4398			}
4399			default:
4400				panic("invalid subpage value %d",
4401				      page_index->subpage);
4402				break;
4403			}
4404   			break;
4405		}
4406		default:
4407			panic("invalid page value %d",
4408			      page_index->page_code & SMPH_PC_MASK);
4409			break;
4410    	}
4411	}
4412
4413	return (CTL_RETVAL_COMPLETE);
4414}
4415
4416/*
4417 * LUN allocation.
4418 *
4419 * Requirements:
4420 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4421 *   wants us to allocate the LUN and he can block.
4422 * - ctl_softc is always set
4423 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4424 *
4425 * Returns 0 for success, non-zero (errno) for failure.
4426 */
4427static int
4428ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4429	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4430{
4431	struct ctl_lun *nlun, *lun;
4432	struct ctl_port *port;
4433	struct scsi_vpd_id_descriptor *desc;
4434	struct scsi_vpd_id_t10 *t10id;
4435	const char *eui, *naa, *scsiname, *vendor;
4436	int lun_number, i, lun_malloced;
4437	int devidlen, idlen1, idlen2 = 0, len;
4438
4439	if (be_lun == NULL)
4440		return (EINVAL);
4441
4442	/*
4443	 * We currently only support Direct Access or Processor LUN types.
4444	 */
4445	switch (be_lun->lun_type) {
4446	case T_DIRECT:
4447		break;
4448	case T_PROCESSOR:
4449		break;
4450	case T_SEQUENTIAL:
4451	case T_CHANGER:
4452	default:
4453		be_lun->lun_config_status(be_lun->be_lun,
4454					  CTL_LUN_CONFIG_FAILURE);
4455		break;
4456	}
4457	if (ctl_lun == NULL) {
4458		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4459		lun_malloced = 1;
4460	} else {
4461		lun_malloced = 0;
4462		lun = ctl_lun;
4463	}
4464
4465	memset(lun, 0, sizeof(*lun));
4466	if (lun_malloced)
4467		lun->flags = CTL_LUN_MALLOCED;
4468
4469	/* Generate LUN ID. */
4470	devidlen = max(CTL_DEVID_MIN_LEN,
4471	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4472	idlen1 = sizeof(*t10id) + devidlen;
4473	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4474	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4475	if (scsiname != NULL) {
4476		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4477		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4478	}
4479	eui = ctl_get_opt(&be_lun->options, "eui");
4480	if (eui != NULL) {
4481		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4482	}
4483	naa = ctl_get_opt(&be_lun->options, "naa");
4484	if (naa != NULL) {
4485		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4486	}
4487	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4488	    M_CTL, M_WAITOK | M_ZERO);
4489	lun->lun_devid->len = len;
4490	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4491	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4492	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4493	desc->length = idlen1;
4494	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4495	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4496	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4497		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4498	} else {
4499		strncpy(t10id->vendor, vendor,
4500		    min(sizeof(t10id->vendor), strlen(vendor)));
4501	}
4502	strncpy((char *)t10id->vendor_spec_id,
4503	    (char *)be_lun->device_id, devidlen);
4504	if (scsiname != NULL) {
4505		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4506		    desc->length);
4507		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4508		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4509		    SVPD_ID_TYPE_SCSI_NAME;
4510		desc->length = idlen2;
4511		strlcpy(desc->identifier, scsiname, idlen2);
4512	}
4513	if (eui != NULL) {
4514		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4515		    desc->length);
4516		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4517		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4518		    SVPD_ID_TYPE_EUI64;
4519		desc->length = 8;
4520		scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4521	}
4522	if (naa != NULL) {
4523		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4524		    desc->length);
4525		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4526		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4527		    SVPD_ID_TYPE_NAA;
4528		desc->length = 8;
4529		scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4530	}
4531
4532	mtx_lock(&ctl_softc->ctl_lock);
4533	/*
4534	 * See if the caller requested a particular LUN number.  If so, see
4535	 * if it is available.  Otherwise, allocate the first available LUN.
4536	 */
4537	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4538		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4539		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4540			mtx_unlock(&ctl_softc->ctl_lock);
4541			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4542				printf("ctl: requested LUN ID %d is higher "
4543				       "than CTL_MAX_LUNS - 1 (%d)\n",
4544				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4545			} else {
4546				/*
4547				 * XXX KDM return an error, or just assign
4548				 * another LUN ID in this case??
4549				 */
4550				printf("ctl: requested LUN ID %d is already "
4551				       "in use\n", be_lun->req_lun_id);
4552			}
4553			if (lun->flags & CTL_LUN_MALLOCED)
4554				free(lun, M_CTL);
4555			be_lun->lun_config_status(be_lun->be_lun,
4556						  CTL_LUN_CONFIG_FAILURE);
4557			return (ENOSPC);
4558		}
4559		lun_number = be_lun->req_lun_id;
4560	} else {
4561		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4562		if (lun_number == -1) {
4563			mtx_unlock(&ctl_softc->ctl_lock);
4564			printf("ctl: can't allocate LUN on target %ju, out of "
4565			       "LUNs\n", (uintmax_t)target_id.id);
4566			if (lun->flags & CTL_LUN_MALLOCED)
4567				free(lun, M_CTL);
4568			be_lun->lun_config_status(be_lun->be_lun,
4569						  CTL_LUN_CONFIG_FAILURE);
4570			return (ENOSPC);
4571		}
4572	}
4573	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4574
4575	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4576	lun->target = target_id;
4577	lun->lun = lun_number;
4578	lun->be_lun = be_lun;
4579	/*
4580	 * The processor LUN is always enabled.  Disk LUNs come on line
4581	 * disabled, and must be enabled by the backend.
4582	 */
4583	lun->flags |= CTL_LUN_DISABLED;
4584	lun->backend = be_lun->be;
4585	be_lun->ctl_lun = lun;
4586	be_lun->lun_id = lun_number;
4587	atomic_add_int(&be_lun->be->num_luns, 1);
4588	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4589		lun->flags |= CTL_LUN_STOPPED;
4590
4591	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4592		lun->flags |= CTL_LUN_INOPERABLE;
4593
4594	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4595		lun->flags |= CTL_LUN_PRIMARY_SC;
4596
4597	lun->ctl_softc = ctl_softc;
4598	TAILQ_INIT(&lun->ooa_queue);
4599	TAILQ_INIT(&lun->blocked_queue);
4600	STAILQ_INIT(&lun->error_list);
4601
4602	/*
4603	 * Initialize the mode page index.
4604	 */
4605	ctl_init_page_index(lun);
4606
4607	/*
4608	 * Set the poweron UA for all initiators on this LUN only.
4609	 */
4610	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4611		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4612
4613	/*
4614	 * Now, before we insert this lun on the lun list, set the lun
4615	 * inventory changed UA for all other luns.
4616	 */
4617	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4618		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4619			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4620		}
4621	}
4622
4623	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4624
4625	ctl_softc->ctl_luns[lun_number] = lun;
4626
4627	ctl_softc->num_luns++;
4628
4629	/* Setup statistics gathering */
4630	lun->stats.device_type = be_lun->lun_type;
4631	lun->stats.lun_number = lun_number;
4632	if (lun->stats.device_type == T_DIRECT)
4633		lun->stats.blocksize = be_lun->blocksize;
4634	else
4635		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4636	for (i = 0;i < CTL_MAX_PORTS;i++)
4637		lun->stats.ports[i].targ_port = i;
4638
4639	mtx_unlock(&ctl_softc->ctl_lock);
4640
4641	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4642
4643	/*
4644	 * Run through each registered FETD and bring it online if it isn't
4645	 * already.  Enable the target ID if it hasn't been enabled, and
4646	 * enable this particular LUN.
4647	 */
4648	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4649		int retval;
4650
4651		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4652		if (retval != 0) {
4653			printf("ctl_alloc_lun: FETD %s port %d returned error "
4654			       "%d for lun_enable on target %ju lun %d\n",
4655			       port->port_name, port->targ_port, retval,
4656			       (uintmax_t)target_id.id, lun_number);
4657		} else
4658			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4659	}
4660	return (0);
4661}
4662
4663/*
4664 * Delete a LUN.
4665 * Assumptions:
4666 * - LUN has already been marked invalid and any pending I/O has been taken
4667 *   care of.
4668 */
4669static int
4670ctl_free_lun(struct ctl_lun *lun)
4671{
4672	struct ctl_softc *softc;
4673#if 0
4674	struct ctl_port *port;
4675#endif
4676	struct ctl_lun *nlun;
4677	int i;
4678
4679	softc = lun->ctl_softc;
4680
4681	mtx_assert(&softc->ctl_lock, MA_OWNED);
4682
4683	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4684
4685	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4686
4687	softc->ctl_luns[lun->lun] = NULL;
4688
4689	if (!TAILQ_EMPTY(&lun->ooa_queue))
4690		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4691
4692	softc->num_luns--;
4693
4694	/*
4695	 * XXX KDM this scheme only works for a single target/multiple LUN
4696	 * setup.  It needs to be revamped for a multiple target scheme.
4697	 *
4698	 * XXX KDM this results in port->lun_disable() getting called twice,
4699	 * once when ctl_disable_lun() is called, and a second time here.
4700	 * We really need to re-think the LUN disable semantics.  There
4701	 * should probably be several steps/levels to LUN removal:
4702	 *  - disable
4703	 *  - invalidate
4704	 *  - free
4705 	 *
4706	 * Right now we only have a disable method when communicating to
4707	 * the front end ports, at least for individual LUNs.
4708	 */
4709#if 0
4710	STAILQ_FOREACH(port, &softc->port_list, links) {
4711		int retval;
4712
4713		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4714					 lun->lun);
4715		if (retval != 0) {
4716			printf("ctl_free_lun: FETD %s port %d returned error "
4717			       "%d for lun_disable on target %ju lun %jd\n",
4718			       port->port_name, port->targ_port, retval,
4719			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4720		}
4721
4722		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4723			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4724
4725			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4726			if (retval != 0) {
4727				printf("ctl_free_lun: FETD %s port %d "
4728				       "returned error %d for targ_disable on "
4729				       "target %ju\n", port->port_name,
4730				       port->targ_port, retval,
4731				       (uintmax_t)lun->target.id);
4732			} else
4733				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4734
4735			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4736				continue;
4737
4738#if 0
4739			port->port_offline(port->onoff_arg);
4740			port->status &= ~CTL_PORT_STATUS_ONLINE;
4741#endif
4742		}
4743	}
4744#endif
4745
4746	/*
4747	 * Tell the backend to free resources, if this LUN has a backend.
4748	 */
4749	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4750	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4751
4752	mtx_destroy(&lun->lun_lock);
4753	free(lun->lun_devid, M_CTL);
4754	if (lun->flags & CTL_LUN_MALLOCED)
4755		free(lun, M_CTL);
4756
4757	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4758		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4759			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4760		}
4761	}
4762
4763	return (0);
4764}
4765
4766static void
4767ctl_create_lun(struct ctl_be_lun *be_lun)
4768{
4769	struct ctl_softc *ctl_softc;
4770
4771	ctl_softc = control_softc;
4772
4773	/*
4774	 * ctl_alloc_lun() should handle all potential failure cases.
4775	 */
4776	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4777}
4778
4779int
4780ctl_add_lun(struct ctl_be_lun *be_lun)
4781{
4782	struct ctl_softc *ctl_softc = control_softc;
4783
4784	mtx_lock(&ctl_softc->ctl_lock);
4785	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4786	mtx_unlock(&ctl_softc->ctl_lock);
4787	wakeup(&ctl_softc->pending_lun_queue);
4788
4789	return (0);
4790}
4791
4792int
4793ctl_enable_lun(struct ctl_be_lun *be_lun)
4794{
4795	struct ctl_softc *ctl_softc;
4796	struct ctl_port *port, *nport;
4797	struct ctl_lun *lun;
4798	int retval;
4799
4800	ctl_softc = control_softc;
4801
4802	lun = (struct ctl_lun *)be_lun->ctl_lun;
4803
4804	mtx_lock(&ctl_softc->ctl_lock);
4805	mtx_lock(&lun->lun_lock);
4806	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4807		/*
4808		 * eh?  Why did we get called if the LUN is already
4809		 * enabled?
4810		 */
4811		mtx_unlock(&lun->lun_lock);
4812		mtx_unlock(&ctl_softc->ctl_lock);
4813		return (0);
4814	}
4815	lun->flags &= ~CTL_LUN_DISABLED;
4816	mtx_unlock(&lun->lun_lock);
4817
4818	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4819		nport = STAILQ_NEXT(port, links);
4820
4821		/*
4822		 * Drop the lock while we call the FETD's enable routine.
4823		 * This can lead to a callback into CTL (at least in the
4824		 * case of the internal initiator frontend.
4825		 */
4826		mtx_unlock(&ctl_softc->ctl_lock);
4827		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4828		mtx_lock(&ctl_softc->ctl_lock);
4829		if (retval != 0) {
4830			printf("%s: FETD %s port %d returned error "
4831			       "%d for lun_enable on target %ju lun %jd\n",
4832			       __func__, port->port_name, port->targ_port, retval,
4833			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4834		}
4835#if 0
4836		 else {
4837            /* NOTE:  TODO:  why does lun enable affect port status? */
4838			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4839		}
4840#endif
4841	}
4842
4843	mtx_unlock(&ctl_softc->ctl_lock);
4844
4845	return (0);
4846}
4847
4848int
4849ctl_disable_lun(struct ctl_be_lun *be_lun)
4850{
4851	struct ctl_softc *ctl_softc;
4852	struct ctl_port *port;
4853	struct ctl_lun *lun;
4854	int retval;
4855
4856	ctl_softc = control_softc;
4857
4858	lun = (struct ctl_lun *)be_lun->ctl_lun;
4859
4860	mtx_lock(&ctl_softc->ctl_lock);
4861	mtx_lock(&lun->lun_lock);
4862	if (lun->flags & CTL_LUN_DISABLED) {
4863		mtx_unlock(&lun->lun_lock);
4864		mtx_unlock(&ctl_softc->ctl_lock);
4865		return (0);
4866	}
4867	lun->flags |= CTL_LUN_DISABLED;
4868	mtx_unlock(&lun->lun_lock);
4869
4870	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4871		mtx_unlock(&ctl_softc->ctl_lock);
4872		/*
4873		 * Drop the lock before we call the frontend's disable
4874		 * routine, to avoid lock order reversals.
4875		 *
4876		 * XXX KDM what happens if the frontend list changes while
4877		 * we're traversing it?  It's unlikely, but should be handled.
4878		 */
4879		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4880					 lun->lun);
4881		mtx_lock(&ctl_softc->ctl_lock);
4882		if (retval != 0) {
4883			printf("ctl_alloc_lun: FETD %s port %d returned error "
4884			       "%d for lun_disable on target %ju lun %jd\n",
4885			       port->port_name, port->targ_port, retval,
4886			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4887		}
4888	}
4889
4890	mtx_unlock(&ctl_softc->ctl_lock);
4891
4892	return (0);
4893}
4894
4895int
4896ctl_start_lun(struct ctl_be_lun *be_lun)
4897{
4898	struct ctl_softc *ctl_softc;
4899	struct ctl_lun *lun;
4900
4901	ctl_softc = control_softc;
4902
4903	lun = (struct ctl_lun *)be_lun->ctl_lun;
4904
4905	mtx_lock(&lun->lun_lock);
4906	lun->flags &= ~CTL_LUN_STOPPED;
4907	mtx_unlock(&lun->lun_lock);
4908
4909	return (0);
4910}
4911
4912int
4913ctl_stop_lun(struct ctl_be_lun *be_lun)
4914{
4915	struct ctl_softc *ctl_softc;
4916	struct ctl_lun *lun;
4917
4918	ctl_softc = control_softc;
4919
4920	lun = (struct ctl_lun *)be_lun->ctl_lun;
4921
4922	mtx_lock(&lun->lun_lock);
4923	lun->flags |= CTL_LUN_STOPPED;
4924	mtx_unlock(&lun->lun_lock);
4925
4926	return (0);
4927}
4928
4929int
4930ctl_lun_offline(struct ctl_be_lun *be_lun)
4931{
4932	struct ctl_softc *ctl_softc;
4933	struct ctl_lun *lun;
4934
4935	ctl_softc = control_softc;
4936
4937	lun = (struct ctl_lun *)be_lun->ctl_lun;
4938
4939	mtx_lock(&lun->lun_lock);
4940	lun->flags |= CTL_LUN_OFFLINE;
4941	mtx_unlock(&lun->lun_lock);
4942
4943	return (0);
4944}
4945
4946int
4947ctl_lun_online(struct ctl_be_lun *be_lun)
4948{
4949	struct ctl_softc *ctl_softc;
4950	struct ctl_lun *lun;
4951
4952	ctl_softc = control_softc;
4953
4954	lun = (struct ctl_lun *)be_lun->ctl_lun;
4955
4956	mtx_lock(&lun->lun_lock);
4957	lun->flags &= ~CTL_LUN_OFFLINE;
4958	mtx_unlock(&lun->lun_lock);
4959
4960	return (0);
4961}
4962
4963int
4964ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4965{
4966	struct ctl_softc *ctl_softc;
4967	struct ctl_lun *lun;
4968
4969	ctl_softc = control_softc;
4970
4971	lun = (struct ctl_lun *)be_lun->ctl_lun;
4972
4973	mtx_lock(&lun->lun_lock);
4974
4975	/*
4976	 * The LUN needs to be disabled before it can be marked invalid.
4977	 */
4978	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4979		mtx_unlock(&lun->lun_lock);
4980		return (-1);
4981	}
4982	/*
4983	 * Mark the LUN invalid.
4984	 */
4985	lun->flags |= CTL_LUN_INVALID;
4986
4987	/*
4988	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4989	 * If we have something in the OOA queue, we'll free it when the
4990	 * last I/O completes.
4991	 */
4992	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4993		mtx_unlock(&lun->lun_lock);
4994		mtx_lock(&ctl_softc->ctl_lock);
4995		ctl_free_lun(lun);
4996		mtx_unlock(&ctl_softc->ctl_lock);
4997	} else
4998		mtx_unlock(&lun->lun_lock);
4999
5000	return (0);
5001}
5002
5003int
5004ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5005{
5006	struct ctl_softc *ctl_softc;
5007	struct ctl_lun *lun;
5008
5009	ctl_softc = control_softc;
5010	lun = (struct ctl_lun *)be_lun->ctl_lun;
5011
5012	mtx_lock(&lun->lun_lock);
5013	lun->flags |= CTL_LUN_INOPERABLE;
5014	mtx_unlock(&lun->lun_lock);
5015
5016	return (0);
5017}
5018
5019int
5020ctl_lun_operable(struct ctl_be_lun *be_lun)
5021{
5022	struct ctl_softc *ctl_softc;
5023	struct ctl_lun *lun;
5024
5025	ctl_softc = control_softc;
5026	lun = (struct ctl_lun *)be_lun->ctl_lun;
5027
5028	mtx_lock(&lun->lun_lock);
5029	lun->flags &= ~CTL_LUN_INOPERABLE;
5030	mtx_unlock(&lun->lun_lock);
5031
5032	return (0);
5033}
5034
5035int
5036ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5037		   int lock)
5038{
5039	struct ctl_softc *softc;
5040	struct ctl_lun *lun;
5041	struct copan_aps_subpage *current_sp;
5042	struct ctl_page_index *page_index;
5043	int i;
5044
5045	softc = control_softc;
5046
5047	mtx_lock(&softc->ctl_lock);
5048
5049	lun = (struct ctl_lun *)be_lun->ctl_lun;
5050	mtx_lock(&lun->lun_lock);
5051
5052	page_index = NULL;
5053	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5054		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5055		     APS_PAGE_CODE)
5056			continue;
5057
5058		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5059			continue;
5060		page_index = &lun->mode_pages.index[i];
5061	}
5062
5063	if (page_index == NULL) {
5064		mtx_unlock(&lun->lun_lock);
5065		mtx_unlock(&softc->ctl_lock);
5066		printf("%s: APS subpage not found for lun %ju!\n", __func__,
5067		       (uintmax_t)lun->lun);
5068		return (1);
5069	}
5070#if 0
5071	if ((softc->aps_locked_lun != 0)
5072	 && (softc->aps_locked_lun != lun->lun)) {
5073		printf("%s: attempt to lock LUN %llu when %llu is already "
5074		       "locked\n");
5075		mtx_unlock(&lun->lun_lock);
5076		mtx_unlock(&softc->ctl_lock);
5077		return (1);
5078	}
5079#endif
5080
5081	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5082		(page_index->page_len * CTL_PAGE_CURRENT));
5083
5084	if (lock != 0) {
5085		current_sp->lock_active = APS_LOCK_ACTIVE;
5086		softc->aps_locked_lun = lun->lun;
5087	} else {
5088		current_sp->lock_active = 0;
5089		softc->aps_locked_lun = 0;
5090	}
5091
5092
5093	/*
5094	 * If we're in HA mode, try to send the lock message to the other
5095	 * side.
5096	 */
5097	if (ctl_is_single == 0) {
5098		int isc_retval;
5099		union ctl_ha_msg lock_msg;
5100
5101		lock_msg.hdr.nexus = *nexus;
5102		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5103		if (lock != 0)
5104			lock_msg.aps.lock_flag = 1;
5105		else
5106			lock_msg.aps.lock_flag = 0;
5107		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5108					 sizeof(lock_msg), 0);
5109		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5110			printf("%s: APS (lock=%d) error returned from "
5111			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5112			mtx_unlock(&lun->lun_lock);
5113			mtx_unlock(&softc->ctl_lock);
5114			return (1);
5115		}
5116	}
5117
5118	mtx_unlock(&lun->lun_lock);
5119	mtx_unlock(&softc->ctl_lock);
5120
5121	return (0);
5122}
5123
5124void
5125ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5126{
5127	struct ctl_lun *lun;
5128	struct ctl_softc *softc;
5129	int i;
5130
5131	softc = control_softc;
5132
5133	lun = (struct ctl_lun *)be_lun->ctl_lun;
5134
5135	mtx_lock(&lun->lun_lock);
5136
5137	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5138		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
5139
5140	mtx_unlock(&lun->lun_lock);
5141}
5142
5143/*
5144 * Backend "memory move is complete" callback for requests that never
5145 * make it down to say RAIDCore's configuration code.
5146 */
5147int
5148ctl_config_move_done(union ctl_io *io)
5149{
5150	int retval;
5151
5152	retval = CTL_RETVAL_COMPLETE;
5153
5154
5155	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5156	/*
5157	 * XXX KDM this shouldn't happen, but what if it does?
5158	 */
5159	if (io->io_hdr.io_type != CTL_IO_SCSI)
5160		panic("I/O type isn't CTL_IO_SCSI!");
5161
5162	if ((io->io_hdr.port_status == 0)
5163	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5164	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5165		io->io_hdr.status = CTL_SUCCESS;
5166	else if ((io->io_hdr.port_status != 0)
5167	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5168	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5169		/*
5170		 * For hardware error sense keys, the sense key
5171		 * specific value is defined to be a retry count,
5172		 * but we use it to pass back an internal FETD
5173		 * error code.  XXX KDM  Hopefully the FETD is only
5174		 * using 16 bits for an error code, since that's
5175		 * all the space we have in the sks field.
5176		 */
5177		ctl_set_internal_failure(&io->scsiio,
5178					 /*sks_valid*/ 1,
5179					 /*retry_count*/
5180					 io->io_hdr.port_status);
5181		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5182			free(io->scsiio.kern_data_ptr, M_CTL);
5183		ctl_done(io);
5184		goto bailout;
5185	}
5186
5187	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5188	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5189	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5190		/*
5191		 * XXX KDM just assuming a single pointer here, and not a
5192		 * S/G list.  If we start using S/G lists for config data,
5193		 * we'll need to know how to clean them up here as well.
5194		 */
5195		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5196			free(io->scsiio.kern_data_ptr, M_CTL);
5197		/* Hopefully the user has already set the status... */
5198		ctl_done(io);
5199	} else {
5200		/*
5201		 * XXX KDM now we need to continue data movement.  Some
5202		 * options:
5203		 * - call ctl_scsiio() again?  We don't do this for data
5204		 *   writes, because for those at least we know ahead of
5205		 *   time where the write will go and how long it is.  For
5206		 *   config writes, though, that information is largely
5207		 *   contained within the write itself, thus we need to
5208		 *   parse out the data again.
5209		 *
5210		 * - Call some other function once the data is in?
5211		 */
5212
5213		/*
5214		 * XXX KDM call ctl_scsiio() again for now, and check flag
5215		 * bits to see whether we're allocated or not.
5216		 */
5217		retval = ctl_scsiio(&io->scsiio);
5218	}
5219bailout:
5220	return (retval);
5221}
5222
5223/*
5224 * This gets called by a backend driver when it is done with a
5225 * data_submit method.
5226 */
5227void
5228ctl_data_submit_done(union ctl_io *io)
5229{
5230	/*
5231	 * If the IO_CONT flag is set, we need to call the supplied
5232	 * function to continue processing the I/O, instead of completing
5233	 * the I/O just yet.
5234	 *
5235	 * If there is an error, though, we don't want to keep processing.
5236	 * Instead, just send status back to the initiator.
5237	 */
5238	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5239	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5240	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5241	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5242		io->scsiio.io_cont(io);
5243		return;
5244	}
5245	ctl_done(io);
5246}
5247
5248/*
5249 * This gets called by a backend driver when it is done with a
5250 * configuration write.
5251 */
5252void
5253ctl_config_write_done(union ctl_io *io)
5254{
5255	/*
5256	 * If the IO_CONT flag is set, we need to call the supplied
5257	 * function to continue processing the I/O, instead of completing
5258	 * the I/O just yet.
5259	 *
5260	 * If there is an error, though, we don't want to keep processing.
5261	 * Instead, just send status back to the initiator.
5262	 */
5263	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5264	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5265	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5266		io->scsiio.io_cont(io);
5267		return;
5268	}
5269	/*
5270	 * Since a configuration write can be done for commands that actually
5271	 * have data allocated, like write buffer, and commands that have
5272	 * no data, like start/stop unit, we need to check here.
5273	 */
5274	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5275		free(io->scsiio.kern_data_ptr, M_CTL);
5276	ctl_done(io);
5277}
5278
5279/*
5280 * SCSI release command.
5281 */
5282int
5283ctl_scsi_release(struct ctl_scsiio *ctsio)
5284{
5285	int length, longid, thirdparty_id, resv_id;
5286	struct ctl_softc *ctl_softc;
5287	struct ctl_lun *lun;
5288
5289	length = 0;
5290	resv_id = 0;
5291
5292	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5293
5294	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5295	ctl_softc = control_softc;
5296
5297	switch (ctsio->cdb[0]) {
5298	case RELEASE_10: {
5299		struct scsi_release_10 *cdb;
5300
5301		cdb = (struct scsi_release_10 *)ctsio->cdb;
5302
5303		if (cdb->byte2 & SR10_LONGID)
5304			longid = 1;
5305		else
5306			thirdparty_id = cdb->thirdparty_id;
5307
5308		resv_id = cdb->resv_id;
5309		length = scsi_2btoul(cdb->length);
5310		break;
5311	}
5312	}
5313
5314
5315	/*
5316	 * XXX KDM right now, we only support LUN reservation.  We don't
5317	 * support 3rd party reservations, or extent reservations, which
5318	 * might actually need the parameter list.  If we've gotten this
5319	 * far, we've got a LUN reservation.  Anything else got kicked out
5320	 * above.  So, according to SPC, ignore the length.
5321	 */
5322	length = 0;
5323
5324	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5325	 && (length > 0)) {
5326		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5327		ctsio->kern_data_len = length;
5328		ctsio->kern_total_len = length;
5329		ctsio->kern_data_resid = 0;
5330		ctsio->kern_rel_offset = 0;
5331		ctsio->kern_sg_entries = 0;
5332		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5333		ctsio->be_move_done = ctl_config_move_done;
5334		ctl_datamove((union ctl_io *)ctsio);
5335
5336		return (CTL_RETVAL_COMPLETE);
5337	}
5338
5339	if (length > 0)
5340		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5341
5342	mtx_lock(&lun->lun_lock);
5343
5344	/*
5345	 * According to SPC, it is not an error for an intiator to attempt
5346	 * to release a reservation on a LUN that isn't reserved, or that
5347	 * is reserved by another initiator.  The reservation can only be
5348	 * released, though, by the initiator who made it or by one of
5349	 * several reset type events.
5350	 */
5351	if (lun->flags & CTL_LUN_RESERVED) {
5352		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5353		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5354		 && (ctsio->io_hdr.nexus.targ_target.id ==
5355		     lun->rsv_nexus.targ_target.id)) {
5356			lun->flags &= ~CTL_LUN_RESERVED;
5357		}
5358	}
5359
5360	mtx_unlock(&lun->lun_lock);
5361
5362	ctsio->scsi_status = SCSI_STATUS_OK;
5363	ctsio->io_hdr.status = CTL_SUCCESS;
5364
5365	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5366		free(ctsio->kern_data_ptr, M_CTL);
5367		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5368	}
5369
5370	ctl_done((union ctl_io *)ctsio);
5371	return (CTL_RETVAL_COMPLETE);
5372}
5373
5374int
5375ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5376{
5377	int extent, thirdparty, longid;
5378	int resv_id, length;
5379	uint64_t thirdparty_id;
5380	struct ctl_softc *ctl_softc;
5381	struct ctl_lun *lun;
5382
5383	extent = 0;
5384	thirdparty = 0;
5385	longid = 0;
5386	resv_id = 0;
5387	length = 0;
5388	thirdparty_id = 0;
5389
5390	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5391
5392	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5393	ctl_softc = control_softc;
5394
5395	switch (ctsio->cdb[0]) {
5396	case RESERVE_10: {
5397		struct scsi_reserve_10 *cdb;
5398
5399		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5400
5401		if (cdb->byte2 & SR10_LONGID)
5402			longid = 1;
5403		else
5404			thirdparty_id = cdb->thirdparty_id;
5405
5406		resv_id = cdb->resv_id;
5407		length = scsi_2btoul(cdb->length);
5408		break;
5409	}
5410	}
5411
5412	/*
5413	 * XXX KDM right now, we only support LUN reservation.  We don't
5414	 * support 3rd party reservations, or extent reservations, which
5415	 * might actually need the parameter list.  If we've gotten this
5416	 * far, we've got a LUN reservation.  Anything else got kicked out
5417	 * above.  So, according to SPC, ignore the length.
5418	 */
5419	length = 0;
5420
5421	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5422	 && (length > 0)) {
5423		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5424		ctsio->kern_data_len = length;
5425		ctsio->kern_total_len = length;
5426		ctsio->kern_data_resid = 0;
5427		ctsio->kern_rel_offset = 0;
5428		ctsio->kern_sg_entries = 0;
5429		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5430		ctsio->be_move_done = ctl_config_move_done;
5431		ctl_datamove((union ctl_io *)ctsio);
5432
5433		return (CTL_RETVAL_COMPLETE);
5434	}
5435
5436	if (length > 0)
5437		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5438
5439	mtx_lock(&lun->lun_lock);
5440	if (lun->flags & CTL_LUN_RESERVED) {
5441		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5442		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5443		 || (ctsio->io_hdr.nexus.targ_target.id !=
5444		     lun->rsv_nexus.targ_target.id)) {
5445			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5446			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5447			goto bailout;
5448		}
5449	}
5450
5451	lun->flags |= CTL_LUN_RESERVED;
5452	lun->rsv_nexus = ctsio->io_hdr.nexus;
5453
5454	ctsio->scsi_status = SCSI_STATUS_OK;
5455	ctsio->io_hdr.status = CTL_SUCCESS;
5456
5457bailout:
5458	mtx_unlock(&lun->lun_lock);
5459
5460	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5461		free(ctsio->kern_data_ptr, M_CTL);
5462		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5463	}
5464
5465	ctl_done((union ctl_io *)ctsio);
5466	return (CTL_RETVAL_COMPLETE);
5467}
5468
5469int
5470ctl_start_stop(struct ctl_scsiio *ctsio)
5471{
5472	struct scsi_start_stop_unit *cdb;
5473	struct ctl_lun *lun;
5474	struct ctl_softc *ctl_softc;
5475	int retval;
5476
5477	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5478
5479	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5480	ctl_softc = control_softc;
5481	retval = 0;
5482
5483	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5484
5485	/*
5486	 * XXX KDM
5487	 * We don't support the immediate bit on a stop unit.  In order to
5488	 * do that, we would need to code up a way to know that a stop is
5489	 * pending, and hold off any new commands until it completes, one
5490	 * way or another.  Then we could accept or reject those commands
5491	 * depending on its status.  We would almost need to do the reverse
5492	 * of what we do below for an immediate start -- return the copy of
5493	 * the ctl_io to the FETD with status to send to the host (and to
5494	 * free the copy!) and then free the original I/O once the stop
5495	 * actually completes.  That way, the OOA queue mechanism can work
5496	 * to block commands that shouldn't proceed.  Another alternative
5497	 * would be to put the copy in the queue in place of the original,
5498	 * and return the original back to the caller.  That could be
5499	 * slightly safer..
5500	 */
5501	if ((cdb->byte2 & SSS_IMMED)
5502	 && ((cdb->how & SSS_START) == 0)) {
5503		ctl_set_invalid_field(ctsio,
5504				      /*sks_valid*/ 1,
5505				      /*command*/ 1,
5506				      /*field*/ 1,
5507				      /*bit_valid*/ 1,
5508				      /*bit*/ 0);
5509		ctl_done((union ctl_io *)ctsio);
5510		return (CTL_RETVAL_COMPLETE);
5511	}
5512
5513	if ((lun->flags & CTL_LUN_PR_RESERVED)
5514	 && ((cdb->how & SSS_START)==0)) {
5515		uint32_t residx;
5516
5517		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5518		if (!lun->per_res[residx].registered
5519		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5520
5521			ctl_set_reservation_conflict(ctsio);
5522			ctl_done((union ctl_io *)ctsio);
5523			return (CTL_RETVAL_COMPLETE);
5524		}
5525	}
5526
5527	/*
5528	 * If there is no backend on this device, we can't start or stop
5529	 * it.  In theory we shouldn't get any start/stop commands in the
5530	 * first place at this level if the LUN doesn't have a backend.
5531	 * That should get stopped by the command decode code.
5532	 */
5533	if (lun->backend == NULL) {
5534		ctl_set_invalid_opcode(ctsio);
5535		ctl_done((union ctl_io *)ctsio);
5536		return (CTL_RETVAL_COMPLETE);
5537	}
5538
5539	/*
5540	 * XXX KDM Copan-specific offline behavior.
5541	 * Figure out a reasonable way to port this?
5542	 */
5543#ifdef NEEDTOPORT
5544	mtx_lock(&lun->lun_lock);
5545
5546	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5547	 && (lun->flags & CTL_LUN_OFFLINE)) {
5548		/*
5549		 * If the LUN is offline, and the on/offline bit isn't set,
5550		 * reject the start or stop.  Otherwise, let it through.
5551		 */
5552		mtx_unlock(&lun->lun_lock);
5553		ctl_set_lun_not_ready(ctsio);
5554		ctl_done((union ctl_io *)ctsio);
5555	} else {
5556		mtx_unlock(&lun->lun_lock);
5557#endif /* NEEDTOPORT */
5558		/*
5559		 * This could be a start or a stop when we're online,
5560		 * or a stop/offline or start/online.  A start or stop when
5561		 * we're offline is covered in the case above.
5562		 */
5563		/*
5564		 * In the non-immediate case, we send the request to
5565		 * the backend and return status to the user when
5566		 * it is done.
5567		 *
5568		 * In the immediate case, we allocate a new ctl_io
5569		 * to hold a copy of the request, and send that to
5570		 * the backend.  We then set good status on the
5571		 * user's request and return it immediately.
5572		 */
5573		if (cdb->byte2 & SSS_IMMED) {
5574			union ctl_io *new_io;
5575
5576			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5577			if (new_io == NULL) {
5578				ctl_set_busy(ctsio);
5579				ctl_done((union ctl_io *)ctsio);
5580			} else {
5581				ctl_copy_io((union ctl_io *)ctsio,
5582					    new_io);
5583				retval = lun->backend->config_write(new_io);
5584				ctl_set_success(ctsio);
5585				ctl_done((union ctl_io *)ctsio);
5586			}
5587		} else {
5588			retval = lun->backend->config_write(
5589				(union ctl_io *)ctsio);
5590		}
5591#ifdef NEEDTOPORT
5592	}
5593#endif
5594	return (retval);
5595}
5596
5597/*
5598 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5599 * we don't really do anything with the LBA and length fields if the user
5600 * passes them in.  Instead we'll just flush out the cache for the entire
5601 * LUN.
5602 */
5603int
5604ctl_sync_cache(struct ctl_scsiio *ctsio)
5605{
5606	struct ctl_lun *lun;
5607	struct ctl_softc *ctl_softc;
5608	uint64_t starting_lba;
5609	uint32_t block_count;
5610	int retval;
5611
5612	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5613
5614	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5615	ctl_softc = control_softc;
5616	retval = 0;
5617
5618	switch (ctsio->cdb[0]) {
5619	case SYNCHRONIZE_CACHE: {
5620		struct scsi_sync_cache *cdb;
5621		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5622
5623		starting_lba = scsi_4btoul(cdb->begin_lba);
5624		block_count = scsi_2btoul(cdb->lb_count);
5625		break;
5626	}
5627	case SYNCHRONIZE_CACHE_16: {
5628		struct scsi_sync_cache_16 *cdb;
5629		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5630
5631		starting_lba = scsi_8btou64(cdb->begin_lba);
5632		block_count = scsi_4btoul(cdb->lb_count);
5633		break;
5634	}
5635	default:
5636		ctl_set_invalid_opcode(ctsio);
5637		ctl_done((union ctl_io *)ctsio);
5638		goto bailout;
5639		break; /* NOTREACHED */
5640	}
5641
5642	/*
5643	 * We check the LBA and length, but don't do anything with them.
5644	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5645	 * get flushed.  This check will just help satisfy anyone who wants
5646	 * to see an error for an out of range LBA.
5647	 */
5648	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5649		ctl_set_lba_out_of_range(ctsio);
5650		ctl_done((union ctl_io *)ctsio);
5651		goto bailout;
5652	}
5653
5654	/*
5655	 * If this LUN has no backend, we can't flush the cache anyway.
5656	 */
5657	if (lun->backend == NULL) {
5658		ctl_set_invalid_opcode(ctsio);
5659		ctl_done((union ctl_io *)ctsio);
5660		goto bailout;
5661	}
5662
5663	/*
5664	 * Check to see whether we're configured to send the SYNCHRONIZE
5665	 * CACHE command directly to the back end.
5666	 */
5667	mtx_lock(&lun->lun_lock);
5668	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5669	 && (++(lun->sync_count) >= lun->sync_interval)) {
5670		lun->sync_count = 0;
5671		mtx_unlock(&lun->lun_lock);
5672		retval = lun->backend->config_write((union ctl_io *)ctsio);
5673	} else {
5674		mtx_unlock(&lun->lun_lock);
5675		ctl_set_success(ctsio);
5676		ctl_done((union ctl_io *)ctsio);
5677	}
5678
5679bailout:
5680
5681	return (retval);
5682}
5683
5684int
5685ctl_format(struct ctl_scsiio *ctsio)
5686{
5687	struct scsi_format *cdb;
5688	struct ctl_lun *lun;
5689	struct ctl_softc *ctl_softc;
5690	int length, defect_list_len;
5691
5692	CTL_DEBUG_PRINT(("ctl_format\n"));
5693
5694	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5695	ctl_softc = control_softc;
5696
5697	cdb = (struct scsi_format *)ctsio->cdb;
5698
5699	length = 0;
5700	if (cdb->byte2 & SF_FMTDATA) {
5701		if (cdb->byte2 & SF_LONGLIST)
5702			length = sizeof(struct scsi_format_header_long);
5703		else
5704			length = sizeof(struct scsi_format_header_short);
5705	}
5706
5707	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5708	 && (length > 0)) {
5709		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5710		ctsio->kern_data_len = length;
5711		ctsio->kern_total_len = length;
5712		ctsio->kern_data_resid = 0;
5713		ctsio->kern_rel_offset = 0;
5714		ctsio->kern_sg_entries = 0;
5715		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5716		ctsio->be_move_done = ctl_config_move_done;
5717		ctl_datamove((union ctl_io *)ctsio);
5718
5719		return (CTL_RETVAL_COMPLETE);
5720	}
5721
5722	defect_list_len = 0;
5723
5724	if (cdb->byte2 & SF_FMTDATA) {
5725		if (cdb->byte2 & SF_LONGLIST) {
5726			struct scsi_format_header_long *header;
5727
5728			header = (struct scsi_format_header_long *)
5729				ctsio->kern_data_ptr;
5730
5731			defect_list_len = scsi_4btoul(header->defect_list_len);
5732			if (defect_list_len != 0) {
5733				ctl_set_invalid_field(ctsio,
5734						      /*sks_valid*/ 1,
5735						      /*command*/ 0,
5736						      /*field*/ 2,
5737						      /*bit_valid*/ 0,
5738						      /*bit*/ 0);
5739				goto bailout;
5740			}
5741		} else {
5742			struct scsi_format_header_short *header;
5743
5744			header = (struct scsi_format_header_short *)
5745				ctsio->kern_data_ptr;
5746
5747			defect_list_len = scsi_2btoul(header->defect_list_len);
5748			if (defect_list_len != 0) {
5749				ctl_set_invalid_field(ctsio,
5750						      /*sks_valid*/ 1,
5751						      /*command*/ 0,
5752						      /*field*/ 2,
5753						      /*bit_valid*/ 0,
5754						      /*bit*/ 0);
5755				goto bailout;
5756			}
5757		}
5758	}
5759
5760	/*
5761	 * The format command will clear out the "Medium format corrupted"
5762	 * status if set by the configuration code.  That status is really
5763	 * just a way to notify the host that we have lost the media, and
5764	 * get them to issue a command that will basically make them think
5765	 * they're blowing away the media.
5766	 */
5767	mtx_lock(&lun->lun_lock);
5768	lun->flags &= ~CTL_LUN_INOPERABLE;
5769	mtx_unlock(&lun->lun_lock);
5770
5771	ctsio->scsi_status = SCSI_STATUS_OK;
5772	ctsio->io_hdr.status = CTL_SUCCESS;
5773bailout:
5774
5775	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5776		free(ctsio->kern_data_ptr, M_CTL);
5777		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5778	}
5779
5780	ctl_done((union ctl_io *)ctsio);
5781	return (CTL_RETVAL_COMPLETE);
5782}
5783
5784int
5785ctl_read_buffer(struct ctl_scsiio *ctsio)
5786{
5787	struct scsi_read_buffer *cdb;
5788	struct ctl_lun *lun;
5789	int buffer_offset, len;
5790	static uint8_t descr[4];
5791	static uint8_t echo_descr[4] = { 0 };
5792
5793	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5794
5795	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5796	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5797
5798	if (lun->flags & CTL_LUN_PR_RESERVED) {
5799		uint32_t residx;
5800
5801		/*
5802		 * XXX KDM need a lock here.
5803		 */
5804		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5805		if ((lun->res_type == SPR_TYPE_EX_AC
5806		  && residx != lun->pr_res_idx)
5807		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5808		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5809		  && !lun->per_res[residx].registered)) {
5810			ctl_set_reservation_conflict(ctsio);
5811			ctl_done((union ctl_io *)ctsio);
5812			return (CTL_RETVAL_COMPLETE);
5813	        }
5814	}
5815
5816	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5817	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5818	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5819		ctl_set_invalid_field(ctsio,
5820				      /*sks_valid*/ 1,
5821				      /*command*/ 1,
5822				      /*field*/ 1,
5823				      /*bit_valid*/ 1,
5824				      /*bit*/ 4);
5825		ctl_done((union ctl_io *)ctsio);
5826		return (CTL_RETVAL_COMPLETE);
5827	}
5828
5829	len = scsi_3btoul(cdb->length);
5830	buffer_offset = scsi_3btoul(cdb->offset);
5831
5832	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5833		ctl_set_invalid_field(ctsio,
5834				      /*sks_valid*/ 1,
5835				      /*command*/ 1,
5836				      /*field*/ 6,
5837				      /*bit_valid*/ 0,
5838				      /*bit*/ 0);
5839		ctl_done((union ctl_io *)ctsio);
5840		return (CTL_RETVAL_COMPLETE);
5841	}
5842
5843	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5844		descr[0] = 0;
5845		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5846		ctsio->kern_data_ptr = descr;
5847		len = min(len, sizeof(descr));
5848	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5849		ctsio->kern_data_ptr = echo_descr;
5850		len = min(len, sizeof(echo_descr));
5851	} else
5852		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5853	ctsio->kern_data_len = len;
5854	ctsio->kern_total_len = len;
5855	ctsio->kern_data_resid = 0;
5856	ctsio->kern_rel_offset = 0;
5857	ctsio->kern_sg_entries = 0;
5858	ctsio->be_move_done = ctl_config_move_done;
5859	ctl_datamove((union ctl_io *)ctsio);
5860
5861	return (CTL_RETVAL_COMPLETE);
5862}
5863
5864int
5865ctl_write_buffer(struct ctl_scsiio *ctsio)
5866{
5867	struct scsi_write_buffer *cdb;
5868	struct ctl_lun *lun;
5869	int buffer_offset, len;
5870
5871	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5872
5873	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5874	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5875
5876	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5877		ctl_set_invalid_field(ctsio,
5878				      /*sks_valid*/ 1,
5879				      /*command*/ 1,
5880				      /*field*/ 1,
5881				      /*bit_valid*/ 1,
5882				      /*bit*/ 4);
5883		ctl_done((union ctl_io *)ctsio);
5884		return (CTL_RETVAL_COMPLETE);
5885	}
5886
5887	len = scsi_3btoul(cdb->length);
5888	buffer_offset = scsi_3btoul(cdb->offset);
5889
5890	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5891		ctl_set_invalid_field(ctsio,
5892				      /*sks_valid*/ 1,
5893				      /*command*/ 1,
5894				      /*field*/ 6,
5895				      /*bit_valid*/ 0,
5896				      /*bit*/ 0);
5897		ctl_done((union ctl_io *)ctsio);
5898		return (CTL_RETVAL_COMPLETE);
5899	}
5900
5901	/*
5902	 * If we've got a kernel request that hasn't been malloced yet,
5903	 * malloc it and tell the caller the data buffer is here.
5904	 */
5905	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5906		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5907		ctsio->kern_data_len = len;
5908		ctsio->kern_total_len = len;
5909		ctsio->kern_data_resid = 0;
5910		ctsio->kern_rel_offset = 0;
5911		ctsio->kern_sg_entries = 0;
5912		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5913		ctsio->be_move_done = ctl_config_move_done;
5914		ctl_datamove((union ctl_io *)ctsio);
5915
5916		return (CTL_RETVAL_COMPLETE);
5917	}
5918
5919	ctl_done((union ctl_io *)ctsio);
5920
5921	return (CTL_RETVAL_COMPLETE);
5922}
5923
5924int
5925ctl_write_same(struct ctl_scsiio *ctsio)
5926{
5927	struct ctl_lun *lun;
5928	struct ctl_lba_len_flags *lbalen;
5929	uint64_t lba;
5930	uint32_t num_blocks;
5931	int len, retval;
5932	uint8_t byte2;
5933
5934	retval = CTL_RETVAL_COMPLETE;
5935
5936	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5937
5938	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5939
5940	switch (ctsio->cdb[0]) {
5941	case WRITE_SAME_10: {
5942		struct scsi_write_same_10 *cdb;
5943
5944		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5945
5946		lba = scsi_4btoul(cdb->addr);
5947		num_blocks = scsi_2btoul(cdb->length);
5948		byte2 = cdb->byte2;
5949		break;
5950	}
5951	case WRITE_SAME_16: {
5952		struct scsi_write_same_16 *cdb;
5953
5954		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5955
5956		lba = scsi_8btou64(cdb->addr);
5957		num_blocks = scsi_4btoul(cdb->length);
5958		byte2 = cdb->byte2;
5959		break;
5960	}
5961	default:
5962		/*
5963		 * We got a command we don't support.  This shouldn't
5964		 * happen, commands should be filtered out above us.
5965		 */
5966		ctl_set_invalid_opcode(ctsio);
5967		ctl_done((union ctl_io *)ctsio);
5968
5969		return (CTL_RETVAL_COMPLETE);
5970		break; /* NOTREACHED */
5971	}
5972
5973	/*
5974	 * The first check is to make sure we're in bounds, the second
5975	 * check is to catch wrap-around problems.  If the lba + num blocks
5976	 * is less than the lba, then we've wrapped around and the block
5977	 * range is invalid anyway.
5978	 */
5979	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5980	 || ((lba + num_blocks) < lba)) {
5981		ctl_set_lba_out_of_range(ctsio);
5982		ctl_done((union ctl_io *)ctsio);
5983		return (CTL_RETVAL_COMPLETE);
5984	}
5985
5986	/* Zero number of blocks means "to the last logical block" */
5987	if (num_blocks == 0) {
5988		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5989			ctl_set_invalid_field(ctsio,
5990					      /*sks_valid*/ 0,
5991					      /*command*/ 1,
5992					      /*field*/ 0,
5993					      /*bit_valid*/ 0,
5994					      /*bit*/ 0);
5995			ctl_done((union ctl_io *)ctsio);
5996			return (CTL_RETVAL_COMPLETE);
5997		}
5998		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5999	}
6000
6001	len = lun->be_lun->blocksize;
6002
6003	/*
6004	 * If we've got a kernel request that hasn't been malloced yet,
6005	 * malloc it and tell the caller the data buffer is here.
6006	 */
6007	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6008		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6009		ctsio->kern_data_len = len;
6010		ctsio->kern_total_len = len;
6011		ctsio->kern_data_resid = 0;
6012		ctsio->kern_rel_offset = 0;
6013		ctsio->kern_sg_entries = 0;
6014		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6015		ctsio->be_move_done = ctl_config_move_done;
6016		ctl_datamove((union ctl_io *)ctsio);
6017
6018		return (CTL_RETVAL_COMPLETE);
6019	}
6020
6021	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6022	lbalen->lba = lba;
6023	lbalen->len = num_blocks;
6024	lbalen->flags = byte2;
6025	retval = lun->backend->config_write((union ctl_io *)ctsio);
6026
6027	return (retval);
6028}
6029
6030int
6031ctl_unmap(struct ctl_scsiio *ctsio)
6032{
6033	struct ctl_lun *lun;
6034	struct scsi_unmap *cdb;
6035	struct ctl_ptr_len_flags *ptrlen;
6036	struct scsi_unmap_header *hdr;
6037	struct scsi_unmap_desc *buf, *end;
6038	uint64_t lba;
6039	uint32_t num_blocks;
6040	int len, retval;
6041	uint8_t byte2;
6042
6043	retval = CTL_RETVAL_COMPLETE;
6044
6045	CTL_DEBUG_PRINT(("ctl_unmap\n"));
6046
6047	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6048	cdb = (struct scsi_unmap *)ctsio->cdb;
6049
6050	len = scsi_2btoul(cdb->length);
6051	byte2 = cdb->byte2;
6052
6053	/*
6054	 * If we've got a kernel request that hasn't been malloced yet,
6055	 * malloc it and tell the caller the data buffer is here.
6056	 */
6057	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6058		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6059		ctsio->kern_data_len = len;
6060		ctsio->kern_total_len = len;
6061		ctsio->kern_data_resid = 0;
6062		ctsio->kern_rel_offset = 0;
6063		ctsio->kern_sg_entries = 0;
6064		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6065		ctsio->be_move_done = ctl_config_move_done;
6066		ctl_datamove((union ctl_io *)ctsio);
6067
6068		return (CTL_RETVAL_COMPLETE);
6069	}
6070
6071	len = ctsio->kern_total_len - ctsio->kern_data_resid;
6072	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6073	if (len < sizeof (*hdr) ||
6074	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6075	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6076	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6077		ctl_set_invalid_field(ctsio,
6078				      /*sks_valid*/ 0,
6079				      /*command*/ 0,
6080				      /*field*/ 0,
6081				      /*bit_valid*/ 0,
6082				      /*bit*/ 0);
6083		ctl_done((union ctl_io *)ctsio);
6084		return (CTL_RETVAL_COMPLETE);
6085	}
6086	len = scsi_2btoul(hdr->desc_length);
6087	buf = (struct scsi_unmap_desc *)(hdr + 1);
6088	end = buf + len / sizeof(*buf);
6089
6090	ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6091	ptrlen->ptr = (void *)buf;
6092	ptrlen->len = len;
6093	ptrlen->flags = byte2;
6094
6095	for (; buf < end; buf++) {
6096		lba = scsi_8btou64(buf->lba);
6097		num_blocks = scsi_4btoul(buf->length);
6098		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6099		 || ((lba + num_blocks) < lba)) {
6100			ctl_set_lba_out_of_range(ctsio);
6101			ctl_done((union ctl_io *)ctsio);
6102			return (CTL_RETVAL_COMPLETE);
6103		}
6104	}
6105
6106	retval = lun->backend->config_write((union ctl_io *)ctsio);
6107
6108	return (retval);
6109}
6110
6111/*
6112 * Note that this function currently doesn't actually do anything inside
6113 * CTL to enforce things if the DQue bit is turned on.
6114 *
6115 * Also note that this function can't be used in the default case, because
6116 * the DQue bit isn't set in the changeable mask for the control mode page
6117 * anyway.  This is just here as an example for how to implement a page
6118 * handler, and a placeholder in case we want to allow the user to turn
6119 * tagged queueing on and off.
6120 *
6121 * The D_SENSE bit handling is functional, however, and will turn
6122 * descriptor sense on and off for a given LUN.
6123 */
6124int
6125ctl_control_page_handler(struct ctl_scsiio *ctsio,
6126			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6127{
6128	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6129	struct ctl_lun *lun;
6130	struct ctl_softc *softc;
6131	int set_ua;
6132	uint32_t initidx;
6133
6134	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6135	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6136	set_ua = 0;
6137
6138	user_cp = (struct scsi_control_page *)page_ptr;
6139	current_cp = (struct scsi_control_page *)
6140		(page_index->page_data + (page_index->page_len *
6141		CTL_PAGE_CURRENT));
6142	saved_cp = (struct scsi_control_page *)
6143		(page_index->page_data + (page_index->page_len *
6144		CTL_PAGE_SAVED));
6145
6146	softc = control_softc;
6147
6148	mtx_lock(&lun->lun_lock);
6149	if (((current_cp->rlec & SCP_DSENSE) == 0)
6150	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6151		/*
6152		 * Descriptor sense is currently turned off and the user
6153		 * wants to turn it on.
6154		 */
6155		current_cp->rlec |= SCP_DSENSE;
6156		saved_cp->rlec |= SCP_DSENSE;
6157		lun->flags |= CTL_LUN_SENSE_DESC;
6158		set_ua = 1;
6159	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6160		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6161		/*
6162		 * Descriptor sense is currently turned on, and the user
6163		 * wants to turn it off.
6164		 */
6165		current_cp->rlec &= ~SCP_DSENSE;
6166		saved_cp->rlec &= ~SCP_DSENSE;
6167		lun->flags &= ~CTL_LUN_SENSE_DESC;
6168		set_ua = 1;
6169	}
6170	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
6171		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6172#ifdef NEEDTOPORT
6173			csevent_log(CSC_CTL | CSC_SHELF_SW |
6174				    CTL_UNTAG_TO_UNTAG,
6175				    csevent_LogType_Trace,
6176				    csevent_Severity_Information,
6177				    csevent_AlertLevel_Green,
6178				    csevent_FRU_Firmware,
6179				    csevent_FRU_Unknown,
6180				    "Received untagged to untagged transition");
6181#endif /* NEEDTOPORT */
6182		} else {
6183#ifdef NEEDTOPORT
6184			csevent_log(CSC_CTL | CSC_SHELF_SW |
6185				    CTL_UNTAG_TO_TAG,
6186				    csevent_LogType_ConfigChange,
6187				    csevent_Severity_Information,
6188				    csevent_AlertLevel_Green,
6189				    csevent_FRU_Firmware,
6190				    csevent_FRU_Unknown,
6191				    "Received untagged to tagged "
6192				    "queueing transition");
6193#endif /* NEEDTOPORT */
6194
6195			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6196			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6197			set_ua = 1;
6198		}
6199	} else {
6200		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6201#ifdef NEEDTOPORT
6202			csevent_log(CSC_CTL | CSC_SHELF_SW |
6203				    CTL_TAG_TO_UNTAG,
6204				    csevent_LogType_ConfigChange,
6205				    csevent_Severity_Warning,
6206				    csevent_AlertLevel_Yellow,
6207				    csevent_FRU_Firmware,
6208				    csevent_FRU_Unknown,
6209				    "Received tagged queueing to untagged "
6210				    "transition");
6211#endif /* NEEDTOPORT */
6212
6213			current_cp->queue_flags |= SCP_QUEUE_DQUE;
6214			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
6215			set_ua = 1;
6216		} else {
6217#ifdef NEEDTOPORT
6218			csevent_log(CSC_CTL | CSC_SHELF_SW |
6219				    CTL_TAG_TO_TAG,
6220				    csevent_LogType_Trace,
6221				    csevent_Severity_Information,
6222				    csevent_AlertLevel_Green,
6223				    csevent_FRU_Firmware,
6224				    csevent_FRU_Unknown,
6225				    "Received tagged queueing to tagged "
6226				    "queueing transition");
6227#endif /* NEEDTOPORT */
6228		}
6229	}
6230	if (set_ua != 0) {
6231		int i;
6232		/*
6233		 * Let other initiators know that the mode
6234		 * parameters for this LUN have changed.
6235		 */
6236		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6237			if (i == initidx)
6238				continue;
6239
6240			lun->pending_sense[i].ua_pending |=
6241				CTL_UA_MODE_CHANGE;
6242		}
6243	}
6244	mtx_unlock(&lun->lun_lock);
6245
6246	return (0);
6247}
6248
6249int
6250ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6251		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6252{
6253	return (0);
6254}
6255
6256int
6257ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6258			   struct ctl_page_index *page_index, int pc)
6259{
6260	struct copan_power_subpage *page;
6261
6262	page = (struct copan_power_subpage *)page_index->page_data +
6263		(page_index->page_len * pc);
6264
6265	switch (pc) {
6266	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6267		/*
6268		 * We don't update the changable bits for this page.
6269		 */
6270		break;
6271	case SMS_PAGE_CTRL_CURRENT >> 6:
6272	case SMS_PAGE_CTRL_DEFAULT >> 6:
6273	case SMS_PAGE_CTRL_SAVED >> 6:
6274#ifdef NEEDTOPORT
6275		ctl_update_power_subpage(page);
6276#endif
6277		break;
6278	default:
6279#ifdef NEEDTOPORT
6280		EPRINT(0, "Invalid PC %d!!", pc);
6281#endif
6282		break;
6283	}
6284	return (0);
6285}
6286
6287
6288int
6289ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6290		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6291{
6292	struct copan_aps_subpage *user_sp;
6293	struct copan_aps_subpage *current_sp;
6294	union ctl_modepage_info *modepage_info;
6295	struct ctl_softc *softc;
6296	struct ctl_lun *lun;
6297	int retval;
6298
6299	retval = CTL_RETVAL_COMPLETE;
6300	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6301		     (page_index->page_len * CTL_PAGE_CURRENT));
6302	softc = control_softc;
6303	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6304
6305	user_sp = (struct copan_aps_subpage *)page_ptr;
6306
6307	modepage_info = (union ctl_modepage_info *)
6308		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6309
6310	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6311	modepage_info->header.subpage = page_index->subpage;
6312	modepage_info->aps.lock_active = user_sp->lock_active;
6313
6314	mtx_lock(&softc->ctl_lock);
6315
6316	/*
6317	 * If there is a request to lock the LUN and another LUN is locked
6318	 * this is an error. If the requested LUN is already locked ignore
6319	 * the request. If no LUN is locked attempt to lock it.
6320	 * if there is a request to unlock the LUN and the LUN is currently
6321	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6322	 * if another LUN is locked or no LUN is locked.
6323	 */
6324	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6325		if (softc->aps_locked_lun == lun->lun) {
6326			/*
6327			 * This LUN is already locked, so we're done.
6328			 */
6329			retval = CTL_RETVAL_COMPLETE;
6330		} else if (softc->aps_locked_lun == 0) {
6331			/*
6332			 * No one has the lock, pass the request to the
6333			 * backend.
6334			 */
6335			retval = lun->backend->config_write(
6336				(union ctl_io *)ctsio);
6337		} else {
6338			/*
6339			 * Someone else has the lock, throw out the request.
6340			 */
6341			ctl_set_already_locked(ctsio);
6342			free(ctsio->kern_data_ptr, M_CTL);
6343			ctl_done((union ctl_io *)ctsio);
6344
6345			/*
6346			 * Set the return value so that ctl_do_mode_select()
6347			 * won't try to complete the command.  We already
6348			 * completed it here.
6349			 */
6350			retval = CTL_RETVAL_ERROR;
6351		}
6352	} else if (softc->aps_locked_lun == lun->lun) {
6353		/*
6354		 * This LUN is locked, so pass the unlock request to the
6355		 * backend.
6356		 */
6357		retval = lun->backend->config_write((union ctl_io *)ctsio);
6358	}
6359	mtx_unlock(&softc->ctl_lock);
6360
6361	return (retval);
6362}
6363
6364int
6365ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6366				struct ctl_page_index *page_index,
6367				uint8_t *page_ptr)
6368{
6369	uint8_t *c;
6370	int i;
6371
6372	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6373	ctl_time_io_secs =
6374		(c[0] << 8) |
6375		(c[1] << 0) |
6376		0;
6377	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6378	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6379	printf("page data:");
6380	for (i=0; i<8; i++)
6381		printf(" %.2x",page_ptr[i]);
6382	printf("\n");
6383	return (0);
6384}
6385
6386int
6387ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6388			       struct ctl_page_index *page_index,
6389			       int pc)
6390{
6391	struct copan_debugconf_subpage *page;
6392
6393	page = (struct copan_debugconf_subpage *)page_index->page_data +
6394		(page_index->page_len * pc);
6395
6396	switch (pc) {
6397	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6398	case SMS_PAGE_CTRL_DEFAULT >> 6:
6399	case SMS_PAGE_CTRL_SAVED >> 6:
6400		/*
6401		 * We don't update the changable or default bits for this page.
6402		 */
6403		break;
6404	case SMS_PAGE_CTRL_CURRENT >> 6:
6405		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6406		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6407		break;
6408	default:
6409#ifdef NEEDTOPORT
6410		EPRINT(0, "Invalid PC %d!!", pc);
6411#endif /* NEEDTOPORT */
6412		break;
6413	}
6414	return (0);
6415}
6416
6417
6418static int
6419ctl_do_mode_select(union ctl_io *io)
6420{
6421	struct scsi_mode_page_header *page_header;
6422	struct ctl_page_index *page_index;
6423	struct ctl_scsiio *ctsio;
6424	int control_dev, page_len;
6425	int page_len_offset, page_len_size;
6426	union ctl_modepage_info *modepage_info;
6427	struct ctl_lun *lun;
6428	int *len_left, *len_used;
6429	int retval, i;
6430
6431	ctsio = &io->scsiio;
6432	page_index = NULL;
6433	page_len = 0;
6434	retval = CTL_RETVAL_COMPLETE;
6435
6436	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6437
6438	if (lun->be_lun->lun_type != T_DIRECT)
6439		control_dev = 1;
6440	else
6441		control_dev = 0;
6442
6443	modepage_info = (union ctl_modepage_info *)
6444		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6445	len_left = &modepage_info->header.len_left;
6446	len_used = &modepage_info->header.len_used;
6447
6448do_next_page:
6449
6450	page_header = (struct scsi_mode_page_header *)
6451		(ctsio->kern_data_ptr + *len_used);
6452
6453	if (*len_left == 0) {
6454		free(ctsio->kern_data_ptr, M_CTL);
6455		ctl_set_success(ctsio);
6456		ctl_done((union ctl_io *)ctsio);
6457		return (CTL_RETVAL_COMPLETE);
6458	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6459
6460		free(ctsio->kern_data_ptr, M_CTL);
6461		ctl_set_param_len_error(ctsio);
6462		ctl_done((union ctl_io *)ctsio);
6463		return (CTL_RETVAL_COMPLETE);
6464
6465	} else if ((page_header->page_code & SMPH_SPF)
6466		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6467
6468		free(ctsio->kern_data_ptr, M_CTL);
6469		ctl_set_param_len_error(ctsio);
6470		ctl_done((union ctl_io *)ctsio);
6471		return (CTL_RETVAL_COMPLETE);
6472	}
6473
6474
6475	/*
6476	 * XXX KDM should we do something with the block descriptor?
6477	 */
6478	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6479
6480		if ((control_dev != 0)
6481		 && (lun->mode_pages.index[i].page_flags &
6482		     CTL_PAGE_FLAG_DISK_ONLY))
6483			continue;
6484
6485		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6486		    (page_header->page_code & SMPH_PC_MASK))
6487			continue;
6488
6489		/*
6490		 * If neither page has a subpage code, then we've got a
6491		 * match.
6492		 */
6493		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6494		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6495			page_index = &lun->mode_pages.index[i];
6496			page_len = page_header->page_length;
6497			break;
6498		}
6499
6500		/*
6501		 * If both pages have subpages, then the subpage numbers
6502		 * have to match.
6503		 */
6504		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6505		  && (page_header->page_code & SMPH_SPF)) {
6506			struct scsi_mode_page_header_sp *sph;
6507
6508			sph = (struct scsi_mode_page_header_sp *)page_header;
6509
6510			if (lun->mode_pages.index[i].subpage ==
6511			    sph->subpage) {
6512				page_index = &lun->mode_pages.index[i];
6513				page_len = scsi_2btoul(sph->page_length);
6514				break;
6515			}
6516		}
6517	}
6518
6519	/*
6520	 * If we couldn't find the page, or if we don't have a mode select
6521	 * handler for it, send back an error to the user.
6522	 */
6523	if ((page_index == NULL)
6524	 || (page_index->select_handler == NULL)) {
6525		ctl_set_invalid_field(ctsio,
6526				      /*sks_valid*/ 1,
6527				      /*command*/ 0,
6528				      /*field*/ *len_used,
6529				      /*bit_valid*/ 0,
6530				      /*bit*/ 0);
6531		free(ctsio->kern_data_ptr, M_CTL);
6532		ctl_done((union ctl_io *)ctsio);
6533		return (CTL_RETVAL_COMPLETE);
6534	}
6535
6536	if (page_index->page_code & SMPH_SPF) {
6537		page_len_offset = 2;
6538		page_len_size = 2;
6539	} else {
6540		page_len_size = 1;
6541		page_len_offset = 1;
6542	}
6543
6544	/*
6545	 * If the length the initiator gives us isn't the one we specify in
6546	 * the mode page header, or if they didn't specify enough data in
6547	 * the CDB to avoid truncating this page, kick out the request.
6548	 */
6549	if ((page_len != (page_index->page_len - page_len_offset -
6550			  page_len_size))
6551	 || (*len_left < page_index->page_len)) {
6552
6553
6554		ctl_set_invalid_field(ctsio,
6555				      /*sks_valid*/ 1,
6556				      /*command*/ 0,
6557				      /*field*/ *len_used + page_len_offset,
6558				      /*bit_valid*/ 0,
6559				      /*bit*/ 0);
6560		free(ctsio->kern_data_ptr, M_CTL);
6561		ctl_done((union ctl_io *)ctsio);
6562		return (CTL_RETVAL_COMPLETE);
6563	}
6564
6565	/*
6566	 * Run through the mode page, checking to make sure that the bits
6567	 * the user changed are actually legal for him to change.
6568	 */
6569	for (i = 0; i < page_index->page_len; i++) {
6570		uint8_t *user_byte, *change_mask, *current_byte;
6571		int bad_bit;
6572		int j;
6573
6574		user_byte = (uint8_t *)page_header + i;
6575		change_mask = page_index->page_data +
6576			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6577		current_byte = page_index->page_data +
6578			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6579
6580		/*
6581		 * Check to see whether the user set any bits in this byte
6582		 * that he is not allowed to set.
6583		 */
6584		if ((*user_byte & ~(*change_mask)) ==
6585		    (*current_byte & ~(*change_mask)))
6586			continue;
6587
6588		/*
6589		 * Go through bit by bit to determine which one is illegal.
6590		 */
6591		bad_bit = 0;
6592		for (j = 7; j >= 0; j--) {
6593			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6594			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6595				bad_bit = i;
6596				break;
6597			}
6598		}
6599		ctl_set_invalid_field(ctsio,
6600				      /*sks_valid*/ 1,
6601				      /*command*/ 0,
6602				      /*field*/ *len_used + i,
6603				      /*bit_valid*/ 1,
6604				      /*bit*/ bad_bit);
6605		free(ctsio->kern_data_ptr, M_CTL);
6606		ctl_done((union ctl_io *)ctsio);
6607		return (CTL_RETVAL_COMPLETE);
6608	}
6609
6610	/*
6611	 * Decrement these before we call the page handler, since we may
6612	 * end up getting called back one way or another before the handler
6613	 * returns to this context.
6614	 */
6615	*len_left -= page_index->page_len;
6616	*len_used += page_index->page_len;
6617
6618	retval = page_index->select_handler(ctsio, page_index,
6619					    (uint8_t *)page_header);
6620
6621	/*
6622	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6623	 * wait until this queued command completes to finish processing
6624	 * the mode page.  If it returns anything other than
6625	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6626	 * already set the sense information, freed the data pointer, and
6627	 * completed the io for us.
6628	 */
6629	if (retval != CTL_RETVAL_COMPLETE)
6630		goto bailout_no_done;
6631
6632	/*
6633	 * If the initiator sent us more than one page, parse the next one.
6634	 */
6635	if (*len_left > 0)
6636		goto do_next_page;
6637
6638	ctl_set_success(ctsio);
6639	free(ctsio->kern_data_ptr, M_CTL);
6640	ctl_done((union ctl_io *)ctsio);
6641
6642bailout_no_done:
6643
6644	return (CTL_RETVAL_COMPLETE);
6645
6646}
6647
6648int
6649ctl_mode_select(struct ctl_scsiio *ctsio)
6650{
6651	int param_len, pf, sp;
6652	int header_size, bd_len;
6653	int len_left, len_used;
6654	struct ctl_page_index *page_index;
6655	struct ctl_lun *lun;
6656	int control_dev, page_len;
6657	union ctl_modepage_info *modepage_info;
6658	int retval;
6659
6660	pf = 0;
6661	sp = 0;
6662	page_len = 0;
6663	len_used = 0;
6664	len_left = 0;
6665	retval = 0;
6666	bd_len = 0;
6667	page_index = NULL;
6668
6669	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6670
6671	if (lun->be_lun->lun_type != T_DIRECT)
6672		control_dev = 1;
6673	else
6674		control_dev = 0;
6675
6676	switch (ctsio->cdb[0]) {
6677	case MODE_SELECT_6: {
6678		struct scsi_mode_select_6 *cdb;
6679
6680		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6681
6682		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6683		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6684
6685		param_len = cdb->length;
6686		header_size = sizeof(struct scsi_mode_header_6);
6687		break;
6688	}
6689	case MODE_SELECT_10: {
6690		struct scsi_mode_select_10 *cdb;
6691
6692		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6693
6694		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6695		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6696
6697		param_len = scsi_2btoul(cdb->length);
6698		header_size = sizeof(struct scsi_mode_header_10);
6699		break;
6700	}
6701	default:
6702		ctl_set_invalid_opcode(ctsio);
6703		ctl_done((union ctl_io *)ctsio);
6704		return (CTL_RETVAL_COMPLETE);
6705		break; /* NOTREACHED */
6706	}
6707
6708	/*
6709	 * From SPC-3:
6710	 * "A parameter list length of zero indicates that the Data-Out Buffer
6711	 * shall be empty. This condition shall not be considered as an error."
6712	 */
6713	if (param_len == 0) {
6714		ctl_set_success(ctsio);
6715		ctl_done((union ctl_io *)ctsio);
6716		return (CTL_RETVAL_COMPLETE);
6717	}
6718
6719	/*
6720	 * Since we'll hit this the first time through, prior to
6721	 * allocation, we don't need to free a data buffer here.
6722	 */
6723	if (param_len < header_size) {
6724		ctl_set_param_len_error(ctsio);
6725		ctl_done((union ctl_io *)ctsio);
6726		return (CTL_RETVAL_COMPLETE);
6727	}
6728
6729	/*
6730	 * Allocate the data buffer and grab the user's data.  In theory,
6731	 * we shouldn't have to sanity check the parameter list length here
6732	 * because the maximum size is 64K.  We should be able to malloc
6733	 * that much without too many problems.
6734	 */
6735	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6736		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6737		ctsio->kern_data_len = param_len;
6738		ctsio->kern_total_len = param_len;
6739		ctsio->kern_data_resid = 0;
6740		ctsio->kern_rel_offset = 0;
6741		ctsio->kern_sg_entries = 0;
6742		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6743		ctsio->be_move_done = ctl_config_move_done;
6744		ctl_datamove((union ctl_io *)ctsio);
6745
6746		return (CTL_RETVAL_COMPLETE);
6747	}
6748
6749	switch (ctsio->cdb[0]) {
6750	case MODE_SELECT_6: {
6751		struct scsi_mode_header_6 *mh6;
6752
6753		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6754		bd_len = mh6->blk_desc_len;
6755		break;
6756	}
6757	case MODE_SELECT_10: {
6758		struct scsi_mode_header_10 *mh10;
6759
6760		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6761		bd_len = scsi_2btoul(mh10->blk_desc_len);
6762		break;
6763	}
6764	default:
6765		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6766		break;
6767	}
6768
6769	if (param_len < (header_size + bd_len)) {
6770		free(ctsio->kern_data_ptr, M_CTL);
6771		ctl_set_param_len_error(ctsio);
6772		ctl_done((union ctl_io *)ctsio);
6773		return (CTL_RETVAL_COMPLETE);
6774	}
6775
6776	/*
6777	 * Set the IO_CONT flag, so that if this I/O gets passed to
6778	 * ctl_config_write_done(), it'll get passed back to
6779	 * ctl_do_mode_select() for further processing, or completion if
6780	 * we're all done.
6781	 */
6782	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6783	ctsio->io_cont = ctl_do_mode_select;
6784
6785	modepage_info = (union ctl_modepage_info *)
6786		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6787
6788	memset(modepage_info, 0, sizeof(*modepage_info));
6789
6790	len_left = param_len - header_size - bd_len;
6791	len_used = header_size + bd_len;
6792
6793	modepage_info->header.len_left = len_left;
6794	modepage_info->header.len_used = len_used;
6795
6796	return (ctl_do_mode_select((union ctl_io *)ctsio));
6797}
6798
6799int
6800ctl_mode_sense(struct ctl_scsiio *ctsio)
6801{
6802	struct ctl_lun *lun;
6803	int pc, page_code, dbd, llba, subpage;
6804	int alloc_len, page_len, header_len, total_len;
6805	struct scsi_mode_block_descr *block_desc;
6806	struct ctl_page_index *page_index;
6807	int control_dev;
6808
6809	dbd = 0;
6810	llba = 0;
6811	block_desc = NULL;
6812	page_index = NULL;
6813
6814	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6815
6816	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6817
6818	if (lun->be_lun->lun_type != T_DIRECT)
6819		control_dev = 1;
6820	else
6821		control_dev = 0;
6822
6823	if (lun->flags & CTL_LUN_PR_RESERVED) {
6824		uint32_t residx;
6825
6826		/*
6827		 * XXX KDM need a lock here.
6828		 */
6829		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6830		if ((lun->res_type == SPR_TYPE_EX_AC
6831		  && residx != lun->pr_res_idx)
6832		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6833		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6834		  && !lun->per_res[residx].registered)) {
6835			ctl_set_reservation_conflict(ctsio);
6836			ctl_done((union ctl_io *)ctsio);
6837			return (CTL_RETVAL_COMPLETE);
6838		}
6839	}
6840
6841	switch (ctsio->cdb[0]) {
6842	case MODE_SENSE_6: {
6843		struct scsi_mode_sense_6 *cdb;
6844
6845		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6846
6847		header_len = sizeof(struct scsi_mode_hdr_6);
6848		if (cdb->byte2 & SMS_DBD)
6849			dbd = 1;
6850		else
6851			header_len += sizeof(struct scsi_mode_block_descr);
6852
6853		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6854		page_code = cdb->page & SMS_PAGE_CODE;
6855		subpage = cdb->subpage;
6856		alloc_len = cdb->length;
6857		break;
6858	}
6859	case MODE_SENSE_10: {
6860		struct scsi_mode_sense_10 *cdb;
6861
6862		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6863
6864		header_len = sizeof(struct scsi_mode_hdr_10);
6865
6866		if (cdb->byte2 & SMS_DBD)
6867			dbd = 1;
6868		else
6869			header_len += sizeof(struct scsi_mode_block_descr);
6870		if (cdb->byte2 & SMS10_LLBAA)
6871			llba = 1;
6872		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6873		page_code = cdb->page & SMS_PAGE_CODE;
6874		subpage = cdb->subpage;
6875		alloc_len = scsi_2btoul(cdb->length);
6876		break;
6877	}
6878	default:
6879		ctl_set_invalid_opcode(ctsio);
6880		ctl_done((union ctl_io *)ctsio);
6881		return (CTL_RETVAL_COMPLETE);
6882		break; /* NOTREACHED */
6883	}
6884
6885	/*
6886	 * We have to make a first pass through to calculate the size of
6887	 * the pages that match the user's query.  Then we allocate enough
6888	 * memory to hold it, and actually copy the data into the buffer.
6889	 */
6890	switch (page_code) {
6891	case SMS_ALL_PAGES_PAGE: {
6892		int i;
6893
6894		page_len = 0;
6895
6896		/*
6897		 * At the moment, values other than 0 and 0xff here are
6898		 * reserved according to SPC-3.
6899		 */
6900		if ((subpage != SMS_SUBPAGE_PAGE_0)
6901		 && (subpage != SMS_SUBPAGE_ALL)) {
6902			ctl_set_invalid_field(ctsio,
6903					      /*sks_valid*/ 1,
6904					      /*command*/ 1,
6905					      /*field*/ 3,
6906					      /*bit_valid*/ 0,
6907					      /*bit*/ 0);
6908			ctl_done((union ctl_io *)ctsio);
6909			return (CTL_RETVAL_COMPLETE);
6910		}
6911
6912		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6913			if ((control_dev != 0)
6914			 && (lun->mode_pages.index[i].page_flags &
6915			     CTL_PAGE_FLAG_DISK_ONLY))
6916				continue;
6917
6918			/*
6919			 * We don't use this subpage if the user didn't
6920			 * request all subpages.
6921			 */
6922			if ((lun->mode_pages.index[i].subpage != 0)
6923			 && (subpage == SMS_SUBPAGE_PAGE_0))
6924				continue;
6925
6926#if 0
6927			printf("found page %#x len %d\n",
6928			       lun->mode_pages.index[i].page_code &
6929			       SMPH_PC_MASK,
6930			       lun->mode_pages.index[i].page_len);
6931#endif
6932			page_len += lun->mode_pages.index[i].page_len;
6933		}
6934		break;
6935	}
6936	default: {
6937		int i;
6938
6939		page_len = 0;
6940
6941		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6942			/* Look for the right page code */
6943			if ((lun->mode_pages.index[i].page_code &
6944			     SMPH_PC_MASK) != page_code)
6945				continue;
6946
6947			/* Look for the right subpage or the subpage wildcard*/
6948			if ((lun->mode_pages.index[i].subpage != subpage)
6949			 && (subpage != SMS_SUBPAGE_ALL))
6950				continue;
6951
6952			/* Make sure the page is supported for this dev type */
6953			if ((control_dev != 0)
6954			 && (lun->mode_pages.index[i].page_flags &
6955			     CTL_PAGE_FLAG_DISK_ONLY))
6956				continue;
6957
6958#if 0
6959			printf("found page %#x len %d\n",
6960			       lun->mode_pages.index[i].page_code &
6961			       SMPH_PC_MASK,
6962			       lun->mode_pages.index[i].page_len);
6963#endif
6964
6965			page_len += lun->mode_pages.index[i].page_len;
6966		}
6967
6968		if (page_len == 0) {
6969			ctl_set_invalid_field(ctsio,
6970					      /*sks_valid*/ 1,
6971					      /*command*/ 1,
6972					      /*field*/ 2,
6973					      /*bit_valid*/ 1,
6974					      /*bit*/ 5);
6975			ctl_done((union ctl_io *)ctsio);
6976			return (CTL_RETVAL_COMPLETE);
6977		}
6978		break;
6979	}
6980	}
6981
6982	total_len = header_len + page_len;
6983#if 0
6984	printf("header_len = %d, page_len = %d, total_len = %d\n",
6985	       header_len, page_len, total_len);
6986#endif
6987
6988	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6989	ctsio->kern_sg_entries = 0;
6990	ctsio->kern_data_resid = 0;
6991	ctsio->kern_rel_offset = 0;
6992	if (total_len < alloc_len) {
6993		ctsio->residual = alloc_len - total_len;
6994		ctsio->kern_data_len = total_len;
6995		ctsio->kern_total_len = total_len;
6996	} else {
6997		ctsio->residual = 0;
6998		ctsio->kern_data_len = alloc_len;
6999		ctsio->kern_total_len = alloc_len;
7000	}
7001
7002	switch (ctsio->cdb[0]) {
7003	case MODE_SENSE_6: {
7004		struct scsi_mode_hdr_6 *header;
7005
7006		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7007
7008		header->datalen = ctl_min(total_len - 1, 254);
7009
7010		if (dbd)
7011			header->block_descr_len = 0;
7012		else
7013			header->block_descr_len =
7014				sizeof(struct scsi_mode_block_descr);
7015		block_desc = (struct scsi_mode_block_descr *)&header[1];
7016		break;
7017	}
7018	case MODE_SENSE_10: {
7019		struct scsi_mode_hdr_10 *header;
7020		int datalen;
7021
7022		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7023
7024		datalen = ctl_min(total_len - 2, 65533);
7025		scsi_ulto2b(datalen, header->datalen);
7026		if (dbd)
7027			scsi_ulto2b(0, header->block_descr_len);
7028		else
7029			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7030				    header->block_descr_len);
7031		block_desc = (struct scsi_mode_block_descr *)&header[1];
7032		break;
7033	}
7034	default:
7035		panic("invalid CDB type %#x", ctsio->cdb[0]);
7036		break; /* NOTREACHED */
7037	}
7038
7039	/*
7040	 * If we've got a disk, use its blocksize in the block
7041	 * descriptor.  Otherwise, just set it to 0.
7042	 */
7043	if (dbd == 0) {
7044		if (control_dev != 0)
7045			scsi_ulto3b(lun->be_lun->blocksize,
7046				    block_desc->block_len);
7047		else
7048			scsi_ulto3b(0, block_desc->block_len);
7049	}
7050
7051	switch (page_code) {
7052	case SMS_ALL_PAGES_PAGE: {
7053		int i, data_used;
7054
7055		data_used = header_len;
7056		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7057			struct ctl_page_index *page_index;
7058
7059			page_index = &lun->mode_pages.index[i];
7060
7061			if ((control_dev != 0)
7062			 && (page_index->page_flags &
7063			    CTL_PAGE_FLAG_DISK_ONLY))
7064				continue;
7065
7066			/*
7067			 * We don't use this subpage if the user didn't
7068			 * request all subpages.  We already checked (above)
7069			 * to make sure the user only specified a subpage
7070			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7071			 */
7072			if ((page_index->subpage != 0)
7073			 && (subpage == SMS_SUBPAGE_PAGE_0))
7074				continue;
7075
7076			/*
7077			 * Call the handler, if it exists, to update the
7078			 * page to the latest values.
7079			 */
7080			if (page_index->sense_handler != NULL)
7081				page_index->sense_handler(ctsio, page_index,pc);
7082
7083			memcpy(ctsio->kern_data_ptr + data_used,
7084			       page_index->page_data +
7085			       (page_index->page_len * pc),
7086			       page_index->page_len);
7087			data_used += page_index->page_len;
7088		}
7089		break;
7090	}
7091	default: {
7092		int i, data_used;
7093
7094		data_used = header_len;
7095
7096		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7097			struct ctl_page_index *page_index;
7098
7099			page_index = &lun->mode_pages.index[i];
7100
7101			/* Look for the right page code */
7102			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7103				continue;
7104
7105			/* Look for the right subpage or the subpage wildcard*/
7106			if ((page_index->subpage != subpage)
7107			 && (subpage != SMS_SUBPAGE_ALL))
7108				continue;
7109
7110			/* Make sure the page is supported for this dev type */
7111			if ((control_dev != 0)
7112			 && (page_index->page_flags &
7113			     CTL_PAGE_FLAG_DISK_ONLY))
7114				continue;
7115
7116			/*
7117			 * Call the handler, if it exists, to update the
7118			 * page to the latest values.
7119			 */
7120			if (page_index->sense_handler != NULL)
7121				page_index->sense_handler(ctsio, page_index,pc);
7122
7123			memcpy(ctsio->kern_data_ptr + data_used,
7124			       page_index->page_data +
7125			       (page_index->page_len * pc),
7126			       page_index->page_len);
7127			data_used += page_index->page_len;
7128		}
7129		break;
7130	}
7131	}
7132
7133	ctsio->scsi_status = SCSI_STATUS_OK;
7134
7135	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7136	ctsio->be_move_done = ctl_config_move_done;
7137	ctl_datamove((union ctl_io *)ctsio);
7138
7139	return (CTL_RETVAL_COMPLETE);
7140}
7141
7142int
7143ctl_read_capacity(struct ctl_scsiio *ctsio)
7144{
7145	struct scsi_read_capacity *cdb;
7146	struct scsi_read_capacity_data *data;
7147	struct ctl_lun *lun;
7148	uint32_t lba;
7149
7150	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7151
7152	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7153
7154	lba = scsi_4btoul(cdb->addr);
7155	if (((cdb->pmi & SRC_PMI) == 0)
7156	 && (lba != 0)) {
7157		ctl_set_invalid_field(/*ctsio*/ ctsio,
7158				      /*sks_valid*/ 1,
7159				      /*command*/ 1,
7160				      /*field*/ 2,
7161				      /*bit_valid*/ 0,
7162				      /*bit*/ 0);
7163		ctl_done((union ctl_io *)ctsio);
7164		return (CTL_RETVAL_COMPLETE);
7165	}
7166
7167	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7168
7169	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7170	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7171	ctsio->residual = 0;
7172	ctsio->kern_data_len = sizeof(*data);
7173	ctsio->kern_total_len = sizeof(*data);
7174	ctsio->kern_data_resid = 0;
7175	ctsio->kern_rel_offset = 0;
7176	ctsio->kern_sg_entries = 0;
7177
7178	/*
7179	 * If the maximum LBA is greater than 0xfffffffe, the user must
7180	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7181	 * serivce action set.
7182	 */
7183	if (lun->be_lun->maxlba > 0xfffffffe)
7184		scsi_ulto4b(0xffffffff, data->addr);
7185	else
7186		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7187
7188	/*
7189	 * XXX KDM this may not be 512 bytes...
7190	 */
7191	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7192
7193	ctsio->scsi_status = SCSI_STATUS_OK;
7194
7195	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7196	ctsio->be_move_done = ctl_config_move_done;
7197	ctl_datamove((union ctl_io *)ctsio);
7198
7199	return (CTL_RETVAL_COMPLETE);
7200}
7201
7202int
7203ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7204{
7205	struct scsi_read_capacity_16 *cdb;
7206	struct scsi_read_capacity_data_long *data;
7207	struct ctl_lun *lun;
7208	uint64_t lba;
7209	uint32_t alloc_len;
7210
7211	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7212
7213	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7214
7215	alloc_len = scsi_4btoul(cdb->alloc_len);
7216	lba = scsi_8btou64(cdb->addr);
7217
7218	if ((cdb->reladr & SRC16_PMI)
7219	 && (lba != 0)) {
7220		ctl_set_invalid_field(/*ctsio*/ ctsio,
7221				      /*sks_valid*/ 1,
7222				      /*command*/ 1,
7223				      /*field*/ 2,
7224				      /*bit_valid*/ 0,
7225				      /*bit*/ 0);
7226		ctl_done((union ctl_io *)ctsio);
7227		return (CTL_RETVAL_COMPLETE);
7228	}
7229
7230	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7231
7232	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7233	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7234
7235	if (sizeof(*data) < alloc_len) {
7236		ctsio->residual = alloc_len - sizeof(*data);
7237		ctsio->kern_data_len = sizeof(*data);
7238		ctsio->kern_total_len = sizeof(*data);
7239	} else {
7240		ctsio->residual = 0;
7241		ctsio->kern_data_len = alloc_len;
7242		ctsio->kern_total_len = alloc_len;
7243	}
7244	ctsio->kern_data_resid = 0;
7245	ctsio->kern_rel_offset = 0;
7246	ctsio->kern_sg_entries = 0;
7247
7248	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7249	/* XXX KDM this may not be 512 bytes... */
7250	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7251	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7252	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7253	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7254		data->lalba_lbp[0] |= SRC16_LBPME;
7255
7256	ctsio->scsi_status = SCSI_STATUS_OK;
7257
7258	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7259	ctsio->be_move_done = ctl_config_move_done;
7260	ctl_datamove((union ctl_io *)ctsio);
7261
7262	return (CTL_RETVAL_COMPLETE);
7263}
7264
7265int
7266ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7267{
7268	struct scsi_maintenance_in *cdb;
7269	int retval;
7270	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7271	int num_target_port_groups, num_target_ports, single;
7272	struct ctl_lun *lun;
7273	struct ctl_softc *softc;
7274	struct ctl_port *port;
7275	struct scsi_target_group_data *rtg_ptr;
7276	struct scsi_target_group_data_extended *rtg_ext_ptr;
7277	struct scsi_target_port_group_descriptor *tpg_desc;
7278
7279	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7280
7281	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7282	softc = control_softc;
7283	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7284
7285	retval = CTL_RETVAL_COMPLETE;
7286
7287	switch (cdb->byte2 & STG_PDF_MASK) {
7288	case STG_PDF_LENGTH:
7289		ext = 0;
7290		break;
7291	case STG_PDF_EXTENDED:
7292		ext = 1;
7293		break;
7294	default:
7295		ctl_set_invalid_field(/*ctsio*/ ctsio,
7296				      /*sks_valid*/ 1,
7297				      /*command*/ 1,
7298				      /*field*/ 2,
7299				      /*bit_valid*/ 1,
7300				      /*bit*/ 5);
7301		ctl_done((union ctl_io *)ctsio);
7302		return(retval);
7303	}
7304
7305	single = ctl_is_single;
7306	if (single)
7307		num_target_port_groups = 1;
7308	else
7309		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7310	num_target_ports = 0;
7311	mtx_lock(&softc->ctl_lock);
7312	STAILQ_FOREACH(port, &softc->port_list, links) {
7313		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7314			continue;
7315		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7316			continue;
7317		num_target_ports++;
7318	}
7319	mtx_unlock(&softc->ctl_lock);
7320
7321	if (ext)
7322		total_len = sizeof(struct scsi_target_group_data_extended);
7323	else
7324		total_len = sizeof(struct scsi_target_group_data);
7325	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7326		num_target_port_groups +
7327	    sizeof(struct scsi_target_port_descriptor) *
7328		num_target_ports * num_target_port_groups;
7329
7330	alloc_len = scsi_4btoul(cdb->length);
7331
7332	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7333
7334	ctsio->kern_sg_entries = 0;
7335
7336	if (total_len < alloc_len) {
7337		ctsio->residual = alloc_len - total_len;
7338		ctsio->kern_data_len = total_len;
7339		ctsio->kern_total_len = total_len;
7340	} else {
7341		ctsio->residual = 0;
7342		ctsio->kern_data_len = alloc_len;
7343		ctsio->kern_total_len = alloc_len;
7344	}
7345	ctsio->kern_data_resid = 0;
7346	ctsio->kern_rel_offset = 0;
7347
7348	if (ext) {
7349		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7350		    ctsio->kern_data_ptr;
7351		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7352		rtg_ext_ptr->format_type = 0x10;
7353		rtg_ext_ptr->implicit_transition_time = 0;
7354		tpg_desc = &rtg_ext_ptr->groups[0];
7355	} else {
7356		rtg_ptr = (struct scsi_target_group_data *)
7357		    ctsio->kern_data_ptr;
7358		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7359		tpg_desc = &rtg_ptr->groups[0];
7360	}
7361
7362	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7363	mtx_lock(&softc->ctl_lock);
7364	for (g = 0; g < num_target_port_groups; g++) {
7365		if (g == pg)
7366			tpg_desc->pref_state = TPG_PRIMARY |
7367			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7368		else
7369			tpg_desc->pref_state =
7370			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7371		tpg_desc->support = TPG_AO_SUP;
7372		if (!single)
7373			tpg_desc->support |= TPG_AN_SUP;
7374		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7375		tpg_desc->status = TPG_IMPLICIT;
7376		pc = 0;
7377		STAILQ_FOREACH(port, &softc->port_list, links) {
7378			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7379				continue;
7380			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7381			    CTL_MAX_LUNS)
7382				continue;
7383			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7384			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7385			    relative_target_port_identifier);
7386			pc++;
7387		}
7388		tpg_desc->target_port_count = pc;
7389		tpg_desc = (struct scsi_target_port_group_descriptor *)
7390		    &tpg_desc->descriptors[pc];
7391	}
7392	mtx_unlock(&softc->ctl_lock);
7393
7394	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7395	ctsio->be_move_done = ctl_config_move_done;
7396
7397	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7398			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7399			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7400			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7401			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7402
7403	ctl_datamove((union ctl_io *)ctsio);
7404	return(retval);
7405}
7406
7407int
7408ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7409{
7410	struct ctl_lun *lun;
7411	struct scsi_report_supported_opcodes *cdb;
7412	const struct ctl_cmd_entry *entry, *sentry;
7413	struct scsi_report_supported_opcodes_all *all;
7414	struct scsi_report_supported_opcodes_descr *descr;
7415	struct scsi_report_supported_opcodes_one *one;
7416	int retval;
7417	int alloc_len, total_len;
7418	int opcode, service_action, i, j, num;
7419
7420	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7421
7422	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7423	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7424
7425	retval = CTL_RETVAL_COMPLETE;
7426
7427	opcode = cdb->requested_opcode;
7428	service_action = scsi_2btoul(cdb->requested_service_action);
7429	switch (cdb->options & RSO_OPTIONS_MASK) {
7430	case RSO_OPTIONS_ALL:
7431		num = 0;
7432		for (i = 0; i < 256; i++) {
7433			entry = &ctl_cmd_table[i];
7434			if (entry->flags & CTL_CMD_FLAG_SA5) {
7435				for (j = 0; j < 32; j++) {
7436					sentry = &((const struct ctl_cmd_entry *)
7437					    entry->execute)[j];
7438					if (ctl_cmd_applicable(
7439					    lun->be_lun->lun_type, sentry))
7440						num++;
7441				}
7442			} else {
7443				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7444				    entry))
7445					num++;
7446			}
7447		}
7448		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7449		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7450		break;
7451	case RSO_OPTIONS_OC:
7452		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7453			ctl_set_invalid_field(/*ctsio*/ ctsio,
7454					      /*sks_valid*/ 1,
7455					      /*command*/ 1,
7456					      /*field*/ 2,
7457					      /*bit_valid*/ 1,
7458					      /*bit*/ 2);
7459			ctl_done((union ctl_io *)ctsio);
7460			return (CTL_RETVAL_COMPLETE);
7461		}
7462		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7463		break;
7464	case RSO_OPTIONS_OC_SA:
7465		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7466		    service_action >= 32) {
7467			ctl_set_invalid_field(/*ctsio*/ ctsio,
7468					      /*sks_valid*/ 1,
7469					      /*command*/ 1,
7470					      /*field*/ 2,
7471					      /*bit_valid*/ 1,
7472					      /*bit*/ 2);
7473			ctl_done((union ctl_io *)ctsio);
7474			return (CTL_RETVAL_COMPLETE);
7475		}
7476		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7477		break;
7478	default:
7479		ctl_set_invalid_field(/*ctsio*/ ctsio,
7480				      /*sks_valid*/ 1,
7481				      /*command*/ 1,
7482				      /*field*/ 2,
7483				      /*bit_valid*/ 1,
7484				      /*bit*/ 2);
7485		ctl_done((union ctl_io *)ctsio);
7486		return (CTL_RETVAL_COMPLETE);
7487	}
7488
7489	alloc_len = scsi_4btoul(cdb->length);
7490
7491	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7492
7493	ctsio->kern_sg_entries = 0;
7494
7495	if (total_len < alloc_len) {
7496		ctsio->residual = alloc_len - total_len;
7497		ctsio->kern_data_len = total_len;
7498		ctsio->kern_total_len = total_len;
7499	} else {
7500		ctsio->residual = 0;
7501		ctsio->kern_data_len = alloc_len;
7502		ctsio->kern_total_len = alloc_len;
7503	}
7504	ctsio->kern_data_resid = 0;
7505	ctsio->kern_rel_offset = 0;
7506
7507	switch (cdb->options & RSO_OPTIONS_MASK) {
7508	case RSO_OPTIONS_ALL:
7509		all = (struct scsi_report_supported_opcodes_all *)
7510		    ctsio->kern_data_ptr;
7511		num = 0;
7512		for (i = 0; i < 256; i++) {
7513			entry = &ctl_cmd_table[i];
7514			if (entry->flags & CTL_CMD_FLAG_SA5) {
7515				for (j = 0; j < 32; j++) {
7516					sentry = &((const struct ctl_cmd_entry *)
7517					    entry->execute)[j];
7518					if (!ctl_cmd_applicable(
7519					    lun->be_lun->lun_type, sentry))
7520						continue;
7521					descr = &all->descr[num++];
7522					descr->opcode = i;
7523					scsi_ulto2b(j, descr->service_action);
7524					descr->flags = RSO_SERVACTV;
7525					scsi_ulto2b(sentry->length,
7526					    descr->cdb_length);
7527				}
7528			} else {
7529				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7530				    entry))
7531					continue;
7532				descr = &all->descr[num++];
7533				descr->opcode = i;
7534				scsi_ulto2b(0, descr->service_action);
7535				descr->flags = 0;
7536				scsi_ulto2b(entry->length, descr->cdb_length);
7537			}
7538		}
7539		scsi_ulto4b(
7540		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7541		    all->length);
7542		break;
7543	case RSO_OPTIONS_OC:
7544		one = (struct scsi_report_supported_opcodes_one *)
7545		    ctsio->kern_data_ptr;
7546		entry = &ctl_cmd_table[opcode];
7547		goto fill_one;
7548	case RSO_OPTIONS_OC_SA:
7549		one = (struct scsi_report_supported_opcodes_one *)
7550		    ctsio->kern_data_ptr;
7551		entry = &ctl_cmd_table[opcode];
7552		entry = &((const struct ctl_cmd_entry *)
7553		    entry->execute)[service_action];
7554fill_one:
7555		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7556			one->support = 3;
7557			scsi_ulto2b(entry->length, one->cdb_length);
7558			one->cdb_usage[0] = opcode;
7559			memcpy(&one->cdb_usage[1], entry->usage,
7560			    entry->length - 1);
7561		} else
7562			one->support = 1;
7563		break;
7564	}
7565
7566	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7567	ctsio->be_move_done = ctl_config_move_done;
7568
7569	ctl_datamove((union ctl_io *)ctsio);
7570	return(retval);
7571}
7572
7573int
7574ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7575{
7576	struct ctl_lun *lun;
7577	struct scsi_report_supported_tmf *cdb;
7578	struct scsi_report_supported_tmf_data *data;
7579	int retval;
7580	int alloc_len, total_len;
7581
7582	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7583
7584	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7585	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7586
7587	retval = CTL_RETVAL_COMPLETE;
7588
7589	total_len = sizeof(struct scsi_report_supported_tmf_data);
7590	alloc_len = scsi_4btoul(cdb->length);
7591
7592	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7593
7594	ctsio->kern_sg_entries = 0;
7595
7596	if (total_len < alloc_len) {
7597		ctsio->residual = alloc_len - total_len;
7598		ctsio->kern_data_len = total_len;
7599		ctsio->kern_total_len = total_len;
7600	} else {
7601		ctsio->residual = 0;
7602		ctsio->kern_data_len = alloc_len;
7603		ctsio->kern_total_len = alloc_len;
7604	}
7605	ctsio->kern_data_resid = 0;
7606	ctsio->kern_rel_offset = 0;
7607
7608	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7609	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7610	data->byte2 |= RST_ITNRS;
7611
7612	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7613	ctsio->be_move_done = ctl_config_move_done;
7614
7615	ctl_datamove((union ctl_io *)ctsio);
7616	return (retval);
7617}
7618
7619int
7620ctl_report_timestamp(struct ctl_scsiio *ctsio)
7621{
7622	struct ctl_lun *lun;
7623	struct scsi_report_timestamp *cdb;
7624	struct scsi_report_timestamp_data *data;
7625	struct timeval tv;
7626	int64_t timestamp;
7627	int retval;
7628	int alloc_len, total_len;
7629
7630	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7631
7632	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7633	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7634
7635	retval = CTL_RETVAL_COMPLETE;
7636
7637	total_len = sizeof(struct scsi_report_timestamp_data);
7638	alloc_len = scsi_4btoul(cdb->length);
7639
7640	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7641
7642	ctsio->kern_sg_entries = 0;
7643
7644	if (total_len < alloc_len) {
7645		ctsio->residual = alloc_len - total_len;
7646		ctsio->kern_data_len = total_len;
7647		ctsio->kern_total_len = total_len;
7648	} else {
7649		ctsio->residual = 0;
7650		ctsio->kern_data_len = alloc_len;
7651		ctsio->kern_total_len = alloc_len;
7652	}
7653	ctsio->kern_data_resid = 0;
7654	ctsio->kern_rel_offset = 0;
7655
7656	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7657	scsi_ulto2b(sizeof(*data) - 2, data->length);
7658	data->origin = RTS_ORIG_OUTSIDE;
7659	getmicrotime(&tv);
7660	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7661	scsi_ulto4b(timestamp >> 16, data->timestamp);
7662	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7663
7664	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7665	ctsio->be_move_done = ctl_config_move_done;
7666
7667	ctl_datamove((union ctl_io *)ctsio);
7668	return (retval);
7669}
7670
7671int
7672ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7673{
7674	struct scsi_per_res_in *cdb;
7675	int alloc_len, total_len = 0;
7676	/* struct scsi_per_res_in_rsrv in_data; */
7677	struct ctl_lun *lun;
7678	struct ctl_softc *softc;
7679
7680	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7681
7682	softc = control_softc;
7683
7684	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7685
7686	alloc_len = scsi_2btoul(cdb->length);
7687
7688	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7689
7690retry:
7691	mtx_lock(&lun->lun_lock);
7692	switch (cdb->action) {
7693	case SPRI_RK: /* read keys */
7694		total_len = sizeof(struct scsi_per_res_in_keys) +
7695			lun->pr_key_count *
7696			sizeof(struct scsi_per_res_key);
7697		break;
7698	case SPRI_RR: /* read reservation */
7699		if (lun->flags & CTL_LUN_PR_RESERVED)
7700			total_len = sizeof(struct scsi_per_res_in_rsrv);
7701		else
7702			total_len = sizeof(struct scsi_per_res_in_header);
7703		break;
7704	case SPRI_RC: /* report capabilities */
7705		total_len = sizeof(struct scsi_per_res_cap);
7706		break;
7707	case SPRI_RS: /* read full status */
7708		total_len = sizeof(struct scsi_per_res_in_header) +
7709		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7710		    lun->pr_key_count;
7711		break;
7712	default:
7713		panic("Invalid PR type %x", cdb->action);
7714	}
7715	mtx_unlock(&lun->lun_lock);
7716
7717	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7718
7719	if (total_len < alloc_len) {
7720		ctsio->residual = alloc_len - total_len;
7721		ctsio->kern_data_len = total_len;
7722		ctsio->kern_total_len = total_len;
7723	} else {
7724		ctsio->residual = 0;
7725		ctsio->kern_data_len = alloc_len;
7726		ctsio->kern_total_len = alloc_len;
7727	}
7728
7729	ctsio->kern_data_resid = 0;
7730	ctsio->kern_rel_offset = 0;
7731	ctsio->kern_sg_entries = 0;
7732
7733	mtx_lock(&lun->lun_lock);
7734	switch (cdb->action) {
7735	case SPRI_RK: { // read keys
7736        struct scsi_per_res_in_keys *res_keys;
7737		int i, key_count;
7738
7739		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7740
7741		/*
7742		 * We had to drop the lock to allocate our buffer, which
7743		 * leaves time for someone to come in with another
7744		 * persistent reservation.  (That is unlikely, though,
7745		 * since this should be the only persistent reservation
7746		 * command active right now.)
7747		 */
7748		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7749		    (lun->pr_key_count *
7750		     sizeof(struct scsi_per_res_key)))){
7751			mtx_unlock(&lun->lun_lock);
7752			free(ctsio->kern_data_ptr, M_CTL);
7753			printf("%s: reservation length changed, retrying\n",
7754			       __func__);
7755			goto retry;
7756		}
7757
7758		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7759
7760		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7761			     lun->pr_key_count, res_keys->header.length);
7762
7763		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7764			if (!lun->per_res[i].registered)
7765				continue;
7766
7767			/*
7768			 * We used lun->pr_key_count to calculate the
7769			 * size to allocate.  If it turns out the number of
7770			 * initiators with the registered flag set is
7771			 * larger than that (i.e. they haven't been kept in
7772			 * sync), we've got a problem.
7773			 */
7774			if (key_count >= lun->pr_key_count) {
7775#ifdef NEEDTOPORT
7776				csevent_log(CSC_CTL | CSC_SHELF_SW |
7777					    CTL_PR_ERROR,
7778					    csevent_LogType_Fault,
7779					    csevent_AlertLevel_Yellow,
7780					    csevent_FRU_ShelfController,
7781					    csevent_FRU_Firmware,
7782				        csevent_FRU_Unknown,
7783					    "registered keys %d >= key "
7784					    "count %d", key_count,
7785					    lun->pr_key_count);
7786#endif
7787				key_count++;
7788				continue;
7789			}
7790			memcpy(res_keys->keys[key_count].key,
7791			       lun->per_res[i].res_key.key,
7792			       ctl_min(sizeof(res_keys->keys[key_count].key),
7793			       sizeof(lun->per_res[i].res_key)));
7794			key_count++;
7795		}
7796		break;
7797	}
7798	case SPRI_RR: { // read reservation
7799		struct scsi_per_res_in_rsrv *res;
7800		int tmp_len, header_only;
7801
7802		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7803
7804		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7805
7806		if (lun->flags & CTL_LUN_PR_RESERVED)
7807		{
7808			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7809			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7810				    res->header.length);
7811			header_only = 0;
7812		} else {
7813			tmp_len = sizeof(struct scsi_per_res_in_header);
7814			scsi_ulto4b(0, res->header.length);
7815			header_only = 1;
7816		}
7817
7818		/*
7819		 * We had to drop the lock to allocate our buffer, which
7820		 * leaves time for someone to come in with another
7821		 * persistent reservation.  (That is unlikely, though,
7822		 * since this should be the only persistent reservation
7823		 * command active right now.)
7824		 */
7825		if (tmp_len != total_len) {
7826			mtx_unlock(&lun->lun_lock);
7827			free(ctsio->kern_data_ptr, M_CTL);
7828			printf("%s: reservation status changed, retrying\n",
7829			       __func__);
7830			goto retry;
7831		}
7832
7833		/*
7834		 * No reservation held, so we're done.
7835		 */
7836		if (header_only != 0)
7837			break;
7838
7839		/*
7840		 * If the registration is an All Registrants type, the key
7841		 * is 0, since it doesn't really matter.
7842		 */
7843		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7844			memcpy(res->data.reservation,
7845			       &lun->per_res[lun->pr_res_idx].res_key,
7846			       sizeof(struct scsi_per_res_key));
7847		}
7848		res->data.scopetype = lun->res_type;
7849		break;
7850	}
7851	case SPRI_RC:     //report capabilities
7852	{
7853		struct scsi_per_res_cap *res_cap;
7854		uint16_t type_mask;
7855
7856		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7857		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7858		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_3;
7859		type_mask = SPRI_TM_WR_EX_AR |
7860			    SPRI_TM_EX_AC_RO |
7861			    SPRI_TM_WR_EX_RO |
7862			    SPRI_TM_EX_AC |
7863			    SPRI_TM_WR_EX |
7864			    SPRI_TM_EX_AC_AR;
7865		scsi_ulto2b(type_mask, res_cap->type_mask);
7866		break;
7867	}
7868	case SPRI_RS: { // read full status
7869		struct scsi_per_res_in_full *res_status;
7870		struct scsi_per_res_in_full_desc *res_desc;
7871		struct ctl_port *port;
7872		int i, len;
7873
7874		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7875
7876		/*
7877		 * We had to drop the lock to allocate our buffer, which
7878		 * leaves time for someone to come in with another
7879		 * persistent reservation.  (That is unlikely, though,
7880		 * since this should be the only persistent reservation
7881		 * command active right now.)
7882		 */
7883		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7884		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7885		     lun->pr_key_count)){
7886			mtx_unlock(&lun->lun_lock);
7887			free(ctsio->kern_data_ptr, M_CTL);
7888			printf("%s: reservation length changed, retrying\n",
7889			       __func__);
7890			goto retry;
7891		}
7892
7893		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7894
7895		res_desc = &res_status->desc[0];
7896		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7897			if (!lun->per_res[i].registered)
7898				continue;
7899
7900			memcpy(&res_desc->res_key, &lun->per_res[i].res_key.key,
7901			    sizeof(res_desc->res_key));
7902			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7903			    (lun->pr_res_idx == i ||
7904			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7905				res_desc->flags = SPRI_FULL_R_HOLDER;
7906				res_desc->scopetype = lun->res_type;
7907			}
7908			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7909			    res_desc->rel_trgt_port_id);
7910			len = 0;
7911			port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7912			if (port != NULL)
7913				len = ctl_create_iid(port,
7914				    i % CTL_MAX_INIT_PER_PORT,
7915				    res_desc->transport_id);
7916			scsi_ulto4b(len, res_desc->additional_length);
7917			res_desc = (struct scsi_per_res_in_full_desc *)
7918			    &res_desc->transport_id[len];
7919		}
7920		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7921		    res_status->header.length);
7922		break;
7923	}
7924	default:
7925		/*
7926		 * This is a bug, because we just checked for this above,
7927		 * and should have returned an error.
7928		 */
7929		panic("Invalid PR type %x", cdb->action);
7930		break; /* NOTREACHED */
7931	}
7932	mtx_unlock(&lun->lun_lock);
7933
7934	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7935	ctsio->be_move_done = ctl_config_move_done;
7936
7937	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7938			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7939			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7940			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7941			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7942
7943	ctl_datamove((union ctl_io *)ctsio);
7944
7945	return (CTL_RETVAL_COMPLETE);
7946}
7947
7948/*
7949 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7950 * it should return.
7951 */
7952static int
7953ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7954		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7955		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7956		struct scsi_per_res_out_parms* param)
7957{
7958	union ctl_ha_msg persis_io;
7959	int retval, i;
7960	int isc_retval;
7961
7962	retval = 0;
7963
7964	mtx_lock(&lun->lun_lock);
7965	if (sa_res_key == 0) {
7966		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7967			/* validate scope and type */
7968			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7969			     SPR_LU_SCOPE) {
7970				mtx_unlock(&lun->lun_lock);
7971				ctl_set_invalid_field(/*ctsio*/ ctsio,
7972						      /*sks_valid*/ 1,
7973						      /*command*/ 1,
7974						      /*field*/ 2,
7975						      /*bit_valid*/ 1,
7976						      /*bit*/ 4);
7977				ctl_done((union ctl_io *)ctsio);
7978				return (1);
7979			}
7980
7981		        if (type>8 || type==2 || type==4 || type==0) {
7982				mtx_unlock(&lun->lun_lock);
7983				ctl_set_invalid_field(/*ctsio*/ ctsio,
7984       	           				      /*sks_valid*/ 1,
7985						      /*command*/ 1,
7986						      /*field*/ 2,
7987						      /*bit_valid*/ 1,
7988						      /*bit*/ 0);
7989				ctl_done((union ctl_io *)ctsio);
7990				return (1);
7991		        }
7992
7993			/* temporarily unregister this nexus */
7994			lun->per_res[residx].registered = 0;
7995
7996			/*
7997			 * Unregister everybody else and build UA for
7998			 * them
7999			 */
8000			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8001				if (lun->per_res[i].registered == 0)
8002					continue;
8003
8004				if (!persis_offset
8005				 && i <CTL_MAX_INITIATORS)
8006					lun->pending_sense[i].ua_pending |=
8007						CTL_UA_REG_PREEMPT;
8008				else if (persis_offset
8009				      && i >= persis_offset)
8010					lun->pending_sense[i-persis_offset
8011						].ua_pending |=
8012						CTL_UA_REG_PREEMPT;
8013				lun->per_res[i].registered = 0;
8014				memset(&lun->per_res[i].res_key, 0,
8015				       sizeof(struct scsi_per_res_key));
8016			}
8017			lun->per_res[residx].registered = 1;
8018			lun->pr_key_count = 1;
8019			lun->res_type = type;
8020			if (lun->res_type != SPR_TYPE_WR_EX_AR
8021			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8022				lun->pr_res_idx = residx;
8023
8024			/* send msg to other side */
8025			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8026			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8027			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8028			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8029			persis_io.pr.pr_info.res_type = type;
8030			memcpy(persis_io.pr.pr_info.sa_res_key,
8031			       param->serv_act_res_key,
8032			       sizeof(param->serv_act_res_key));
8033			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8034			     &persis_io, sizeof(persis_io), 0)) >
8035			     CTL_HA_STATUS_SUCCESS) {
8036				printf("CTL:Persis Out error returned "
8037				       "from ctl_ha_msg_send %d\n",
8038				       isc_retval);
8039			}
8040		} else {
8041			/* not all registrants */
8042			mtx_unlock(&lun->lun_lock);
8043			free(ctsio->kern_data_ptr, M_CTL);
8044			ctl_set_invalid_field(ctsio,
8045					      /*sks_valid*/ 1,
8046					      /*command*/ 0,
8047					      /*field*/ 8,
8048					      /*bit_valid*/ 0,
8049					      /*bit*/ 0);
8050			ctl_done((union ctl_io *)ctsio);
8051			return (1);
8052		}
8053	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8054		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8055		int found = 0;
8056
8057		if (res_key == sa_res_key) {
8058			/* special case */
8059			/*
8060			 * The spec implies this is not good but doesn't
8061			 * say what to do. There are two choices either
8062			 * generate a res conflict or check condition
8063			 * with illegal field in parameter data. Since
8064			 * that is what is done when the sa_res_key is
8065			 * zero I'll take that approach since this has
8066			 * to do with the sa_res_key.
8067			 */
8068			mtx_unlock(&lun->lun_lock);
8069			free(ctsio->kern_data_ptr, M_CTL);
8070			ctl_set_invalid_field(ctsio,
8071					      /*sks_valid*/ 1,
8072					      /*command*/ 0,
8073					      /*field*/ 8,
8074					      /*bit_valid*/ 0,
8075					      /*bit*/ 0);
8076			ctl_done((union ctl_io *)ctsio);
8077			return (1);
8078		}
8079
8080		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8081			if (lun->per_res[i].registered
8082			 && memcmp(param->serv_act_res_key,
8083			    lun->per_res[i].res_key.key,
8084			    sizeof(struct scsi_per_res_key)) != 0)
8085				continue;
8086
8087			found = 1;
8088			lun->per_res[i].registered = 0;
8089			memset(&lun->per_res[i].res_key, 0,
8090			       sizeof(struct scsi_per_res_key));
8091			lun->pr_key_count--;
8092
8093			if (!persis_offset
8094			 && i < CTL_MAX_INITIATORS)
8095				lun->pending_sense[i].ua_pending |=
8096					CTL_UA_REG_PREEMPT;
8097			else if (persis_offset
8098			      && i >= persis_offset)
8099				lun->pending_sense[i-persis_offset].ua_pending|=
8100					CTL_UA_REG_PREEMPT;
8101		}
8102		if (!found) {
8103			mtx_unlock(&lun->lun_lock);
8104			free(ctsio->kern_data_ptr, M_CTL);
8105			ctl_set_reservation_conflict(ctsio);
8106			ctl_done((union ctl_io *)ctsio);
8107			return (CTL_RETVAL_COMPLETE);
8108		}
8109		/* send msg to other side */
8110		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8111		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8112		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8113		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8114		persis_io.pr.pr_info.res_type = type;
8115		memcpy(persis_io.pr.pr_info.sa_res_key,
8116		       param->serv_act_res_key,
8117		       sizeof(param->serv_act_res_key));
8118		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8119		     &persis_io, sizeof(persis_io), 0)) >
8120		     CTL_HA_STATUS_SUCCESS) {
8121			printf("CTL:Persis Out error returned from "
8122			       "ctl_ha_msg_send %d\n", isc_retval);
8123		}
8124	} else {
8125		/* Reserved but not all registrants */
8126		/* sa_res_key is res holder */
8127		if (memcmp(param->serv_act_res_key,
8128                   lun->per_res[lun->pr_res_idx].res_key.key,
8129                   sizeof(struct scsi_per_res_key)) == 0) {
8130			/* validate scope and type */
8131			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8132			     SPR_LU_SCOPE) {
8133				mtx_unlock(&lun->lun_lock);
8134				ctl_set_invalid_field(/*ctsio*/ ctsio,
8135						      /*sks_valid*/ 1,
8136						      /*command*/ 1,
8137						      /*field*/ 2,
8138						      /*bit_valid*/ 1,
8139						      /*bit*/ 4);
8140				ctl_done((union ctl_io *)ctsio);
8141				return (1);
8142			}
8143
8144			if (type>8 || type==2 || type==4 || type==0) {
8145				mtx_unlock(&lun->lun_lock);
8146				ctl_set_invalid_field(/*ctsio*/ ctsio,
8147						      /*sks_valid*/ 1,
8148						      /*command*/ 1,
8149						      /*field*/ 2,
8150						      /*bit_valid*/ 1,
8151						      /*bit*/ 0);
8152				ctl_done((union ctl_io *)ctsio);
8153				return (1);
8154			}
8155
8156			/*
8157			 * Do the following:
8158			 * if sa_res_key != res_key remove all
8159			 * registrants w/sa_res_key and generate UA
8160			 * for these registrants(Registrations
8161			 * Preempted) if it wasn't an exclusive
8162			 * reservation generate UA(Reservations
8163			 * Preempted) for all other registered nexuses
8164			 * if the type has changed. Establish the new
8165			 * reservation and holder. If res_key and
8166			 * sa_res_key are the same do the above
8167			 * except don't unregister the res holder.
8168			 */
8169
8170			/*
8171			 * Temporarily unregister so it won't get
8172			 * removed or UA generated
8173			 */
8174			lun->per_res[residx].registered = 0;
8175			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8176				if (lun->per_res[i].registered == 0)
8177					continue;
8178
8179				if (memcmp(param->serv_act_res_key,
8180				    lun->per_res[i].res_key.key,
8181				    sizeof(struct scsi_per_res_key)) == 0) {
8182					lun->per_res[i].registered = 0;
8183					memset(&lun->per_res[i].res_key,
8184					       0,
8185					       sizeof(struct scsi_per_res_key));
8186					lun->pr_key_count--;
8187
8188					if (!persis_offset
8189					 && i < CTL_MAX_INITIATORS)
8190						lun->pending_sense[i
8191							].ua_pending |=
8192							CTL_UA_REG_PREEMPT;
8193					else if (persis_offset
8194					      && i >= persis_offset)
8195						lun->pending_sense[
8196						  i-persis_offset].ua_pending |=
8197						  CTL_UA_REG_PREEMPT;
8198				} else if (type != lun->res_type
8199					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8200					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8201						if (!persis_offset
8202						 && i < CTL_MAX_INITIATORS)
8203							lun->pending_sense[i
8204							].ua_pending |=
8205							CTL_UA_RES_RELEASE;
8206						else if (persis_offset
8207						      && i >= persis_offset)
8208							lun->pending_sense[
8209							i-persis_offset
8210							].ua_pending |=
8211							CTL_UA_RES_RELEASE;
8212				}
8213			}
8214			lun->per_res[residx].registered = 1;
8215			lun->res_type = type;
8216			if (lun->res_type != SPR_TYPE_WR_EX_AR
8217			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8218				lun->pr_res_idx = residx;
8219			else
8220				lun->pr_res_idx =
8221					CTL_PR_ALL_REGISTRANTS;
8222
8223			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8224			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8225			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8226			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8227			persis_io.pr.pr_info.res_type = type;
8228			memcpy(persis_io.pr.pr_info.sa_res_key,
8229			       param->serv_act_res_key,
8230			       sizeof(param->serv_act_res_key));
8231			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8232			     &persis_io, sizeof(persis_io), 0)) >
8233			     CTL_HA_STATUS_SUCCESS) {
8234				printf("CTL:Persis Out error returned "
8235				       "from ctl_ha_msg_send %d\n",
8236				       isc_retval);
8237			}
8238		} else {
8239			/*
8240			 * sa_res_key is not the res holder just
8241			 * remove registrants
8242			 */
8243			int found=0;
8244
8245			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8246				if (memcmp(param->serv_act_res_key,
8247				    lun->per_res[i].res_key.key,
8248				    sizeof(struct scsi_per_res_key)) != 0)
8249					continue;
8250
8251				found = 1;
8252				lun->per_res[i].registered = 0;
8253				memset(&lun->per_res[i].res_key, 0,
8254				       sizeof(struct scsi_per_res_key));
8255				lun->pr_key_count--;
8256
8257				if (!persis_offset
8258				 && i < CTL_MAX_INITIATORS)
8259					lun->pending_sense[i].ua_pending |=
8260						CTL_UA_REG_PREEMPT;
8261				else if (persis_offset
8262				      && i >= persis_offset)
8263					lun->pending_sense[
8264						i-persis_offset].ua_pending |=
8265						CTL_UA_REG_PREEMPT;
8266			}
8267
8268			if (!found) {
8269				mtx_unlock(&lun->lun_lock);
8270				free(ctsio->kern_data_ptr, M_CTL);
8271				ctl_set_reservation_conflict(ctsio);
8272				ctl_done((union ctl_io *)ctsio);
8273		        	return (1);
8274			}
8275			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8276			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8277			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8278			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8279			persis_io.pr.pr_info.res_type = type;
8280			memcpy(persis_io.pr.pr_info.sa_res_key,
8281			       param->serv_act_res_key,
8282			       sizeof(param->serv_act_res_key));
8283			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8284			     &persis_io, sizeof(persis_io), 0)) >
8285			     CTL_HA_STATUS_SUCCESS) {
8286				printf("CTL:Persis Out error returned "
8287				       "from ctl_ha_msg_send %d\n",
8288				isc_retval);
8289			}
8290		}
8291	}
8292
8293	lun->PRGeneration++;
8294	mtx_unlock(&lun->lun_lock);
8295
8296	return (retval);
8297}
8298
8299static void
8300ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8301{
8302	int i;
8303
8304	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8305	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8306	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8307		   msg->pr.pr_info.sa_res_key,
8308		   sizeof(struct scsi_per_res_key)) != 0) {
8309		uint64_t sa_res_key;
8310		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8311
8312		if (sa_res_key == 0) {
8313			/* temporarily unregister this nexus */
8314			lun->per_res[msg->pr.pr_info.residx].registered = 0;
8315
8316			/*
8317			 * Unregister everybody else and build UA for
8318			 * them
8319			 */
8320			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8321				if (lun->per_res[i].registered == 0)
8322					continue;
8323
8324				if (!persis_offset
8325				 && i < CTL_MAX_INITIATORS)
8326					lun->pending_sense[i].ua_pending |=
8327						CTL_UA_REG_PREEMPT;
8328				else if (persis_offset && i >= persis_offset)
8329					lun->pending_sense[i -
8330						persis_offset].ua_pending |=
8331						CTL_UA_REG_PREEMPT;
8332				lun->per_res[i].registered = 0;
8333				memset(&lun->per_res[i].res_key, 0,
8334				       sizeof(struct scsi_per_res_key));
8335			}
8336
8337			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8338			lun->pr_key_count = 1;
8339			lun->res_type = msg->pr.pr_info.res_type;
8340			if (lun->res_type != SPR_TYPE_WR_EX_AR
8341			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8342				lun->pr_res_idx = msg->pr.pr_info.residx;
8343		} else {
8344		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8345				if (memcmp(msg->pr.pr_info.sa_res_key,
8346		                   lun->per_res[i].res_key.key,
8347		                   sizeof(struct scsi_per_res_key)) != 0)
8348					continue;
8349
8350				lun->per_res[i].registered = 0;
8351				memset(&lun->per_res[i].res_key, 0,
8352				       sizeof(struct scsi_per_res_key));
8353				lun->pr_key_count--;
8354
8355				if (!persis_offset
8356				 && i < persis_offset)
8357					lun->pending_sense[i].ua_pending |=
8358						CTL_UA_REG_PREEMPT;
8359				else if (persis_offset
8360				      && i >= persis_offset)
8361					lun->pending_sense[i -
8362						persis_offset].ua_pending |=
8363						CTL_UA_REG_PREEMPT;
8364			}
8365		}
8366	} else {
8367		/*
8368		 * Temporarily unregister so it won't get removed
8369		 * or UA generated
8370		 */
8371		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8372		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8373			if (lun->per_res[i].registered == 0)
8374				continue;
8375
8376			if (memcmp(msg->pr.pr_info.sa_res_key,
8377	                   lun->per_res[i].res_key.key,
8378	                   sizeof(struct scsi_per_res_key)) == 0) {
8379				lun->per_res[i].registered = 0;
8380				memset(&lun->per_res[i].res_key, 0,
8381				       sizeof(struct scsi_per_res_key));
8382				lun->pr_key_count--;
8383				if (!persis_offset
8384				 && i < CTL_MAX_INITIATORS)
8385					lun->pending_sense[i].ua_pending |=
8386						CTL_UA_REG_PREEMPT;
8387				else if (persis_offset
8388				      && i >= persis_offset)
8389					lun->pending_sense[i -
8390						persis_offset].ua_pending |=
8391						CTL_UA_REG_PREEMPT;
8392			} else if (msg->pr.pr_info.res_type != lun->res_type
8393				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8394				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8395					if (!persis_offset
8396					 && i < persis_offset)
8397						lun->pending_sense[i
8398							].ua_pending |=
8399							CTL_UA_RES_RELEASE;
8400					else if (persis_offset
8401					      && i >= persis_offset)
8402					lun->pending_sense[i -
8403						persis_offset].ua_pending |=
8404						CTL_UA_RES_RELEASE;
8405			}
8406		}
8407		lun->per_res[msg->pr.pr_info.residx].registered = 1;
8408		lun->res_type = msg->pr.pr_info.res_type;
8409		if (lun->res_type != SPR_TYPE_WR_EX_AR
8410		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8411			lun->pr_res_idx = msg->pr.pr_info.residx;
8412		else
8413			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8414	}
8415	lun->PRGeneration++;
8416
8417}
8418
8419
8420int
8421ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8422{
8423	int retval;
8424	int isc_retval;
8425	u_int32_t param_len;
8426	struct scsi_per_res_out *cdb;
8427	struct ctl_lun *lun;
8428	struct scsi_per_res_out_parms* param;
8429	struct ctl_softc *softc;
8430	uint32_t residx;
8431	uint64_t res_key, sa_res_key;
8432	uint8_t type;
8433	union ctl_ha_msg persis_io;
8434	int    i;
8435
8436	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8437
8438	retval = CTL_RETVAL_COMPLETE;
8439
8440	softc = control_softc;
8441
8442	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8443	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8444
8445	/*
8446	 * We only support whole-LUN scope.  The scope & type are ignored for
8447	 * register, register and ignore existing key and clear.
8448	 * We sometimes ignore scope and type on preempts too!!
8449	 * Verify reservation type here as well.
8450	 */
8451	type = cdb->scope_type & SPR_TYPE_MASK;
8452	if ((cdb->action == SPRO_RESERVE)
8453	 || (cdb->action == SPRO_RELEASE)) {
8454		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8455			ctl_set_invalid_field(/*ctsio*/ ctsio,
8456					      /*sks_valid*/ 1,
8457					      /*command*/ 1,
8458					      /*field*/ 2,
8459					      /*bit_valid*/ 1,
8460					      /*bit*/ 4);
8461			ctl_done((union ctl_io *)ctsio);
8462			return (CTL_RETVAL_COMPLETE);
8463		}
8464
8465		if (type>8 || type==2 || type==4 || type==0) {
8466			ctl_set_invalid_field(/*ctsio*/ ctsio,
8467					      /*sks_valid*/ 1,
8468					      /*command*/ 1,
8469					      /*field*/ 2,
8470					      /*bit_valid*/ 1,
8471					      /*bit*/ 0);
8472			ctl_done((union ctl_io *)ctsio);
8473			return (CTL_RETVAL_COMPLETE);
8474		}
8475	}
8476
8477	param_len = scsi_4btoul(cdb->length);
8478
8479	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8480		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8481		ctsio->kern_data_len = param_len;
8482		ctsio->kern_total_len = param_len;
8483		ctsio->kern_data_resid = 0;
8484		ctsio->kern_rel_offset = 0;
8485		ctsio->kern_sg_entries = 0;
8486		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8487		ctsio->be_move_done = ctl_config_move_done;
8488		ctl_datamove((union ctl_io *)ctsio);
8489
8490		return (CTL_RETVAL_COMPLETE);
8491	}
8492
8493	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8494
8495	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8496	res_key = scsi_8btou64(param->res_key.key);
8497	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8498
8499	/*
8500	 * Validate the reservation key here except for SPRO_REG_IGNO
8501	 * This must be done for all other service actions
8502	 */
8503	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8504		mtx_lock(&lun->lun_lock);
8505		if (lun->per_res[residx].registered) {
8506		    if (memcmp(param->res_key.key,
8507			       lun->per_res[residx].res_key.key,
8508			       ctl_min(sizeof(param->res_key),
8509			       sizeof(lun->per_res[residx].res_key))) != 0) {
8510				/*
8511				 * The current key passed in doesn't match
8512				 * the one the initiator previously
8513				 * registered.
8514				 */
8515				mtx_unlock(&lun->lun_lock);
8516				free(ctsio->kern_data_ptr, M_CTL);
8517				ctl_set_reservation_conflict(ctsio);
8518				ctl_done((union ctl_io *)ctsio);
8519				return (CTL_RETVAL_COMPLETE);
8520			}
8521		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8522			/*
8523			 * We are not registered
8524			 */
8525			mtx_unlock(&lun->lun_lock);
8526			free(ctsio->kern_data_ptr, M_CTL);
8527			ctl_set_reservation_conflict(ctsio);
8528			ctl_done((union ctl_io *)ctsio);
8529			return (CTL_RETVAL_COMPLETE);
8530		} else if (res_key != 0) {
8531			/*
8532			 * We are not registered and trying to register but
8533			 * the register key isn't zero.
8534			 */
8535			mtx_unlock(&lun->lun_lock);
8536			free(ctsio->kern_data_ptr, M_CTL);
8537			ctl_set_reservation_conflict(ctsio);
8538			ctl_done((union ctl_io *)ctsio);
8539			return (CTL_RETVAL_COMPLETE);
8540		}
8541		mtx_unlock(&lun->lun_lock);
8542	}
8543
8544	switch (cdb->action & SPRO_ACTION_MASK) {
8545	case SPRO_REGISTER:
8546	case SPRO_REG_IGNO: {
8547
8548#if 0
8549		printf("Registration received\n");
8550#endif
8551
8552		/*
8553		 * We don't support any of these options, as we report in
8554		 * the read capabilities request (see
8555		 * ctl_persistent_reserve_in(), above).
8556		 */
8557		if ((param->flags & SPR_SPEC_I_PT)
8558		 || (param->flags & SPR_ALL_TG_PT)
8559		 || (param->flags & SPR_APTPL)) {
8560			int bit_ptr;
8561
8562			if (param->flags & SPR_APTPL)
8563				bit_ptr = 0;
8564			else if (param->flags & SPR_ALL_TG_PT)
8565				bit_ptr = 2;
8566			else /* SPR_SPEC_I_PT */
8567				bit_ptr = 3;
8568
8569			free(ctsio->kern_data_ptr, M_CTL);
8570			ctl_set_invalid_field(ctsio,
8571					      /*sks_valid*/ 1,
8572					      /*command*/ 0,
8573					      /*field*/ 20,
8574					      /*bit_valid*/ 1,
8575					      /*bit*/ bit_ptr);
8576			ctl_done((union ctl_io *)ctsio);
8577			return (CTL_RETVAL_COMPLETE);
8578		}
8579
8580		mtx_lock(&lun->lun_lock);
8581
8582		/*
8583		 * The initiator wants to clear the
8584		 * key/unregister.
8585		 */
8586		if (sa_res_key == 0) {
8587			if ((res_key == 0
8588			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8589			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8590			  && !lun->per_res[residx].registered)) {
8591				mtx_unlock(&lun->lun_lock);
8592				goto done;
8593			}
8594
8595			lun->per_res[residx].registered = 0;
8596			memset(&lun->per_res[residx].res_key,
8597			       0, sizeof(lun->per_res[residx].res_key));
8598			lun->pr_key_count--;
8599
8600			if (residx == lun->pr_res_idx) {
8601				lun->flags &= ~CTL_LUN_PR_RESERVED;
8602				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8603
8604				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8605				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8606				 && lun->pr_key_count) {
8607					/*
8608					 * If the reservation is a registrants
8609					 * only type we need to generate a UA
8610					 * for other registered inits.  The
8611					 * sense code should be RESERVATIONS
8612					 * RELEASED
8613					 */
8614
8615					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8616						if (lun->per_res[
8617						    i+persis_offset].registered
8618						    == 0)
8619							continue;
8620						lun->pending_sense[i
8621							].ua_pending |=
8622							CTL_UA_RES_RELEASE;
8623					}
8624				}
8625				lun->res_type = 0;
8626			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8627				if (lun->pr_key_count==0) {
8628					lun->flags &= ~CTL_LUN_PR_RESERVED;
8629					lun->res_type = 0;
8630					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8631				}
8632			}
8633			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8634			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8635			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8636			persis_io.pr.pr_info.residx = residx;
8637			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8638			     &persis_io, sizeof(persis_io), 0 )) >
8639			     CTL_HA_STATUS_SUCCESS) {
8640				printf("CTL:Persis Out error returned from "
8641				       "ctl_ha_msg_send %d\n", isc_retval);
8642			}
8643		} else /* sa_res_key != 0 */ {
8644
8645			/*
8646			 * If we aren't registered currently then increment
8647			 * the key count and set the registered flag.
8648			 */
8649			if (!lun->per_res[residx].registered) {
8650				lun->pr_key_count++;
8651				lun->per_res[residx].registered = 1;
8652			}
8653
8654			memcpy(&lun->per_res[residx].res_key,
8655			       param->serv_act_res_key,
8656			       ctl_min(sizeof(param->serv_act_res_key),
8657			       sizeof(lun->per_res[residx].res_key)));
8658
8659			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8660			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8661			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8662			persis_io.pr.pr_info.residx = residx;
8663			memcpy(persis_io.pr.pr_info.sa_res_key,
8664			       param->serv_act_res_key,
8665			       sizeof(param->serv_act_res_key));
8666			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8667			     &persis_io, sizeof(persis_io), 0)) >
8668			     CTL_HA_STATUS_SUCCESS) {
8669				printf("CTL:Persis Out error returned from "
8670				       "ctl_ha_msg_send %d\n", isc_retval);
8671			}
8672		}
8673		lun->PRGeneration++;
8674		mtx_unlock(&lun->lun_lock);
8675
8676		break;
8677	}
8678	case SPRO_RESERVE:
8679#if 0
8680                printf("Reserve executed type %d\n", type);
8681#endif
8682		mtx_lock(&lun->lun_lock);
8683		if (lun->flags & CTL_LUN_PR_RESERVED) {
8684			/*
8685			 * if this isn't the reservation holder and it's
8686			 * not a "all registrants" type or if the type is
8687			 * different then we have a conflict
8688			 */
8689			if ((lun->pr_res_idx != residx
8690			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8691			 || lun->res_type != type) {
8692				mtx_unlock(&lun->lun_lock);
8693				free(ctsio->kern_data_ptr, M_CTL);
8694				ctl_set_reservation_conflict(ctsio);
8695				ctl_done((union ctl_io *)ctsio);
8696				return (CTL_RETVAL_COMPLETE);
8697			}
8698			mtx_unlock(&lun->lun_lock);
8699		} else /* create a reservation */ {
8700			/*
8701			 * If it's not an "all registrants" type record
8702			 * reservation holder
8703			 */
8704			if (type != SPR_TYPE_WR_EX_AR
8705			 && type != SPR_TYPE_EX_AC_AR)
8706				lun->pr_res_idx = residx; /* Res holder */
8707			else
8708				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8709
8710			lun->flags |= CTL_LUN_PR_RESERVED;
8711			lun->res_type = type;
8712
8713			mtx_unlock(&lun->lun_lock);
8714
8715			/* send msg to other side */
8716			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8717			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8718			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8719			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8720			persis_io.pr.pr_info.res_type = type;
8721			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8722			     &persis_io, sizeof(persis_io), 0)) >
8723			     CTL_HA_STATUS_SUCCESS) {
8724				printf("CTL:Persis Out error returned from "
8725				       "ctl_ha_msg_send %d\n", isc_retval);
8726			}
8727		}
8728		break;
8729
8730	case SPRO_RELEASE:
8731		mtx_lock(&lun->lun_lock);
8732		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8733			/* No reservation exists return good status */
8734			mtx_unlock(&lun->lun_lock);
8735			goto done;
8736		}
8737		/*
8738		 * Is this nexus a reservation holder?
8739		 */
8740		if (lun->pr_res_idx != residx
8741		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8742			/*
8743			 * not a res holder return good status but
8744			 * do nothing
8745			 */
8746			mtx_unlock(&lun->lun_lock);
8747			goto done;
8748		}
8749
8750		if (lun->res_type != type) {
8751			mtx_unlock(&lun->lun_lock);
8752			free(ctsio->kern_data_ptr, M_CTL);
8753			ctl_set_illegal_pr_release(ctsio);
8754			ctl_done((union ctl_io *)ctsio);
8755			return (CTL_RETVAL_COMPLETE);
8756		}
8757
8758		/* okay to release */
8759		lun->flags &= ~CTL_LUN_PR_RESERVED;
8760		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8761		lun->res_type = 0;
8762
8763		/*
8764		 * if this isn't an exclusive access
8765		 * res generate UA for all other
8766		 * registrants.
8767		 */
8768		if (type != SPR_TYPE_EX_AC
8769		 && type != SPR_TYPE_WR_EX) {
8770			/*
8771			 * temporarily unregister so we don't generate UA
8772			 */
8773			lun->per_res[residx].registered = 0;
8774
8775			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8776				if (lun->per_res[i+persis_offset].registered
8777				    == 0)
8778					continue;
8779				lun->pending_sense[i].ua_pending |=
8780					CTL_UA_RES_RELEASE;
8781			}
8782
8783			lun->per_res[residx].registered = 1;
8784		}
8785		mtx_unlock(&lun->lun_lock);
8786		/* Send msg to other side */
8787		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8788		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8789		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8790		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8791		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8792			printf("CTL:Persis Out error returned from "
8793			       "ctl_ha_msg_send %d\n", isc_retval);
8794		}
8795		break;
8796
8797	case SPRO_CLEAR:
8798		/* send msg to other side */
8799
8800		mtx_lock(&lun->lun_lock);
8801		lun->flags &= ~CTL_LUN_PR_RESERVED;
8802		lun->res_type = 0;
8803		lun->pr_key_count = 0;
8804		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8805
8806
8807		memset(&lun->per_res[residx].res_key,
8808		       0, sizeof(lun->per_res[residx].res_key));
8809		lun->per_res[residx].registered = 0;
8810
8811		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8812			if (lun->per_res[i].registered) {
8813				if (!persis_offset && i < CTL_MAX_INITIATORS)
8814					lun->pending_sense[i].ua_pending |=
8815						CTL_UA_RES_PREEMPT;
8816				else if (persis_offset && i >= persis_offset)
8817					lun->pending_sense[i-persis_offset
8818					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8819
8820				memset(&lun->per_res[i].res_key,
8821				       0, sizeof(struct scsi_per_res_key));
8822				lun->per_res[i].registered = 0;
8823			}
8824		lun->PRGeneration++;
8825		mtx_unlock(&lun->lun_lock);
8826		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8827		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8828		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8829		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8830		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8831			printf("CTL:Persis Out error returned from "
8832			       "ctl_ha_msg_send %d\n", isc_retval);
8833		}
8834		break;
8835
8836	case SPRO_PREEMPT: {
8837		int nretval;
8838
8839		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8840					  residx, ctsio, cdb, param);
8841		if (nretval != 0)
8842			return (CTL_RETVAL_COMPLETE);
8843		break;
8844	}
8845	default:
8846		panic("Invalid PR type %x", cdb->action);
8847	}
8848
8849done:
8850	free(ctsio->kern_data_ptr, M_CTL);
8851	ctl_set_success(ctsio);
8852	ctl_done((union ctl_io *)ctsio);
8853
8854	return (retval);
8855}
8856
8857/*
8858 * This routine is for handling a message from the other SC pertaining to
8859 * persistent reserve out. All the error checking will have been done
8860 * so only perorming the action need be done here to keep the two
8861 * in sync.
8862 */
8863static void
8864ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8865{
8866	struct ctl_lun *lun;
8867	struct ctl_softc *softc;
8868	int i;
8869	uint32_t targ_lun;
8870
8871	softc = control_softc;
8872
8873	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8874	lun = softc->ctl_luns[targ_lun];
8875	mtx_lock(&lun->lun_lock);
8876	switch(msg->pr.pr_info.action) {
8877	case CTL_PR_REG_KEY:
8878		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8879			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8880			lun->pr_key_count++;
8881		}
8882		lun->PRGeneration++;
8883		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8884		       msg->pr.pr_info.sa_res_key,
8885		       sizeof(struct scsi_per_res_key));
8886		break;
8887
8888	case CTL_PR_UNREG_KEY:
8889		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8890		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8891		       0, sizeof(struct scsi_per_res_key));
8892		lun->pr_key_count--;
8893
8894		/* XXX Need to see if the reservation has been released */
8895		/* if so do we need to generate UA? */
8896		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8897			lun->flags &= ~CTL_LUN_PR_RESERVED;
8898			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8899
8900			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8901			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8902			 && lun->pr_key_count) {
8903				/*
8904				 * If the reservation is a registrants
8905				 * only type we need to generate a UA
8906				 * for other registered inits.  The
8907				 * sense code should be RESERVATIONS
8908				 * RELEASED
8909				 */
8910
8911				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8912					if (lun->per_res[i+
8913					    persis_offset].registered == 0)
8914						continue;
8915
8916					lun->pending_sense[i
8917						].ua_pending |=
8918						CTL_UA_RES_RELEASE;
8919				}
8920			}
8921			lun->res_type = 0;
8922		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8923			if (lun->pr_key_count==0) {
8924				lun->flags &= ~CTL_LUN_PR_RESERVED;
8925				lun->res_type = 0;
8926				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8927			}
8928		}
8929		lun->PRGeneration++;
8930		break;
8931
8932	case CTL_PR_RESERVE:
8933		lun->flags |= CTL_LUN_PR_RESERVED;
8934		lun->res_type = msg->pr.pr_info.res_type;
8935		lun->pr_res_idx = msg->pr.pr_info.residx;
8936
8937		break;
8938
8939	case CTL_PR_RELEASE:
8940		/*
8941		 * if this isn't an exclusive access res generate UA for all
8942		 * other registrants.
8943		 */
8944		if (lun->res_type != SPR_TYPE_EX_AC
8945		 && lun->res_type != SPR_TYPE_WR_EX) {
8946			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8947				if (lun->per_res[i+persis_offset].registered)
8948					lun->pending_sense[i].ua_pending |=
8949						CTL_UA_RES_RELEASE;
8950		}
8951
8952		lun->flags &= ~CTL_LUN_PR_RESERVED;
8953		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8954		lun->res_type = 0;
8955		break;
8956
8957	case CTL_PR_PREEMPT:
8958		ctl_pro_preempt_other(lun, msg);
8959		break;
8960	case CTL_PR_CLEAR:
8961		lun->flags &= ~CTL_LUN_PR_RESERVED;
8962		lun->res_type = 0;
8963		lun->pr_key_count = 0;
8964		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8965
8966		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8967			if (lun->per_res[i].registered == 0)
8968				continue;
8969			if (!persis_offset
8970			 && i < CTL_MAX_INITIATORS)
8971				lun->pending_sense[i].ua_pending |=
8972					CTL_UA_RES_PREEMPT;
8973			else if (persis_offset
8974			      && i >= persis_offset)
8975   				lun->pending_sense[i-persis_offset].ua_pending|=
8976					CTL_UA_RES_PREEMPT;
8977			memset(&lun->per_res[i].res_key, 0,
8978			       sizeof(struct scsi_per_res_key));
8979			lun->per_res[i].registered = 0;
8980		}
8981		lun->PRGeneration++;
8982		break;
8983	}
8984
8985	mtx_unlock(&lun->lun_lock);
8986}
8987
8988int
8989ctl_read_write(struct ctl_scsiio *ctsio)
8990{
8991	struct ctl_lun *lun;
8992	struct ctl_lba_len_flags *lbalen;
8993	uint64_t lba;
8994	uint32_t num_blocks;
8995	int fua, dpo;
8996	int retval;
8997	int isread;
8998
8999	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9000
9001	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
9002
9003	fua = 0;
9004	dpo = 0;
9005
9006	retval = CTL_RETVAL_COMPLETE;
9007
9008	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
9009	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
9010	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
9011		uint32_t residx;
9012
9013		/*
9014		 * XXX KDM need a lock here.
9015		 */
9016		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9017		if ((lun->res_type == SPR_TYPE_EX_AC
9018		  && residx != lun->pr_res_idx)
9019		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
9020		   || lun->res_type == SPR_TYPE_EX_AC_AR)
9021		  && !lun->per_res[residx].registered)) {
9022			ctl_set_reservation_conflict(ctsio);
9023			ctl_done((union ctl_io *)ctsio);
9024			return (CTL_RETVAL_COMPLETE);
9025	        }
9026	}
9027
9028	switch (ctsio->cdb[0]) {
9029	case READ_6:
9030	case WRITE_6: {
9031		struct scsi_rw_6 *cdb;
9032
9033		cdb = (struct scsi_rw_6 *)ctsio->cdb;
9034
9035		lba = scsi_3btoul(cdb->addr);
9036		/* only 5 bits are valid in the most significant address byte */
9037		lba &= 0x1fffff;
9038		num_blocks = cdb->length;
9039		/*
9040		 * This is correct according to SBC-2.
9041		 */
9042		if (num_blocks == 0)
9043			num_blocks = 256;
9044		break;
9045	}
9046	case READ_10:
9047	case WRITE_10: {
9048		struct scsi_rw_10 *cdb;
9049
9050		cdb = (struct scsi_rw_10 *)ctsio->cdb;
9051
9052		if (cdb->byte2 & SRW10_FUA)
9053			fua = 1;
9054		if (cdb->byte2 & SRW10_DPO)
9055			dpo = 1;
9056
9057		lba = scsi_4btoul(cdb->addr);
9058		num_blocks = scsi_2btoul(cdb->length);
9059		break;
9060	}
9061	case WRITE_VERIFY_10: {
9062		struct scsi_write_verify_10 *cdb;
9063
9064		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9065
9066		/*
9067		 * XXX KDM we should do actual write verify support at some
9068		 * point.  This is obviously fake, we're just translating
9069		 * things to a write.  So we don't even bother checking the
9070		 * BYTCHK field, since we don't do any verification.  If
9071		 * the user asks for it, we'll just pretend we did it.
9072		 */
9073		if (cdb->byte2 & SWV_DPO)
9074			dpo = 1;
9075
9076		lba = scsi_4btoul(cdb->addr);
9077		num_blocks = scsi_2btoul(cdb->length);
9078		break;
9079	}
9080	case READ_12:
9081	case WRITE_12: {
9082		struct scsi_rw_12 *cdb;
9083
9084		cdb = (struct scsi_rw_12 *)ctsio->cdb;
9085
9086		if (cdb->byte2 & SRW12_FUA)
9087			fua = 1;
9088		if (cdb->byte2 & SRW12_DPO)
9089			dpo = 1;
9090		lba = scsi_4btoul(cdb->addr);
9091		num_blocks = scsi_4btoul(cdb->length);
9092		break;
9093	}
9094	case WRITE_VERIFY_12: {
9095		struct scsi_write_verify_12 *cdb;
9096
9097		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9098
9099		if (cdb->byte2 & SWV_DPO)
9100			dpo = 1;
9101
9102		lba = scsi_4btoul(cdb->addr);
9103		num_blocks = scsi_4btoul(cdb->length);
9104
9105		break;
9106	}
9107	case READ_16:
9108	case WRITE_16: {
9109		struct scsi_rw_16 *cdb;
9110
9111		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9112
9113		if (cdb->byte2 & SRW12_FUA)
9114			fua = 1;
9115		if (cdb->byte2 & SRW12_DPO)
9116			dpo = 1;
9117
9118		lba = scsi_8btou64(cdb->addr);
9119		num_blocks = scsi_4btoul(cdb->length);
9120		break;
9121	}
9122	case WRITE_VERIFY_16: {
9123		struct scsi_write_verify_16 *cdb;
9124
9125		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9126
9127		if (cdb->byte2 & SWV_DPO)
9128			dpo = 1;
9129
9130		lba = scsi_8btou64(cdb->addr);
9131		num_blocks = scsi_4btoul(cdb->length);
9132		break;
9133	}
9134	default:
9135		/*
9136		 * We got a command we don't support.  This shouldn't
9137		 * happen, commands should be filtered out above us.
9138		 */
9139		ctl_set_invalid_opcode(ctsio);
9140		ctl_done((union ctl_io *)ctsio);
9141
9142		return (CTL_RETVAL_COMPLETE);
9143		break; /* NOTREACHED */
9144	}
9145
9146	/*
9147	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
9148	 * interesting for us, but if RAIDCore is in write-back mode,
9149	 * getting it to do write-through for a particular transaction may
9150	 * not be possible.
9151	 */
9152
9153	/*
9154	 * The first check is to make sure we're in bounds, the second
9155	 * check is to catch wrap-around problems.  If the lba + num blocks
9156	 * is less than the lba, then we've wrapped around and the block
9157	 * range is invalid anyway.
9158	 */
9159	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9160	 || ((lba + num_blocks) < lba)) {
9161		ctl_set_lba_out_of_range(ctsio);
9162		ctl_done((union ctl_io *)ctsio);
9163		return (CTL_RETVAL_COMPLETE);
9164	}
9165
9166	/*
9167	 * According to SBC-3, a transfer length of 0 is not an error.
9168	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9169	 * translates to 256 blocks for those commands.
9170	 */
9171	if (num_blocks == 0) {
9172		ctl_set_success(ctsio);
9173		ctl_done((union ctl_io *)ctsio);
9174		return (CTL_RETVAL_COMPLETE);
9175	}
9176
9177	lbalen = (struct ctl_lba_len_flags *)
9178	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9179	lbalen->lba = lba;
9180	lbalen->len = num_blocks;
9181	lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
9182
9183	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9184	ctsio->kern_rel_offset = 0;
9185
9186	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9187
9188	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9189
9190	return (retval);
9191}
9192
9193static int
9194ctl_cnw_cont(union ctl_io *io)
9195{
9196	struct ctl_scsiio *ctsio;
9197	struct ctl_lun *lun;
9198	struct ctl_lba_len_flags *lbalen;
9199	int retval;
9200
9201	ctsio = &io->scsiio;
9202	ctsio->io_hdr.status = CTL_STATUS_NONE;
9203	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9204	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9205	lbalen = (struct ctl_lba_len_flags *)
9206	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9207	lbalen->flags = CTL_LLF_WRITE;
9208
9209	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9210	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9211	return (retval);
9212}
9213
9214int
9215ctl_cnw(struct ctl_scsiio *ctsio)
9216{
9217	struct ctl_lun *lun;
9218	struct ctl_lba_len_flags *lbalen;
9219	uint64_t lba;
9220	uint32_t num_blocks;
9221	int fua, dpo;
9222	int retval;
9223
9224	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9225
9226	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9227
9228	fua = 0;
9229	dpo = 0;
9230
9231	retval = CTL_RETVAL_COMPLETE;
9232
9233	switch (ctsio->cdb[0]) {
9234	case COMPARE_AND_WRITE: {
9235		struct scsi_compare_and_write *cdb;
9236
9237		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9238
9239		if (cdb->byte2 & SRW10_FUA)
9240			fua = 1;
9241		if (cdb->byte2 & SRW10_DPO)
9242			dpo = 1;
9243		lba = scsi_8btou64(cdb->addr);
9244		num_blocks = cdb->length;
9245		break;
9246	}
9247	default:
9248		/*
9249		 * We got a command we don't support.  This shouldn't
9250		 * happen, commands should be filtered out above us.
9251		 */
9252		ctl_set_invalid_opcode(ctsio);
9253		ctl_done((union ctl_io *)ctsio);
9254
9255		return (CTL_RETVAL_COMPLETE);
9256		break; /* NOTREACHED */
9257	}
9258
9259	/*
9260	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
9261	 * interesting for us, but if RAIDCore is in write-back mode,
9262	 * getting it to do write-through for a particular transaction may
9263	 * not be possible.
9264	 */
9265
9266	/*
9267	 * The first check is to make sure we're in bounds, the second
9268	 * check is to catch wrap-around problems.  If the lba + num blocks
9269	 * is less than the lba, then we've wrapped around and the block
9270	 * range is invalid anyway.
9271	 */
9272	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9273	 || ((lba + num_blocks) < lba)) {
9274		ctl_set_lba_out_of_range(ctsio);
9275		ctl_done((union ctl_io *)ctsio);
9276		return (CTL_RETVAL_COMPLETE);
9277	}
9278
9279	/*
9280	 * According to SBC-3, a transfer length of 0 is not an error.
9281	 */
9282	if (num_blocks == 0) {
9283		ctl_set_success(ctsio);
9284		ctl_done((union ctl_io *)ctsio);
9285		return (CTL_RETVAL_COMPLETE);
9286	}
9287
9288	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9289	ctsio->kern_rel_offset = 0;
9290
9291	/*
9292	 * Set the IO_CONT flag, so that if this I/O gets passed to
9293	 * ctl_data_submit_done(), it'll get passed back to
9294	 * ctl_ctl_cnw_cont() for further processing.
9295	 */
9296	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9297	ctsio->io_cont = ctl_cnw_cont;
9298
9299	lbalen = (struct ctl_lba_len_flags *)
9300	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9301	lbalen->lba = lba;
9302	lbalen->len = num_blocks;
9303	lbalen->flags = CTL_LLF_COMPARE;
9304
9305	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9306	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9307	return (retval);
9308}
9309
9310int
9311ctl_verify(struct ctl_scsiio *ctsio)
9312{
9313	struct ctl_lun *lun;
9314	struct ctl_lba_len_flags *lbalen;
9315	uint64_t lba;
9316	uint32_t num_blocks;
9317	int bytchk, dpo;
9318	int retval;
9319
9320	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9321
9322	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9323
9324	bytchk = 0;
9325	dpo = 0;
9326	retval = CTL_RETVAL_COMPLETE;
9327
9328	switch (ctsio->cdb[0]) {
9329	case VERIFY_10: {
9330		struct scsi_verify_10 *cdb;
9331
9332		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9333		if (cdb->byte2 & SVFY_BYTCHK)
9334			bytchk = 1;
9335		if (cdb->byte2 & SVFY_DPO)
9336			dpo = 1;
9337		lba = scsi_4btoul(cdb->addr);
9338		num_blocks = scsi_2btoul(cdb->length);
9339		break;
9340	}
9341	case VERIFY_12: {
9342		struct scsi_verify_12 *cdb;
9343
9344		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9345		if (cdb->byte2 & SVFY_BYTCHK)
9346			bytchk = 1;
9347		if (cdb->byte2 & SVFY_DPO)
9348			dpo = 1;
9349		lba = scsi_4btoul(cdb->addr);
9350		num_blocks = scsi_4btoul(cdb->length);
9351		break;
9352	}
9353	case VERIFY_16: {
9354		struct scsi_rw_16 *cdb;
9355
9356		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9357		if (cdb->byte2 & SVFY_BYTCHK)
9358			bytchk = 1;
9359		if (cdb->byte2 & SVFY_DPO)
9360			dpo = 1;
9361		lba = scsi_8btou64(cdb->addr);
9362		num_blocks = scsi_4btoul(cdb->length);
9363		break;
9364	}
9365	default:
9366		/*
9367		 * We got a command we don't support.  This shouldn't
9368		 * happen, commands should be filtered out above us.
9369		 */
9370		ctl_set_invalid_opcode(ctsio);
9371		ctl_done((union ctl_io *)ctsio);
9372		return (CTL_RETVAL_COMPLETE);
9373	}
9374
9375	/*
9376	 * The first check is to make sure we're in bounds, the second
9377	 * check is to catch wrap-around problems.  If the lba + num blocks
9378	 * is less than the lba, then we've wrapped around and the block
9379	 * range is invalid anyway.
9380	 */
9381	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9382	 || ((lba + num_blocks) < lba)) {
9383		ctl_set_lba_out_of_range(ctsio);
9384		ctl_done((union ctl_io *)ctsio);
9385		return (CTL_RETVAL_COMPLETE);
9386	}
9387
9388	/*
9389	 * According to SBC-3, a transfer length of 0 is not an error.
9390	 */
9391	if (num_blocks == 0) {
9392		ctl_set_success(ctsio);
9393		ctl_done((union ctl_io *)ctsio);
9394		return (CTL_RETVAL_COMPLETE);
9395	}
9396
9397	lbalen = (struct ctl_lba_len_flags *)
9398	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9399	lbalen->lba = lba;
9400	lbalen->len = num_blocks;
9401	if (bytchk) {
9402		lbalen->flags = CTL_LLF_COMPARE;
9403		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9404	} else {
9405		lbalen->flags = CTL_LLF_VERIFY;
9406		ctsio->kern_total_len = 0;
9407	}
9408	ctsio->kern_rel_offset = 0;
9409
9410	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9411	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9412	return (retval);
9413}
9414
9415int
9416ctl_report_luns(struct ctl_scsiio *ctsio)
9417{
9418	struct scsi_report_luns *cdb;
9419	struct scsi_report_luns_data *lun_data;
9420	struct ctl_lun *lun, *request_lun;
9421	int num_luns, retval;
9422	uint32_t alloc_len, lun_datalen;
9423	int num_filled, well_known;
9424	uint32_t initidx, targ_lun_id, lun_id;
9425
9426	retval = CTL_RETVAL_COMPLETE;
9427	well_known = 0;
9428
9429	cdb = (struct scsi_report_luns *)ctsio->cdb;
9430
9431	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9432
9433	mtx_lock(&control_softc->ctl_lock);
9434	num_luns = control_softc->num_luns;
9435	mtx_unlock(&control_softc->ctl_lock);
9436
9437	switch (cdb->select_report) {
9438	case RPL_REPORT_DEFAULT:
9439	case RPL_REPORT_ALL:
9440		break;
9441	case RPL_REPORT_WELLKNOWN:
9442		well_known = 1;
9443		num_luns = 0;
9444		break;
9445	default:
9446		ctl_set_invalid_field(ctsio,
9447				      /*sks_valid*/ 1,
9448				      /*command*/ 1,
9449				      /*field*/ 2,
9450				      /*bit_valid*/ 0,
9451				      /*bit*/ 0);
9452		ctl_done((union ctl_io *)ctsio);
9453		return (retval);
9454		break; /* NOTREACHED */
9455	}
9456
9457	alloc_len = scsi_4btoul(cdb->length);
9458	/*
9459	 * The initiator has to allocate at least 16 bytes for this request,
9460	 * so he can at least get the header and the first LUN.  Otherwise
9461	 * we reject the request (per SPC-3 rev 14, section 6.21).
9462	 */
9463	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9464	    sizeof(struct scsi_report_luns_lundata))) {
9465		ctl_set_invalid_field(ctsio,
9466				      /*sks_valid*/ 1,
9467				      /*command*/ 1,
9468				      /*field*/ 6,
9469				      /*bit_valid*/ 0,
9470				      /*bit*/ 0);
9471		ctl_done((union ctl_io *)ctsio);
9472		return (retval);
9473	}
9474
9475	request_lun = (struct ctl_lun *)
9476		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9477
9478	lun_datalen = sizeof(*lun_data) +
9479		(num_luns * sizeof(struct scsi_report_luns_lundata));
9480
9481	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9482	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9483	ctsio->kern_sg_entries = 0;
9484
9485	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9486
9487	mtx_lock(&control_softc->ctl_lock);
9488	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9489		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9490		if (lun_id >= CTL_MAX_LUNS)
9491			continue;
9492		lun = control_softc->ctl_luns[lun_id];
9493		if (lun == NULL)
9494			continue;
9495
9496		if (targ_lun_id <= 0xff) {
9497			/*
9498			 * Peripheral addressing method, bus number 0.
9499			 */
9500			lun_data->luns[num_filled].lundata[0] =
9501				RPL_LUNDATA_ATYP_PERIPH;
9502			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9503			num_filled++;
9504		} else if (targ_lun_id <= 0x3fff) {
9505			/*
9506			 * Flat addressing method.
9507			 */
9508			lun_data->luns[num_filled].lundata[0] =
9509				RPL_LUNDATA_ATYP_FLAT |
9510				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9511#ifdef OLDCTLHEADERS
9512				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9513				(targ_lun_id & SRLD_BUS_LUN_MASK);
9514#endif
9515			lun_data->luns[num_filled].lundata[1] =
9516#ifdef OLDCTLHEADERS
9517				targ_lun_id >> SRLD_BUS_LUN_BITS;
9518#endif
9519				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9520			num_filled++;
9521		} else {
9522			printf("ctl_report_luns: bogus LUN number %jd, "
9523			       "skipping\n", (intmax_t)targ_lun_id);
9524		}
9525		/*
9526		 * According to SPC-3, rev 14 section 6.21:
9527		 *
9528		 * "The execution of a REPORT LUNS command to any valid and
9529		 * installed logical unit shall clear the REPORTED LUNS DATA
9530		 * HAS CHANGED unit attention condition for all logical
9531		 * units of that target with respect to the requesting
9532		 * initiator. A valid and installed logical unit is one
9533		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9534		 * INQUIRY data (see 6.4.2)."
9535		 *
9536		 * If request_lun is NULL, the LUN this report luns command
9537		 * was issued to is either disabled or doesn't exist. In that
9538		 * case, we shouldn't clear any pending lun change unit
9539		 * attention.
9540		 */
9541		if (request_lun != NULL) {
9542			mtx_lock(&lun->lun_lock);
9543			lun->pending_sense[initidx].ua_pending &=
9544				~CTL_UA_LUN_CHANGE;
9545			mtx_unlock(&lun->lun_lock);
9546		}
9547	}
9548	mtx_unlock(&control_softc->ctl_lock);
9549
9550	/*
9551	 * It's quite possible that we've returned fewer LUNs than we allocated
9552	 * space for.  Trim it.
9553	 */
9554	lun_datalen = sizeof(*lun_data) +
9555		(num_filled * sizeof(struct scsi_report_luns_lundata));
9556
9557	if (lun_datalen < alloc_len) {
9558		ctsio->residual = alloc_len - lun_datalen;
9559		ctsio->kern_data_len = lun_datalen;
9560		ctsio->kern_total_len = lun_datalen;
9561	} else {
9562		ctsio->residual = 0;
9563		ctsio->kern_data_len = alloc_len;
9564		ctsio->kern_total_len = alloc_len;
9565	}
9566	ctsio->kern_data_resid = 0;
9567	ctsio->kern_rel_offset = 0;
9568	ctsio->kern_sg_entries = 0;
9569
9570	/*
9571	 * We set this to the actual data length, regardless of how much
9572	 * space we actually have to return results.  If the user looks at
9573	 * this value, he'll know whether or not he allocated enough space
9574	 * and reissue the command if necessary.  We don't support well
9575	 * known logical units, so if the user asks for that, return none.
9576	 */
9577	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9578
9579	/*
9580	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9581	 * this request.
9582	 */
9583	ctsio->scsi_status = SCSI_STATUS_OK;
9584
9585	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9586	ctsio->be_move_done = ctl_config_move_done;
9587	ctl_datamove((union ctl_io *)ctsio);
9588
9589	return (retval);
9590}
9591
9592int
9593ctl_request_sense(struct ctl_scsiio *ctsio)
9594{
9595	struct scsi_request_sense *cdb;
9596	struct scsi_sense_data *sense_ptr;
9597	struct ctl_lun *lun;
9598	uint32_t initidx;
9599	int have_error;
9600	scsi_sense_data_type sense_format;
9601
9602	cdb = (struct scsi_request_sense *)ctsio->cdb;
9603
9604	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9605
9606	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9607
9608	/*
9609	 * Determine which sense format the user wants.
9610	 */
9611	if (cdb->byte2 & SRS_DESC)
9612		sense_format = SSD_TYPE_DESC;
9613	else
9614		sense_format = SSD_TYPE_FIXED;
9615
9616	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9617	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9618	ctsio->kern_sg_entries = 0;
9619
9620	/*
9621	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9622	 * larger than the largest allowed value for the length field in the
9623	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9624	 */
9625	ctsio->residual = 0;
9626	ctsio->kern_data_len = cdb->length;
9627	ctsio->kern_total_len = cdb->length;
9628
9629	ctsio->kern_data_resid = 0;
9630	ctsio->kern_rel_offset = 0;
9631	ctsio->kern_sg_entries = 0;
9632
9633	/*
9634	 * If we don't have a LUN, we don't have any pending sense.
9635	 */
9636	if (lun == NULL)
9637		goto no_sense;
9638
9639	have_error = 0;
9640	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9641	/*
9642	 * Check for pending sense, and then for pending unit attentions.
9643	 * Pending sense gets returned first, then pending unit attentions.
9644	 */
9645	mtx_lock(&lun->lun_lock);
9646	if (ctl_is_set(lun->have_ca, initidx)) {
9647		scsi_sense_data_type stored_format;
9648
9649		/*
9650		 * Check to see which sense format was used for the stored
9651		 * sense data.
9652		 */
9653		stored_format = scsi_sense_type(
9654		    &lun->pending_sense[initidx].sense);
9655
9656		/*
9657		 * If the user requested a different sense format than the
9658		 * one we stored, then we need to convert it to the other
9659		 * format.  If we're going from descriptor to fixed format
9660		 * sense data, we may lose things in translation, depending
9661		 * on what options were used.
9662		 *
9663		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9664		 * for some reason we'll just copy it out as-is.
9665		 */
9666		if ((stored_format == SSD_TYPE_FIXED)
9667		 && (sense_format == SSD_TYPE_DESC))
9668			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9669			    &lun->pending_sense[initidx].sense,
9670			    (struct scsi_sense_data_desc *)sense_ptr);
9671		else if ((stored_format == SSD_TYPE_DESC)
9672		      && (sense_format == SSD_TYPE_FIXED))
9673			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9674			    &lun->pending_sense[initidx].sense,
9675			    (struct scsi_sense_data_fixed *)sense_ptr);
9676		else
9677			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
9678			       ctl_min(sizeof(*sense_ptr),
9679			       sizeof(lun->pending_sense[initidx].sense)));
9680
9681		ctl_clear_mask(lun->have_ca, initidx);
9682		have_error = 1;
9683	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
9684		ctl_ua_type ua_type;
9685
9686		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
9687				       sense_ptr, sense_format);
9688		if (ua_type != CTL_UA_NONE) {
9689			have_error = 1;
9690			/* We're reporting this UA, so clear it */
9691			lun->pending_sense[initidx].ua_pending &= ~ua_type;
9692		}
9693	}
9694	mtx_unlock(&lun->lun_lock);
9695
9696	/*
9697	 * We already have a pending error, return it.
9698	 */
9699	if (have_error != 0) {
9700		/*
9701		 * We report the SCSI status as OK, since the status of the
9702		 * request sense command itself is OK.
9703		 */
9704		ctsio->scsi_status = SCSI_STATUS_OK;
9705
9706		/*
9707		 * We report 0 for the sense length, because we aren't doing
9708		 * autosense in this case.  We're reporting sense as
9709		 * parameter data.
9710		 */
9711		ctsio->sense_len = 0;
9712		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9713		ctsio->be_move_done = ctl_config_move_done;
9714		ctl_datamove((union ctl_io *)ctsio);
9715
9716		return (CTL_RETVAL_COMPLETE);
9717	}
9718
9719no_sense:
9720
9721	/*
9722	 * No sense information to report, so we report that everything is
9723	 * okay.
9724	 */
9725	ctl_set_sense_data(sense_ptr,
9726			   lun,
9727			   sense_format,
9728			   /*current_error*/ 1,
9729			   /*sense_key*/ SSD_KEY_NO_SENSE,
9730			   /*asc*/ 0x00,
9731			   /*ascq*/ 0x00,
9732			   SSD_ELEM_NONE);
9733
9734	ctsio->scsi_status = SCSI_STATUS_OK;
9735
9736	/*
9737	 * We report 0 for the sense length, because we aren't doing
9738	 * autosense in this case.  We're reporting sense as parameter data.
9739	 */
9740	ctsio->sense_len = 0;
9741	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9742	ctsio->be_move_done = ctl_config_move_done;
9743	ctl_datamove((union ctl_io *)ctsio);
9744
9745	return (CTL_RETVAL_COMPLETE);
9746}
9747
9748int
9749ctl_tur(struct ctl_scsiio *ctsio)
9750{
9751	struct ctl_lun *lun;
9752
9753	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9754
9755	CTL_DEBUG_PRINT(("ctl_tur\n"));
9756
9757	if (lun == NULL)
9758		return (EINVAL);
9759
9760	ctsio->scsi_status = SCSI_STATUS_OK;
9761	ctsio->io_hdr.status = CTL_SUCCESS;
9762
9763	ctl_done((union ctl_io *)ctsio);
9764
9765	return (CTL_RETVAL_COMPLETE);
9766}
9767
9768#ifdef notyet
9769static int
9770ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9771{
9772
9773}
9774#endif
9775
9776static int
9777ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9778{
9779	struct scsi_vpd_supported_pages *pages;
9780	int sup_page_size;
9781	struct ctl_lun *lun;
9782
9783	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9784
9785	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9786	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9787	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9788	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9789	ctsio->kern_sg_entries = 0;
9790
9791	if (sup_page_size < alloc_len) {
9792		ctsio->residual = alloc_len - sup_page_size;
9793		ctsio->kern_data_len = sup_page_size;
9794		ctsio->kern_total_len = sup_page_size;
9795	} else {
9796		ctsio->residual = 0;
9797		ctsio->kern_data_len = alloc_len;
9798		ctsio->kern_total_len = alloc_len;
9799	}
9800	ctsio->kern_data_resid = 0;
9801	ctsio->kern_rel_offset = 0;
9802	ctsio->kern_sg_entries = 0;
9803
9804	/*
9805	 * The control device is always connected.  The disk device, on the
9806	 * other hand, may not be online all the time.  Need to change this
9807	 * to figure out whether the disk device is actually online or not.
9808	 */
9809	if (lun != NULL)
9810		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9811				lun->be_lun->lun_type;
9812	else
9813		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9814
9815	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9816	/* Supported VPD pages */
9817	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9818	/* Serial Number */
9819	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9820	/* Device Identification */
9821	pages->page_list[2] = SVPD_DEVICE_ID;
9822	/* SCSI Ports */
9823	pages->page_list[3] = SVPD_SCSI_PORTS;
9824	/* Block limits */
9825	pages->page_list[4] = SVPD_BLOCK_LIMITS;
9826	/* Logical Block Provisioning */
9827	pages->page_list[5] = SVPD_LBP;
9828
9829	ctsio->scsi_status = SCSI_STATUS_OK;
9830
9831	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9832	ctsio->be_move_done = ctl_config_move_done;
9833	ctl_datamove((union ctl_io *)ctsio);
9834
9835	return (CTL_RETVAL_COMPLETE);
9836}
9837
9838static int
9839ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9840{
9841	struct scsi_vpd_unit_serial_number *sn_ptr;
9842	struct ctl_lun *lun;
9843
9844	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9845
9846	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9847	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9848	ctsio->kern_sg_entries = 0;
9849
9850	if (sizeof(*sn_ptr) < alloc_len) {
9851		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9852		ctsio->kern_data_len = sizeof(*sn_ptr);
9853		ctsio->kern_total_len = sizeof(*sn_ptr);
9854	} else {
9855		ctsio->residual = 0;
9856		ctsio->kern_data_len = alloc_len;
9857		ctsio->kern_total_len = alloc_len;
9858	}
9859	ctsio->kern_data_resid = 0;
9860	ctsio->kern_rel_offset = 0;
9861	ctsio->kern_sg_entries = 0;
9862
9863	/*
9864	 * The control device is always connected.  The disk device, on the
9865	 * other hand, may not be online all the time.  Need to change this
9866	 * to figure out whether the disk device is actually online or not.
9867	 */
9868	if (lun != NULL)
9869		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9870				  lun->be_lun->lun_type;
9871	else
9872		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9873
9874	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9875	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9876	/*
9877	 * If we don't have a LUN, we just leave the serial number as
9878	 * all spaces.
9879	 */
9880	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9881	if (lun != NULL) {
9882		strncpy((char *)sn_ptr->serial_num,
9883			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9884	}
9885	ctsio->scsi_status = SCSI_STATUS_OK;
9886
9887	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9888	ctsio->be_move_done = ctl_config_move_done;
9889	ctl_datamove((union ctl_io *)ctsio);
9890
9891	return (CTL_RETVAL_COMPLETE);
9892}
9893
9894
9895static int
9896ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9897{
9898	struct scsi_vpd_device_id *devid_ptr;
9899	struct scsi_vpd_id_descriptor *desc;
9900	struct ctl_softc *ctl_softc;
9901	struct ctl_lun *lun;
9902	struct ctl_port *port;
9903	int data_len;
9904	uint8_t proto;
9905
9906	ctl_softc = control_softc;
9907
9908	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9909	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9910
9911	data_len = sizeof(struct scsi_vpd_device_id) +
9912	    sizeof(struct scsi_vpd_id_descriptor) +
9913		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9914	    sizeof(struct scsi_vpd_id_descriptor) +
9915		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9916	if (lun && lun->lun_devid)
9917		data_len += lun->lun_devid->len;
9918	if (port->port_devid)
9919		data_len += port->port_devid->len;
9920	if (port->target_devid)
9921		data_len += port->target_devid->len;
9922
9923	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9924	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9925	ctsio->kern_sg_entries = 0;
9926
9927	if (data_len < alloc_len) {
9928		ctsio->residual = alloc_len - data_len;
9929		ctsio->kern_data_len = data_len;
9930		ctsio->kern_total_len = data_len;
9931	} else {
9932		ctsio->residual = 0;
9933		ctsio->kern_data_len = alloc_len;
9934		ctsio->kern_total_len = alloc_len;
9935	}
9936	ctsio->kern_data_resid = 0;
9937	ctsio->kern_rel_offset = 0;
9938	ctsio->kern_sg_entries = 0;
9939
9940	/*
9941	 * The control device is always connected.  The disk device, on the
9942	 * other hand, may not be online all the time.
9943	 */
9944	if (lun != NULL)
9945		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9946				     lun->be_lun->lun_type;
9947	else
9948		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9949	devid_ptr->page_code = SVPD_DEVICE_ID;
9950	scsi_ulto2b(data_len - 4, devid_ptr->length);
9951
9952	if (port->port_type == CTL_PORT_FC)
9953		proto = SCSI_PROTO_FC << 4;
9954	else if (port->port_type == CTL_PORT_ISCSI)
9955		proto = SCSI_PROTO_ISCSI << 4;
9956	else
9957		proto = SCSI_PROTO_SPI << 4;
9958	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9959
9960	/*
9961	 * We're using a LUN association here.  i.e., this device ID is a
9962	 * per-LUN identifier.
9963	 */
9964	if (lun && lun->lun_devid) {
9965		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9966		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9967		    lun->lun_devid->len);
9968	}
9969
9970	/*
9971	 * This is for the WWPN which is a port association.
9972	 */
9973	if (port->port_devid) {
9974		memcpy(desc, port->port_devid->data, port->port_devid->len);
9975		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9976		    port->port_devid->len);
9977	}
9978
9979	/*
9980	 * This is for the Relative Target Port(type 4h) identifier
9981	 */
9982	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9983	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9984	    SVPD_ID_TYPE_RELTARG;
9985	desc->length = 4;
9986	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9987	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9988	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9989
9990	/*
9991	 * This is for the Target Port Group(type 5h) identifier
9992	 */
9993	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9994	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9995	    SVPD_ID_TYPE_TPORTGRP;
9996	desc->length = 4;
9997	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9998	    &desc->identifier[2]);
9999	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10000	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10001
10002	/*
10003	 * This is for the Target identifier
10004	 */
10005	if (port->target_devid) {
10006		memcpy(desc, port->target_devid->data, port->target_devid->len);
10007	}
10008
10009	ctsio->scsi_status = SCSI_STATUS_OK;
10010	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10011	ctsio->be_move_done = ctl_config_move_done;
10012	ctl_datamove((union ctl_io *)ctsio);
10013
10014	return (CTL_RETVAL_COMPLETE);
10015}
10016
10017static int
10018ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10019{
10020	struct ctl_softc *softc = control_softc;
10021	struct scsi_vpd_scsi_ports *sp;
10022	struct scsi_vpd_port_designation *pd;
10023	struct scsi_vpd_port_designation_cont *pdc;
10024	struct ctl_lun *lun;
10025	struct ctl_port *port;
10026	int data_len, num_target_ports, id_len, g, pg, p;
10027	int num_target_port_groups, single;
10028
10029	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10030
10031	single = ctl_is_single;
10032	if (single)
10033		num_target_port_groups = 1;
10034	else
10035		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10036	num_target_ports = 0;
10037	id_len = 0;
10038	mtx_lock(&softc->ctl_lock);
10039	STAILQ_FOREACH(port, &softc->port_list, links) {
10040		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10041			continue;
10042		if (ctl_map_lun_back(port->targ_port, lun->lun) >=
10043		    CTL_MAX_LUNS)
10044			continue;
10045		num_target_ports++;
10046		if (port->port_devid)
10047			id_len += port->port_devid->len;
10048	}
10049	mtx_unlock(&softc->ctl_lock);
10050
10051	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10052	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10053	     sizeof(struct scsi_vpd_port_designation_cont)) + id_len;
10054	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10055	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10056	ctsio->kern_sg_entries = 0;
10057
10058	if (data_len < alloc_len) {
10059		ctsio->residual = alloc_len - data_len;
10060		ctsio->kern_data_len = data_len;
10061		ctsio->kern_total_len = data_len;
10062	} else {
10063		ctsio->residual = 0;
10064		ctsio->kern_data_len = alloc_len;
10065		ctsio->kern_total_len = alloc_len;
10066	}
10067	ctsio->kern_data_resid = 0;
10068	ctsio->kern_rel_offset = 0;
10069	ctsio->kern_sg_entries = 0;
10070
10071	/*
10072	 * The control device is always connected.  The disk device, on the
10073	 * other hand, may not be online all the time.  Need to change this
10074	 * to figure out whether the disk device is actually online or not.
10075	 */
10076	if (lun != NULL)
10077		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10078				  lun->be_lun->lun_type;
10079	else
10080		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10081
10082	sp->page_code = SVPD_SCSI_PORTS;
10083	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10084	    sp->page_length);
10085	pd = &sp->design[0];
10086
10087	mtx_lock(&softc->ctl_lock);
10088	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10089		pg = 0;
10090	else
10091		pg = 1;
10092	for (g = 0; g < num_target_port_groups; g++) {
10093		STAILQ_FOREACH(port, &softc->port_list, links) {
10094			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10095				continue;
10096			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
10097			    CTL_MAX_LUNS)
10098				continue;
10099			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10100			scsi_ulto2b(p, pd->relative_port_id);
10101			scsi_ulto2b(0, pd->initiator_transportid_length);
10102			pdc = (struct scsi_vpd_port_designation_cont *)
10103			    &pd->initiator_transportid[0];
10104			if (port->port_devid && g == pg) {
10105				id_len = port->port_devid->len;
10106				scsi_ulto2b(port->port_devid->len,
10107				    pdc->target_port_descriptors_length);
10108				memcpy(pdc->target_port_descriptors,
10109				    port->port_devid->data, port->port_devid->len);
10110			} else {
10111				id_len = 0;
10112				scsi_ulto2b(0, pdc->target_port_descriptors_length);
10113			}
10114			pd = (struct scsi_vpd_port_designation *)
10115			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10116		}
10117	}
10118	mtx_unlock(&softc->ctl_lock);
10119
10120	ctsio->scsi_status = SCSI_STATUS_OK;
10121	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10122	ctsio->be_move_done = ctl_config_move_done;
10123	ctl_datamove((union ctl_io *)ctsio);
10124
10125	return (CTL_RETVAL_COMPLETE);
10126}
10127
10128static int
10129ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10130{
10131	struct scsi_vpd_block_limits *bl_ptr;
10132	struct ctl_lun *lun;
10133	int bs;
10134
10135	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10136	bs = lun->be_lun->blocksize;
10137
10138	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10139	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10140	ctsio->kern_sg_entries = 0;
10141
10142	if (sizeof(*bl_ptr) < alloc_len) {
10143		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10144		ctsio->kern_data_len = sizeof(*bl_ptr);
10145		ctsio->kern_total_len = sizeof(*bl_ptr);
10146	} else {
10147		ctsio->residual = 0;
10148		ctsio->kern_data_len = alloc_len;
10149		ctsio->kern_total_len = alloc_len;
10150	}
10151	ctsio->kern_data_resid = 0;
10152	ctsio->kern_rel_offset = 0;
10153	ctsio->kern_sg_entries = 0;
10154
10155	/*
10156	 * The control device is always connected.  The disk device, on the
10157	 * other hand, may not be online all the time.  Need to change this
10158	 * to figure out whether the disk device is actually online or not.
10159	 */
10160	if (lun != NULL)
10161		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10162				  lun->be_lun->lun_type;
10163	else
10164		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10165
10166	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10167	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
10168	bl_ptr->max_cmp_write_len = 0xff;
10169	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10170	scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10171	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10172		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10173		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10174	}
10175	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10176
10177	ctsio->scsi_status = SCSI_STATUS_OK;
10178	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10179	ctsio->be_move_done = ctl_config_move_done;
10180	ctl_datamove((union ctl_io *)ctsio);
10181
10182	return (CTL_RETVAL_COMPLETE);
10183}
10184
10185static int
10186ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10187{
10188	struct scsi_vpd_logical_block_prov *lbp_ptr;
10189	struct ctl_lun *lun;
10190	int bs;
10191
10192	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10193	bs = lun->be_lun->blocksize;
10194
10195	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10196	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10197	ctsio->kern_sg_entries = 0;
10198
10199	if (sizeof(*lbp_ptr) < alloc_len) {
10200		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10201		ctsio->kern_data_len = sizeof(*lbp_ptr);
10202		ctsio->kern_total_len = sizeof(*lbp_ptr);
10203	} else {
10204		ctsio->residual = 0;
10205		ctsio->kern_data_len = alloc_len;
10206		ctsio->kern_total_len = alloc_len;
10207	}
10208	ctsio->kern_data_resid = 0;
10209	ctsio->kern_rel_offset = 0;
10210	ctsio->kern_sg_entries = 0;
10211
10212	/*
10213	 * The control device is always connected.  The disk device, on the
10214	 * other hand, may not be online all the time.  Need to change this
10215	 * to figure out whether the disk device is actually online or not.
10216	 */
10217	if (lun != NULL)
10218		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10219				  lun->be_lun->lun_type;
10220	else
10221		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10222
10223	lbp_ptr->page_code = SVPD_LBP;
10224	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
10225		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10;
10226
10227	ctsio->scsi_status = SCSI_STATUS_OK;
10228	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10229	ctsio->be_move_done = ctl_config_move_done;
10230	ctl_datamove((union ctl_io *)ctsio);
10231
10232	return (CTL_RETVAL_COMPLETE);
10233}
10234
10235static int
10236ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10237{
10238	struct scsi_inquiry *cdb;
10239	struct ctl_lun *lun;
10240	int alloc_len, retval;
10241
10242	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10243	cdb = (struct scsi_inquiry *)ctsio->cdb;
10244
10245	retval = CTL_RETVAL_COMPLETE;
10246
10247	alloc_len = scsi_2btoul(cdb->length);
10248
10249	switch (cdb->page_code) {
10250	case SVPD_SUPPORTED_PAGES:
10251		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10252		break;
10253	case SVPD_UNIT_SERIAL_NUMBER:
10254		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10255		break;
10256	case SVPD_DEVICE_ID:
10257		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10258		break;
10259	case SVPD_SCSI_PORTS:
10260		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10261		break;
10262	case SVPD_BLOCK_LIMITS:
10263		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10264		break;
10265	case SVPD_LBP:
10266		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10267		break;
10268	default:
10269		ctl_set_invalid_field(ctsio,
10270				      /*sks_valid*/ 1,
10271				      /*command*/ 1,
10272				      /*field*/ 2,
10273				      /*bit_valid*/ 0,
10274				      /*bit*/ 0);
10275		ctl_done((union ctl_io *)ctsio);
10276		retval = CTL_RETVAL_COMPLETE;
10277		break;
10278	}
10279
10280	return (retval);
10281}
10282
10283static int
10284ctl_inquiry_std(struct ctl_scsiio *ctsio)
10285{
10286	struct scsi_inquiry_data *inq_ptr;
10287	struct scsi_inquiry *cdb;
10288	struct ctl_softc *ctl_softc;
10289	struct ctl_lun *lun;
10290	char *val;
10291	uint32_t alloc_len;
10292	int is_fc;
10293
10294	ctl_softc = control_softc;
10295
10296	/*
10297	 * Figure out whether we're talking to a Fibre Channel port or not.
10298	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10299	 * SCSI front ends.
10300	 */
10301	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
10302	    CTL_PORT_FC)
10303		is_fc = 0;
10304	else
10305		is_fc = 1;
10306
10307	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10308	cdb = (struct scsi_inquiry *)ctsio->cdb;
10309	alloc_len = scsi_2btoul(cdb->length);
10310
10311	/*
10312	 * We malloc the full inquiry data size here and fill it
10313	 * in.  If the user only asks for less, we'll give him
10314	 * that much.
10315	 */
10316	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
10317	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10318	ctsio->kern_sg_entries = 0;
10319	ctsio->kern_data_resid = 0;
10320	ctsio->kern_rel_offset = 0;
10321
10322	if (sizeof(*inq_ptr) < alloc_len) {
10323		ctsio->residual = alloc_len - sizeof(*inq_ptr);
10324		ctsio->kern_data_len = sizeof(*inq_ptr);
10325		ctsio->kern_total_len = sizeof(*inq_ptr);
10326	} else {
10327		ctsio->residual = 0;
10328		ctsio->kern_data_len = alloc_len;
10329		ctsio->kern_total_len = alloc_len;
10330	}
10331
10332	/*
10333	 * If we have a LUN configured, report it as connected.  Otherwise,
10334	 * report that it is offline or no device is supported, depending
10335	 * on the value of inquiry_pq_no_lun.
10336	 *
10337	 * According to the spec (SPC-4 r34), the peripheral qualifier
10338	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10339	 *
10340	 * "A peripheral device having the specified peripheral device type
10341	 * is not connected to this logical unit. However, the device
10342	 * server is capable of supporting the specified peripheral device
10343	 * type on this logical unit."
10344	 *
10345	 * According to the same spec, the peripheral qualifier
10346	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10347	 *
10348	 * "The device server is not capable of supporting a peripheral
10349	 * device on this logical unit. For this peripheral qualifier the
10350	 * peripheral device type shall be set to 1Fh. All other peripheral
10351	 * device type values are reserved for this peripheral qualifier."
10352	 *
10353	 * Given the text, it would seem that we probably want to report that
10354	 * the LUN is offline here.  There is no LUN connected, but we can
10355	 * support a LUN at the given LUN number.
10356	 *
10357	 * In the real world, though, it sounds like things are a little
10358	 * different:
10359	 *
10360	 * - Linux, when presented with a LUN with the offline peripheral
10361	 *   qualifier, will create an sg driver instance for it.  So when
10362	 *   you attach it to CTL, you wind up with a ton of sg driver
10363	 *   instances.  (One for every LUN that Linux bothered to probe.)
10364	 *   Linux does this despite the fact that it issues a REPORT LUNs
10365	 *   to LUN 0 to get the inventory of supported LUNs.
10366	 *
10367	 * - There is other anecdotal evidence (from Emulex folks) about
10368	 *   arrays that use the offline peripheral qualifier for LUNs that
10369	 *   are on the "passive" path in an active/passive array.
10370	 *
10371	 * So the solution is provide a hopefully reasonable default
10372	 * (return bad/no LUN) and allow the user to change the behavior
10373	 * with a tunable/sysctl variable.
10374	 */
10375	if (lun != NULL)
10376		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10377				  lun->be_lun->lun_type;
10378	else if (ctl_softc->inquiry_pq_no_lun == 0)
10379		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10380	else
10381		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10382
10383	/* RMB in byte 2 is 0 */
10384	inq_ptr->version = SCSI_REV_SPC3;
10385
10386	/*
10387	 * According to SAM-3, even if a device only supports a single
10388	 * level of LUN addressing, it should still set the HISUP bit:
10389	 *
10390	 * 4.9.1 Logical unit numbers overview
10391	 *
10392	 * All logical unit number formats described in this standard are
10393	 * hierarchical in structure even when only a single level in that
10394	 * hierarchy is used. The HISUP bit shall be set to one in the
10395	 * standard INQUIRY data (see SPC-2) when any logical unit number
10396	 * format described in this standard is used.  Non-hierarchical
10397	 * formats are outside the scope of this standard.
10398	 *
10399	 * Therefore we set the HiSup bit here.
10400	 *
10401	 * The reponse format is 2, per SPC-3.
10402	 */
10403	inq_ptr->response_format = SID_HiSup | 2;
10404
10405	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
10406	CTL_DEBUG_PRINT(("additional_length = %d\n",
10407			 inq_ptr->additional_length));
10408
10409	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
10410	/* 16 bit addressing */
10411	if (is_fc == 0)
10412		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10413	/* XXX set the SID_MultiP bit here if we're actually going to
10414	   respond on multiple ports */
10415	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10416
10417	/* 16 bit data bus, synchronous transfers */
10418	/* XXX these flags don't apply for FC */
10419	if (is_fc == 0)
10420		inq_ptr->flags = SID_WBus16 | SID_Sync;
10421	/*
10422	 * XXX KDM do we want to support tagged queueing on the control
10423	 * device at all?
10424	 */
10425	if ((lun == NULL)
10426	 || (lun->be_lun->lun_type != T_PROCESSOR))
10427		inq_ptr->flags |= SID_CmdQue;
10428	/*
10429	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10430	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10431	 * name and 4 bytes for the revision.
10432	 */
10433	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10434	    "vendor")) == NULL) {
10435		strcpy(inq_ptr->vendor, CTL_VENDOR);
10436	} else {
10437		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10438		strncpy(inq_ptr->vendor, val,
10439		    min(sizeof(inq_ptr->vendor), strlen(val)));
10440	}
10441	if (lun == NULL) {
10442		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10443	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10444		switch (lun->be_lun->lun_type) {
10445		case T_DIRECT:
10446			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10447			break;
10448		case T_PROCESSOR:
10449			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
10450			break;
10451		default:
10452			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
10453			break;
10454		}
10455	} else {
10456		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10457		strncpy(inq_ptr->product, val,
10458		    min(sizeof(inq_ptr->product), strlen(val)));
10459	}
10460
10461	/*
10462	 * XXX make this a macro somewhere so it automatically gets
10463	 * incremented when we make changes.
10464	 */
10465	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10466	    "revision")) == NULL) {
10467		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10468	} else {
10469		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10470		strncpy(inq_ptr->revision, val,
10471		    min(sizeof(inq_ptr->revision), strlen(val)));
10472	}
10473
10474	/*
10475	 * For parallel SCSI, we support double transition and single
10476	 * transition clocking.  We also support QAS (Quick Arbitration
10477	 * and Selection) and Information Unit transfers on both the
10478	 * control and array devices.
10479	 */
10480	if (is_fc == 0)
10481		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10482				    SID_SPI_IUS;
10483
10484	/* SAM-3 */
10485	scsi_ulto2b(0x0060, inq_ptr->version1);
10486	/* SPC-3 (no version claimed) XXX should we claim a version? */
10487	scsi_ulto2b(0x0300, inq_ptr->version2);
10488	if (is_fc) {
10489		/* FCP-2 ANSI INCITS.350:2003 */
10490		scsi_ulto2b(0x0917, inq_ptr->version3);
10491	} else {
10492		/* SPI-4 ANSI INCITS.362:200x */
10493		scsi_ulto2b(0x0B56, inq_ptr->version3);
10494	}
10495
10496	if (lun == NULL) {
10497		/* SBC-2 (no version claimed) XXX should we claim a version? */
10498		scsi_ulto2b(0x0320, inq_ptr->version4);
10499	} else {
10500		switch (lun->be_lun->lun_type) {
10501		case T_DIRECT:
10502			/*
10503			 * SBC-2 (no version claimed) XXX should we claim a
10504			 * version?
10505			 */
10506			scsi_ulto2b(0x0320, inq_ptr->version4);
10507			break;
10508		case T_PROCESSOR:
10509		default:
10510			break;
10511		}
10512	}
10513
10514	ctsio->scsi_status = SCSI_STATUS_OK;
10515	if (ctsio->kern_data_len > 0) {
10516		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10517		ctsio->be_move_done = ctl_config_move_done;
10518		ctl_datamove((union ctl_io *)ctsio);
10519	} else {
10520		ctsio->io_hdr.status = CTL_SUCCESS;
10521		ctl_done((union ctl_io *)ctsio);
10522	}
10523
10524	return (CTL_RETVAL_COMPLETE);
10525}
10526
10527int
10528ctl_inquiry(struct ctl_scsiio *ctsio)
10529{
10530	struct scsi_inquiry *cdb;
10531	int retval;
10532
10533	cdb = (struct scsi_inquiry *)ctsio->cdb;
10534
10535	retval = 0;
10536
10537	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10538
10539	/*
10540	 * Right now, we don't support the CmdDt inquiry information.
10541	 * This would be nice to support in the future.  When we do
10542	 * support it, we should change this test so that it checks to make
10543	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10544	 */
10545#ifdef notyet
10546	if (((cdb->byte2 & SI_EVPD)
10547	 && (cdb->byte2 & SI_CMDDT)))
10548#endif
10549	if (cdb->byte2 & SI_CMDDT) {
10550		/*
10551		 * Point to the SI_CMDDT bit.  We might change this
10552		 * when we support SI_CMDDT, but since both bits would be
10553		 * "wrong", this should probably just stay as-is then.
10554		 */
10555		ctl_set_invalid_field(ctsio,
10556				      /*sks_valid*/ 1,
10557				      /*command*/ 1,
10558				      /*field*/ 1,
10559				      /*bit_valid*/ 1,
10560				      /*bit*/ 1);
10561		ctl_done((union ctl_io *)ctsio);
10562		return (CTL_RETVAL_COMPLETE);
10563	}
10564	if (cdb->byte2 & SI_EVPD)
10565		retval = ctl_inquiry_evpd(ctsio);
10566#ifdef notyet
10567	else if (cdb->byte2 & SI_CMDDT)
10568		retval = ctl_inquiry_cmddt(ctsio);
10569#endif
10570	else
10571		retval = ctl_inquiry_std(ctsio);
10572
10573	return (retval);
10574}
10575
10576/*
10577 * For known CDB types, parse the LBA and length.
10578 */
10579static int
10580ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10581{
10582	if (io->io_hdr.io_type != CTL_IO_SCSI)
10583		return (1);
10584
10585	switch (io->scsiio.cdb[0]) {
10586	case COMPARE_AND_WRITE: {
10587		struct scsi_compare_and_write *cdb;
10588
10589		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10590
10591		*lba = scsi_8btou64(cdb->addr);
10592		*len = cdb->length;
10593		break;
10594	}
10595	case READ_6:
10596	case WRITE_6: {
10597		struct scsi_rw_6 *cdb;
10598
10599		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10600
10601		*lba = scsi_3btoul(cdb->addr);
10602		/* only 5 bits are valid in the most significant address byte */
10603		*lba &= 0x1fffff;
10604		*len = cdb->length;
10605		break;
10606	}
10607	case READ_10:
10608	case WRITE_10: {
10609		struct scsi_rw_10 *cdb;
10610
10611		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10612
10613		*lba = scsi_4btoul(cdb->addr);
10614		*len = scsi_2btoul(cdb->length);
10615		break;
10616	}
10617	case WRITE_VERIFY_10: {
10618		struct scsi_write_verify_10 *cdb;
10619
10620		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10621
10622		*lba = scsi_4btoul(cdb->addr);
10623		*len = scsi_2btoul(cdb->length);
10624		break;
10625	}
10626	case READ_12:
10627	case WRITE_12: {
10628		struct scsi_rw_12 *cdb;
10629
10630		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10631
10632		*lba = scsi_4btoul(cdb->addr);
10633		*len = scsi_4btoul(cdb->length);
10634		break;
10635	}
10636	case WRITE_VERIFY_12: {
10637		struct scsi_write_verify_12 *cdb;
10638
10639		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10640
10641		*lba = scsi_4btoul(cdb->addr);
10642		*len = scsi_4btoul(cdb->length);
10643		break;
10644	}
10645	case READ_16:
10646	case WRITE_16: {
10647		struct scsi_rw_16 *cdb;
10648
10649		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10650
10651		*lba = scsi_8btou64(cdb->addr);
10652		*len = scsi_4btoul(cdb->length);
10653		break;
10654	}
10655	case WRITE_VERIFY_16: {
10656		struct scsi_write_verify_16 *cdb;
10657
10658		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10659
10660
10661		*lba = scsi_8btou64(cdb->addr);
10662		*len = scsi_4btoul(cdb->length);
10663		break;
10664	}
10665	case WRITE_SAME_10: {
10666		struct scsi_write_same_10 *cdb;
10667
10668		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10669
10670		*lba = scsi_4btoul(cdb->addr);
10671		*len = scsi_2btoul(cdb->length);
10672		break;
10673	}
10674	case WRITE_SAME_16: {
10675		struct scsi_write_same_16 *cdb;
10676
10677		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10678
10679		*lba = scsi_8btou64(cdb->addr);
10680		*len = scsi_4btoul(cdb->length);
10681		break;
10682	}
10683	case VERIFY_10: {
10684		struct scsi_verify_10 *cdb;
10685
10686		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10687
10688		*lba = scsi_4btoul(cdb->addr);
10689		*len = scsi_2btoul(cdb->length);
10690		break;
10691	}
10692	case VERIFY_12: {
10693		struct scsi_verify_12 *cdb;
10694
10695		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10696
10697		*lba = scsi_4btoul(cdb->addr);
10698		*len = scsi_4btoul(cdb->length);
10699		break;
10700	}
10701	case VERIFY_16: {
10702		struct scsi_verify_16 *cdb;
10703
10704		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10705
10706		*lba = scsi_8btou64(cdb->addr);
10707		*len = scsi_4btoul(cdb->length);
10708		break;
10709	}
10710	default:
10711		return (1);
10712		break; /* NOTREACHED */
10713	}
10714
10715	return (0);
10716}
10717
10718static ctl_action
10719ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10720{
10721	uint64_t endlba1, endlba2;
10722
10723	endlba1 = lba1 + len1 - 1;
10724	endlba2 = lba2 + len2 - 1;
10725
10726	if ((endlba1 < lba2)
10727	 || (endlba2 < lba1))
10728		return (CTL_ACTION_PASS);
10729	else
10730		return (CTL_ACTION_BLOCK);
10731}
10732
10733static ctl_action
10734ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10735{
10736	uint64_t lba1, lba2;
10737	uint32_t len1, len2;
10738	int retval;
10739
10740	retval = ctl_get_lba_len(io1, &lba1, &len1);
10741	if (retval != 0)
10742		return (CTL_ACTION_ERROR);
10743
10744	retval = ctl_get_lba_len(io2, &lba2, &len2);
10745	if (retval != 0)
10746		return (CTL_ACTION_ERROR);
10747
10748	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10749}
10750
10751static ctl_action
10752ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10753{
10754	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10755	ctl_serialize_action *serialize_row;
10756
10757	/*
10758	 * The initiator attempted multiple untagged commands at the same
10759	 * time.  Can't do that.
10760	 */
10761	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10762	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10763	 && ((pending_io->io_hdr.nexus.targ_port ==
10764	      ooa_io->io_hdr.nexus.targ_port)
10765	  && (pending_io->io_hdr.nexus.initid.id ==
10766	      ooa_io->io_hdr.nexus.initid.id))
10767	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10768		return (CTL_ACTION_OVERLAP);
10769
10770	/*
10771	 * The initiator attempted to send multiple tagged commands with
10772	 * the same ID.  (It's fine if different initiators have the same
10773	 * tag ID.)
10774	 *
10775	 * Even if all of those conditions are true, we don't kill the I/O
10776	 * if the command ahead of us has been aborted.  We won't end up
10777	 * sending it to the FETD, and it's perfectly legal to resend a
10778	 * command with the same tag number as long as the previous
10779	 * instance of this tag number has been aborted somehow.
10780	 */
10781	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10782	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10783	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10784	 && ((pending_io->io_hdr.nexus.targ_port ==
10785	      ooa_io->io_hdr.nexus.targ_port)
10786	  && (pending_io->io_hdr.nexus.initid.id ==
10787	      ooa_io->io_hdr.nexus.initid.id))
10788	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10789		return (CTL_ACTION_OVERLAP_TAG);
10790
10791	/*
10792	 * If we get a head of queue tag, SAM-3 says that we should
10793	 * immediately execute it.
10794	 *
10795	 * What happens if this command would normally block for some other
10796	 * reason?  e.g. a request sense with a head of queue tag
10797	 * immediately after a write.  Normally that would block, but this
10798	 * will result in its getting executed immediately...
10799	 *
10800	 * We currently return "pass" instead of "skip", so we'll end up
10801	 * going through the rest of the queue to check for overlapped tags.
10802	 *
10803	 * XXX KDM check for other types of blockage first??
10804	 */
10805	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10806		return (CTL_ACTION_PASS);
10807
10808	/*
10809	 * Ordered tags have to block until all items ahead of them
10810	 * have completed.  If we get called with an ordered tag, we always
10811	 * block, if something else is ahead of us in the queue.
10812	 */
10813	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10814		return (CTL_ACTION_BLOCK);
10815
10816	/*
10817	 * Simple tags get blocked until all head of queue and ordered tags
10818	 * ahead of them have completed.  I'm lumping untagged commands in
10819	 * with simple tags here.  XXX KDM is that the right thing to do?
10820	 */
10821	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10822	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10823	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10824	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10825		return (CTL_ACTION_BLOCK);
10826
10827	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio);
10828	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio);
10829
10830	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10831
10832	switch (serialize_row[pending_entry->seridx]) {
10833	case CTL_SER_BLOCK:
10834		return (CTL_ACTION_BLOCK);
10835		break; /* NOTREACHED */
10836	case CTL_SER_EXTENT:
10837		return (ctl_extent_check(pending_io, ooa_io));
10838		break; /* NOTREACHED */
10839	case CTL_SER_PASS:
10840		return (CTL_ACTION_PASS);
10841		break; /* NOTREACHED */
10842	case CTL_SER_SKIP:
10843		return (CTL_ACTION_SKIP);
10844		break;
10845	default:
10846		panic("invalid serialization value %d",
10847		      serialize_row[pending_entry->seridx]);
10848		break; /* NOTREACHED */
10849	}
10850
10851	return (CTL_ACTION_ERROR);
10852}
10853
10854/*
10855 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10856 * Assumptions:
10857 * - pending_io is generally either incoming, or on the blocked queue
10858 * - starting I/O is the I/O we want to start the check with.
10859 */
10860static ctl_action
10861ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10862	      union ctl_io *starting_io)
10863{
10864	union ctl_io *ooa_io;
10865	ctl_action action;
10866
10867	mtx_assert(&lun->lun_lock, MA_OWNED);
10868
10869	/*
10870	 * Run back along the OOA queue, starting with the current
10871	 * blocked I/O and going through every I/O before it on the
10872	 * queue.  If starting_io is NULL, we'll just end up returning
10873	 * CTL_ACTION_PASS.
10874	 */
10875	for (ooa_io = starting_io; ooa_io != NULL;
10876	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10877	     ooa_links)){
10878
10879		/*
10880		 * This routine just checks to see whether
10881		 * cur_blocked is blocked by ooa_io, which is ahead
10882		 * of it in the queue.  It doesn't queue/dequeue
10883		 * cur_blocked.
10884		 */
10885		action = ctl_check_for_blockage(pending_io, ooa_io);
10886		switch (action) {
10887		case CTL_ACTION_BLOCK:
10888		case CTL_ACTION_OVERLAP:
10889		case CTL_ACTION_OVERLAP_TAG:
10890		case CTL_ACTION_SKIP:
10891		case CTL_ACTION_ERROR:
10892			return (action);
10893			break; /* NOTREACHED */
10894		case CTL_ACTION_PASS:
10895			break;
10896		default:
10897			panic("invalid action %d", action);
10898			break;  /* NOTREACHED */
10899		}
10900	}
10901
10902	return (CTL_ACTION_PASS);
10903}
10904
10905/*
10906 * Assumptions:
10907 * - An I/O has just completed, and has been removed from the per-LUN OOA
10908 *   queue, so some items on the blocked queue may now be unblocked.
10909 */
10910static int
10911ctl_check_blocked(struct ctl_lun *lun)
10912{
10913	union ctl_io *cur_blocked, *next_blocked;
10914
10915	mtx_assert(&lun->lun_lock, MA_OWNED);
10916
10917	/*
10918	 * Run forward from the head of the blocked queue, checking each
10919	 * entry against the I/Os prior to it on the OOA queue to see if
10920	 * there is still any blockage.
10921	 *
10922	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10923	 * with our removing a variable on it while it is traversing the
10924	 * list.
10925	 */
10926	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10927	     cur_blocked != NULL; cur_blocked = next_blocked) {
10928		union ctl_io *prev_ooa;
10929		ctl_action action;
10930
10931		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10932							  blocked_links);
10933
10934		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10935						      ctl_ooaq, ooa_links);
10936
10937		/*
10938		 * If cur_blocked happens to be the first item in the OOA
10939		 * queue now, prev_ooa will be NULL, and the action
10940		 * returned will just be CTL_ACTION_PASS.
10941		 */
10942		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10943
10944		switch (action) {
10945		case CTL_ACTION_BLOCK:
10946			/* Nothing to do here, still blocked */
10947			break;
10948		case CTL_ACTION_OVERLAP:
10949		case CTL_ACTION_OVERLAP_TAG:
10950			/*
10951			 * This shouldn't happen!  In theory we've already
10952			 * checked this command for overlap...
10953			 */
10954			break;
10955		case CTL_ACTION_PASS:
10956		case CTL_ACTION_SKIP: {
10957			struct ctl_softc *softc;
10958			const struct ctl_cmd_entry *entry;
10959			uint32_t initidx;
10960			int isc_retval;
10961
10962			/*
10963			 * The skip case shouldn't happen, this transaction
10964			 * should have never made it onto the blocked queue.
10965			 */
10966			/*
10967			 * This I/O is no longer blocked, we can remove it
10968			 * from the blocked queue.  Since this is a TAILQ
10969			 * (doubly linked list), we can do O(1) removals
10970			 * from any place on the list.
10971			 */
10972			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10973				     blocked_links);
10974			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10975
10976			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10977				/*
10978				 * Need to send IO back to original side to
10979				 * run
10980				 */
10981				union ctl_ha_msg msg_info;
10982
10983				msg_info.hdr.original_sc =
10984					cur_blocked->io_hdr.original_sc;
10985				msg_info.hdr.serializing_sc = cur_blocked;
10986				msg_info.hdr.msg_type = CTL_MSG_R2R;
10987				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10988				     &msg_info, sizeof(msg_info), 0)) >
10989				     CTL_HA_STATUS_SUCCESS) {
10990					printf("CTL:Check Blocked error from "
10991					       "ctl_ha_msg_send %d\n",
10992					       isc_retval);
10993				}
10994				break;
10995			}
10996			entry = ctl_get_cmd_entry(&cur_blocked->scsiio);
10997			softc = control_softc;
10998
10999			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11000
11001			/*
11002			 * Check this I/O for LUN state changes that may
11003			 * have happened while this command was blocked.
11004			 * The LUN state may have been changed by a command
11005			 * ahead of us in the queue, so we need to re-check
11006			 * for any states that can be caused by SCSI
11007			 * commands.
11008			 */
11009			if (ctl_scsiio_lun_check(softc, lun, entry,
11010						 &cur_blocked->scsiio) == 0) {
11011				cur_blocked->io_hdr.flags |=
11012				                      CTL_FLAG_IS_WAS_ON_RTR;
11013				ctl_enqueue_rtr(cur_blocked);
11014			} else
11015				ctl_done(cur_blocked);
11016			break;
11017		}
11018		default:
11019			/*
11020			 * This probably shouldn't happen -- we shouldn't
11021			 * get CTL_ACTION_ERROR, or anything else.
11022			 */
11023			break;
11024		}
11025	}
11026
11027	return (CTL_RETVAL_COMPLETE);
11028}
11029
11030/*
11031 * This routine (with one exception) checks LUN flags that can be set by
11032 * commands ahead of us in the OOA queue.  These flags have to be checked
11033 * when a command initially comes in, and when we pull a command off the
11034 * blocked queue and are preparing to execute it.  The reason we have to
11035 * check these flags for commands on the blocked queue is that the LUN
11036 * state may have been changed by a command ahead of us while we're on the
11037 * blocked queue.
11038 *
11039 * Ordering is somewhat important with these checks, so please pay
11040 * careful attention to the placement of any new checks.
11041 */
11042static int
11043ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11044    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11045{
11046	int retval;
11047
11048	retval = 0;
11049
11050	mtx_assert(&lun->lun_lock, MA_OWNED);
11051
11052	/*
11053	 * If this shelf is a secondary shelf controller, we have to reject
11054	 * any media access commands.
11055	 */
11056#if 0
11057	/* No longer needed for HA */
11058	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11059	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11060		ctl_set_lun_standby(ctsio);
11061		retval = 1;
11062		goto bailout;
11063	}
11064#endif
11065
11066	/*
11067	 * Check for a reservation conflict.  If this command isn't allowed
11068	 * even on reserved LUNs, and if this initiator isn't the one who
11069	 * reserved us, reject the command with a reservation conflict.
11070	 */
11071	if ((lun->flags & CTL_LUN_RESERVED)
11072	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11073		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
11074		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
11075		 || (ctsio->io_hdr.nexus.targ_target.id !=
11076		     lun->rsv_nexus.targ_target.id)) {
11077			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11078			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11079			retval = 1;
11080			goto bailout;
11081		}
11082	}
11083
11084	if ( (lun->flags & CTL_LUN_PR_RESERVED)
11085	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11086		uint32_t residx;
11087
11088		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11089		/*
11090		 * if we aren't registered or it's a res holder type
11091		 * reservation and this isn't the res holder then set a
11092		 * conflict.
11093		 * NOTE: Commands which might be allowed on write exclusive
11094		 * type reservations are checked in the particular command
11095		 * for a conflict. Read and SSU are the only ones.
11096		 */
11097		if (!lun->per_res[residx].registered
11098		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11099			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11100			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11101			retval = 1;
11102			goto bailout;
11103		}
11104
11105	}
11106
11107	if ((lun->flags & CTL_LUN_OFFLINE)
11108	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11109		ctl_set_lun_not_ready(ctsio);
11110		retval = 1;
11111		goto bailout;
11112	}
11113
11114	/*
11115	 * If the LUN is stopped, see if this particular command is allowed
11116	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11117	 */
11118	if ((lun->flags & CTL_LUN_STOPPED)
11119	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11120		/* "Logical unit not ready, initializing cmd. required" */
11121		ctl_set_lun_stopped(ctsio);
11122		retval = 1;
11123		goto bailout;
11124	}
11125
11126	if ((lun->flags & CTL_LUN_INOPERABLE)
11127	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11128		/* "Medium format corrupted" */
11129		ctl_set_medium_format_corrupted(ctsio);
11130		retval = 1;
11131		goto bailout;
11132	}
11133
11134bailout:
11135	return (retval);
11136
11137}
11138
11139static void
11140ctl_failover_io(union ctl_io *io, int have_lock)
11141{
11142	ctl_set_busy(&io->scsiio);
11143	ctl_done(io);
11144}
11145
11146static void
11147ctl_failover(void)
11148{
11149	struct ctl_lun *lun;
11150	struct ctl_softc *ctl_softc;
11151	union ctl_io *next_io, *pending_io;
11152	union ctl_io *io;
11153	int lun_idx;
11154	int i;
11155
11156	ctl_softc = control_softc;
11157
11158	mtx_lock(&ctl_softc->ctl_lock);
11159	/*
11160	 * Remove any cmds from the other SC from the rtr queue.  These
11161	 * will obviously only be for LUNs for which we're the primary.
11162	 * We can't send status or get/send data for these commands.
11163	 * Since they haven't been executed yet, we can just remove them.
11164	 * We'll either abort them or delete them below, depending on
11165	 * which HA mode we're in.
11166	 */
11167#ifdef notyet
11168	mtx_lock(&ctl_softc->queue_lock);
11169	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11170	     io != NULL; io = next_io) {
11171		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11172		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11173			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11174				      ctl_io_hdr, links);
11175	}
11176	mtx_unlock(&ctl_softc->queue_lock);
11177#endif
11178
11179	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11180		lun = ctl_softc->ctl_luns[lun_idx];
11181		if (lun==NULL)
11182			continue;
11183
11184		/*
11185		 * Processor LUNs are primary on both sides.
11186		 * XXX will this always be true?
11187		 */
11188		if (lun->be_lun->lun_type == T_PROCESSOR)
11189			continue;
11190
11191		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11192		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11193			printf("FAILOVER: primary lun %d\n", lun_idx);
11194		        /*
11195			 * Remove all commands from the other SC. First from the
11196			 * blocked queue then from the ooa queue. Once we have
11197			 * removed them. Call ctl_check_blocked to see if there
11198			 * is anything that can run.
11199			 */
11200			for (io = (union ctl_io *)TAILQ_FIRST(
11201			     &lun->blocked_queue); io != NULL; io = next_io) {
11202
11203		        	next_io = (union ctl_io *)TAILQ_NEXT(
11204				    &io->io_hdr, blocked_links);
11205
11206				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11207					TAILQ_REMOVE(&lun->blocked_queue,
11208						     &io->io_hdr,blocked_links);
11209					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11210					TAILQ_REMOVE(&lun->ooa_queue,
11211						     &io->io_hdr, ooa_links);
11212
11213					ctl_free_io(io);
11214				}
11215			}
11216
11217			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11218	     		     io != NULL; io = next_io) {
11219
11220		        	next_io = (union ctl_io *)TAILQ_NEXT(
11221				    &io->io_hdr, ooa_links);
11222
11223				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11224
11225					TAILQ_REMOVE(&lun->ooa_queue,
11226						&io->io_hdr,
11227					     	ooa_links);
11228
11229					ctl_free_io(io);
11230				}
11231			}
11232			ctl_check_blocked(lun);
11233		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11234			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11235
11236			printf("FAILOVER: primary lun %d\n", lun_idx);
11237			/*
11238			 * Abort all commands from the other SC.  We can't
11239			 * send status back for them now.  These should get
11240			 * cleaned up when they are completed or come out
11241			 * for a datamove operation.
11242			 */
11243			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11244	     		     io != NULL; io = next_io) {
11245		        	next_io = (union ctl_io *)TAILQ_NEXT(
11246					&io->io_hdr, ooa_links);
11247
11248				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11249					io->io_hdr.flags |= CTL_FLAG_ABORT;
11250			}
11251		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11252			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11253
11254			printf("FAILOVER: secondary lun %d\n", lun_idx);
11255
11256			lun->flags |= CTL_LUN_PRIMARY_SC;
11257
11258			/*
11259			 * We send all I/O that was sent to this controller
11260			 * and redirected to the other side back with
11261			 * busy status, and have the initiator retry it.
11262			 * Figuring out how much data has been transferred,
11263			 * etc. and picking up where we left off would be
11264			 * very tricky.
11265			 *
11266			 * XXX KDM need to remove I/O from the blocked
11267			 * queue as well!
11268			 */
11269			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11270			     &lun->ooa_queue); pending_io != NULL;
11271			     pending_io = next_io) {
11272
11273				next_io =  (union ctl_io *)TAILQ_NEXT(
11274					&pending_io->io_hdr, ooa_links);
11275
11276				pending_io->io_hdr.flags &=
11277					~CTL_FLAG_SENT_2OTHER_SC;
11278
11279				if (pending_io->io_hdr.flags &
11280				    CTL_FLAG_IO_ACTIVE) {
11281					pending_io->io_hdr.flags |=
11282						CTL_FLAG_FAILOVER;
11283				} else {
11284					ctl_set_busy(&pending_io->scsiio);
11285					ctl_done(pending_io);
11286				}
11287			}
11288
11289			/*
11290			 * Build Unit Attention
11291			 */
11292			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11293				lun->pending_sense[i].ua_pending |=
11294				                     CTL_UA_ASYM_ACC_CHANGE;
11295			}
11296		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11297			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11298			printf("FAILOVER: secondary lun %d\n", lun_idx);
11299			/*
11300			 * if the first io on the OOA is not on the RtR queue
11301			 * add it.
11302			 */
11303			lun->flags |= CTL_LUN_PRIMARY_SC;
11304
11305			pending_io = (union ctl_io *)TAILQ_FIRST(
11306			    &lun->ooa_queue);
11307			if (pending_io==NULL) {
11308				printf("Nothing on OOA queue\n");
11309				continue;
11310			}
11311
11312			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11313			if ((pending_io->io_hdr.flags &
11314			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11315				pending_io->io_hdr.flags |=
11316				    CTL_FLAG_IS_WAS_ON_RTR;
11317				ctl_enqueue_rtr(pending_io);
11318			}
11319#if 0
11320			else
11321			{
11322				printf("Tag 0x%04x is running\n",
11323				      pending_io->scsiio.tag_num);
11324			}
11325#endif
11326
11327			next_io = (union ctl_io *)TAILQ_NEXT(
11328			    &pending_io->io_hdr, ooa_links);
11329			for (pending_io=next_io; pending_io != NULL;
11330			     pending_io = next_io) {
11331				pending_io->io_hdr.flags &=
11332				    ~CTL_FLAG_SENT_2OTHER_SC;
11333				next_io = (union ctl_io *)TAILQ_NEXT(
11334					&pending_io->io_hdr, ooa_links);
11335				if (pending_io->io_hdr.flags &
11336				    CTL_FLAG_IS_WAS_ON_RTR) {
11337#if 0
11338				        printf("Tag 0x%04x is running\n",
11339				      		pending_io->scsiio.tag_num);
11340#endif
11341					continue;
11342				}
11343
11344				switch (ctl_check_ooa(lun, pending_io,
11345			            (union ctl_io *)TAILQ_PREV(
11346				    &pending_io->io_hdr, ctl_ooaq,
11347				    ooa_links))) {
11348
11349				case CTL_ACTION_BLOCK:
11350					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11351							  &pending_io->io_hdr,
11352							  blocked_links);
11353					pending_io->io_hdr.flags |=
11354					    CTL_FLAG_BLOCKED;
11355					break;
11356				case CTL_ACTION_PASS:
11357				case CTL_ACTION_SKIP:
11358					pending_io->io_hdr.flags |=
11359					    CTL_FLAG_IS_WAS_ON_RTR;
11360					ctl_enqueue_rtr(pending_io);
11361					break;
11362				case CTL_ACTION_OVERLAP:
11363					ctl_set_overlapped_cmd(
11364					    (struct ctl_scsiio *)pending_io);
11365					ctl_done(pending_io);
11366					break;
11367				case CTL_ACTION_OVERLAP_TAG:
11368					ctl_set_overlapped_tag(
11369					    (struct ctl_scsiio *)pending_io,
11370					    pending_io->scsiio.tag_num & 0xff);
11371					ctl_done(pending_io);
11372					break;
11373				case CTL_ACTION_ERROR:
11374				default:
11375					ctl_set_internal_failure(
11376						(struct ctl_scsiio *)pending_io,
11377						0,  // sks_valid
11378						0); //retry count
11379					ctl_done(pending_io);
11380					break;
11381				}
11382			}
11383
11384			/*
11385			 * Build Unit Attention
11386			 */
11387			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11388				lun->pending_sense[i].ua_pending |=
11389				                     CTL_UA_ASYM_ACC_CHANGE;
11390			}
11391		} else {
11392			panic("Unhandled HA mode failover, LUN flags = %#x, "
11393			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11394		}
11395	}
11396	ctl_pause_rtr = 0;
11397	mtx_unlock(&ctl_softc->ctl_lock);
11398}
11399
11400static int
11401ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11402{
11403	struct ctl_lun *lun;
11404	const struct ctl_cmd_entry *entry;
11405	uint32_t initidx, targ_lun;
11406	int retval;
11407
11408	retval = 0;
11409
11410	lun = NULL;
11411
11412	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11413	if ((targ_lun < CTL_MAX_LUNS)
11414	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11415		lun = ctl_softc->ctl_luns[targ_lun];
11416		/*
11417		 * If the LUN is invalid, pretend that it doesn't exist.
11418		 * It will go away as soon as all pending I/O has been
11419		 * completed.
11420		 */
11421		if (lun->flags & CTL_LUN_DISABLED) {
11422			lun = NULL;
11423		} else {
11424			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11425			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11426				lun->be_lun;
11427			if (lun->be_lun->lun_type == T_PROCESSOR) {
11428				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11429			}
11430
11431			/*
11432			 * Every I/O goes into the OOA queue for a
11433			 * particular LUN, and stays there until completion.
11434			 */
11435			mtx_lock(&lun->lun_lock);
11436			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11437			    ooa_links);
11438		}
11439	} else {
11440		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11441		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11442	}
11443
11444	/* Get command entry and return error if it is unsuppotyed. */
11445	entry = ctl_validate_command(ctsio);
11446	if (entry == NULL) {
11447		if (lun)
11448			mtx_unlock(&lun->lun_lock);
11449		return (retval);
11450	}
11451
11452	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11453	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11454
11455	/*
11456	 * Check to see whether we can send this command to LUNs that don't
11457	 * exist.  This should pretty much only be the case for inquiry
11458	 * and request sense.  Further checks, below, really require having
11459	 * a LUN, so we can't really check the command anymore.  Just put
11460	 * it on the rtr queue.
11461	 */
11462	if (lun == NULL) {
11463		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11464			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11465			ctl_enqueue_rtr((union ctl_io *)ctsio);
11466			return (retval);
11467		}
11468
11469		ctl_set_unsupported_lun(ctsio);
11470		ctl_done((union ctl_io *)ctsio);
11471		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11472		return (retval);
11473	} else {
11474		/*
11475		 * Make sure we support this particular command on this LUN.
11476		 * e.g., we don't support writes to the control LUN.
11477		 */
11478		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11479			mtx_unlock(&lun->lun_lock);
11480			ctl_set_invalid_opcode(ctsio);
11481			ctl_done((union ctl_io *)ctsio);
11482			return (retval);
11483		}
11484	}
11485
11486	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11487
11488	/*
11489	 * If we've got a request sense, it'll clear the contingent
11490	 * allegiance condition.  Otherwise, if we have a CA condition for
11491	 * this initiator, clear it, because it sent down a command other
11492	 * than request sense.
11493	 */
11494	if ((ctsio->cdb[0] != REQUEST_SENSE)
11495	 && (ctl_is_set(lun->have_ca, initidx)))
11496		ctl_clear_mask(lun->have_ca, initidx);
11497
11498	/*
11499	 * If the command has this flag set, it handles its own unit
11500	 * attention reporting, we shouldn't do anything.  Otherwise we
11501	 * check for any pending unit attentions, and send them back to the
11502	 * initiator.  We only do this when a command initially comes in,
11503	 * not when we pull it off the blocked queue.
11504	 *
11505	 * According to SAM-3, section 5.3.2, the order that things get
11506	 * presented back to the host is basically unit attentions caused
11507	 * by some sort of reset event, busy status, reservation conflicts
11508	 * or task set full, and finally any other status.
11509	 *
11510	 * One issue here is that some of the unit attentions we report
11511	 * don't fall into the "reset" category (e.g. "reported luns data
11512	 * has changed").  So reporting it here, before the reservation
11513	 * check, may be technically wrong.  I guess the only thing to do
11514	 * would be to check for and report the reset events here, and then
11515	 * check for the other unit attention types after we check for a
11516	 * reservation conflict.
11517	 *
11518	 * XXX KDM need to fix this
11519	 */
11520	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11521		ctl_ua_type ua_type;
11522
11523		ua_type = lun->pending_sense[initidx].ua_pending;
11524		if (ua_type != CTL_UA_NONE) {
11525			scsi_sense_data_type sense_format;
11526
11527			if (lun != NULL)
11528				sense_format = (lun->flags &
11529				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11530				    SSD_TYPE_FIXED;
11531			else
11532				sense_format = SSD_TYPE_FIXED;
11533
11534			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11535					       sense_format);
11536			if (ua_type != CTL_UA_NONE) {
11537				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11538				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11539						       CTL_AUTOSENSE;
11540				ctsio->sense_len = SSD_FULL_SIZE;
11541				lun->pending_sense[initidx].ua_pending &=
11542					~ua_type;
11543				mtx_unlock(&lun->lun_lock);
11544				ctl_done((union ctl_io *)ctsio);
11545				return (retval);
11546			}
11547		}
11548	}
11549
11550
11551	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11552		mtx_unlock(&lun->lun_lock);
11553		ctl_done((union ctl_io *)ctsio);
11554		return (retval);
11555	}
11556
11557	/*
11558	 * XXX CHD this is where we want to send IO to other side if
11559	 * this LUN is secondary on this SC. We will need to make a copy
11560	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11561	 * the copy we send as FROM_OTHER.
11562	 * We also need to stuff the address of the original IO so we can
11563	 * find it easily. Something similar will need be done on the other
11564	 * side so when we are done we can find the copy.
11565	 */
11566	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11567		union ctl_ha_msg msg_info;
11568		int isc_retval;
11569
11570		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11571
11572		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11573		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11574#if 0
11575		printf("1. ctsio %p\n", ctsio);
11576#endif
11577		msg_info.hdr.serializing_sc = NULL;
11578		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11579		msg_info.scsi.tag_num = ctsio->tag_num;
11580		msg_info.scsi.tag_type = ctsio->tag_type;
11581		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11582
11583		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11584
11585		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11586		    (void *)&msg_info, sizeof(msg_info), 0)) >
11587		    CTL_HA_STATUS_SUCCESS) {
11588			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11589			       isc_retval);
11590			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11591		} else {
11592#if 0
11593			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11594#endif
11595		}
11596
11597		/*
11598		 * XXX KDM this I/O is off the incoming queue, but hasn't
11599		 * been inserted on any other queue.  We may need to come
11600		 * up with a holding queue while we wait for serialization
11601		 * so that we have an idea of what we're waiting for from
11602		 * the other side.
11603		 */
11604		mtx_unlock(&lun->lun_lock);
11605		return (retval);
11606	}
11607
11608	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11609			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11610			      ctl_ooaq, ooa_links))) {
11611	case CTL_ACTION_BLOCK:
11612		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11613		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11614				  blocked_links);
11615		mtx_unlock(&lun->lun_lock);
11616		return (retval);
11617	case CTL_ACTION_PASS:
11618	case CTL_ACTION_SKIP:
11619		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11620		mtx_unlock(&lun->lun_lock);
11621		ctl_enqueue_rtr((union ctl_io *)ctsio);
11622		break;
11623	case CTL_ACTION_OVERLAP:
11624		mtx_unlock(&lun->lun_lock);
11625		ctl_set_overlapped_cmd(ctsio);
11626		ctl_done((union ctl_io *)ctsio);
11627		break;
11628	case CTL_ACTION_OVERLAP_TAG:
11629		mtx_unlock(&lun->lun_lock);
11630		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11631		ctl_done((union ctl_io *)ctsio);
11632		break;
11633	case CTL_ACTION_ERROR:
11634	default:
11635		mtx_unlock(&lun->lun_lock);
11636		ctl_set_internal_failure(ctsio,
11637					 /*sks_valid*/ 0,
11638					 /*retry_count*/ 0);
11639		ctl_done((union ctl_io *)ctsio);
11640		break;
11641	}
11642	return (retval);
11643}
11644
11645const struct ctl_cmd_entry *
11646ctl_get_cmd_entry(struct ctl_scsiio *ctsio)
11647{
11648	const struct ctl_cmd_entry *entry;
11649	int service_action;
11650
11651	entry = &ctl_cmd_table[ctsio->cdb[0]];
11652	if (entry->flags & CTL_CMD_FLAG_SA5) {
11653		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11654		entry = &((const struct ctl_cmd_entry *)
11655		    entry->execute)[service_action];
11656	}
11657	return (entry);
11658}
11659
11660const struct ctl_cmd_entry *
11661ctl_validate_command(struct ctl_scsiio *ctsio)
11662{
11663	const struct ctl_cmd_entry *entry;
11664	int i;
11665	uint8_t diff;
11666
11667	entry = ctl_get_cmd_entry(ctsio);
11668	if (entry->execute == NULL) {
11669		ctl_set_invalid_opcode(ctsio);
11670		ctl_done((union ctl_io *)ctsio);
11671		return (NULL);
11672	}
11673	KASSERT(entry->length > 0,
11674	    ("Not defined length for command 0x%02x/0x%02x",
11675	     ctsio->cdb[0], ctsio->cdb[1]));
11676	for (i = 1; i < entry->length; i++) {
11677		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11678		if (diff == 0)
11679			continue;
11680		ctl_set_invalid_field(ctsio,
11681				      /*sks_valid*/ 1,
11682				      /*command*/ 1,
11683				      /*field*/ i,
11684				      /*bit_valid*/ 1,
11685				      /*bit*/ fls(diff) - 1);
11686		ctl_done((union ctl_io *)ctsio);
11687		return (NULL);
11688	}
11689	return (entry);
11690}
11691
11692static int
11693ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11694{
11695
11696	switch (lun_type) {
11697	case T_PROCESSOR:
11698		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11699		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11700			return (0);
11701		break;
11702	case T_DIRECT:
11703		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11704		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11705			return (0);
11706		break;
11707	default:
11708		return (0);
11709	}
11710	return (1);
11711}
11712
11713static int
11714ctl_scsiio(struct ctl_scsiio *ctsio)
11715{
11716	int retval;
11717	const struct ctl_cmd_entry *entry;
11718
11719	retval = CTL_RETVAL_COMPLETE;
11720
11721	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11722
11723	entry = ctl_get_cmd_entry(ctsio);
11724
11725	/*
11726	 * If this I/O has been aborted, just send it straight to
11727	 * ctl_done() without executing it.
11728	 */
11729	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11730		ctl_done((union ctl_io *)ctsio);
11731		goto bailout;
11732	}
11733
11734	/*
11735	 * All the checks should have been handled by ctl_scsiio_precheck().
11736	 * We should be clear now to just execute the I/O.
11737	 */
11738	retval = entry->execute(ctsio);
11739
11740bailout:
11741	return (retval);
11742}
11743
11744/*
11745 * Since we only implement one target right now, a bus reset simply resets
11746 * our single target.
11747 */
11748static int
11749ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11750{
11751	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11752}
11753
11754static int
11755ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11756		 ctl_ua_type ua_type)
11757{
11758	struct ctl_lun *lun;
11759	int retval;
11760
11761	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11762		union ctl_ha_msg msg_info;
11763
11764		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11765		msg_info.hdr.nexus = io->io_hdr.nexus;
11766		if (ua_type==CTL_UA_TARG_RESET)
11767			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11768		else
11769			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11770		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11771		msg_info.hdr.original_sc = NULL;
11772		msg_info.hdr.serializing_sc = NULL;
11773		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11774		    (void *)&msg_info, sizeof(msg_info), 0)) {
11775		}
11776	}
11777	retval = 0;
11778
11779	mtx_lock(&ctl_softc->ctl_lock);
11780	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11781		retval += ctl_lun_reset(lun, io, ua_type);
11782	mtx_unlock(&ctl_softc->ctl_lock);
11783
11784	return (retval);
11785}
11786
11787/*
11788 * The LUN should always be set.  The I/O is optional, and is used to
11789 * distinguish between I/Os sent by this initiator, and by other
11790 * initiators.  We set unit attention for initiators other than this one.
11791 * SAM-3 is vague on this point.  It does say that a unit attention should
11792 * be established for other initiators when a LUN is reset (see section
11793 * 5.7.3), but it doesn't specifically say that the unit attention should
11794 * be established for this particular initiator when a LUN is reset.  Here
11795 * is the relevant text, from SAM-3 rev 8:
11796 *
11797 * 5.7.2 When a SCSI initiator port aborts its own tasks
11798 *
11799 * When a SCSI initiator port causes its own task(s) to be aborted, no
11800 * notification that the task(s) have been aborted shall be returned to
11801 * the SCSI initiator port other than the completion response for the
11802 * command or task management function action that caused the task(s) to
11803 * be aborted and notification(s) associated with related effects of the
11804 * action (e.g., a reset unit attention condition).
11805 *
11806 * XXX KDM for now, we're setting unit attention for all initiators.
11807 */
11808static int
11809ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11810{
11811	union ctl_io *xio;
11812#if 0
11813	uint32_t initindex;
11814#endif
11815	int i;
11816
11817	mtx_lock(&lun->lun_lock);
11818	/*
11819	 * Run through the OOA queue and abort each I/O.
11820	 */
11821#if 0
11822	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11823#endif
11824	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11825	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11826		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11827	}
11828
11829	/*
11830	 * This version sets unit attention for every
11831	 */
11832#if 0
11833	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11834	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11835		if (initindex == i)
11836			continue;
11837		lun->pending_sense[i].ua_pending |= ua_type;
11838	}
11839#endif
11840
11841	/*
11842	 * A reset (any kind, really) clears reservations established with
11843	 * RESERVE/RELEASE.  It does not clear reservations established
11844	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11845	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11846	 * reservations made with the RESERVE/RELEASE commands, because
11847	 * those commands are obsolete in SPC-3.
11848	 */
11849	lun->flags &= ~CTL_LUN_RESERVED;
11850
11851	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11852		ctl_clear_mask(lun->have_ca, i);
11853		lun->pending_sense[i].ua_pending |= ua_type;
11854	}
11855	mtx_unlock(&lun->lun_lock);
11856
11857	return (0);
11858}
11859
11860static int
11861ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11862    int other_sc)
11863{
11864	union ctl_io *xio;
11865	int found;
11866
11867	mtx_assert(&lun->lun_lock, MA_OWNED);
11868
11869	/*
11870	 * Run through the OOA queue and attempt to find the given I/O.
11871	 * The target port, initiator ID, tag type and tag number have to
11872	 * match the values that we got from the initiator.  If we have an
11873	 * untagged command to abort, simply abort the first untagged command
11874	 * we come to.  We only allow one untagged command at a time of course.
11875	 */
11876	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11877	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11878
11879		if ((targ_port == UINT32_MAX ||
11880		     targ_port == xio->io_hdr.nexus.targ_port) &&
11881		    (init_id == UINT32_MAX ||
11882		     init_id == xio->io_hdr.nexus.initid.id)) {
11883			if (targ_port != xio->io_hdr.nexus.targ_port ||
11884			    init_id != xio->io_hdr.nexus.initid.id)
11885				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11886			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11887			found = 1;
11888			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11889				union ctl_ha_msg msg_info;
11890
11891				msg_info.hdr.nexus = xio->io_hdr.nexus;
11892				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11893				msg_info.task.tag_num = xio->scsiio.tag_num;
11894				msg_info.task.tag_type = xio->scsiio.tag_type;
11895				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11896				msg_info.hdr.original_sc = NULL;
11897				msg_info.hdr.serializing_sc = NULL;
11898				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11899				    (void *)&msg_info, sizeof(msg_info), 0);
11900			}
11901		}
11902	}
11903	return (found);
11904}
11905
11906static int
11907ctl_abort_task_set(union ctl_io *io)
11908{
11909	struct ctl_softc *softc = control_softc;
11910	struct ctl_lun *lun;
11911	uint32_t targ_lun;
11912
11913	/*
11914	 * Look up the LUN.
11915	 */
11916	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11917	mtx_lock(&softc->ctl_lock);
11918	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
11919		lun = softc->ctl_luns[targ_lun];
11920	else {
11921		mtx_unlock(&softc->ctl_lock);
11922		return (1);
11923	}
11924
11925	mtx_lock(&lun->lun_lock);
11926	mtx_unlock(&softc->ctl_lock);
11927	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11928		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11929		    io->io_hdr.nexus.initid.id,
11930		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11931	} else { /* CTL_TASK_CLEAR_TASK_SET */
11932		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11933		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11934	}
11935	mtx_unlock(&lun->lun_lock);
11936	return (0);
11937}
11938
11939static int
11940ctl_i_t_nexus_reset(union ctl_io *io)
11941{
11942	struct ctl_softc *softc = control_softc;
11943	struct ctl_lun *lun;
11944	uint32_t initindex;
11945
11946	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11947	mtx_lock(&softc->ctl_lock);
11948	STAILQ_FOREACH(lun, &softc->lun_list, links) {
11949		mtx_lock(&lun->lun_lock);
11950		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11951		    io->io_hdr.nexus.initid.id,
11952		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11953		ctl_clear_mask(lun->have_ca, initindex);
11954		lun->pending_sense[initindex].ua_pending |= CTL_UA_I_T_NEXUS_LOSS;
11955		mtx_unlock(&lun->lun_lock);
11956	}
11957	mtx_unlock(&softc->ctl_lock);
11958	return (0);
11959}
11960
11961static int
11962ctl_abort_task(union ctl_io *io)
11963{
11964	union ctl_io *xio;
11965	struct ctl_lun *lun;
11966	struct ctl_softc *ctl_softc;
11967#if 0
11968	struct sbuf sb;
11969	char printbuf[128];
11970#endif
11971	int found;
11972	uint32_t targ_lun;
11973
11974	ctl_softc = control_softc;
11975	found = 0;
11976
11977	/*
11978	 * Look up the LUN.
11979	 */
11980	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11981	mtx_lock(&ctl_softc->ctl_lock);
11982	if ((targ_lun < CTL_MAX_LUNS)
11983	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11984		lun = ctl_softc->ctl_luns[targ_lun];
11985	else {
11986		mtx_unlock(&ctl_softc->ctl_lock);
11987		return (1);
11988	}
11989
11990#if 0
11991	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11992	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11993#endif
11994
11995	mtx_lock(&lun->lun_lock);
11996	mtx_unlock(&ctl_softc->ctl_lock);
11997	/*
11998	 * Run through the OOA queue and attempt to find the given I/O.
11999	 * The target port, initiator ID, tag type and tag number have to
12000	 * match the values that we got from the initiator.  If we have an
12001	 * untagged command to abort, simply abort the first untagged command
12002	 * we come to.  We only allow one untagged command at a time of course.
12003	 */
12004#if 0
12005	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12006#endif
12007	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12008	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12009#if 0
12010		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12011
12012		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12013			    lun->lun, xio->scsiio.tag_num,
12014			    xio->scsiio.tag_type,
12015			    (xio->io_hdr.blocked_links.tqe_prev
12016			    == NULL) ? "" : " BLOCKED",
12017			    (xio->io_hdr.flags &
12018			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12019			    (xio->io_hdr.flags &
12020			    CTL_FLAG_ABORT) ? " ABORT" : "",
12021			    (xio->io_hdr.flags &
12022			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12023		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12024		sbuf_finish(&sb);
12025		printf("%s\n", sbuf_data(&sb));
12026#endif
12027
12028		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12029		 && (xio->io_hdr.nexus.initid.id ==
12030		     io->io_hdr.nexus.initid.id)) {
12031			/*
12032			 * If the abort says that the task is untagged, the
12033			 * task in the queue must be untagged.  Otherwise,
12034			 * we just check to see whether the tag numbers
12035			 * match.  This is because the QLogic firmware
12036			 * doesn't pass back the tag type in an abort
12037			 * request.
12038			 */
12039#if 0
12040			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12041			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12042			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12043#endif
12044			/*
12045			 * XXX KDM we've got problems with FC, because it
12046			 * doesn't send down a tag type with aborts.  So we
12047			 * can only really go by the tag number...
12048			 * This may cause problems with parallel SCSI.
12049			 * Need to figure that out!!
12050			 */
12051			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12052				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12053				found = 1;
12054				if ((io->io_hdr.flags &
12055				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12056				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12057					union ctl_ha_msg msg_info;
12058
12059					io->io_hdr.flags |=
12060					                CTL_FLAG_SENT_2OTHER_SC;
12061					msg_info.hdr.nexus = io->io_hdr.nexus;
12062					msg_info.task.task_action =
12063						CTL_TASK_ABORT_TASK;
12064					msg_info.task.tag_num =
12065						io->taskio.tag_num;
12066					msg_info.task.tag_type =
12067						io->taskio.tag_type;
12068					msg_info.hdr.msg_type =
12069						CTL_MSG_MANAGE_TASKS;
12070					msg_info.hdr.original_sc = NULL;
12071					msg_info.hdr.serializing_sc = NULL;
12072#if 0
12073					printf("Sent Abort to other side\n");
12074#endif
12075					if (CTL_HA_STATUS_SUCCESS !=
12076					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12077		    				(void *)&msg_info,
12078						sizeof(msg_info), 0)) {
12079					}
12080				}
12081#if 0
12082				printf("ctl_abort_task: found I/O to abort\n");
12083#endif
12084				break;
12085			}
12086		}
12087	}
12088	mtx_unlock(&lun->lun_lock);
12089
12090	if (found == 0) {
12091		/*
12092		 * This isn't really an error.  It's entirely possible for
12093		 * the abort and command completion to cross on the wire.
12094		 * This is more of an informative/diagnostic error.
12095		 */
12096#if 0
12097		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12098		       "%d:%d:%d:%d tag %d type %d\n",
12099		       io->io_hdr.nexus.initid.id,
12100		       io->io_hdr.nexus.targ_port,
12101		       io->io_hdr.nexus.targ_target.id,
12102		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12103		       io->taskio.tag_type);
12104#endif
12105	}
12106	return (0);
12107}
12108
12109static void
12110ctl_run_task(union ctl_io *io)
12111{
12112	struct ctl_softc *ctl_softc = control_softc;
12113	int retval = 1;
12114	const char *task_desc;
12115
12116	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12117
12118	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12119	    ("ctl_run_task: Unextected io_type %d\n",
12120	     io->io_hdr.io_type));
12121
12122	task_desc = ctl_scsi_task_string(&io->taskio);
12123	if (task_desc != NULL) {
12124#ifdef NEEDTOPORT
12125		csevent_log(CSC_CTL | CSC_SHELF_SW |
12126			    CTL_TASK_REPORT,
12127			    csevent_LogType_Trace,
12128			    csevent_Severity_Information,
12129			    csevent_AlertLevel_Green,
12130			    csevent_FRU_Firmware,
12131			    csevent_FRU_Unknown,
12132			    "CTL: received task: %s",task_desc);
12133#endif
12134	} else {
12135#ifdef NEEDTOPORT
12136		csevent_log(CSC_CTL | CSC_SHELF_SW |
12137			    CTL_TASK_REPORT,
12138			    csevent_LogType_Trace,
12139			    csevent_Severity_Information,
12140			    csevent_AlertLevel_Green,
12141			    csevent_FRU_Firmware,
12142			    csevent_FRU_Unknown,
12143			    "CTL: received unknown task "
12144			    "type: %d (%#x)",
12145			    io->taskio.task_action,
12146			    io->taskio.task_action);
12147#endif
12148	}
12149	switch (io->taskio.task_action) {
12150	case CTL_TASK_ABORT_TASK:
12151		retval = ctl_abort_task(io);
12152		break;
12153	case CTL_TASK_ABORT_TASK_SET:
12154	case CTL_TASK_CLEAR_TASK_SET:
12155		retval = ctl_abort_task_set(io);
12156		break;
12157	case CTL_TASK_CLEAR_ACA:
12158		break;
12159	case CTL_TASK_I_T_NEXUS_RESET:
12160		retval = ctl_i_t_nexus_reset(io);
12161		break;
12162	case CTL_TASK_LUN_RESET: {
12163		struct ctl_lun *lun;
12164		uint32_t targ_lun;
12165
12166		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12167		mtx_lock(&ctl_softc->ctl_lock);
12168		if ((targ_lun < CTL_MAX_LUNS)
12169		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12170			lun = ctl_softc->ctl_luns[targ_lun];
12171		else {
12172			mtx_unlock(&ctl_softc->ctl_lock);
12173			retval = 1;
12174			break;
12175		}
12176
12177		if (!(io->io_hdr.flags &
12178		    CTL_FLAG_FROM_OTHER_SC)) {
12179			union ctl_ha_msg msg_info;
12180
12181			io->io_hdr.flags |=
12182				CTL_FLAG_SENT_2OTHER_SC;
12183			msg_info.hdr.msg_type =
12184				CTL_MSG_MANAGE_TASKS;
12185			msg_info.hdr.nexus = io->io_hdr.nexus;
12186			msg_info.task.task_action =
12187				CTL_TASK_LUN_RESET;
12188			msg_info.hdr.original_sc = NULL;
12189			msg_info.hdr.serializing_sc = NULL;
12190			if (CTL_HA_STATUS_SUCCESS !=
12191			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12192			    (void *)&msg_info,
12193			    sizeof(msg_info), 0)) {
12194			}
12195		}
12196
12197		retval = ctl_lun_reset(lun, io,
12198				       CTL_UA_LUN_RESET);
12199		mtx_unlock(&ctl_softc->ctl_lock);
12200		break;
12201	}
12202	case CTL_TASK_TARGET_RESET:
12203		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12204		break;
12205	case CTL_TASK_BUS_RESET:
12206		retval = ctl_bus_reset(ctl_softc, io);
12207		break;
12208	case CTL_TASK_PORT_LOGIN:
12209		break;
12210	case CTL_TASK_PORT_LOGOUT:
12211		break;
12212	default:
12213		printf("ctl_run_task: got unknown task management event %d\n",
12214		       io->taskio.task_action);
12215		break;
12216	}
12217	if (retval == 0)
12218		io->io_hdr.status = CTL_SUCCESS;
12219	else
12220		io->io_hdr.status = CTL_ERROR;
12221	ctl_done(io);
12222}
12223
12224/*
12225 * For HA operation.  Handle commands that come in from the other
12226 * controller.
12227 */
12228static void
12229ctl_handle_isc(union ctl_io *io)
12230{
12231	int free_io;
12232	struct ctl_lun *lun;
12233	struct ctl_softc *ctl_softc;
12234	uint32_t targ_lun;
12235
12236	ctl_softc = control_softc;
12237
12238	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12239	lun = ctl_softc->ctl_luns[targ_lun];
12240
12241	switch (io->io_hdr.msg_type) {
12242	case CTL_MSG_SERIALIZE:
12243		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12244		break;
12245	case CTL_MSG_R2R: {
12246		const struct ctl_cmd_entry *entry;
12247
12248		/*
12249		 * This is only used in SER_ONLY mode.
12250		 */
12251		free_io = 0;
12252		entry = ctl_get_cmd_entry(&io->scsiio);
12253		mtx_lock(&lun->lun_lock);
12254		if (ctl_scsiio_lun_check(ctl_softc, lun,
12255		    entry, (struct ctl_scsiio *)io) != 0) {
12256			mtx_unlock(&lun->lun_lock);
12257			ctl_done(io);
12258			break;
12259		}
12260		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12261		mtx_unlock(&lun->lun_lock);
12262		ctl_enqueue_rtr(io);
12263		break;
12264	}
12265	case CTL_MSG_FINISH_IO:
12266		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12267			free_io = 0;
12268			ctl_done(io);
12269		} else {
12270			free_io = 1;
12271			mtx_lock(&lun->lun_lock);
12272			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12273				     ooa_links);
12274			ctl_check_blocked(lun);
12275			mtx_unlock(&lun->lun_lock);
12276		}
12277		break;
12278	case CTL_MSG_PERS_ACTION:
12279		ctl_hndl_per_res_out_on_other_sc(
12280			(union ctl_ha_msg *)&io->presio.pr_msg);
12281		free_io = 1;
12282		break;
12283	case CTL_MSG_BAD_JUJU:
12284		free_io = 0;
12285		ctl_done(io);
12286		break;
12287	case CTL_MSG_DATAMOVE:
12288		/* Only used in XFER mode */
12289		free_io = 0;
12290		ctl_datamove_remote(io);
12291		break;
12292	case CTL_MSG_DATAMOVE_DONE:
12293		/* Only used in XFER mode */
12294		free_io = 0;
12295		io->scsiio.be_move_done(io);
12296		break;
12297	default:
12298		free_io = 1;
12299		printf("%s: Invalid message type %d\n",
12300		       __func__, io->io_hdr.msg_type);
12301		break;
12302	}
12303	if (free_io)
12304		ctl_free_io(io);
12305
12306}
12307
12308
12309/*
12310 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12311 * there is no match.
12312 */
12313static ctl_lun_error_pattern
12314ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12315{
12316	const struct ctl_cmd_entry *entry;
12317	ctl_lun_error_pattern filtered_pattern, pattern;
12318
12319	pattern = desc->error_pattern;
12320
12321	/*
12322	 * XXX KDM we need more data passed into this function to match a
12323	 * custom pattern, and we actually need to implement custom pattern
12324	 * matching.
12325	 */
12326	if (pattern & CTL_LUN_PAT_CMD)
12327		return (CTL_LUN_PAT_CMD);
12328
12329	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12330		return (CTL_LUN_PAT_ANY);
12331
12332	entry = ctl_get_cmd_entry(ctsio);
12333
12334	filtered_pattern = entry->pattern & pattern;
12335
12336	/*
12337	 * If the user requested specific flags in the pattern (e.g.
12338	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12339	 * flags.
12340	 *
12341	 * If the user did not specify any flags, it doesn't matter whether
12342	 * or not the command supports the flags.
12343	 */
12344	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12345	     (pattern & ~CTL_LUN_PAT_MASK))
12346		return (CTL_LUN_PAT_NONE);
12347
12348	/*
12349	 * If the user asked for a range check, see if the requested LBA
12350	 * range overlaps with this command's LBA range.
12351	 */
12352	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12353		uint64_t lba1;
12354		uint32_t len1;
12355		ctl_action action;
12356		int retval;
12357
12358		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12359		if (retval != 0)
12360			return (CTL_LUN_PAT_NONE);
12361
12362		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12363					      desc->lba_range.len);
12364		/*
12365		 * A "pass" means that the LBA ranges don't overlap, so
12366		 * this doesn't match the user's range criteria.
12367		 */
12368		if (action == CTL_ACTION_PASS)
12369			return (CTL_LUN_PAT_NONE);
12370	}
12371
12372	return (filtered_pattern);
12373}
12374
12375static void
12376ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12377{
12378	struct ctl_error_desc *desc, *desc2;
12379
12380	mtx_assert(&lun->lun_lock, MA_OWNED);
12381
12382	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12383		ctl_lun_error_pattern pattern;
12384		/*
12385		 * Check to see whether this particular command matches
12386		 * the pattern in the descriptor.
12387		 */
12388		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12389		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12390			continue;
12391
12392		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12393		case CTL_LUN_INJ_ABORTED:
12394			ctl_set_aborted(&io->scsiio);
12395			break;
12396		case CTL_LUN_INJ_MEDIUM_ERR:
12397			ctl_set_medium_error(&io->scsiio);
12398			break;
12399		case CTL_LUN_INJ_UA:
12400			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12401			 * OCCURRED */
12402			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12403			break;
12404		case CTL_LUN_INJ_CUSTOM:
12405			/*
12406			 * We're assuming the user knows what he is doing.
12407			 * Just copy the sense information without doing
12408			 * checks.
12409			 */
12410			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12411			      ctl_min(sizeof(desc->custom_sense),
12412				      sizeof(io->scsiio.sense_data)));
12413			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12414			io->scsiio.sense_len = SSD_FULL_SIZE;
12415			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12416			break;
12417		case CTL_LUN_INJ_NONE:
12418		default:
12419			/*
12420			 * If this is an error injection type we don't know
12421			 * about, clear the continuous flag (if it is set)
12422			 * so it will get deleted below.
12423			 */
12424			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12425			break;
12426		}
12427		/*
12428		 * By default, each error injection action is a one-shot
12429		 */
12430		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12431			continue;
12432
12433		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12434
12435		free(desc, M_CTL);
12436	}
12437}
12438
12439#ifdef CTL_IO_DELAY
12440static void
12441ctl_datamove_timer_wakeup(void *arg)
12442{
12443	union ctl_io *io;
12444
12445	io = (union ctl_io *)arg;
12446
12447	ctl_datamove(io);
12448}
12449#endif /* CTL_IO_DELAY */
12450
12451void
12452ctl_datamove(union ctl_io *io)
12453{
12454	void (*fe_datamove)(union ctl_io *io);
12455
12456	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12457
12458	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12459
12460#ifdef CTL_TIME_IO
12461	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12462		char str[256];
12463		char path_str[64];
12464		struct sbuf sb;
12465
12466		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12467		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12468
12469		sbuf_cat(&sb, path_str);
12470		switch (io->io_hdr.io_type) {
12471		case CTL_IO_SCSI:
12472			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12473			sbuf_printf(&sb, "\n");
12474			sbuf_cat(&sb, path_str);
12475			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12476				    io->scsiio.tag_num, io->scsiio.tag_type);
12477			break;
12478		case CTL_IO_TASK:
12479			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12480				    "Tag Type: %d\n", io->taskio.task_action,
12481				    io->taskio.tag_num, io->taskio.tag_type);
12482			break;
12483		default:
12484			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12485			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12486			break;
12487		}
12488		sbuf_cat(&sb, path_str);
12489		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12490			    (intmax_t)time_uptime - io->io_hdr.start_time);
12491		sbuf_finish(&sb);
12492		printf("%s", sbuf_data(&sb));
12493	}
12494#endif /* CTL_TIME_IO */
12495
12496#ifdef CTL_IO_DELAY
12497	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12498		struct ctl_lun *lun;
12499
12500		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12501
12502		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12503	} else {
12504		struct ctl_lun *lun;
12505
12506		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12507		if ((lun != NULL)
12508		 && (lun->delay_info.datamove_delay > 0)) {
12509			struct callout *callout;
12510
12511			callout = (struct callout *)&io->io_hdr.timer_bytes;
12512			callout_init(callout, /*mpsafe*/ 1);
12513			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12514			callout_reset(callout,
12515				      lun->delay_info.datamove_delay * hz,
12516				      ctl_datamove_timer_wakeup, io);
12517			if (lun->delay_info.datamove_type ==
12518			    CTL_DELAY_TYPE_ONESHOT)
12519				lun->delay_info.datamove_delay = 0;
12520			return;
12521		}
12522	}
12523#endif
12524
12525	/*
12526	 * This command has been aborted.  Set the port status, so we fail
12527	 * the data move.
12528	 */
12529	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12530		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12531		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12532		       io->io_hdr.nexus.targ_port,
12533		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12534		       io->io_hdr.nexus.targ_lun);
12535		io->io_hdr.port_status = 31337;
12536		/*
12537		 * Note that the backend, in this case, will get the
12538		 * callback in its context.  In other cases it may get
12539		 * called in the frontend's interrupt thread context.
12540		 */
12541		io->scsiio.be_move_done(io);
12542		return;
12543	}
12544
12545	/*
12546	 * If we're in XFER mode and this I/O is from the other shelf
12547	 * controller, we need to send the DMA to the other side to
12548	 * actually transfer the data to/from the host.  In serialize only
12549	 * mode the transfer happens below CTL and ctl_datamove() is only
12550	 * called on the machine that originally received the I/O.
12551	 */
12552	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12553	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12554		union ctl_ha_msg msg;
12555		uint32_t sg_entries_sent;
12556		int do_sg_copy;
12557		int i;
12558
12559		memset(&msg, 0, sizeof(msg));
12560		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12561		msg.hdr.original_sc = io->io_hdr.original_sc;
12562		msg.hdr.serializing_sc = io;
12563		msg.hdr.nexus = io->io_hdr.nexus;
12564		msg.dt.flags = io->io_hdr.flags;
12565		/*
12566		 * We convert everything into a S/G list here.  We can't
12567		 * pass by reference, only by value between controllers.
12568		 * So we can't pass a pointer to the S/G list, only as many
12569		 * S/G entries as we can fit in here.  If it's possible for
12570		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12571		 * then we need to break this up into multiple transfers.
12572		 */
12573		if (io->scsiio.kern_sg_entries == 0) {
12574			msg.dt.kern_sg_entries = 1;
12575			/*
12576			 * If this is in cached memory, flush the cache
12577			 * before we send the DMA request to the other
12578			 * controller.  We want to do this in either the
12579			 * read or the write case.  The read case is
12580			 * straightforward.  In the write case, we want to
12581			 * make sure nothing is in the local cache that
12582			 * could overwrite the DMAed data.
12583			 */
12584			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12585				/*
12586				 * XXX KDM use bus_dmamap_sync() here.
12587				 */
12588			}
12589
12590			/*
12591			 * Convert to a physical address if this is a
12592			 * virtual address.
12593			 */
12594			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12595				msg.dt.sg_list[0].addr =
12596					io->scsiio.kern_data_ptr;
12597			} else {
12598				/*
12599				 * XXX KDM use busdma here!
12600				 */
12601#if 0
12602				msg.dt.sg_list[0].addr = (void *)
12603					vtophys(io->scsiio.kern_data_ptr);
12604#endif
12605			}
12606
12607			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12608			do_sg_copy = 0;
12609		} else {
12610			struct ctl_sg_entry *sgl;
12611
12612			do_sg_copy = 1;
12613			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12614			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12615			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12616				/*
12617				 * XXX KDM use bus_dmamap_sync() here.
12618				 */
12619			}
12620		}
12621
12622		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12623		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12624		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12625		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12626		msg.dt.sg_sequence = 0;
12627
12628		/*
12629		 * Loop until we've sent all of the S/G entries.  On the
12630		 * other end, we'll recompose these S/G entries into one
12631		 * contiguous list before passing it to the
12632		 */
12633		for (sg_entries_sent = 0; sg_entries_sent <
12634		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12635			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12636				sizeof(msg.dt.sg_list[0])),
12637				msg.dt.kern_sg_entries - sg_entries_sent);
12638
12639			if (do_sg_copy != 0) {
12640				struct ctl_sg_entry *sgl;
12641				int j;
12642
12643				sgl = (struct ctl_sg_entry *)
12644					io->scsiio.kern_data_ptr;
12645				/*
12646				 * If this is in cached memory, flush the cache
12647				 * before we send the DMA request to the other
12648				 * controller.  We want to do this in either
12649				 * the * read or the write case.  The read
12650				 * case is straightforward.  In the write
12651				 * case, we want to make sure nothing is
12652				 * in the local cache that could overwrite
12653				 * the DMAed data.
12654				 */
12655
12656				for (i = sg_entries_sent, j = 0;
12657				     i < msg.dt.cur_sg_entries; i++, j++) {
12658					if ((io->io_hdr.flags &
12659					     CTL_FLAG_NO_DATASYNC) == 0) {
12660						/*
12661						 * XXX KDM use bus_dmamap_sync()
12662						 */
12663					}
12664					if ((io->io_hdr.flags &
12665					     CTL_FLAG_BUS_ADDR) == 0) {
12666						/*
12667						 * XXX KDM use busdma.
12668						 */
12669#if 0
12670						msg.dt.sg_list[j].addr =(void *)
12671						       vtophys(sgl[i].addr);
12672#endif
12673					} else {
12674						msg.dt.sg_list[j].addr =
12675							sgl[i].addr;
12676					}
12677					msg.dt.sg_list[j].len = sgl[i].len;
12678				}
12679			}
12680
12681			sg_entries_sent += msg.dt.cur_sg_entries;
12682			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12683				msg.dt.sg_last = 1;
12684			else
12685				msg.dt.sg_last = 0;
12686
12687			/*
12688			 * XXX KDM drop and reacquire the lock here?
12689			 */
12690			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12691			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12692				/*
12693				 * XXX do something here.
12694				 */
12695			}
12696
12697			msg.dt.sent_sg_entries = sg_entries_sent;
12698		}
12699		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12700		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12701			ctl_failover_io(io, /*have_lock*/ 0);
12702
12703	} else {
12704
12705		/*
12706		 * Lookup the fe_datamove() function for this particular
12707		 * front end.
12708		 */
12709		fe_datamove =
12710		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12711
12712		fe_datamove(io);
12713	}
12714}
12715
12716static void
12717ctl_send_datamove_done(union ctl_io *io, int have_lock)
12718{
12719	union ctl_ha_msg msg;
12720	int isc_status;
12721
12722	memset(&msg, 0, sizeof(msg));
12723
12724	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12725	msg.hdr.original_sc = io;
12726	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12727	msg.hdr.nexus = io->io_hdr.nexus;
12728	msg.hdr.status = io->io_hdr.status;
12729	msg.scsi.tag_num = io->scsiio.tag_num;
12730	msg.scsi.tag_type = io->scsiio.tag_type;
12731	msg.scsi.scsi_status = io->scsiio.scsi_status;
12732	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12733	       sizeof(io->scsiio.sense_data));
12734	msg.scsi.sense_len = io->scsiio.sense_len;
12735	msg.scsi.sense_residual = io->scsiio.sense_residual;
12736	msg.scsi.fetd_status = io->io_hdr.port_status;
12737	msg.scsi.residual = io->scsiio.residual;
12738	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12739
12740	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12741		ctl_failover_io(io, /*have_lock*/ have_lock);
12742		return;
12743	}
12744
12745	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12746	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12747		/* XXX do something if this fails */
12748	}
12749
12750}
12751
12752/*
12753 * The DMA to the remote side is done, now we need to tell the other side
12754 * we're done so it can continue with its data movement.
12755 */
12756static void
12757ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12758{
12759	union ctl_io *io;
12760
12761	io = rq->context;
12762
12763	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12764		printf("%s: ISC DMA write failed with error %d", __func__,
12765		       rq->ret);
12766		ctl_set_internal_failure(&io->scsiio,
12767					 /*sks_valid*/ 1,
12768					 /*retry_count*/ rq->ret);
12769	}
12770
12771	ctl_dt_req_free(rq);
12772
12773	/*
12774	 * In this case, we had to malloc the memory locally.  Free it.
12775	 */
12776	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12777		int i;
12778		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12779			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12780	}
12781	/*
12782	 * The data is in local and remote memory, so now we need to send
12783	 * status (good or back) back to the other side.
12784	 */
12785	ctl_send_datamove_done(io, /*have_lock*/ 0);
12786}
12787
12788/*
12789 * We've moved the data from the host/controller into local memory.  Now we
12790 * need to push it over to the remote controller's memory.
12791 */
12792static int
12793ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12794{
12795	int retval;
12796
12797	retval = 0;
12798
12799	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12800					  ctl_datamove_remote_write_cb);
12801
12802	return (retval);
12803}
12804
12805static void
12806ctl_datamove_remote_write(union ctl_io *io)
12807{
12808	int retval;
12809	void (*fe_datamove)(union ctl_io *io);
12810
12811	/*
12812	 * - Get the data from the host/HBA into local memory.
12813	 * - DMA memory from the local controller to the remote controller.
12814	 * - Send status back to the remote controller.
12815	 */
12816
12817	retval = ctl_datamove_remote_sgl_setup(io);
12818	if (retval != 0)
12819		return;
12820
12821	/* Switch the pointer over so the FETD knows what to do */
12822	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12823
12824	/*
12825	 * Use a custom move done callback, since we need to send completion
12826	 * back to the other controller, not to the backend on this side.
12827	 */
12828	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12829
12830	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12831
12832	fe_datamove(io);
12833
12834	return;
12835
12836}
12837
12838static int
12839ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12840{
12841#if 0
12842	char str[256];
12843	char path_str[64];
12844	struct sbuf sb;
12845#endif
12846
12847	/*
12848	 * In this case, we had to malloc the memory locally.  Free it.
12849	 */
12850	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12851		int i;
12852		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12853			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12854	}
12855
12856#if 0
12857	scsi_path_string(io, path_str, sizeof(path_str));
12858	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12859	sbuf_cat(&sb, path_str);
12860	scsi_command_string(&io->scsiio, NULL, &sb);
12861	sbuf_printf(&sb, "\n");
12862	sbuf_cat(&sb, path_str);
12863	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12864		    io->scsiio.tag_num, io->scsiio.tag_type);
12865	sbuf_cat(&sb, path_str);
12866	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12867		    io->io_hdr.flags, io->io_hdr.status);
12868	sbuf_finish(&sb);
12869	printk("%s", sbuf_data(&sb));
12870#endif
12871
12872
12873	/*
12874	 * The read is done, now we need to send status (good or bad) back
12875	 * to the other side.
12876	 */
12877	ctl_send_datamove_done(io, /*have_lock*/ 0);
12878
12879	return (0);
12880}
12881
12882static void
12883ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12884{
12885	union ctl_io *io;
12886	void (*fe_datamove)(union ctl_io *io);
12887
12888	io = rq->context;
12889
12890	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12891		printf("%s: ISC DMA read failed with error %d", __func__,
12892		       rq->ret);
12893		ctl_set_internal_failure(&io->scsiio,
12894					 /*sks_valid*/ 1,
12895					 /*retry_count*/ rq->ret);
12896	}
12897
12898	ctl_dt_req_free(rq);
12899
12900	/* Switch the pointer over so the FETD knows what to do */
12901	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12902
12903	/*
12904	 * Use a custom move done callback, since we need to send completion
12905	 * back to the other controller, not to the backend on this side.
12906	 */
12907	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12908
12909	/* XXX KDM add checks like the ones in ctl_datamove? */
12910
12911	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12912
12913	fe_datamove(io);
12914}
12915
12916static int
12917ctl_datamove_remote_sgl_setup(union ctl_io *io)
12918{
12919	struct ctl_sg_entry *local_sglist, *remote_sglist;
12920	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12921	struct ctl_softc *softc;
12922	int retval;
12923	int i;
12924
12925	retval = 0;
12926	softc = control_softc;
12927
12928	local_sglist = io->io_hdr.local_sglist;
12929	local_dma_sglist = io->io_hdr.local_dma_sglist;
12930	remote_sglist = io->io_hdr.remote_sglist;
12931	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12932
12933	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12934		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12935			local_sglist[i].len = remote_sglist[i].len;
12936
12937			/*
12938			 * XXX Detect the situation where the RS-level I/O
12939			 * redirector on the other side has already read the
12940			 * data off of the AOR RS on this side, and
12941			 * transferred it to remote (mirror) memory on the
12942			 * other side.  Since we already have the data in
12943			 * memory here, we just need to use it.
12944			 *
12945			 * XXX KDM this can probably be removed once we
12946			 * get the cache device code in and take the
12947			 * current AOR implementation out.
12948			 */
12949#ifdef NEEDTOPORT
12950			if ((remote_sglist[i].addr >=
12951			     (void *)vtophys(softc->mirr->addr))
12952			 && (remote_sglist[i].addr <
12953			     ((void *)vtophys(softc->mirr->addr) +
12954			     CacheMirrorOffset))) {
12955				local_sglist[i].addr = remote_sglist[i].addr -
12956					CacheMirrorOffset;
12957				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12958				     CTL_FLAG_DATA_IN)
12959					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12960			} else {
12961				local_sglist[i].addr = remote_sglist[i].addr +
12962					CacheMirrorOffset;
12963			}
12964#endif
12965#if 0
12966			printf("%s: local %p, remote %p, len %d\n",
12967			       __func__, local_sglist[i].addr,
12968			       remote_sglist[i].addr, local_sglist[i].len);
12969#endif
12970		}
12971	} else {
12972		uint32_t len_to_go;
12973
12974		/*
12975		 * In this case, we don't have automatically allocated
12976		 * memory for this I/O on this controller.  This typically
12977		 * happens with internal CTL I/O -- e.g. inquiry, mode
12978		 * sense, etc.  Anything coming from RAIDCore will have
12979		 * a mirror area available.
12980		 */
12981		len_to_go = io->scsiio.kern_data_len;
12982
12983		/*
12984		 * Clear the no datasync flag, we have to use malloced
12985		 * buffers.
12986		 */
12987		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12988
12989		/*
12990		 * The difficult thing here is that the size of the various
12991		 * S/G segments may be different than the size from the
12992		 * remote controller.  That'll make it harder when DMAing
12993		 * the data back to the other side.
12994		 */
12995		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12996		     sizeof(io->io_hdr.remote_sglist[0])) &&
12997		     (len_to_go > 0); i++) {
12998			local_sglist[i].len = ctl_min(len_to_go, 131072);
12999			CTL_SIZE_8B(local_dma_sglist[i].len,
13000				    local_sglist[i].len);
13001			local_sglist[i].addr =
13002				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13003
13004			local_dma_sglist[i].addr = local_sglist[i].addr;
13005
13006			if (local_sglist[i].addr == NULL) {
13007				int j;
13008
13009				printf("malloc failed for %zd bytes!",
13010				       local_dma_sglist[i].len);
13011				for (j = 0; j < i; j++) {
13012					free(local_sglist[j].addr, M_CTL);
13013				}
13014				ctl_set_internal_failure(&io->scsiio,
13015							 /*sks_valid*/ 1,
13016							 /*retry_count*/ 4857);
13017				retval = 1;
13018				goto bailout_error;
13019
13020			}
13021			/* XXX KDM do we need a sync here? */
13022
13023			len_to_go -= local_sglist[i].len;
13024		}
13025		/*
13026		 * Reset the number of S/G entries accordingly.  The
13027		 * original number of S/G entries is available in
13028		 * rem_sg_entries.
13029		 */
13030		io->scsiio.kern_sg_entries = i;
13031
13032#if 0
13033		printf("%s: kern_sg_entries = %d\n", __func__,
13034		       io->scsiio.kern_sg_entries);
13035		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13036			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13037			       local_sglist[i].addr, local_sglist[i].len,
13038			       local_dma_sglist[i].len);
13039#endif
13040	}
13041
13042
13043	return (retval);
13044
13045bailout_error:
13046
13047	ctl_send_datamove_done(io, /*have_lock*/ 0);
13048
13049	return (retval);
13050}
13051
13052static int
13053ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13054			 ctl_ha_dt_cb callback)
13055{
13056	struct ctl_ha_dt_req *rq;
13057	struct ctl_sg_entry *remote_sglist, *local_sglist;
13058	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13059	uint32_t local_used, remote_used, total_used;
13060	int retval;
13061	int i, j;
13062
13063	retval = 0;
13064
13065	rq = ctl_dt_req_alloc();
13066
13067	/*
13068	 * If we failed to allocate the request, and if the DMA didn't fail
13069	 * anyway, set busy status.  This is just a resource allocation
13070	 * failure.
13071	 */
13072	if ((rq == NULL)
13073	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13074		ctl_set_busy(&io->scsiio);
13075
13076	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13077
13078		if (rq != NULL)
13079			ctl_dt_req_free(rq);
13080
13081		/*
13082		 * The data move failed.  We need to return status back
13083		 * to the other controller.  No point in trying to DMA
13084		 * data to the remote controller.
13085		 */
13086
13087		ctl_send_datamove_done(io, /*have_lock*/ 0);
13088
13089		retval = 1;
13090
13091		goto bailout;
13092	}
13093
13094	local_sglist = io->io_hdr.local_sglist;
13095	local_dma_sglist = io->io_hdr.local_dma_sglist;
13096	remote_sglist = io->io_hdr.remote_sglist;
13097	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13098	local_used = 0;
13099	remote_used = 0;
13100	total_used = 0;
13101
13102	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13103		rq->ret = CTL_HA_STATUS_SUCCESS;
13104		rq->context = io;
13105		callback(rq);
13106		goto bailout;
13107	}
13108
13109	/*
13110	 * Pull/push the data over the wire from/to the other controller.
13111	 * This takes into account the possibility that the local and
13112	 * remote sglists may not be identical in terms of the size of
13113	 * the elements and the number of elements.
13114	 *
13115	 * One fundamental assumption here is that the length allocated for
13116	 * both the local and remote sglists is identical.  Otherwise, we've
13117	 * essentially got a coding error of some sort.
13118	 */
13119	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13120		int isc_ret;
13121		uint32_t cur_len, dma_length;
13122		uint8_t *tmp_ptr;
13123
13124		rq->id = CTL_HA_DATA_CTL;
13125		rq->command = command;
13126		rq->context = io;
13127
13128		/*
13129		 * Both pointers should be aligned.  But it is possible
13130		 * that the allocation length is not.  They should both
13131		 * also have enough slack left over at the end, though,
13132		 * to round up to the next 8 byte boundary.
13133		 */
13134		cur_len = ctl_min(local_sglist[i].len - local_used,
13135				  remote_sglist[j].len - remote_used);
13136
13137		/*
13138		 * In this case, we have a size issue and need to decrease
13139		 * the size, except in the case where we actually have less
13140		 * than 8 bytes left.  In that case, we need to increase
13141		 * the DMA length to get the last bit.
13142		 */
13143		if ((cur_len & 0x7) != 0) {
13144			if (cur_len > 0x7) {
13145				cur_len = cur_len - (cur_len & 0x7);
13146				dma_length = cur_len;
13147			} else {
13148				CTL_SIZE_8B(dma_length, cur_len);
13149			}
13150
13151		} else
13152			dma_length = cur_len;
13153
13154		/*
13155		 * If we had to allocate memory for this I/O, instead of using
13156		 * the non-cached mirror memory, we'll need to flush the cache
13157		 * before trying to DMA to the other controller.
13158		 *
13159		 * We could end up doing this multiple times for the same
13160		 * segment if we have a larger local segment than remote
13161		 * segment.  That shouldn't be an issue.
13162		 */
13163		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13164			/*
13165			 * XXX KDM use bus_dmamap_sync() here.
13166			 */
13167		}
13168
13169		rq->size = dma_length;
13170
13171		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13172		tmp_ptr += local_used;
13173
13174		/* Use physical addresses when talking to ISC hardware */
13175		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13176			/* XXX KDM use busdma */
13177#if 0
13178			rq->local = vtophys(tmp_ptr);
13179#endif
13180		} else
13181			rq->local = tmp_ptr;
13182
13183		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13184		tmp_ptr += remote_used;
13185		rq->remote = tmp_ptr;
13186
13187		rq->callback = NULL;
13188
13189		local_used += cur_len;
13190		if (local_used >= local_sglist[i].len) {
13191			i++;
13192			local_used = 0;
13193		}
13194
13195		remote_used += cur_len;
13196		if (remote_used >= remote_sglist[j].len) {
13197			j++;
13198			remote_used = 0;
13199		}
13200		total_used += cur_len;
13201
13202		if (total_used >= io->scsiio.kern_data_len)
13203			rq->callback = callback;
13204
13205		if ((rq->size & 0x7) != 0) {
13206			printf("%s: warning: size %d is not on 8b boundary\n",
13207			       __func__, rq->size);
13208		}
13209		if (((uintptr_t)rq->local & 0x7) != 0) {
13210			printf("%s: warning: local %p not on 8b boundary\n",
13211			       __func__, rq->local);
13212		}
13213		if (((uintptr_t)rq->remote & 0x7) != 0) {
13214			printf("%s: warning: remote %p not on 8b boundary\n",
13215			       __func__, rq->local);
13216		}
13217#if 0
13218		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13219		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13220		       rq->local, rq->remote, rq->size);
13221#endif
13222
13223		isc_ret = ctl_dt_single(rq);
13224		if (isc_ret == CTL_HA_STATUS_WAIT)
13225			continue;
13226
13227		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13228			rq->ret = CTL_HA_STATUS_SUCCESS;
13229		} else {
13230			rq->ret = isc_ret;
13231		}
13232		callback(rq);
13233		goto bailout;
13234	}
13235
13236bailout:
13237	return (retval);
13238
13239}
13240
13241static void
13242ctl_datamove_remote_read(union ctl_io *io)
13243{
13244	int retval;
13245	int i;
13246
13247	/*
13248	 * This will send an error to the other controller in the case of a
13249	 * failure.
13250	 */
13251	retval = ctl_datamove_remote_sgl_setup(io);
13252	if (retval != 0)
13253		return;
13254
13255	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13256					  ctl_datamove_remote_read_cb);
13257	if ((retval != 0)
13258	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13259		/*
13260		 * Make sure we free memory if there was an error..  The
13261		 * ctl_datamove_remote_xfer() function will send the
13262		 * datamove done message, or call the callback with an
13263		 * error if there is a problem.
13264		 */
13265		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13266			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13267	}
13268
13269	return;
13270}
13271
13272/*
13273 * Process a datamove request from the other controller.  This is used for
13274 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13275 * first.  Once that is complete, the data gets DMAed into the remote
13276 * controller's memory.  For reads, we DMA from the remote controller's
13277 * memory into our memory first, and then move it out to the FETD.
13278 */
13279static void
13280ctl_datamove_remote(union ctl_io *io)
13281{
13282	struct ctl_softc *softc;
13283
13284	softc = control_softc;
13285
13286	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13287
13288	/*
13289	 * Note that we look for an aborted I/O here, but don't do some of
13290	 * the other checks that ctl_datamove() normally does.
13291	 * We don't need to run the datamove delay code, since that should
13292	 * have been done if need be on the other controller.
13293	 */
13294	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13295		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13296		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13297		       io->io_hdr.nexus.targ_port,
13298		       io->io_hdr.nexus.targ_target.id,
13299		       io->io_hdr.nexus.targ_lun);
13300		io->io_hdr.port_status = 31338;
13301		ctl_send_datamove_done(io, /*have_lock*/ 0);
13302		return;
13303	}
13304
13305	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13306		ctl_datamove_remote_write(io);
13307	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13308		ctl_datamove_remote_read(io);
13309	} else {
13310		union ctl_ha_msg msg;
13311		struct scsi_sense_data *sense;
13312		uint8_t sks[3];
13313		int retry_count;
13314
13315		memset(&msg, 0, sizeof(msg));
13316
13317		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13318		msg.hdr.status = CTL_SCSI_ERROR;
13319		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13320
13321		retry_count = 4243;
13322
13323		sense = &msg.scsi.sense_data;
13324		sks[0] = SSD_SCS_VALID;
13325		sks[1] = (retry_count >> 8) & 0xff;
13326		sks[2] = retry_count & 0xff;
13327
13328		/* "Internal target failure" */
13329		scsi_set_sense_data(sense,
13330				    /*sense_format*/ SSD_TYPE_NONE,
13331				    /*current_error*/ 1,
13332				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13333				    /*asc*/ 0x44,
13334				    /*ascq*/ 0x00,
13335				    /*type*/ SSD_ELEM_SKS,
13336				    /*size*/ sizeof(sks),
13337				    /*data*/ sks,
13338				    SSD_ELEM_NONE);
13339
13340		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13341		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13342			ctl_failover_io(io, /*have_lock*/ 1);
13343			return;
13344		}
13345
13346		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13347		    CTL_HA_STATUS_SUCCESS) {
13348			/* XXX KDM what to do if this fails? */
13349		}
13350		return;
13351	}
13352
13353}
13354
13355static int
13356ctl_process_done(union ctl_io *io)
13357{
13358	struct ctl_lun *lun;
13359	struct ctl_softc *ctl_softc;
13360	void (*fe_done)(union ctl_io *io);
13361	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13362
13363	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13364
13365	fe_done =
13366	    control_softc->ctl_ports[targ_port]->fe_done;
13367
13368#ifdef CTL_TIME_IO
13369	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13370		char str[256];
13371		char path_str[64];
13372		struct sbuf sb;
13373
13374		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13375		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13376
13377		sbuf_cat(&sb, path_str);
13378		switch (io->io_hdr.io_type) {
13379		case CTL_IO_SCSI:
13380			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13381			sbuf_printf(&sb, "\n");
13382			sbuf_cat(&sb, path_str);
13383			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13384				    io->scsiio.tag_num, io->scsiio.tag_type);
13385			break;
13386		case CTL_IO_TASK:
13387			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13388				    "Tag Type: %d\n", io->taskio.task_action,
13389				    io->taskio.tag_num, io->taskio.tag_type);
13390			break;
13391		default:
13392			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13393			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13394			break;
13395		}
13396		sbuf_cat(&sb, path_str);
13397		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13398			    (intmax_t)time_uptime - io->io_hdr.start_time);
13399		sbuf_finish(&sb);
13400		printf("%s", sbuf_data(&sb));
13401	}
13402#endif /* CTL_TIME_IO */
13403
13404	switch (io->io_hdr.io_type) {
13405	case CTL_IO_SCSI:
13406		break;
13407	case CTL_IO_TASK:
13408		if (bootverbose || verbose > 0)
13409			ctl_io_error_print(io, NULL);
13410		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13411			ctl_free_io(io);
13412		else
13413			fe_done(io);
13414		return (CTL_RETVAL_COMPLETE);
13415		break;
13416	default:
13417		printf("ctl_process_done: invalid io type %d\n",
13418		       io->io_hdr.io_type);
13419		panic("ctl_process_done: invalid io type %d\n",
13420		      io->io_hdr.io_type);
13421		break; /* NOTREACHED */
13422	}
13423
13424	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13425	if (lun == NULL) {
13426		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13427				 io->io_hdr.nexus.targ_mapped_lun));
13428		fe_done(io);
13429		goto bailout;
13430	}
13431	ctl_softc = lun->ctl_softc;
13432
13433	mtx_lock(&lun->lun_lock);
13434
13435	/*
13436	 * Check to see if we have any errors to inject here.  We only
13437	 * inject errors for commands that don't already have errors set.
13438	 */
13439	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13440	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13441		ctl_inject_error(lun, io);
13442
13443	/*
13444	 * XXX KDM how do we treat commands that aren't completed
13445	 * successfully?
13446	 *
13447	 * XXX KDM should we also track I/O latency?
13448	 */
13449	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13450	    io->io_hdr.io_type == CTL_IO_SCSI) {
13451#ifdef CTL_TIME_IO
13452		struct bintime cur_bt;
13453#endif
13454		int type;
13455
13456		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13457		    CTL_FLAG_DATA_IN)
13458			type = CTL_STATS_READ;
13459		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13460		    CTL_FLAG_DATA_OUT)
13461			type = CTL_STATS_WRITE;
13462		else
13463			type = CTL_STATS_NO_IO;
13464
13465		lun->stats.ports[targ_port].bytes[type] +=
13466		    io->scsiio.kern_total_len;
13467		lun->stats.ports[targ_port].operations[type]++;
13468#ifdef CTL_TIME_IO
13469		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13470		   &io->io_hdr.dma_bt);
13471		lun->stats.ports[targ_port].num_dmas[type] +=
13472		    io->io_hdr.num_dmas;
13473		getbintime(&cur_bt);
13474		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13475		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13476#endif
13477	}
13478
13479	/*
13480	 * Remove this from the OOA queue.
13481	 */
13482	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13483
13484	/*
13485	 * Run through the blocked queue on this LUN and see if anything
13486	 * has become unblocked, now that this transaction is done.
13487	 */
13488	ctl_check_blocked(lun);
13489
13490	/*
13491	 * If the LUN has been invalidated, free it if there is nothing
13492	 * left on its OOA queue.
13493	 */
13494	if ((lun->flags & CTL_LUN_INVALID)
13495	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13496		mtx_unlock(&lun->lun_lock);
13497		mtx_lock(&ctl_softc->ctl_lock);
13498		ctl_free_lun(lun);
13499		mtx_unlock(&ctl_softc->ctl_lock);
13500	} else
13501		mtx_unlock(&lun->lun_lock);
13502
13503	/*
13504	 * If this command has been aborted, make sure we set the status
13505	 * properly.  The FETD is responsible for freeing the I/O and doing
13506	 * whatever it needs to do to clean up its state.
13507	 */
13508	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13509		ctl_set_task_aborted(&io->scsiio);
13510
13511	/*
13512	 * We print out status for every task management command.  For SCSI
13513	 * commands, we filter out any unit attention errors; they happen
13514	 * on every boot, and would clutter up the log.  Note:  task
13515	 * management commands aren't printed here, they are printed above,
13516	 * since they should never even make it down here.
13517	 */
13518	switch (io->io_hdr.io_type) {
13519	case CTL_IO_SCSI: {
13520		int error_code, sense_key, asc, ascq;
13521
13522		sense_key = 0;
13523
13524		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13525		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13526			/*
13527			 * Since this is just for printing, no need to
13528			 * show errors here.
13529			 */
13530			scsi_extract_sense_len(&io->scsiio.sense_data,
13531					       io->scsiio.sense_len,
13532					       &error_code,
13533					       &sense_key,
13534					       &asc,
13535					       &ascq,
13536					       /*show_errors*/ 0);
13537		}
13538
13539		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13540		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13541		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13542		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13543
13544			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13545				ctl_softc->skipped_prints++;
13546			} else {
13547				uint32_t skipped_prints;
13548
13549				skipped_prints = ctl_softc->skipped_prints;
13550
13551				ctl_softc->skipped_prints = 0;
13552				ctl_softc->last_print_jiffies = time_uptime;
13553
13554				if (skipped_prints > 0) {
13555#ifdef NEEDTOPORT
13556					csevent_log(CSC_CTL | CSC_SHELF_SW |
13557					    CTL_ERROR_REPORT,
13558					    csevent_LogType_Trace,
13559					    csevent_Severity_Information,
13560					    csevent_AlertLevel_Green,
13561					    csevent_FRU_Firmware,
13562					    csevent_FRU_Unknown,
13563					    "High CTL error volume, %d prints "
13564					    "skipped", skipped_prints);
13565#endif
13566				}
13567				if (bootverbose || verbose > 0)
13568					ctl_io_error_print(io, NULL);
13569			}
13570		}
13571		break;
13572	}
13573	case CTL_IO_TASK:
13574		if (bootverbose || verbose > 0)
13575			ctl_io_error_print(io, NULL);
13576		break;
13577	default:
13578		break;
13579	}
13580
13581	/*
13582	 * Tell the FETD or the other shelf controller we're done with this
13583	 * command.  Note that only SCSI commands get to this point.  Task
13584	 * management commands are completed above.
13585	 *
13586	 * We only send status to the other controller if we're in XFER
13587	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13588	 * received the I/O (from CTL's perspective), and so the status is
13589	 * generated there.
13590	 *
13591	 * XXX KDM if we hold the lock here, we could cause a deadlock
13592	 * if the frontend comes back in in this context to queue
13593	 * something.
13594	 */
13595	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13596	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13597		union ctl_ha_msg msg;
13598
13599		memset(&msg, 0, sizeof(msg));
13600		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13601		msg.hdr.original_sc = io->io_hdr.original_sc;
13602		msg.hdr.nexus = io->io_hdr.nexus;
13603		msg.hdr.status = io->io_hdr.status;
13604		msg.scsi.scsi_status = io->scsiio.scsi_status;
13605		msg.scsi.tag_num = io->scsiio.tag_num;
13606		msg.scsi.tag_type = io->scsiio.tag_type;
13607		msg.scsi.sense_len = io->scsiio.sense_len;
13608		msg.scsi.sense_residual = io->scsiio.sense_residual;
13609		msg.scsi.residual = io->scsiio.residual;
13610		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13611		       sizeof(io->scsiio.sense_data));
13612		/*
13613		 * We copy this whether or not this is an I/O-related
13614		 * command.  Otherwise, we'd have to go and check to see
13615		 * whether it's a read/write command, and it really isn't
13616		 * worth it.
13617		 */
13618		memcpy(&msg.scsi.lbalen,
13619		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13620		       sizeof(msg.scsi.lbalen));
13621
13622		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13623				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13624			/* XXX do something here */
13625		}
13626
13627		ctl_free_io(io);
13628	} else
13629		fe_done(io);
13630
13631bailout:
13632
13633	return (CTL_RETVAL_COMPLETE);
13634}
13635
13636/*
13637 * Front end should call this if it doesn't do autosense.  When the request
13638 * sense comes back in from the initiator, we'll dequeue this and send it.
13639 */
13640int
13641ctl_queue_sense(union ctl_io *io)
13642{
13643	struct ctl_lun *lun;
13644	struct ctl_softc *ctl_softc;
13645	uint32_t initidx, targ_lun;
13646
13647	ctl_softc = control_softc;
13648
13649	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13650
13651	/*
13652	 * LUN lookup will likely move to the ctl_work_thread() once we
13653	 * have our new queueing infrastructure (that doesn't put things on
13654	 * a per-LUN queue initially).  That is so that we can handle
13655	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13656	 * can't deal with that right now.
13657	 */
13658	mtx_lock(&ctl_softc->ctl_lock);
13659
13660	/*
13661	 * If we don't have a LUN for this, just toss the sense
13662	 * information.
13663	 */
13664	targ_lun = io->io_hdr.nexus.targ_lun;
13665	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13666	if ((targ_lun < CTL_MAX_LUNS)
13667	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13668		lun = ctl_softc->ctl_luns[targ_lun];
13669	else
13670		goto bailout;
13671
13672	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13673
13674	mtx_lock(&lun->lun_lock);
13675	/*
13676	 * Already have CA set for this LUN...toss the sense information.
13677	 */
13678	if (ctl_is_set(lun->have_ca, initidx)) {
13679		mtx_unlock(&lun->lun_lock);
13680		goto bailout;
13681	}
13682
13683	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
13684	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
13685	       sizeof(io->scsiio.sense_data)));
13686	ctl_set_mask(lun->have_ca, initidx);
13687	mtx_unlock(&lun->lun_lock);
13688
13689bailout:
13690	mtx_unlock(&ctl_softc->ctl_lock);
13691
13692	ctl_free_io(io);
13693
13694	return (CTL_RETVAL_COMPLETE);
13695}
13696
13697/*
13698 * Primary command inlet from frontend ports.  All SCSI and task I/O
13699 * requests must go through this function.
13700 */
13701int
13702ctl_queue(union ctl_io *io)
13703{
13704	struct ctl_softc *ctl_softc;
13705
13706	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13707
13708	ctl_softc = control_softc;
13709
13710#ifdef CTL_TIME_IO
13711	io->io_hdr.start_time = time_uptime;
13712	getbintime(&io->io_hdr.start_bt);
13713#endif /* CTL_TIME_IO */
13714
13715	/* Map FE-specific LUN ID into global one. */
13716	io->io_hdr.nexus.targ_mapped_lun =
13717	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13718
13719	switch (io->io_hdr.io_type) {
13720	case CTL_IO_SCSI:
13721	case CTL_IO_TASK:
13722		ctl_enqueue_incoming(io);
13723		break;
13724	default:
13725		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13726		return (EINVAL);
13727	}
13728
13729	return (CTL_RETVAL_COMPLETE);
13730}
13731
13732#ifdef CTL_IO_DELAY
13733static void
13734ctl_done_timer_wakeup(void *arg)
13735{
13736	union ctl_io *io;
13737
13738	io = (union ctl_io *)arg;
13739	ctl_done(io);
13740}
13741#endif /* CTL_IO_DELAY */
13742
13743void
13744ctl_done(union ctl_io *io)
13745{
13746	struct ctl_softc *ctl_softc;
13747
13748	ctl_softc = control_softc;
13749
13750	/*
13751	 * Enable this to catch duplicate completion issues.
13752	 */
13753#if 0
13754	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13755		printf("%s: type %d msg %d cdb %x iptl: "
13756		       "%d:%d:%d:%d tag 0x%04x "
13757		       "flag %#x status %x\n",
13758			__func__,
13759			io->io_hdr.io_type,
13760			io->io_hdr.msg_type,
13761			io->scsiio.cdb[0],
13762			io->io_hdr.nexus.initid.id,
13763			io->io_hdr.nexus.targ_port,
13764			io->io_hdr.nexus.targ_target.id,
13765			io->io_hdr.nexus.targ_lun,
13766			(io->io_hdr.io_type ==
13767			CTL_IO_TASK) ?
13768			io->taskio.tag_num :
13769			io->scsiio.tag_num,
13770		        io->io_hdr.flags,
13771			io->io_hdr.status);
13772	} else
13773		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13774#endif
13775
13776	/*
13777	 * This is an internal copy of an I/O, and should not go through
13778	 * the normal done processing logic.
13779	 */
13780	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13781		return;
13782
13783	/*
13784	 * We need to send a msg to the serializing shelf to finish the IO
13785	 * as well.  We don't send a finish message to the other shelf if
13786	 * this is a task management command.  Task management commands
13787	 * aren't serialized in the OOA queue, but rather just executed on
13788	 * both shelf controllers for commands that originated on that
13789	 * controller.
13790	 */
13791	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13792	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13793		union ctl_ha_msg msg_io;
13794
13795		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13796		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13797		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13798		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13799		}
13800		/* continue on to finish IO */
13801	}
13802#ifdef CTL_IO_DELAY
13803	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13804		struct ctl_lun *lun;
13805
13806		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13807
13808		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13809	} else {
13810		struct ctl_lun *lun;
13811
13812		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13813
13814		if ((lun != NULL)
13815		 && (lun->delay_info.done_delay > 0)) {
13816			struct callout *callout;
13817
13818			callout = (struct callout *)&io->io_hdr.timer_bytes;
13819			callout_init(callout, /*mpsafe*/ 1);
13820			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13821			callout_reset(callout,
13822				      lun->delay_info.done_delay * hz,
13823				      ctl_done_timer_wakeup, io);
13824			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13825				lun->delay_info.done_delay = 0;
13826			return;
13827		}
13828	}
13829#endif /* CTL_IO_DELAY */
13830
13831	ctl_enqueue_done(io);
13832}
13833
13834int
13835ctl_isc(struct ctl_scsiio *ctsio)
13836{
13837	struct ctl_lun *lun;
13838	int retval;
13839
13840	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13841
13842	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13843
13844	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13845
13846	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13847
13848	return (retval);
13849}
13850
13851
13852static void
13853ctl_work_thread(void *arg)
13854{
13855	struct ctl_thread *thr = (struct ctl_thread *)arg;
13856	struct ctl_softc *softc = thr->ctl_softc;
13857	union ctl_io *io;
13858	int retval;
13859
13860	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13861
13862	for (;;) {
13863		retval = 0;
13864
13865		/*
13866		 * We handle the queues in this order:
13867		 * - ISC
13868		 * - done queue (to free up resources, unblock other commands)
13869		 * - RtR queue
13870		 * - incoming queue
13871		 *
13872		 * If those queues are empty, we break out of the loop and
13873		 * go to sleep.
13874		 */
13875		mtx_lock(&thr->queue_lock);
13876		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13877		if (io != NULL) {
13878			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13879			mtx_unlock(&thr->queue_lock);
13880			ctl_handle_isc(io);
13881			continue;
13882		}
13883		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13884		if (io != NULL) {
13885			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13886			/* clear any blocked commands, call fe_done */
13887			mtx_unlock(&thr->queue_lock);
13888			retval = ctl_process_done(io);
13889			continue;
13890		}
13891		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13892		if (io != NULL) {
13893			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13894			mtx_unlock(&thr->queue_lock);
13895			if (io->io_hdr.io_type == CTL_IO_TASK)
13896				ctl_run_task(io);
13897			else
13898				ctl_scsiio_precheck(softc, &io->scsiio);
13899			continue;
13900		}
13901		if (!ctl_pause_rtr) {
13902			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13903			if (io != NULL) {
13904				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13905				mtx_unlock(&thr->queue_lock);
13906				retval = ctl_scsiio(&io->scsiio);
13907				if (retval != CTL_RETVAL_COMPLETE)
13908					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13909				continue;
13910			}
13911		}
13912
13913		/* Sleep until we have something to do. */
13914		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13915	}
13916}
13917
13918static void
13919ctl_lun_thread(void *arg)
13920{
13921	struct ctl_softc *softc = (struct ctl_softc *)arg;
13922	struct ctl_be_lun *be_lun;
13923	int retval;
13924
13925	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13926
13927	for (;;) {
13928		retval = 0;
13929		mtx_lock(&softc->ctl_lock);
13930		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13931		if (be_lun != NULL) {
13932			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13933			mtx_unlock(&softc->ctl_lock);
13934			ctl_create_lun(be_lun);
13935			continue;
13936		}
13937
13938		/* Sleep until we have something to do. */
13939		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13940		    PDROP | PRIBIO, "-", 0);
13941	}
13942}
13943
13944static void
13945ctl_enqueue_incoming(union ctl_io *io)
13946{
13947	struct ctl_softc *softc = control_softc;
13948	struct ctl_thread *thr;
13949	u_int idx;
13950
13951	idx = (io->io_hdr.nexus.targ_port * 127 +
13952	       io->io_hdr.nexus.initid.id) % worker_threads;
13953	thr = &softc->threads[idx];
13954	mtx_lock(&thr->queue_lock);
13955	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13956	mtx_unlock(&thr->queue_lock);
13957	wakeup(thr);
13958}
13959
13960static void
13961ctl_enqueue_rtr(union ctl_io *io)
13962{
13963	struct ctl_softc *softc = control_softc;
13964	struct ctl_thread *thr;
13965
13966	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13967	mtx_lock(&thr->queue_lock);
13968	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13969	mtx_unlock(&thr->queue_lock);
13970	wakeup(thr);
13971}
13972
13973static void
13974ctl_enqueue_done(union ctl_io *io)
13975{
13976	struct ctl_softc *softc = control_softc;
13977	struct ctl_thread *thr;
13978
13979	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13980	mtx_lock(&thr->queue_lock);
13981	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13982	mtx_unlock(&thr->queue_lock);
13983	wakeup(thr);
13984}
13985
13986static void
13987ctl_enqueue_isc(union ctl_io *io)
13988{
13989	struct ctl_softc *softc = control_softc;
13990	struct ctl_thread *thr;
13991
13992	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13993	mtx_lock(&thr->queue_lock);
13994	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13995	mtx_unlock(&thr->queue_lock);
13996	wakeup(thr);
13997}
13998
13999/* Initialization and failover */
14000
14001void
14002ctl_init_isc_msg(void)
14003{
14004	printf("CTL: Still calling this thing\n");
14005}
14006
14007/*
14008 * Init component
14009 * 	Initializes component into configuration defined by bootMode
14010 *	(see hasc-sv.c)
14011 *  	returns hasc_Status:
14012 * 		OK
14013 *		ERROR - fatal error
14014 */
14015static ctl_ha_comp_status
14016ctl_isc_init(struct ctl_ha_component *c)
14017{
14018	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14019
14020	c->status = ret;
14021	return ret;
14022}
14023
14024/* Start component
14025 * 	Starts component in state requested. If component starts successfully,
14026 *	it must set its own state to the requestrd state
14027 *	When requested state is HASC_STATE_HA, the component may refine it
14028 * 	by adding _SLAVE or _MASTER flags.
14029 *	Currently allowed state transitions are:
14030 *	UNKNOWN->HA		- initial startup
14031 *	UNKNOWN->SINGLE - initial startup when no parter detected
14032 *	HA->SINGLE		- failover
14033 * returns ctl_ha_comp_status:
14034 * 		OK	- component successfully started in requested state
14035 *		FAILED  - could not start the requested state, failover may
14036 * 			  be possible
14037 *		ERROR	- fatal error detected, no future startup possible
14038 */
14039static ctl_ha_comp_status
14040ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14041{
14042	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14043
14044	printf("%s: go\n", __func__);
14045
14046	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14047	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14048		ctl_is_single = 0;
14049		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14050		    != CTL_HA_STATUS_SUCCESS) {
14051			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14052			ret = CTL_HA_COMP_STATUS_ERROR;
14053		}
14054	} else if (CTL_HA_STATE_IS_HA(c->state)
14055		&& CTL_HA_STATE_IS_SINGLE(state)){
14056		// HA->SINGLE transition
14057	        ctl_failover();
14058		ctl_is_single = 1;
14059	} else {
14060		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14061		       c->state, state);
14062		ret = CTL_HA_COMP_STATUS_ERROR;
14063	}
14064	if (CTL_HA_STATE_IS_SINGLE(state))
14065		ctl_is_single = 1;
14066
14067	c->state = state;
14068	c->status = ret;
14069	return ret;
14070}
14071
14072/*
14073 * Quiesce component
14074 * The component must clear any error conditions (set status to OK) and
14075 * prepare itself to another Start call
14076 * returns ctl_ha_comp_status:
14077 * 	OK
14078 *	ERROR
14079 */
14080static ctl_ha_comp_status
14081ctl_isc_quiesce(struct ctl_ha_component *c)
14082{
14083	int ret = CTL_HA_COMP_STATUS_OK;
14084
14085	ctl_pause_rtr = 1;
14086	c->status = ret;
14087	return ret;
14088}
14089
14090struct ctl_ha_component ctl_ha_component_ctlisc =
14091{
14092	.name = "CTL ISC",
14093	.state = CTL_HA_STATE_UNKNOWN,
14094	.init = ctl_isc_init,
14095	.start = ctl_isc_start,
14096	.quiesce = ctl_isc_quiesce
14097};
14098
14099/*
14100 *  vim: ts=8
14101 */
14102