ctl.c revision 268689
1/*-
2 * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3 * Copyright (c) 2012 The FreeBSD Foundation
4 * All rights reserved.
5 *
6 * Portions of this software were developed by Edward Tomasz Napierala
7 * under sponsorship from the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions, and the following disclaimer,
14 *    without modification.
15 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16 *    substantially similar to the "NO WARRANTY" disclaimer below
17 *    ("Disclaimer") and any redistribution must be conditioned upon
18 *    including a substantially similar Disclaimer requirement for further
19 *    binary redistribution.
20 *
21 * NO WARRANTY
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGES.
33 *
34 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
35 */
36/*
37 * CAM Target Layer, a SCSI device emulation subsystem.
38 *
39 * Author: Ken Merry <ken@FreeBSD.org>
40 */
41
42#define _CTL_C
43
44#include <sys/cdefs.h>
45__FBSDID("$FreeBSD: stable/10/sys/cam/ctl/ctl.c 268689 2014-07-15 17:11:42Z mav $");
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/kernel.h>
50#include <sys/types.h>
51#include <sys/kthread.h>
52#include <sys/bio.h>
53#include <sys/fcntl.h>
54#include <sys/lock.h>
55#include <sys/module.h>
56#include <sys/mutex.h>
57#include <sys/condvar.h>
58#include <sys/malloc.h>
59#include <sys/conf.h>
60#include <sys/ioccom.h>
61#include <sys/queue.h>
62#include <sys/sbuf.h>
63#include <sys/smp.h>
64#include <sys/endian.h>
65#include <sys/sysctl.h>
66
67#include <cam/cam.h>
68#include <cam/scsi/scsi_all.h>
69#include <cam/scsi/scsi_da.h>
70#include <cam/ctl/ctl_io.h>
71#include <cam/ctl/ctl.h>
72#include <cam/ctl/ctl_frontend.h>
73#include <cam/ctl/ctl_frontend_internal.h>
74#include <cam/ctl/ctl_util.h>
75#include <cam/ctl/ctl_backend.h>
76#include <cam/ctl/ctl_ioctl.h>
77#include <cam/ctl/ctl_ha.h>
78#include <cam/ctl/ctl_private.h>
79#include <cam/ctl/ctl_debug.h>
80#include <cam/ctl/ctl_scsi_all.h>
81#include <cam/ctl/ctl_error.h>
82
83struct ctl_softc *control_softc = NULL;
84
85/*
86 * Size and alignment macros needed for Copan-specific HA hardware.  These
87 * can go away when the HA code is re-written, and uses busdma for any
88 * hardware.
89 */
90#define	CTL_ALIGN_8B(target, source, type)				\
91	if (((uint32_t)source & 0x7) != 0)				\
92		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
93	else								\
94		target = (type)source;
95
96#define	CTL_SIZE_8B(target, size)					\
97	if ((size & 0x7) != 0)						\
98		target = size + (0x8 - (size & 0x7));			\
99	else								\
100		target = size;
101
102#define CTL_ALIGN_8B_MARGIN	16
103
104/*
105 * Template mode pages.
106 */
107
108/*
109 * Note that these are default values only.  The actual values will be
110 * filled in when the user does a mode sense.
111 */
112static struct copan_power_subpage power_page_default = {
113	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
114	/*subpage*/ PWR_SUBPAGE_CODE,
115	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
116			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
117	/*page_version*/ PWR_VERSION,
118	/* total_luns */ 26,
119	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
120	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
121		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
122		      0, 0, 0, 0, 0, 0}
123};
124
125static struct copan_power_subpage power_page_changeable = {
126	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
127	/*subpage*/ PWR_SUBPAGE_CODE,
128	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
129			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
130	/*page_version*/ 0,
131	/* total_luns */ 0,
132	/* max_active_luns*/ 0,
133	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
134		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
135		      0, 0, 0, 0, 0, 0}
136};
137
138static struct copan_aps_subpage aps_page_default = {
139	APS_PAGE_CODE | SMPH_SPF, //page_code
140	APS_SUBPAGE_CODE, //subpage
141	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
142	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
143	APS_VERSION, //page_version
144	0, //lock_active
145	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
146	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
147	0, 0, 0, 0, 0} //reserved
148};
149
150static struct copan_aps_subpage aps_page_changeable = {
151	APS_PAGE_CODE | SMPH_SPF, //page_code
152	APS_SUBPAGE_CODE, //subpage
153	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
154	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
155	0, //page_version
156	0, //lock_active
157	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
158	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
159	0, 0, 0, 0, 0} //reserved
160};
161
162static struct copan_debugconf_subpage debugconf_page_default = {
163	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
164	DBGCNF_SUBPAGE_CODE,		/* subpage */
165	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
166	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
167	DBGCNF_VERSION,			/* page_version */
168	{CTL_TIME_IO_DEFAULT_SECS>>8,
169	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
170};
171
172static struct copan_debugconf_subpage debugconf_page_changeable = {
173	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
174	DBGCNF_SUBPAGE_CODE,		/* subpage */
175	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
176	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
177	0,				/* page_version */
178	{0xff,0xff},			/* ctl_time_io_secs */
179};
180
181static struct scsi_format_page format_page_default = {
182	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
183	/*page_length*/sizeof(struct scsi_format_page) - 2,
184	/*tracks_per_zone*/ {0, 0},
185	/*alt_sectors_per_zone*/ {0, 0},
186	/*alt_tracks_per_zone*/ {0, 0},
187	/*alt_tracks_per_lun*/ {0, 0},
188	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
189			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
190	/*bytes_per_sector*/ {0, 0},
191	/*interleave*/ {0, 0},
192	/*track_skew*/ {0, 0},
193	/*cylinder_skew*/ {0, 0},
194	/*flags*/ SFP_HSEC,
195	/*reserved*/ {0, 0, 0}
196};
197
198static struct scsi_format_page format_page_changeable = {
199	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
200	/*page_length*/sizeof(struct scsi_format_page) - 2,
201	/*tracks_per_zone*/ {0, 0},
202	/*alt_sectors_per_zone*/ {0, 0},
203	/*alt_tracks_per_zone*/ {0, 0},
204	/*alt_tracks_per_lun*/ {0, 0},
205	/*sectors_per_track*/ {0, 0},
206	/*bytes_per_sector*/ {0, 0},
207	/*interleave*/ {0, 0},
208	/*track_skew*/ {0, 0},
209	/*cylinder_skew*/ {0, 0},
210	/*flags*/ 0,
211	/*reserved*/ {0, 0, 0}
212};
213
214static struct scsi_rigid_disk_page rigid_disk_page_default = {
215	/*page_code*/SMS_RIGID_DISK_PAGE,
216	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
217	/*cylinders*/ {0, 0, 0},
218	/*heads*/ CTL_DEFAULT_HEADS,
219	/*start_write_precomp*/ {0, 0, 0},
220	/*start_reduced_current*/ {0, 0, 0},
221	/*step_rate*/ {0, 0},
222	/*landing_zone_cylinder*/ {0, 0, 0},
223	/*rpl*/ SRDP_RPL_DISABLED,
224	/*rotational_offset*/ 0,
225	/*reserved1*/ 0,
226	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
227			   CTL_DEFAULT_ROTATION_RATE & 0xff},
228	/*reserved2*/ {0, 0}
229};
230
231static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
232	/*page_code*/SMS_RIGID_DISK_PAGE,
233	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
234	/*cylinders*/ {0, 0, 0},
235	/*heads*/ 0,
236	/*start_write_precomp*/ {0, 0, 0},
237	/*start_reduced_current*/ {0, 0, 0},
238	/*step_rate*/ {0, 0},
239	/*landing_zone_cylinder*/ {0, 0, 0},
240	/*rpl*/ 0,
241	/*rotational_offset*/ 0,
242	/*reserved1*/ 0,
243	/*rotation_rate*/ {0, 0},
244	/*reserved2*/ {0, 0}
245};
246
247static struct scsi_caching_page caching_page_default = {
248	/*page_code*/SMS_CACHING_PAGE,
249	/*page_length*/sizeof(struct scsi_caching_page) - 2,
250	/*flags1*/ SCP_DISC | SCP_WCE,
251	/*ret_priority*/ 0,
252	/*disable_pf_transfer_len*/ {0xff, 0xff},
253	/*min_prefetch*/ {0, 0},
254	/*max_prefetch*/ {0xff, 0xff},
255	/*max_pf_ceiling*/ {0xff, 0xff},
256	/*flags2*/ 0,
257	/*cache_segments*/ 0,
258	/*cache_seg_size*/ {0, 0},
259	/*reserved*/ 0,
260	/*non_cache_seg_size*/ {0, 0, 0}
261};
262
263static struct scsi_caching_page caching_page_changeable = {
264	/*page_code*/SMS_CACHING_PAGE,
265	/*page_length*/sizeof(struct scsi_caching_page) - 2,
266	/*flags1*/ 0,
267	/*ret_priority*/ 0,
268	/*disable_pf_transfer_len*/ {0, 0},
269	/*min_prefetch*/ {0, 0},
270	/*max_prefetch*/ {0, 0},
271	/*max_pf_ceiling*/ {0, 0},
272	/*flags2*/ 0,
273	/*cache_segments*/ 0,
274	/*cache_seg_size*/ {0, 0},
275	/*reserved*/ 0,
276	/*non_cache_seg_size*/ {0, 0, 0}
277};
278
279static struct scsi_control_page control_page_default = {
280	/*page_code*/SMS_CONTROL_MODE_PAGE,
281	/*page_length*/sizeof(struct scsi_control_page) - 2,
282	/*rlec*/0,
283	/*queue_flags*/0,
284	/*eca_and_aen*/0,
285	/*reserved*/0,
286	/*aen_holdoff_period*/{0, 0}
287};
288
289static struct scsi_control_page control_page_changeable = {
290	/*page_code*/SMS_CONTROL_MODE_PAGE,
291	/*page_length*/sizeof(struct scsi_control_page) - 2,
292	/*rlec*/SCP_DSENSE,
293	/*queue_flags*/0,
294	/*eca_and_aen*/0,
295	/*reserved*/0,
296	/*aen_holdoff_period*/{0, 0}
297};
298
299
300/*
301 * XXX KDM move these into the softc.
302 */
303static int rcv_sync_msg;
304static int persis_offset;
305static uint8_t ctl_pause_rtr;
306static int     ctl_is_single = 1;
307static int     index_to_aps_page;
308
309SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
310static int worker_threads = -1;
311TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
312SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
313    &worker_threads, 1, "Number of worker threads");
314static int verbose = 0;
315TUNABLE_INT("kern.cam.ctl.verbose", &verbose);
316SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
317    &verbose, 0, "Show SCSI errors returned to initiator");
318
319/*
320 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
321 * SCSI Ports (0x88), Block limits (0xB0) and
322 * Logical Block Provisioning (0xB2)
323 */
324#define SCSI_EVPD_NUM_SUPPORTED_PAGES	6
325
326static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
327				  int param);
328static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
329static int ctl_init(void);
330void ctl_shutdown(void);
331static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
332static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
333static void ctl_ioctl_online(void *arg);
334static void ctl_ioctl_offline(void *arg);
335static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
336static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
337static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
338static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
339static int ctl_ioctl_submit_wait(union ctl_io *io);
340static void ctl_ioctl_datamove(union ctl_io *io);
341static void ctl_ioctl_done(union ctl_io *io);
342static void ctl_ioctl_hard_startstop_callback(void *arg,
343					      struct cfi_metatask *metatask);
344static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
345static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
346			      struct ctl_ooa *ooa_hdr,
347			      struct ctl_ooa_entry *kern_entries);
348static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
349		     struct thread *td);
350uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
351uint32_t ctl_port_idx(int port_num);
352static uint32_t ctl_map_lun(int port_num, uint32_t lun);
353static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
354#ifdef unused
355static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
356				   uint32_t targ_target, uint32_t targ_lun,
357				   int can_wait);
358static void ctl_kfree_io(union ctl_io *io);
359#endif /* unused */
360static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
361			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
362static int ctl_free_lun(struct ctl_lun *lun);
363static void ctl_create_lun(struct ctl_be_lun *be_lun);
364/**
365static void ctl_failover_change_pages(struct ctl_softc *softc,
366				      struct ctl_scsiio *ctsio, int master);
367**/
368
369static int ctl_do_mode_select(union ctl_io *io);
370static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
371			   uint64_t res_key, uint64_t sa_res_key,
372			   uint8_t type, uint32_t residx,
373			   struct ctl_scsiio *ctsio,
374			   struct scsi_per_res_out *cdb,
375			   struct scsi_per_res_out_parms* param);
376static void ctl_pro_preempt_other(struct ctl_lun *lun,
377				  union ctl_ha_msg *msg);
378static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
379static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
380static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
381static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
382static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
383					 int alloc_len);
384static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
385					 int alloc_len);
386static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
387static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
388static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
389static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
390static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
391static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
392					 union ctl_io *ooa_io);
393static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
394				union ctl_io *starting_io);
395static int ctl_check_blocked(struct ctl_lun *lun);
396static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
397				struct ctl_lun *lun,
398				const struct ctl_cmd_entry *entry,
399				struct ctl_scsiio *ctsio);
400//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
401static void ctl_failover(void);
402static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
403			       struct ctl_scsiio *ctsio);
404static int ctl_scsiio(struct ctl_scsiio *ctsio);
405
406static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
407static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
408			    ctl_ua_type ua_type);
409static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
410			 ctl_ua_type ua_type);
411static int ctl_abort_task(union ctl_io *io);
412static void ctl_run_task(union ctl_io *io);
413#ifdef CTL_IO_DELAY
414static void ctl_datamove_timer_wakeup(void *arg);
415static void ctl_done_timer_wakeup(void *arg);
416#endif /* CTL_IO_DELAY */
417
418static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
419static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
420static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
421static void ctl_datamove_remote_write(union ctl_io *io);
422static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
423static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
424static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
425static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
426				    ctl_ha_dt_cb callback);
427static void ctl_datamove_remote_read(union ctl_io *io);
428static void ctl_datamove_remote(union ctl_io *io);
429static int ctl_process_done(union ctl_io *io);
430static void ctl_lun_thread(void *arg);
431static void ctl_work_thread(void *arg);
432static void ctl_enqueue_incoming(union ctl_io *io);
433static void ctl_enqueue_rtr(union ctl_io *io);
434static void ctl_enqueue_done(union ctl_io *io);
435static void ctl_enqueue_isc(union ctl_io *io);
436static const struct ctl_cmd_entry *
437    ctl_get_cmd_entry(struct ctl_scsiio *ctsio);
438static const struct ctl_cmd_entry *
439    ctl_validate_command(struct ctl_scsiio *ctsio);
440static int ctl_cmd_applicable(uint8_t lun_type,
441    const struct ctl_cmd_entry *entry);
442
443/*
444 * Load the serialization table.  This isn't very pretty, but is probably
445 * the easiest way to do it.
446 */
447#include "ctl_ser_table.c"
448
449/*
450 * We only need to define open, close and ioctl routines for this driver.
451 */
452static struct cdevsw ctl_cdevsw = {
453	.d_version =	D_VERSION,
454	.d_flags =	0,
455	.d_open =	ctl_open,
456	.d_close =	ctl_close,
457	.d_ioctl =	ctl_ioctl,
458	.d_name =	"ctl",
459};
460
461
462MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
463MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
464
465static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
466
467static moduledata_t ctl_moduledata = {
468	"ctl",
469	ctl_module_event_handler,
470	NULL
471};
472
473DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
474MODULE_VERSION(ctl, 1);
475
476static struct ctl_frontend ioctl_frontend =
477{
478	.name = "ioctl",
479};
480
481static void
482ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
483			    union ctl_ha_msg *msg_info)
484{
485	struct ctl_scsiio *ctsio;
486
487	if (msg_info->hdr.original_sc == NULL) {
488		printf("%s: original_sc == NULL!\n", __func__);
489		/* XXX KDM now what? */
490		return;
491	}
492
493	ctsio = &msg_info->hdr.original_sc->scsiio;
494	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
495	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
496	ctsio->io_hdr.status = msg_info->hdr.status;
497	ctsio->scsi_status = msg_info->scsi.scsi_status;
498	ctsio->sense_len = msg_info->scsi.sense_len;
499	ctsio->sense_residual = msg_info->scsi.sense_residual;
500	ctsio->residual = msg_info->scsi.residual;
501	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
502	       sizeof(ctsio->sense_data));
503	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
504	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
505	ctl_enqueue_isc((union ctl_io *)ctsio);
506}
507
508static void
509ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
510				union ctl_ha_msg *msg_info)
511{
512	struct ctl_scsiio *ctsio;
513
514	if (msg_info->hdr.serializing_sc == NULL) {
515		printf("%s: serializing_sc == NULL!\n", __func__);
516		/* XXX KDM now what? */
517		return;
518	}
519
520	ctsio = &msg_info->hdr.serializing_sc->scsiio;
521#if 0
522	/*
523	 * Attempt to catch the situation where an I/O has
524	 * been freed, and we're using it again.
525	 */
526	if (ctsio->io_hdr.io_type == 0xff) {
527		union ctl_io *tmp_io;
528		tmp_io = (union ctl_io *)ctsio;
529		printf("%s: %p use after free!\n", __func__,
530		       ctsio);
531		printf("%s: type %d msg %d cdb %x iptl: "
532		       "%d:%d:%d:%d tag 0x%04x "
533		       "flag %#x status %x\n",
534			__func__,
535			tmp_io->io_hdr.io_type,
536			tmp_io->io_hdr.msg_type,
537			tmp_io->scsiio.cdb[0],
538			tmp_io->io_hdr.nexus.initid.id,
539			tmp_io->io_hdr.nexus.targ_port,
540			tmp_io->io_hdr.nexus.targ_target.id,
541			tmp_io->io_hdr.nexus.targ_lun,
542			(tmp_io->io_hdr.io_type ==
543			CTL_IO_TASK) ?
544			tmp_io->taskio.tag_num :
545			tmp_io->scsiio.tag_num,
546		        tmp_io->io_hdr.flags,
547			tmp_io->io_hdr.status);
548	}
549#endif
550	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
551	ctl_enqueue_isc((union ctl_io *)ctsio);
552}
553
554/*
555 * ISC (Inter Shelf Communication) event handler.  Events from the HA
556 * subsystem come in here.
557 */
558static void
559ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
560{
561	struct ctl_softc *ctl_softc;
562	union ctl_io *io;
563	struct ctl_prio *presio;
564	ctl_ha_status isc_status;
565
566	ctl_softc = control_softc;
567	io = NULL;
568
569
570#if 0
571	printf("CTL: Isc Msg event %d\n", event);
572#endif
573	if (event == CTL_HA_EVT_MSG_RECV) {
574		union ctl_ha_msg msg_info;
575
576		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
577					     sizeof(msg_info), /*wait*/ 0);
578#if 0
579		printf("CTL: msg_type %d\n", msg_info.msg_type);
580#endif
581		if (isc_status != 0) {
582			printf("Error receiving message, status = %d\n",
583			       isc_status);
584			return;
585		}
586
587		switch (msg_info.hdr.msg_type) {
588		case CTL_MSG_SERIALIZE:
589#if 0
590			printf("Serialize\n");
591#endif
592			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
593			if (io == NULL) {
594				printf("ctl_isc_event_handler: can't allocate "
595				       "ctl_io!\n");
596				/* Bad Juju */
597				/* Need to set busy and send msg back */
598				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
599				msg_info.hdr.status = CTL_SCSI_ERROR;
600				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
601				msg_info.scsi.sense_len = 0;
602			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
603				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
604				}
605				goto bailout;
606			}
607			ctl_zero_io(io);
608			// populate ctsio from msg_info
609			io->io_hdr.io_type = CTL_IO_SCSI;
610			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
611			io->io_hdr.original_sc = msg_info.hdr.original_sc;
612#if 0
613			printf("pOrig %x\n", (int)msg_info.original_sc);
614#endif
615			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
616					    CTL_FLAG_IO_ACTIVE;
617			/*
618			 * If we're in serialization-only mode, we don't
619			 * want to go through full done processing.  Thus
620			 * the COPY flag.
621			 *
622			 * XXX KDM add another flag that is more specific.
623			 */
624			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
625				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
626			io->io_hdr.nexus = msg_info.hdr.nexus;
627#if 0
628			printf("targ %d, port %d, iid %d, lun %d\n",
629			       io->io_hdr.nexus.targ_target.id,
630			       io->io_hdr.nexus.targ_port,
631			       io->io_hdr.nexus.initid.id,
632			       io->io_hdr.nexus.targ_lun);
633#endif
634			io->scsiio.tag_num = msg_info.scsi.tag_num;
635			io->scsiio.tag_type = msg_info.scsi.tag_type;
636			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
637			       CTL_MAX_CDBLEN);
638			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
639				const struct ctl_cmd_entry *entry;
640
641				entry = ctl_get_cmd_entry(&io->scsiio);
642				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
643				io->io_hdr.flags |=
644					entry->flags & CTL_FLAG_DATA_MASK;
645			}
646			ctl_enqueue_isc(io);
647			break;
648
649		/* Performed on the Originating SC, XFER mode only */
650		case CTL_MSG_DATAMOVE: {
651			struct ctl_sg_entry *sgl;
652			int i, j;
653
654			io = msg_info.hdr.original_sc;
655			if (io == NULL) {
656				printf("%s: original_sc == NULL!\n", __func__);
657				/* XXX KDM do something here */
658				break;
659			}
660			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
661			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
662			/*
663			 * Keep track of this, we need to send it back over
664			 * when the datamove is complete.
665			 */
666			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
667
668			if (msg_info.dt.sg_sequence == 0) {
669				/*
670				 * XXX KDM we use the preallocated S/G list
671				 * here, but we'll need to change this to
672				 * dynamic allocation if we need larger S/G
673				 * lists.
674				 */
675				if (msg_info.dt.kern_sg_entries >
676				    sizeof(io->io_hdr.remote_sglist) /
677				    sizeof(io->io_hdr.remote_sglist[0])) {
678					printf("%s: number of S/G entries "
679					    "needed %u > allocated num %zd\n",
680					    __func__,
681					    msg_info.dt.kern_sg_entries,
682					    sizeof(io->io_hdr.remote_sglist)/
683					    sizeof(io->io_hdr.remote_sglist[0]));
684
685					/*
686					 * XXX KDM send a message back to
687					 * the other side to shut down the
688					 * DMA.  The error will come back
689					 * through via the normal channel.
690					 */
691					break;
692				}
693				sgl = io->io_hdr.remote_sglist;
694				memset(sgl, 0,
695				       sizeof(io->io_hdr.remote_sglist));
696
697				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
698
699				io->scsiio.kern_sg_entries =
700					msg_info.dt.kern_sg_entries;
701				io->scsiio.rem_sg_entries =
702					msg_info.dt.kern_sg_entries;
703				io->scsiio.kern_data_len =
704					msg_info.dt.kern_data_len;
705				io->scsiio.kern_total_len =
706					msg_info.dt.kern_total_len;
707				io->scsiio.kern_data_resid =
708					msg_info.dt.kern_data_resid;
709				io->scsiio.kern_rel_offset =
710					msg_info.dt.kern_rel_offset;
711				/*
712				 * Clear out per-DMA flags.
713				 */
714				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
715				/*
716				 * Add per-DMA flags that are set for this
717				 * particular DMA request.
718				 */
719				io->io_hdr.flags |= msg_info.dt.flags &
720						    CTL_FLAG_RDMA_MASK;
721			} else
722				sgl = (struct ctl_sg_entry *)
723					io->scsiio.kern_data_ptr;
724
725			for (i = msg_info.dt.sent_sg_entries, j = 0;
726			     i < (msg_info.dt.sent_sg_entries +
727			     msg_info.dt.cur_sg_entries); i++, j++) {
728				sgl[i].addr = msg_info.dt.sg_list[j].addr;
729				sgl[i].len = msg_info.dt.sg_list[j].len;
730
731#if 0
732				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
733				       __func__,
734				       msg_info.dt.sg_list[j].addr,
735				       msg_info.dt.sg_list[j].len,
736				       sgl[i].addr, sgl[i].len, j, i);
737#endif
738			}
739#if 0
740			memcpy(&sgl[msg_info.dt.sent_sg_entries],
741			       msg_info.dt.sg_list,
742			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
743#endif
744
745			/*
746			 * If this is the last piece of the I/O, we've got
747			 * the full S/G list.  Queue processing in the thread.
748			 * Otherwise wait for the next piece.
749			 */
750			if (msg_info.dt.sg_last != 0)
751				ctl_enqueue_isc(io);
752			break;
753		}
754		/* Performed on the Serializing (primary) SC, XFER mode only */
755		case CTL_MSG_DATAMOVE_DONE: {
756			if (msg_info.hdr.serializing_sc == NULL) {
757				printf("%s: serializing_sc == NULL!\n",
758				       __func__);
759				/* XXX KDM now what? */
760				break;
761			}
762			/*
763			 * We grab the sense information here in case
764			 * there was a failure, so we can return status
765			 * back to the initiator.
766			 */
767			io = msg_info.hdr.serializing_sc;
768			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
769			io->io_hdr.status = msg_info.hdr.status;
770			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
771			io->scsiio.sense_len = msg_info.scsi.sense_len;
772			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
773			io->io_hdr.port_status = msg_info.scsi.fetd_status;
774			io->scsiio.residual = msg_info.scsi.residual;
775			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
776			       sizeof(io->scsiio.sense_data));
777			ctl_enqueue_isc(io);
778			break;
779		}
780
781		/* Preformed on Originating SC, SER_ONLY mode */
782		case CTL_MSG_R2R:
783			io = msg_info.hdr.original_sc;
784			if (io == NULL) {
785				printf("%s: Major Bummer\n", __func__);
786				return;
787			} else {
788#if 0
789				printf("pOrig %x\n",(int) ctsio);
790#endif
791			}
792			io->io_hdr.msg_type = CTL_MSG_R2R;
793			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
794			ctl_enqueue_isc(io);
795			break;
796
797		/*
798		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
799		 * mode.
800		 * Performed on the Originating (i.e. secondary) SC in XFER
801		 * mode
802		 */
803		case CTL_MSG_FINISH_IO:
804			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
805				ctl_isc_handler_finish_xfer(ctl_softc,
806							    &msg_info);
807			else
808				ctl_isc_handler_finish_ser_only(ctl_softc,
809								&msg_info);
810			break;
811
812		/* Preformed on Originating SC */
813		case CTL_MSG_BAD_JUJU:
814			io = msg_info.hdr.original_sc;
815			if (io == NULL) {
816				printf("%s: Bad JUJU!, original_sc is NULL!\n",
817				       __func__);
818				break;
819			}
820			ctl_copy_sense_data(&msg_info, io);
821			/*
822			 * IO should have already been cleaned up on other
823			 * SC so clear this flag so we won't send a message
824			 * back to finish the IO there.
825			 */
826			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
827			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
828
829			/* io = msg_info.hdr.serializing_sc; */
830			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
831			ctl_enqueue_isc(io);
832			break;
833
834		/* Handle resets sent from the other side */
835		case CTL_MSG_MANAGE_TASKS: {
836			struct ctl_taskio *taskio;
837			taskio = (struct ctl_taskio *)ctl_alloc_io(
838				(void *)ctl_softc->othersc_pool);
839			if (taskio == NULL) {
840				printf("ctl_isc_event_handler: can't allocate "
841				       "ctl_io!\n");
842				/* Bad Juju */
843				/* should I just call the proper reset func
844				   here??? */
845				goto bailout;
846			}
847			ctl_zero_io((union ctl_io *)taskio);
848			taskio->io_hdr.io_type = CTL_IO_TASK;
849			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
850			taskio->io_hdr.nexus = msg_info.hdr.nexus;
851			taskio->task_action = msg_info.task.task_action;
852			taskio->tag_num = msg_info.task.tag_num;
853			taskio->tag_type = msg_info.task.tag_type;
854#ifdef CTL_TIME_IO
855			taskio->io_hdr.start_time = time_uptime;
856			getbintime(&taskio->io_hdr.start_bt);
857#if 0
858			cs_prof_gettime(&taskio->io_hdr.start_ticks);
859#endif
860#endif /* CTL_TIME_IO */
861			ctl_run_task((union ctl_io *)taskio);
862			break;
863		}
864		/* Persistent Reserve action which needs attention */
865		case CTL_MSG_PERS_ACTION:
866			presio = (struct ctl_prio *)ctl_alloc_io(
867				(void *)ctl_softc->othersc_pool);
868			if (presio == NULL) {
869				printf("ctl_isc_event_handler: can't allocate "
870				       "ctl_io!\n");
871				/* Bad Juju */
872				/* Need to set busy and send msg back */
873				goto bailout;
874			}
875			ctl_zero_io((union ctl_io *)presio);
876			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
877			presio->pr_msg = msg_info.pr;
878			ctl_enqueue_isc((union ctl_io *)presio);
879			break;
880		case CTL_MSG_SYNC_FE:
881			rcv_sync_msg = 1;
882			break;
883		case CTL_MSG_APS_LOCK: {
884			// It's quicker to execute this then to
885			// queue it.
886			struct ctl_lun *lun;
887			struct ctl_page_index *page_index;
888			struct copan_aps_subpage *current_sp;
889			uint32_t targ_lun;
890
891			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
892			lun = ctl_softc->ctl_luns[targ_lun];
893			mtx_lock(&lun->lun_lock);
894			page_index = &lun->mode_pages.index[index_to_aps_page];
895			current_sp = (struct copan_aps_subpage *)
896				     (page_index->page_data +
897				     (page_index->page_len * CTL_PAGE_CURRENT));
898
899			current_sp->lock_active = msg_info.aps.lock_flag;
900			mtx_unlock(&lun->lun_lock);
901		        break;
902		}
903		default:
904		        printf("How did I get here?\n");
905		}
906	} else if (event == CTL_HA_EVT_MSG_SENT) {
907		if (param != CTL_HA_STATUS_SUCCESS) {
908			printf("Bad status from ctl_ha_msg_send status %d\n",
909			       param);
910		}
911		return;
912	} else if (event == CTL_HA_EVT_DISCONNECT) {
913		printf("CTL: Got a disconnect from Isc\n");
914		return;
915	} else {
916		printf("ctl_isc_event_handler: Unknown event %d\n", event);
917		return;
918	}
919
920bailout:
921	return;
922}
923
924static void
925ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
926{
927	struct scsi_sense_data *sense;
928
929	sense = &dest->scsiio.sense_data;
930	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
931	dest->scsiio.scsi_status = src->scsi.scsi_status;
932	dest->scsiio.sense_len = src->scsi.sense_len;
933	dest->io_hdr.status = src->hdr.status;
934}
935
936static int
937ctl_init(void)
938{
939	struct ctl_softc *softc;
940	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
941	struct ctl_port *port;
942        uint8_t sc_id =0;
943	int i, error, retval;
944	//int isc_retval;
945
946	retval = 0;
947	ctl_pause_rtr = 0;
948        rcv_sync_msg = 0;
949
950	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
951			       M_WAITOK | M_ZERO);
952	softc = control_softc;
953
954	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
955			      "cam/ctl");
956
957	softc->dev->si_drv1 = softc;
958
959	/*
960	 * By default, return a "bad LUN" peripheral qualifier for unknown
961	 * LUNs.  The user can override this default using the tunable or
962	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
963	 */
964	softc->inquiry_pq_no_lun = 1;
965	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
966			  &softc->inquiry_pq_no_lun);
967	sysctl_ctx_init(&softc->sysctl_ctx);
968	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
969		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
970		CTLFLAG_RD, 0, "CAM Target Layer");
971
972	if (softc->sysctl_tree == NULL) {
973		printf("%s: unable to allocate sysctl tree\n", __func__);
974		destroy_dev(softc->dev);
975		free(control_softc, M_DEVBUF);
976		control_softc = NULL;
977		return (ENOMEM);
978	}
979
980	SYSCTL_ADD_INT(&softc->sysctl_ctx,
981		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
982		       "inquiry_pq_no_lun", CTLFLAG_RW,
983		       &softc->inquiry_pq_no_lun, 0,
984		       "Report no lun possible for invalid LUNs");
985
986	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
987	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
988	softc->open_count = 0;
989
990	/*
991	 * Default to actually sending a SYNCHRONIZE CACHE command down to
992	 * the drive.
993	 */
994	softc->flags = CTL_FLAG_REAL_SYNC;
995
996	/*
997	 * In Copan's HA scheme, the "master" and "slave" roles are
998	 * figured out through the slot the controller is in.  Although it
999	 * is an active/active system, someone has to be in charge.
1000 	 */
1001#ifdef NEEDTOPORT
1002        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1003#endif
1004
1005        if (sc_id == 0) {
1006		softc->flags |= CTL_FLAG_MASTER_SHELF;
1007		persis_offset = 0;
1008	} else
1009		persis_offset = CTL_MAX_INITIATORS;
1010
1011	/*
1012	 * XXX KDM need to figure out where we want to get our target ID
1013	 * and WWID.  Is it different on each port?
1014	 */
1015	softc->target.id = 0;
1016	softc->target.wwid[0] = 0x12345678;
1017	softc->target.wwid[1] = 0x87654321;
1018	STAILQ_INIT(&softc->lun_list);
1019	STAILQ_INIT(&softc->pending_lun_queue);
1020	STAILQ_INIT(&softc->fe_list);
1021	STAILQ_INIT(&softc->port_list);
1022	STAILQ_INIT(&softc->be_list);
1023	STAILQ_INIT(&softc->io_pools);
1024
1025	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1026			    &internal_pool)!= 0){
1027		printf("ctl: can't allocate %d entry internal pool, "
1028		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1029		return (ENOMEM);
1030	}
1031
1032	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1033			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1034		printf("ctl: can't allocate %d entry emergency pool, "
1035		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1036		ctl_pool_free(internal_pool);
1037		return (ENOMEM);
1038	}
1039
1040	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1041	                    &other_pool) != 0)
1042	{
1043		printf("ctl: can't allocate %d entry other SC pool, "
1044		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1045		ctl_pool_free(internal_pool);
1046		ctl_pool_free(emergency_pool);
1047		return (ENOMEM);
1048	}
1049
1050	softc->internal_pool = internal_pool;
1051	softc->emergency_pool = emergency_pool;
1052	softc->othersc_pool = other_pool;
1053
1054	if (worker_threads <= 0)
1055		worker_threads = max(1, mp_ncpus / 4);
1056	if (worker_threads > CTL_MAX_THREADS)
1057		worker_threads = CTL_MAX_THREADS;
1058
1059	for (i = 0; i < worker_threads; i++) {
1060		struct ctl_thread *thr = &softc->threads[i];
1061
1062		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1063		thr->ctl_softc = softc;
1064		STAILQ_INIT(&thr->incoming_queue);
1065		STAILQ_INIT(&thr->rtr_queue);
1066		STAILQ_INIT(&thr->done_queue);
1067		STAILQ_INIT(&thr->isc_queue);
1068
1069		error = kproc_kthread_add(ctl_work_thread, thr,
1070		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1071		if (error != 0) {
1072			printf("error creating CTL work thread!\n");
1073			ctl_pool_free(internal_pool);
1074			ctl_pool_free(emergency_pool);
1075			ctl_pool_free(other_pool);
1076			return (error);
1077		}
1078	}
1079	error = kproc_kthread_add(ctl_lun_thread, softc,
1080	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1081	if (error != 0) {
1082		printf("error creating CTL lun thread!\n");
1083		ctl_pool_free(internal_pool);
1084		ctl_pool_free(emergency_pool);
1085		ctl_pool_free(other_pool);
1086		return (error);
1087	}
1088	if (bootverbose)
1089		printf("ctl: CAM Target Layer loaded\n");
1090
1091	/*
1092	 * Initialize the initiator and portname mappings
1093	 */
1094	memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1095
1096	/*
1097	 * Initialize the ioctl front end.
1098	 */
1099	ctl_frontend_register(&ioctl_frontend);
1100	port = &softc->ioctl_info.port;
1101	port->frontend = &ioctl_frontend;
1102	sprintf(softc->ioctl_info.port_name, "ioctl");
1103	port->port_type = CTL_PORT_IOCTL;
1104	port->num_requested_ctl_io = 100;
1105	port->port_name = softc->ioctl_info.port_name;
1106	port->port_online = ctl_ioctl_online;
1107	port->port_offline = ctl_ioctl_offline;
1108	port->onoff_arg = &softc->ioctl_info;
1109	port->lun_enable = ctl_ioctl_lun_enable;
1110	port->lun_disable = ctl_ioctl_lun_disable;
1111	port->targ_lun_arg = &softc->ioctl_info;
1112	port->fe_datamove = ctl_ioctl_datamove;
1113	port->fe_done = ctl_ioctl_done;
1114	port->max_targets = 15;
1115	port->max_target_id = 15;
1116
1117	if (ctl_port_register(&softc->ioctl_info.port,
1118	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1119		printf("ctl: ioctl front end registration failed, will "
1120		       "continue anyway\n");
1121	}
1122
1123#ifdef CTL_IO_DELAY
1124	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1125		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1126		       sizeof(struct callout), CTL_TIMER_BYTES);
1127		return (EINVAL);
1128	}
1129#endif /* CTL_IO_DELAY */
1130
1131	return (0);
1132}
1133
1134void
1135ctl_shutdown(void)
1136{
1137	struct ctl_softc *softc;
1138	struct ctl_lun *lun, *next_lun;
1139	struct ctl_io_pool *pool;
1140
1141	softc = (struct ctl_softc *)control_softc;
1142
1143	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1144		printf("ctl: ioctl front end deregistration failed\n");
1145
1146	mtx_lock(&softc->ctl_lock);
1147
1148	/*
1149	 * Free up each LUN.
1150	 */
1151	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1152		next_lun = STAILQ_NEXT(lun, links);
1153		ctl_free_lun(lun);
1154	}
1155
1156	mtx_unlock(&softc->ctl_lock);
1157
1158	ctl_frontend_deregister(&ioctl_frontend);
1159
1160	/*
1161	 * This will rip the rug out from under any FETDs or anyone else
1162	 * that has a pool allocated.  Since we increment our module
1163	 * refcount any time someone outside the main CTL module allocates
1164	 * a pool, we shouldn't have any problems here.  The user won't be
1165	 * able to unload the CTL module until client modules have
1166	 * successfully unloaded.
1167	 */
1168	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1169		ctl_pool_free(pool);
1170
1171#if 0
1172	ctl_shutdown_thread(softc->work_thread);
1173	mtx_destroy(&softc->queue_lock);
1174#endif
1175
1176	mtx_destroy(&softc->pool_lock);
1177	mtx_destroy(&softc->ctl_lock);
1178
1179	destroy_dev(softc->dev);
1180
1181	sysctl_ctx_free(&softc->sysctl_ctx);
1182
1183	free(control_softc, M_DEVBUF);
1184	control_softc = NULL;
1185
1186	if (bootverbose)
1187		printf("ctl: CAM Target Layer unloaded\n");
1188}
1189
1190static int
1191ctl_module_event_handler(module_t mod, int what, void *arg)
1192{
1193
1194	switch (what) {
1195	case MOD_LOAD:
1196		return (ctl_init());
1197	case MOD_UNLOAD:
1198		return (EBUSY);
1199	default:
1200		return (EOPNOTSUPP);
1201	}
1202}
1203
1204/*
1205 * XXX KDM should we do some access checks here?  Bump a reference count to
1206 * prevent a CTL module from being unloaded while someone has it open?
1207 */
1208static int
1209ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1210{
1211	return (0);
1212}
1213
1214static int
1215ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1216{
1217	return (0);
1218}
1219
1220int
1221ctl_port_enable(ctl_port_type port_type)
1222{
1223	struct ctl_softc *softc;
1224	struct ctl_port *port;
1225
1226	if (ctl_is_single == 0) {
1227		union ctl_ha_msg msg_info;
1228		int isc_retval;
1229
1230#if 0
1231		printf("%s: HA mode, synchronizing frontend enable\n",
1232		        __func__);
1233#endif
1234		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1235	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1236		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1237			printf("Sync msg send error retval %d\n", isc_retval);
1238		}
1239		if (!rcv_sync_msg) {
1240			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1241			        sizeof(msg_info), 1);
1242		}
1243#if 0
1244        	printf("CTL:Frontend Enable\n");
1245	} else {
1246		printf("%s: single mode, skipping frontend synchronization\n",
1247		        __func__);
1248#endif
1249	}
1250
1251	softc = control_softc;
1252
1253	STAILQ_FOREACH(port, &softc->port_list, links) {
1254		if (port_type & port->port_type)
1255		{
1256#if 0
1257			printf("port %d\n", port->targ_port);
1258#endif
1259			ctl_port_online(port);
1260		}
1261	}
1262
1263	return (0);
1264}
1265
1266int
1267ctl_port_disable(ctl_port_type port_type)
1268{
1269	struct ctl_softc *softc;
1270	struct ctl_port *port;
1271
1272	softc = control_softc;
1273
1274	STAILQ_FOREACH(port, &softc->port_list, links) {
1275		if (port_type & port->port_type)
1276			ctl_port_offline(port);
1277	}
1278
1279	return (0);
1280}
1281
1282/*
1283 * Returns 0 for success, 1 for failure.
1284 * Currently the only failure mode is if there aren't enough entries
1285 * allocated.  So, in case of a failure, look at num_entries_dropped,
1286 * reallocate and try again.
1287 */
1288int
1289ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1290	      int *num_entries_filled, int *num_entries_dropped,
1291	      ctl_port_type port_type, int no_virtual)
1292{
1293	struct ctl_softc *softc;
1294	struct ctl_port *port;
1295	int entries_dropped, entries_filled;
1296	int retval;
1297	int i;
1298
1299	softc = control_softc;
1300
1301	retval = 0;
1302	entries_filled = 0;
1303	entries_dropped = 0;
1304
1305	i = 0;
1306	mtx_lock(&softc->ctl_lock);
1307	STAILQ_FOREACH(port, &softc->port_list, links) {
1308		struct ctl_port_entry *entry;
1309
1310		if ((port->port_type & port_type) == 0)
1311			continue;
1312
1313		if ((no_virtual != 0)
1314		 && (port->virtual_port != 0))
1315			continue;
1316
1317		if (entries_filled >= num_entries_alloced) {
1318			entries_dropped++;
1319			continue;
1320		}
1321		entry = &entries[i];
1322
1323		entry->port_type = port->port_type;
1324		strlcpy(entry->port_name, port->port_name,
1325			sizeof(entry->port_name));
1326		entry->physical_port = port->physical_port;
1327		entry->virtual_port = port->virtual_port;
1328		entry->wwnn = port->wwnn;
1329		entry->wwpn = port->wwpn;
1330
1331		i++;
1332		entries_filled++;
1333	}
1334
1335	mtx_unlock(&softc->ctl_lock);
1336
1337	if (entries_dropped > 0)
1338		retval = 1;
1339
1340	*num_entries_dropped = entries_dropped;
1341	*num_entries_filled = entries_filled;
1342
1343	return (retval);
1344}
1345
1346static void
1347ctl_ioctl_online(void *arg)
1348{
1349	struct ctl_ioctl_info *ioctl_info;
1350
1351	ioctl_info = (struct ctl_ioctl_info *)arg;
1352
1353	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1354}
1355
1356static void
1357ctl_ioctl_offline(void *arg)
1358{
1359	struct ctl_ioctl_info *ioctl_info;
1360
1361	ioctl_info = (struct ctl_ioctl_info *)arg;
1362
1363	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1364}
1365
1366/*
1367 * Remove an initiator by port number and initiator ID.
1368 * Returns 0 for success, 1 for failure.
1369 */
1370int
1371ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1372{
1373	struct ctl_softc *softc;
1374
1375	softc = control_softc;
1376
1377	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1378
1379	if ((targ_port < 0)
1380	 || (targ_port > CTL_MAX_PORTS)) {
1381		printf("%s: invalid port number %d\n", __func__, targ_port);
1382		return (1);
1383	}
1384	if (iid > CTL_MAX_INIT_PER_PORT) {
1385		printf("%s: initiator ID %u > maximun %u!\n",
1386		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1387		return (1);
1388	}
1389
1390	mtx_lock(&softc->ctl_lock);
1391
1392	softc->wwpn_iid[targ_port][iid].in_use = 0;
1393
1394	mtx_unlock(&softc->ctl_lock);
1395
1396	return (0);
1397}
1398
1399/*
1400 * Add an initiator to the initiator map.
1401 * Returns 0 for success, 1 for failure.
1402 */
1403int
1404ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1405{
1406	struct ctl_softc *softc;
1407	int retval;
1408
1409	softc = control_softc;
1410
1411	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1412
1413	retval = 0;
1414
1415	if ((targ_port < 0)
1416	 || (targ_port > CTL_MAX_PORTS)) {
1417		printf("%s: invalid port number %d\n", __func__, targ_port);
1418		return (1);
1419	}
1420	if (iid > CTL_MAX_INIT_PER_PORT) {
1421		printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1422		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1423		return (1);
1424	}
1425
1426	mtx_lock(&softc->ctl_lock);
1427
1428	if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1429		/*
1430		 * We don't treat this as an error.
1431		 */
1432		if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1433			printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1434			       __func__, targ_port, iid, (uintmax_t)wwpn);
1435			goto bailout;
1436		}
1437
1438		/*
1439		 * This is an error, but what do we do about it?  The
1440		 * driver is telling us we have a new WWPN for this
1441		 * initiator ID, so we pretty much need to use it.
1442		 */
1443		printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1444		       "still at that address\n", __func__, targ_port, iid,
1445		       (uintmax_t)wwpn,
1446		       (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1447
1448		/*
1449		 * XXX KDM clear have_ca and ua_pending on each LUN for
1450		 * this initiator.
1451		 */
1452	}
1453	softc->wwpn_iid[targ_port][iid].in_use = 1;
1454	softc->wwpn_iid[targ_port][iid].iid = iid;
1455	softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1456	softc->wwpn_iid[targ_port][iid].port = targ_port;
1457
1458bailout:
1459
1460	mtx_unlock(&softc->ctl_lock);
1461
1462	return (retval);
1463}
1464
1465static int
1466ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1467{
1468	return (0);
1469}
1470
1471static int
1472ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1473{
1474	return (0);
1475}
1476
1477/*
1478 * Data movement routine for the CTL ioctl frontend port.
1479 */
1480static int
1481ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1482{
1483	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1484	struct ctl_sg_entry ext_entry, kern_entry;
1485	int ext_sglen, ext_sg_entries, kern_sg_entries;
1486	int ext_sg_start, ext_offset;
1487	int len_to_copy, len_copied;
1488	int kern_watermark, ext_watermark;
1489	int ext_sglist_malloced;
1490	int i, j;
1491
1492	ext_sglist_malloced = 0;
1493	ext_sg_start = 0;
1494	ext_offset = 0;
1495
1496	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1497
1498	/*
1499	 * If this flag is set, fake the data transfer.
1500	 */
1501	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1502		ctsio->ext_data_filled = ctsio->ext_data_len;
1503		goto bailout;
1504	}
1505
1506	/*
1507	 * To simplify things here, if we have a single buffer, stick it in
1508	 * a S/G entry and just make it a single entry S/G list.
1509	 */
1510	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1511		int len_seen;
1512
1513		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1514
1515		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1516							   M_WAITOK);
1517		ext_sglist_malloced = 1;
1518		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1519				   ext_sglen) != 0) {
1520			ctl_set_internal_failure(ctsio,
1521						 /*sks_valid*/ 0,
1522						 /*retry_count*/ 0);
1523			goto bailout;
1524		}
1525		ext_sg_entries = ctsio->ext_sg_entries;
1526		len_seen = 0;
1527		for (i = 0; i < ext_sg_entries; i++) {
1528			if ((len_seen + ext_sglist[i].len) >=
1529			     ctsio->ext_data_filled) {
1530				ext_sg_start = i;
1531				ext_offset = ctsio->ext_data_filled - len_seen;
1532				break;
1533			}
1534			len_seen += ext_sglist[i].len;
1535		}
1536	} else {
1537		ext_sglist = &ext_entry;
1538		ext_sglist->addr = ctsio->ext_data_ptr;
1539		ext_sglist->len = ctsio->ext_data_len;
1540		ext_sg_entries = 1;
1541		ext_sg_start = 0;
1542		ext_offset = ctsio->ext_data_filled;
1543	}
1544
1545	if (ctsio->kern_sg_entries > 0) {
1546		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1547		kern_sg_entries = ctsio->kern_sg_entries;
1548	} else {
1549		kern_sglist = &kern_entry;
1550		kern_sglist->addr = ctsio->kern_data_ptr;
1551		kern_sglist->len = ctsio->kern_data_len;
1552		kern_sg_entries = 1;
1553	}
1554
1555
1556	kern_watermark = 0;
1557	ext_watermark = ext_offset;
1558	len_copied = 0;
1559	for (i = ext_sg_start, j = 0;
1560	     i < ext_sg_entries && j < kern_sg_entries;) {
1561		uint8_t *ext_ptr, *kern_ptr;
1562
1563		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1564				      kern_sglist[j].len - kern_watermark);
1565
1566		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1567		ext_ptr = ext_ptr + ext_watermark;
1568		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1569			/*
1570			 * XXX KDM fix this!
1571			 */
1572			panic("need to implement bus address support");
1573#if 0
1574			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1575#endif
1576		} else
1577			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1578		kern_ptr = kern_ptr + kern_watermark;
1579
1580		kern_watermark += len_to_copy;
1581		ext_watermark += len_to_copy;
1582
1583		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1584		     CTL_FLAG_DATA_IN) {
1585			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1586					 "bytes to user\n", len_to_copy));
1587			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1588					 "to %p\n", kern_ptr, ext_ptr));
1589			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1590				ctl_set_internal_failure(ctsio,
1591							 /*sks_valid*/ 0,
1592							 /*retry_count*/ 0);
1593				goto bailout;
1594			}
1595		} else {
1596			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1597					 "bytes from user\n", len_to_copy));
1598			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1599					 "to %p\n", ext_ptr, kern_ptr));
1600			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1601				ctl_set_internal_failure(ctsio,
1602							 /*sks_valid*/ 0,
1603							 /*retry_count*/0);
1604				goto bailout;
1605			}
1606		}
1607
1608		len_copied += len_to_copy;
1609
1610		if (ext_sglist[i].len == ext_watermark) {
1611			i++;
1612			ext_watermark = 0;
1613		}
1614
1615		if (kern_sglist[j].len == kern_watermark) {
1616			j++;
1617			kern_watermark = 0;
1618		}
1619	}
1620
1621	ctsio->ext_data_filled += len_copied;
1622
1623	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1624			 "kern_sg_entries: %d\n", ext_sg_entries,
1625			 kern_sg_entries));
1626	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1627			 "kern_data_len = %d\n", ctsio->ext_data_len,
1628			 ctsio->kern_data_len));
1629
1630
1631	/* XXX KDM set residual?? */
1632bailout:
1633
1634	if (ext_sglist_malloced != 0)
1635		free(ext_sglist, M_CTL);
1636
1637	return (CTL_RETVAL_COMPLETE);
1638}
1639
1640/*
1641 * Serialize a command that went down the "wrong" side, and so was sent to
1642 * this controller for execution.  The logic is a little different than the
1643 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1644 * sent back to the other side, but in the success case, we execute the
1645 * command on this side (XFER mode) or tell the other side to execute it
1646 * (SER_ONLY mode).
1647 */
1648static int
1649ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1650{
1651	struct ctl_softc *ctl_softc;
1652	union ctl_ha_msg msg_info;
1653	struct ctl_lun *lun;
1654	int retval = 0;
1655	uint32_t targ_lun;
1656
1657	ctl_softc = control_softc;
1658
1659	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1660	lun = ctl_softc->ctl_luns[targ_lun];
1661	if (lun==NULL)
1662	{
1663		/*
1664		 * Why isn't LUN defined? The other side wouldn't
1665		 * send a cmd if the LUN is undefined.
1666		 */
1667		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1668
1669		/* "Logical unit not supported" */
1670		ctl_set_sense_data(&msg_info.scsi.sense_data,
1671				   lun,
1672				   /*sense_format*/SSD_TYPE_NONE,
1673				   /*current_error*/ 1,
1674				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1675				   /*asc*/ 0x25,
1676				   /*ascq*/ 0x00,
1677				   SSD_ELEM_NONE);
1678
1679		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1680		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1681		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1682		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1683		msg_info.hdr.serializing_sc = NULL;
1684		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1685	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1686				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1687		}
1688		return(1);
1689
1690	}
1691
1692	mtx_lock(&lun->lun_lock);
1693    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1694
1695	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1696		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1697		 ooa_links))) {
1698	case CTL_ACTION_BLOCK:
1699		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1700		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1701				  blocked_links);
1702		break;
1703	case CTL_ACTION_PASS:
1704	case CTL_ACTION_SKIP:
1705		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1706			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1707			ctl_enqueue_rtr((union ctl_io *)ctsio);
1708		} else {
1709
1710			/* send msg back to other side */
1711			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1712			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1713			msg_info.hdr.msg_type = CTL_MSG_R2R;
1714#if 0
1715			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1716#endif
1717		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1718			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1719			}
1720		}
1721		break;
1722	case CTL_ACTION_OVERLAP:
1723		/* OVERLAPPED COMMANDS ATTEMPTED */
1724		ctl_set_sense_data(&msg_info.scsi.sense_data,
1725				   lun,
1726				   /*sense_format*/SSD_TYPE_NONE,
1727				   /*current_error*/ 1,
1728				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1729				   /*asc*/ 0x4E,
1730				   /*ascq*/ 0x00,
1731				   SSD_ELEM_NONE);
1732
1733		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1734		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1735		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1736		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1737		msg_info.hdr.serializing_sc = NULL;
1738		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1739#if 0
1740		printf("BAD JUJU:Major Bummer Overlap\n");
1741#endif
1742		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1743		retval = 1;
1744		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1745		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1746		}
1747		break;
1748	case CTL_ACTION_OVERLAP_TAG:
1749		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1750		ctl_set_sense_data(&msg_info.scsi.sense_data,
1751				   lun,
1752				   /*sense_format*/SSD_TYPE_NONE,
1753				   /*current_error*/ 1,
1754				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1755				   /*asc*/ 0x4D,
1756				   /*ascq*/ ctsio->tag_num & 0xff,
1757				   SSD_ELEM_NONE);
1758
1759		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1760		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1761		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1762		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1763		msg_info.hdr.serializing_sc = NULL;
1764		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1765#if 0
1766		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1767#endif
1768		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1769		retval = 1;
1770		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1771		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1772		}
1773		break;
1774	case CTL_ACTION_ERROR:
1775	default:
1776		/* "Internal target failure" */
1777		ctl_set_sense_data(&msg_info.scsi.sense_data,
1778				   lun,
1779				   /*sense_format*/SSD_TYPE_NONE,
1780				   /*current_error*/ 1,
1781				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1782				   /*asc*/ 0x44,
1783				   /*ascq*/ 0x00,
1784				   SSD_ELEM_NONE);
1785
1786		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1787		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1788		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1789		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1790		msg_info.hdr.serializing_sc = NULL;
1791		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1792#if 0
1793		printf("BAD JUJU:Major Bummer HW Error\n");
1794#endif
1795		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1796		retval = 1;
1797		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1798		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1799		}
1800		break;
1801	}
1802	mtx_unlock(&lun->lun_lock);
1803	return (retval);
1804}
1805
1806static int
1807ctl_ioctl_submit_wait(union ctl_io *io)
1808{
1809	struct ctl_fe_ioctl_params params;
1810	ctl_fe_ioctl_state last_state;
1811	int done, retval;
1812
1813	retval = 0;
1814
1815	bzero(&params, sizeof(params));
1816
1817	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1818	cv_init(&params.sem, "ctlioccv");
1819	params.state = CTL_IOCTL_INPROG;
1820	last_state = params.state;
1821
1822	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1823
1824	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1825
1826	/* This shouldn't happen */
1827	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1828		return (retval);
1829
1830	done = 0;
1831
1832	do {
1833		mtx_lock(&params.ioctl_mtx);
1834		/*
1835		 * Check the state here, and don't sleep if the state has
1836		 * already changed (i.e. wakeup has already occured, but we
1837		 * weren't waiting yet).
1838		 */
1839		if (params.state == last_state) {
1840			/* XXX KDM cv_wait_sig instead? */
1841			cv_wait(&params.sem, &params.ioctl_mtx);
1842		}
1843		last_state = params.state;
1844
1845		switch (params.state) {
1846		case CTL_IOCTL_INPROG:
1847			/* Why did we wake up? */
1848			/* XXX KDM error here? */
1849			mtx_unlock(&params.ioctl_mtx);
1850			break;
1851		case CTL_IOCTL_DATAMOVE:
1852			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1853
1854			/*
1855			 * change last_state back to INPROG to avoid
1856			 * deadlock on subsequent data moves.
1857			 */
1858			params.state = last_state = CTL_IOCTL_INPROG;
1859
1860			mtx_unlock(&params.ioctl_mtx);
1861			ctl_ioctl_do_datamove(&io->scsiio);
1862			/*
1863			 * Note that in some cases, most notably writes,
1864			 * this will queue the I/O and call us back later.
1865			 * In other cases, generally reads, this routine
1866			 * will immediately call back and wake us up,
1867			 * probably using our own context.
1868			 */
1869			io->scsiio.be_move_done(io);
1870			break;
1871		case CTL_IOCTL_DONE:
1872			mtx_unlock(&params.ioctl_mtx);
1873			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1874			done = 1;
1875			break;
1876		default:
1877			mtx_unlock(&params.ioctl_mtx);
1878			/* XXX KDM error here? */
1879			break;
1880		}
1881	} while (done == 0);
1882
1883	mtx_destroy(&params.ioctl_mtx);
1884	cv_destroy(&params.sem);
1885
1886	return (CTL_RETVAL_COMPLETE);
1887}
1888
1889static void
1890ctl_ioctl_datamove(union ctl_io *io)
1891{
1892	struct ctl_fe_ioctl_params *params;
1893
1894	params = (struct ctl_fe_ioctl_params *)
1895		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1896
1897	mtx_lock(&params->ioctl_mtx);
1898	params->state = CTL_IOCTL_DATAMOVE;
1899	cv_broadcast(&params->sem);
1900	mtx_unlock(&params->ioctl_mtx);
1901}
1902
1903static void
1904ctl_ioctl_done(union ctl_io *io)
1905{
1906	struct ctl_fe_ioctl_params *params;
1907
1908	params = (struct ctl_fe_ioctl_params *)
1909		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1910
1911	mtx_lock(&params->ioctl_mtx);
1912	params->state = CTL_IOCTL_DONE;
1913	cv_broadcast(&params->sem);
1914	mtx_unlock(&params->ioctl_mtx);
1915}
1916
1917static void
1918ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1919{
1920	struct ctl_fe_ioctl_startstop_info *sd_info;
1921
1922	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1923
1924	sd_info->hs_info.status = metatask->status;
1925	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1926	sd_info->hs_info.luns_complete =
1927		metatask->taskinfo.startstop.luns_complete;
1928	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1929
1930	cv_broadcast(&sd_info->sem);
1931}
1932
1933static void
1934ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1935{
1936	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1937
1938	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1939
1940	mtx_lock(fe_bbr_info->lock);
1941	fe_bbr_info->bbr_info->status = metatask->status;
1942	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1943	fe_bbr_info->wakeup_done = 1;
1944	mtx_unlock(fe_bbr_info->lock);
1945
1946	cv_broadcast(&fe_bbr_info->sem);
1947}
1948
1949/*
1950 * Returns 0 for success, errno for failure.
1951 */
1952static int
1953ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1954		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1955{
1956	union ctl_io *io;
1957	int retval;
1958
1959	retval = 0;
1960
1961	mtx_lock(&lun->lun_lock);
1962	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
1963	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
1964	     ooa_links)) {
1965		struct ctl_ooa_entry *entry;
1966
1967		/*
1968		 * If we've got more than we can fit, just count the
1969		 * remaining entries.
1970		 */
1971		if (*cur_fill_num >= ooa_hdr->alloc_num)
1972			continue;
1973
1974		entry = &kern_entries[*cur_fill_num];
1975
1976		entry->tag_num = io->scsiio.tag_num;
1977		entry->lun_num = lun->lun;
1978#ifdef CTL_TIME_IO
1979		entry->start_bt = io->io_hdr.start_bt;
1980#endif
1981		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
1982		entry->cdb_len = io->scsiio.cdb_len;
1983		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
1984			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
1985
1986		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
1987			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
1988
1989		if (io->io_hdr.flags & CTL_FLAG_ABORT)
1990			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
1991
1992		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
1993			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
1994
1995		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
1996			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
1997	}
1998	mtx_unlock(&lun->lun_lock);
1999
2000	return (retval);
2001}
2002
2003static void *
2004ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2005		 size_t error_str_len)
2006{
2007	void *kptr;
2008
2009	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2010
2011	if (copyin(user_addr, kptr, len) != 0) {
2012		snprintf(error_str, error_str_len, "Error copying %d bytes "
2013			 "from user address %p to kernel address %p", len,
2014			 user_addr, kptr);
2015		free(kptr, M_CTL);
2016		return (NULL);
2017	}
2018
2019	return (kptr);
2020}
2021
2022static void
2023ctl_free_args(int num_args, struct ctl_be_arg *args)
2024{
2025	int i;
2026
2027	if (args == NULL)
2028		return;
2029
2030	for (i = 0; i < num_args; i++) {
2031		free(args[i].kname, M_CTL);
2032		free(args[i].kvalue, M_CTL);
2033	}
2034
2035	free(args, M_CTL);
2036}
2037
2038static struct ctl_be_arg *
2039ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2040		char *error_str, size_t error_str_len)
2041{
2042	struct ctl_be_arg *args;
2043	int i;
2044
2045	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2046				error_str, error_str_len);
2047
2048	if (args == NULL)
2049		goto bailout;
2050
2051	for (i = 0; i < num_args; i++) {
2052		args[i].kname = NULL;
2053		args[i].kvalue = NULL;
2054	}
2055
2056	for (i = 0; i < num_args; i++) {
2057		uint8_t *tmpptr;
2058
2059		args[i].kname = ctl_copyin_alloc(args[i].name,
2060			args[i].namelen, error_str, error_str_len);
2061		if (args[i].kname == NULL)
2062			goto bailout;
2063
2064		if (args[i].kname[args[i].namelen - 1] != '\0') {
2065			snprintf(error_str, error_str_len, "Argument %d "
2066				 "name is not NUL-terminated", i);
2067			goto bailout;
2068		}
2069
2070		if (args[i].flags & CTL_BEARG_RD) {
2071			tmpptr = ctl_copyin_alloc(args[i].value,
2072				args[i].vallen, error_str, error_str_len);
2073			if (tmpptr == NULL)
2074				goto bailout;
2075			if ((args[i].flags & CTL_BEARG_ASCII)
2076			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2077				snprintf(error_str, error_str_len, "Argument "
2078				    "%d value is not NUL-terminated", i);
2079				goto bailout;
2080			}
2081			args[i].kvalue = tmpptr;
2082		} else {
2083			args[i].kvalue = malloc(args[i].vallen,
2084			    M_CTL, M_WAITOK | M_ZERO);
2085		}
2086	}
2087
2088	return (args);
2089bailout:
2090
2091	ctl_free_args(num_args, args);
2092
2093	return (NULL);
2094}
2095
2096static void
2097ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2098{
2099	int i;
2100
2101	for (i = 0; i < num_args; i++) {
2102		if (args[i].flags & CTL_BEARG_WR)
2103			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2104	}
2105}
2106
2107/*
2108 * Escape characters that are illegal or not recommended in XML.
2109 */
2110int
2111ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2112{
2113	int retval;
2114
2115	retval = 0;
2116
2117	for (; *str; str++) {
2118		switch (*str) {
2119		case '&':
2120			retval = sbuf_printf(sb, "&amp;");
2121			break;
2122		case '>':
2123			retval = sbuf_printf(sb, "&gt;");
2124			break;
2125		case '<':
2126			retval = sbuf_printf(sb, "&lt;");
2127			break;
2128		default:
2129			retval = sbuf_putc(sb, *str);
2130			break;
2131		}
2132
2133		if (retval != 0)
2134			break;
2135
2136	}
2137
2138	return (retval);
2139}
2140
2141static int
2142ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2143	  struct thread *td)
2144{
2145	struct ctl_softc *softc;
2146	int retval;
2147
2148	softc = control_softc;
2149
2150	retval = 0;
2151
2152	switch (cmd) {
2153	case CTL_IO: {
2154		union ctl_io *io;
2155		void *pool_tmp;
2156
2157		/*
2158		 * If we haven't been "enabled", don't allow any SCSI I/O
2159		 * to this FETD.
2160		 */
2161		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2162			retval = EPERM;
2163			break;
2164		}
2165
2166		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2167		if (io == NULL) {
2168			printf("ctl_ioctl: can't allocate ctl_io!\n");
2169			retval = ENOSPC;
2170			break;
2171		}
2172
2173		/*
2174		 * Need to save the pool reference so it doesn't get
2175		 * spammed by the user's ctl_io.
2176		 */
2177		pool_tmp = io->io_hdr.pool;
2178
2179		memcpy(io, (void *)addr, sizeof(*io));
2180
2181		io->io_hdr.pool = pool_tmp;
2182		/*
2183		 * No status yet, so make sure the status is set properly.
2184		 */
2185		io->io_hdr.status = CTL_STATUS_NONE;
2186
2187		/*
2188		 * The user sets the initiator ID, target and LUN IDs.
2189		 */
2190		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2191		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2192		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2193		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2194			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2195
2196		retval = ctl_ioctl_submit_wait(io);
2197
2198		if (retval != 0) {
2199			ctl_free_io(io);
2200			break;
2201		}
2202
2203		memcpy((void *)addr, io, sizeof(*io));
2204
2205		/* return this to our pool */
2206		ctl_free_io(io);
2207
2208		break;
2209	}
2210	case CTL_ENABLE_PORT:
2211	case CTL_DISABLE_PORT:
2212	case CTL_SET_PORT_WWNS: {
2213		struct ctl_port *port;
2214		struct ctl_port_entry *entry;
2215
2216		entry = (struct ctl_port_entry *)addr;
2217
2218		mtx_lock(&softc->ctl_lock);
2219		STAILQ_FOREACH(port, &softc->port_list, links) {
2220			int action, done;
2221
2222			action = 0;
2223			done = 0;
2224
2225			if ((entry->port_type == CTL_PORT_NONE)
2226			 && (entry->targ_port == port->targ_port)) {
2227				/*
2228				 * If the user only wants to enable or
2229				 * disable or set WWNs on a specific port,
2230				 * do the operation and we're done.
2231				 */
2232				action = 1;
2233				done = 1;
2234			} else if (entry->port_type & port->port_type) {
2235				/*
2236				 * Compare the user's type mask with the
2237				 * particular frontend type to see if we
2238				 * have a match.
2239				 */
2240				action = 1;
2241				done = 0;
2242
2243				/*
2244				 * Make sure the user isn't trying to set
2245				 * WWNs on multiple ports at the same time.
2246				 */
2247				if (cmd == CTL_SET_PORT_WWNS) {
2248					printf("%s: Can't set WWNs on "
2249					       "multiple ports\n", __func__);
2250					retval = EINVAL;
2251					break;
2252				}
2253			}
2254			if (action != 0) {
2255				/*
2256				 * XXX KDM we have to drop the lock here,
2257				 * because the online/offline operations
2258				 * can potentially block.  We need to
2259				 * reference count the frontends so they
2260				 * can't go away,
2261				 */
2262				mtx_unlock(&softc->ctl_lock);
2263
2264				if (cmd == CTL_ENABLE_PORT) {
2265					struct ctl_lun *lun;
2266
2267					STAILQ_FOREACH(lun, &softc->lun_list,
2268						       links) {
2269						port->lun_enable(port->targ_lun_arg,
2270						    lun->target,
2271						    lun->lun);
2272					}
2273
2274					ctl_port_online(port);
2275				} else if (cmd == CTL_DISABLE_PORT) {
2276					struct ctl_lun *lun;
2277
2278					ctl_port_offline(port);
2279
2280					STAILQ_FOREACH(lun, &softc->lun_list,
2281						       links) {
2282						port->lun_disable(
2283						    port->targ_lun_arg,
2284						    lun->target,
2285						    lun->lun);
2286					}
2287				}
2288
2289				mtx_lock(&softc->ctl_lock);
2290
2291				if (cmd == CTL_SET_PORT_WWNS)
2292					ctl_port_set_wwns(port,
2293					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2294					    1 : 0, entry->wwnn,
2295					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2296					    1 : 0, entry->wwpn);
2297			}
2298			if (done != 0)
2299				break;
2300		}
2301		mtx_unlock(&softc->ctl_lock);
2302		break;
2303	}
2304	case CTL_GET_PORT_LIST: {
2305		struct ctl_port *port;
2306		struct ctl_port_list *list;
2307		int i;
2308
2309		list = (struct ctl_port_list *)addr;
2310
2311		if (list->alloc_len != (list->alloc_num *
2312		    sizeof(struct ctl_port_entry))) {
2313			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2314			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2315			       "%zu\n", __func__, list->alloc_len,
2316			       list->alloc_num, sizeof(struct ctl_port_entry));
2317			retval = EINVAL;
2318			break;
2319		}
2320		list->fill_len = 0;
2321		list->fill_num = 0;
2322		list->dropped_num = 0;
2323		i = 0;
2324		mtx_lock(&softc->ctl_lock);
2325		STAILQ_FOREACH(port, &softc->port_list, links) {
2326			struct ctl_port_entry entry, *list_entry;
2327
2328			if (list->fill_num >= list->alloc_num) {
2329				list->dropped_num++;
2330				continue;
2331			}
2332
2333			entry.port_type = port->port_type;
2334			strlcpy(entry.port_name, port->port_name,
2335				sizeof(entry.port_name));
2336			entry.targ_port = port->targ_port;
2337			entry.physical_port = port->physical_port;
2338			entry.virtual_port = port->virtual_port;
2339			entry.wwnn = port->wwnn;
2340			entry.wwpn = port->wwpn;
2341			if (port->status & CTL_PORT_STATUS_ONLINE)
2342				entry.online = 1;
2343			else
2344				entry.online = 0;
2345
2346			list_entry = &list->entries[i];
2347
2348			retval = copyout(&entry, list_entry, sizeof(entry));
2349			if (retval != 0) {
2350				printf("%s: CTL_GET_PORT_LIST: copyout "
2351				       "returned %d\n", __func__, retval);
2352				break;
2353			}
2354			i++;
2355			list->fill_num++;
2356			list->fill_len += sizeof(entry);
2357		}
2358		mtx_unlock(&softc->ctl_lock);
2359
2360		/*
2361		 * If this is non-zero, we had a copyout fault, so there's
2362		 * probably no point in attempting to set the status inside
2363		 * the structure.
2364		 */
2365		if (retval != 0)
2366			break;
2367
2368		if (list->dropped_num > 0)
2369			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2370		else
2371			list->status = CTL_PORT_LIST_OK;
2372		break;
2373	}
2374	case CTL_DUMP_OOA: {
2375		struct ctl_lun *lun;
2376		union ctl_io *io;
2377		char printbuf[128];
2378		struct sbuf sb;
2379
2380		mtx_lock(&softc->ctl_lock);
2381		printf("Dumping OOA queues:\n");
2382		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2383			mtx_lock(&lun->lun_lock);
2384			for (io = (union ctl_io *)TAILQ_FIRST(
2385			     &lun->ooa_queue); io != NULL;
2386			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2387			     ooa_links)) {
2388				sbuf_new(&sb, printbuf, sizeof(printbuf),
2389					 SBUF_FIXEDLEN);
2390				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2391					    (intmax_t)lun->lun,
2392					    io->scsiio.tag_num,
2393					    (io->io_hdr.flags &
2394					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2395					    (io->io_hdr.flags &
2396					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2397					    (io->io_hdr.flags &
2398					    CTL_FLAG_ABORT) ? " ABORT" : "",
2399			                    (io->io_hdr.flags &
2400		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2401				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2402				sbuf_finish(&sb);
2403				printf("%s\n", sbuf_data(&sb));
2404			}
2405			mtx_unlock(&lun->lun_lock);
2406		}
2407		printf("OOA queues dump done\n");
2408		mtx_unlock(&softc->ctl_lock);
2409		break;
2410	}
2411	case CTL_GET_OOA: {
2412		struct ctl_lun *lun;
2413		struct ctl_ooa *ooa_hdr;
2414		struct ctl_ooa_entry *entries;
2415		uint32_t cur_fill_num;
2416
2417		ooa_hdr = (struct ctl_ooa *)addr;
2418
2419		if ((ooa_hdr->alloc_len == 0)
2420		 || (ooa_hdr->alloc_num == 0)) {
2421			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2422			       "must be non-zero\n", __func__,
2423			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2424			retval = EINVAL;
2425			break;
2426		}
2427
2428		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2429		    sizeof(struct ctl_ooa_entry))) {
2430			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2431			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2432			       __func__, ooa_hdr->alloc_len,
2433			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2434			retval = EINVAL;
2435			break;
2436		}
2437
2438		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2439		if (entries == NULL) {
2440			printf("%s: could not allocate %d bytes for OOA "
2441			       "dump\n", __func__, ooa_hdr->alloc_len);
2442			retval = ENOMEM;
2443			break;
2444		}
2445
2446		mtx_lock(&softc->ctl_lock);
2447		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2448		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2449		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2450			mtx_unlock(&softc->ctl_lock);
2451			free(entries, M_CTL);
2452			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2453			       __func__, (uintmax_t)ooa_hdr->lun_num);
2454			retval = EINVAL;
2455			break;
2456		}
2457
2458		cur_fill_num = 0;
2459
2460		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2461			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2462				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2463					ooa_hdr, entries);
2464				if (retval != 0)
2465					break;
2466			}
2467			if (retval != 0) {
2468				mtx_unlock(&softc->ctl_lock);
2469				free(entries, M_CTL);
2470				break;
2471			}
2472		} else {
2473			lun = softc->ctl_luns[ooa_hdr->lun_num];
2474
2475			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2476						    entries);
2477		}
2478		mtx_unlock(&softc->ctl_lock);
2479
2480		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2481		ooa_hdr->fill_len = ooa_hdr->fill_num *
2482			sizeof(struct ctl_ooa_entry);
2483		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2484		if (retval != 0) {
2485			printf("%s: error copying out %d bytes for OOA dump\n",
2486			       __func__, ooa_hdr->fill_len);
2487		}
2488
2489		getbintime(&ooa_hdr->cur_bt);
2490
2491		if (cur_fill_num > ooa_hdr->alloc_num) {
2492			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2493			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2494		} else {
2495			ooa_hdr->dropped_num = 0;
2496			ooa_hdr->status = CTL_OOA_OK;
2497		}
2498
2499		free(entries, M_CTL);
2500		break;
2501	}
2502	case CTL_CHECK_OOA: {
2503		union ctl_io *io;
2504		struct ctl_lun *lun;
2505		struct ctl_ooa_info *ooa_info;
2506
2507
2508		ooa_info = (struct ctl_ooa_info *)addr;
2509
2510		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2511			ooa_info->status = CTL_OOA_INVALID_LUN;
2512			break;
2513		}
2514		mtx_lock(&softc->ctl_lock);
2515		lun = softc->ctl_luns[ooa_info->lun_id];
2516		if (lun == NULL) {
2517			mtx_unlock(&softc->ctl_lock);
2518			ooa_info->status = CTL_OOA_INVALID_LUN;
2519			break;
2520		}
2521		mtx_lock(&lun->lun_lock);
2522		mtx_unlock(&softc->ctl_lock);
2523		ooa_info->num_entries = 0;
2524		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2525		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2526		     &io->io_hdr, ooa_links)) {
2527			ooa_info->num_entries++;
2528		}
2529		mtx_unlock(&lun->lun_lock);
2530
2531		ooa_info->status = CTL_OOA_SUCCESS;
2532
2533		break;
2534	}
2535	case CTL_HARD_START:
2536	case CTL_HARD_STOP: {
2537		struct ctl_fe_ioctl_startstop_info ss_info;
2538		struct cfi_metatask *metatask;
2539		struct mtx hs_mtx;
2540
2541		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2542
2543		cv_init(&ss_info.sem, "hard start/stop cv" );
2544
2545		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2546		if (metatask == NULL) {
2547			retval = ENOMEM;
2548			mtx_destroy(&hs_mtx);
2549			break;
2550		}
2551
2552		if (cmd == CTL_HARD_START)
2553			metatask->tasktype = CFI_TASK_STARTUP;
2554		else
2555			metatask->tasktype = CFI_TASK_SHUTDOWN;
2556
2557		metatask->callback = ctl_ioctl_hard_startstop_callback;
2558		metatask->callback_arg = &ss_info;
2559
2560		cfi_action(metatask);
2561
2562		/* Wait for the callback */
2563		mtx_lock(&hs_mtx);
2564		cv_wait_sig(&ss_info.sem, &hs_mtx);
2565		mtx_unlock(&hs_mtx);
2566
2567		/*
2568		 * All information has been copied from the metatask by the
2569		 * time cv_broadcast() is called, so we free the metatask here.
2570		 */
2571		cfi_free_metatask(metatask);
2572
2573		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2574
2575		mtx_destroy(&hs_mtx);
2576		break;
2577	}
2578	case CTL_BBRREAD: {
2579		struct ctl_bbrread_info *bbr_info;
2580		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2581		struct mtx bbr_mtx;
2582		struct cfi_metatask *metatask;
2583
2584		bbr_info = (struct ctl_bbrread_info *)addr;
2585
2586		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2587
2588		bzero(&bbr_mtx, sizeof(bbr_mtx));
2589		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2590
2591		fe_bbr_info.bbr_info = bbr_info;
2592		fe_bbr_info.lock = &bbr_mtx;
2593
2594		cv_init(&fe_bbr_info.sem, "BBR read cv");
2595		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2596
2597		if (metatask == NULL) {
2598			mtx_destroy(&bbr_mtx);
2599			cv_destroy(&fe_bbr_info.sem);
2600			retval = ENOMEM;
2601			break;
2602		}
2603		metatask->tasktype = CFI_TASK_BBRREAD;
2604		metatask->callback = ctl_ioctl_bbrread_callback;
2605		metatask->callback_arg = &fe_bbr_info;
2606		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2607		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2608		metatask->taskinfo.bbrread.len = bbr_info->len;
2609
2610		cfi_action(metatask);
2611
2612		mtx_lock(&bbr_mtx);
2613		while (fe_bbr_info.wakeup_done == 0)
2614			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2615		mtx_unlock(&bbr_mtx);
2616
2617		bbr_info->status = metatask->status;
2618		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2619		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2620		memcpy(&bbr_info->sense_data,
2621		       &metatask->taskinfo.bbrread.sense_data,
2622		       ctl_min(sizeof(bbr_info->sense_data),
2623			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2624
2625		cfi_free_metatask(metatask);
2626
2627		mtx_destroy(&bbr_mtx);
2628		cv_destroy(&fe_bbr_info.sem);
2629
2630		break;
2631	}
2632	case CTL_DELAY_IO: {
2633		struct ctl_io_delay_info *delay_info;
2634#ifdef CTL_IO_DELAY
2635		struct ctl_lun *lun;
2636#endif /* CTL_IO_DELAY */
2637
2638		delay_info = (struct ctl_io_delay_info *)addr;
2639
2640#ifdef CTL_IO_DELAY
2641		mtx_lock(&softc->ctl_lock);
2642
2643		if ((delay_info->lun_id > CTL_MAX_LUNS)
2644		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2645			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2646		} else {
2647			lun = softc->ctl_luns[delay_info->lun_id];
2648			mtx_lock(&lun->lun_lock);
2649
2650			delay_info->status = CTL_DELAY_STATUS_OK;
2651
2652			switch (delay_info->delay_type) {
2653			case CTL_DELAY_TYPE_CONT:
2654				break;
2655			case CTL_DELAY_TYPE_ONESHOT:
2656				break;
2657			default:
2658				delay_info->status =
2659					CTL_DELAY_STATUS_INVALID_TYPE;
2660				break;
2661			}
2662
2663			switch (delay_info->delay_loc) {
2664			case CTL_DELAY_LOC_DATAMOVE:
2665				lun->delay_info.datamove_type =
2666					delay_info->delay_type;
2667				lun->delay_info.datamove_delay =
2668					delay_info->delay_secs;
2669				break;
2670			case CTL_DELAY_LOC_DONE:
2671				lun->delay_info.done_type =
2672					delay_info->delay_type;
2673				lun->delay_info.done_delay =
2674					delay_info->delay_secs;
2675				break;
2676			default:
2677				delay_info->status =
2678					CTL_DELAY_STATUS_INVALID_LOC;
2679				break;
2680			}
2681			mtx_unlock(&lun->lun_lock);
2682		}
2683
2684		mtx_unlock(&softc->ctl_lock);
2685#else
2686		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2687#endif /* CTL_IO_DELAY */
2688		break;
2689	}
2690	case CTL_REALSYNC_SET: {
2691		int *syncstate;
2692
2693		syncstate = (int *)addr;
2694
2695		mtx_lock(&softc->ctl_lock);
2696		switch (*syncstate) {
2697		case 0:
2698			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2699			break;
2700		case 1:
2701			softc->flags |= CTL_FLAG_REAL_SYNC;
2702			break;
2703		default:
2704			retval = EINVAL;
2705			break;
2706		}
2707		mtx_unlock(&softc->ctl_lock);
2708		break;
2709	}
2710	case CTL_REALSYNC_GET: {
2711		int *syncstate;
2712
2713		syncstate = (int*)addr;
2714
2715		mtx_lock(&softc->ctl_lock);
2716		if (softc->flags & CTL_FLAG_REAL_SYNC)
2717			*syncstate = 1;
2718		else
2719			*syncstate = 0;
2720		mtx_unlock(&softc->ctl_lock);
2721
2722		break;
2723	}
2724	case CTL_SETSYNC:
2725	case CTL_GETSYNC: {
2726		struct ctl_sync_info *sync_info;
2727		struct ctl_lun *lun;
2728
2729		sync_info = (struct ctl_sync_info *)addr;
2730
2731		mtx_lock(&softc->ctl_lock);
2732		lun = softc->ctl_luns[sync_info->lun_id];
2733		if (lun == NULL) {
2734			mtx_unlock(&softc->ctl_lock);
2735			sync_info->status = CTL_GS_SYNC_NO_LUN;
2736		}
2737		/*
2738		 * Get or set the sync interval.  We're not bounds checking
2739		 * in the set case, hopefully the user won't do something
2740		 * silly.
2741		 */
2742		mtx_lock(&lun->lun_lock);
2743		mtx_unlock(&softc->ctl_lock);
2744		if (cmd == CTL_GETSYNC)
2745			sync_info->sync_interval = lun->sync_interval;
2746		else
2747			lun->sync_interval = sync_info->sync_interval;
2748		mtx_unlock(&lun->lun_lock);
2749
2750		sync_info->status = CTL_GS_SYNC_OK;
2751
2752		break;
2753	}
2754	case CTL_GETSTATS: {
2755		struct ctl_stats *stats;
2756		struct ctl_lun *lun;
2757		int i;
2758
2759		stats = (struct ctl_stats *)addr;
2760
2761		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2762		     stats->alloc_len) {
2763			stats->status = CTL_SS_NEED_MORE_SPACE;
2764			stats->num_luns = softc->num_luns;
2765			break;
2766		}
2767		/*
2768		 * XXX KDM no locking here.  If the LUN list changes,
2769		 * things can blow up.
2770		 */
2771		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2772		     i++, lun = STAILQ_NEXT(lun, links)) {
2773			retval = copyout(&lun->stats, &stats->lun_stats[i],
2774					 sizeof(lun->stats));
2775			if (retval != 0)
2776				break;
2777		}
2778		stats->num_luns = softc->num_luns;
2779		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2780				 softc->num_luns;
2781		stats->status = CTL_SS_OK;
2782#ifdef CTL_TIME_IO
2783		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2784#else
2785		stats->flags = CTL_STATS_FLAG_NONE;
2786#endif
2787		getnanouptime(&stats->timestamp);
2788		break;
2789	}
2790	case CTL_ERROR_INJECT: {
2791		struct ctl_error_desc *err_desc, *new_err_desc;
2792		struct ctl_lun *lun;
2793
2794		err_desc = (struct ctl_error_desc *)addr;
2795
2796		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2797				      M_WAITOK | M_ZERO);
2798		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2799
2800		mtx_lock(&softc->ctl_lock);
2801		lun = softc->ctl_luns[err_desc->lun_id];
2802		if (lun == NULL) {
2803			mtx_unlock(&softc->ctl_lock);
2804			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2805			       __func__, (uintmax_t)err_desc->lun_id);
2806			retval = EINVAL;
2807			break;
2808		}
2809		mtx_lock(&lun->lun_lock);
2810		mtx_unlock(&softc->ctl_lock);
2811
2812		/*
2813		 * We could do some checking here to verify the validity
2814		 * of the request, but given the complexity of error
2815		 * injection requests, the checking logic would be fairly
2816		 * complex.
2817		 *
2818		 * For now, if the request is invalid, it just won't get
2819		 * executed and might get deleted.
2820		 */
2821		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2822
2823		/*
2824		 * XXX KDM check to make sure the serial number is unique,
2825		 * in case we somehow manage to wrap.  That shouldn't
2826		 * happen for a very long time, but it's the right thing to
2827		 * do.
2828		 */
2829		new_err_desc->serial = lun->error_serial;
2830		err_desc->serial = lun->error_serial;
2831		lun->error_serial++;
2832
2833		mtx_unlock(&lun->lun_lock);
2834		break;
2835	}
2836	case CTL_ERROR_INJECT_DELETE: {
2837		struct ctl_error_desc *delete_desc, *desc, *desc2;
2838		struct ctl_lun *lun;
2839		int delete_done;
2840
2841		delete_desc = (struct ctl_error_desc *)addr;
2842		delete_done = 0;
2843
2844		mtx_lock(&softc->ctl_lock);
2845		lun = softc->ctl_luns[delete_desc->lun_id];
2846		if (lun == NULL) {
2847			mtx_unlock(&softc->ctl_lock);
2848			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2849			       __func__, (uintmax_t)delete_desc->lun_id);
2850			retval = EINVAL;
2851			break;
2852		}
2853		mtx_lock(&lun->lun_lock);
2854		mtx_unlock(&softc->ctl_lock);
2855		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2856			if (desc->serial != delete_desc->serial)
2857				continue;
2858
2859			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2860				      links);
2861			free(desc, M_CTL);
2862			delete_done = 1;
2863		}
2864		mtx_unlock(&lun->lun_lock);
2865		if (delete_done == 0) {
2866			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2867			       "error serial %ju on LUN %u\n", __func__,
2868			       delete_desc->serial, delete_desc->lun_id);
2869			retval = EINVAL;
2870			break;
2871		}
2872		break;
2873	}
2874	case CTL_DUMP_STRUCTS: {
2875		int i, j, k;
2876		struct ctl_port *port;
2877		struct ctl_frontend *fe;
2878
2879		printf("CTL IID to WWPN map start:\n");
2880		for (i = 0; i < CTL_MAX_PORTS; i++) {
2881			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2882				if (softc->wwpn_iid[i][j].in_use == 0)
2883					continue;
2884
2885				printf("port %d iid %u WWPN %#jx\n",
2886				       softc->wwpn_iid[i][j].port,
2887				       softc->wwpn_iid[i][j].iid,
2888				       (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2889			}
2890		}
2891		printf("CTL IID to WWPN map end\n");
2892		printf("CTL Persistent Reservation information start:\n");
2893		for (i = 0; i < CTL_MAX_LUNS; i++) {
2894			struct ctl_lun *lun;
2895
2896			lun = softc->ctl_luns[i];
2897
2898			if ((lun == NULL)
2899			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2900				continue;
2901
2902			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2903				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2904					if (lun->per_res[j+k].registered == 0)
2905						continue;
2906					printf("LUN %d port %d iid %d key "
2907					       "%#jx\n", i, j, k,
2908					       (uintmax_t)scsi_8btou64(
2909					       lun->per_res[j+k].res_key.key));
2910				}
2911			}
2912		}
2913		printf("CTL Persistent Reservation information end\n");
2914		printf("CTL Ports:\n");
2915		/*
2916		 * XXX KDM calling this without a lock.  We'd likely want
2917		 * to drop the lock before calling the frontend's dump
2918		 * routine anyway.
2919		 */
2920		STAILQ_FOREACH(port, &softc->port_list, links) {
2921			printf("Port %s Frontend %s Type %u pport %d vport %d WWNN "
2922			       "%#jx WWPN %#jx\n", port->port_name,
2923			       port->frontend->name, port->port_type,
2924			       port->physical_port, port->virtual_port,
2925			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2926		}
2927		printf("CTL Port information end\n");
2928		printf("CTL Frontends:\n");
2929		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2930			printf("Frontend %s\n", fe->name);
2931			if (fe->fe_dump != NULL)
2932				fe->fe_dump();
2933		}
2934		printf("CTL Frontend information end\n");
2935		break;
2936	}
2937	case CTL_LUN_REQ: {
2938		struct ctl_lun_req *lun_req;
2939		struct ctl_backend_driver *backend;
2940
2941		lun_req = (struct ctl_lun_req *)addr;
2942
2943		backend = ctl_backend_find(lun_req->backend);
2944		if (backend == NULL) {
2945			lun_req->status = CTL_LUN_ERROR;
2946			snprintf(lun_req->error_str,
2947				 sizeof(lun_req->error_str),
2948				 "Backend \"%s\" not found.",
2949				 lun_req->backend);
2950			break;
2951		}
2952		if (lun_req->num_be_args > 0) {
2953			lun_req->kern_be_args = ctl_copyin_args(
2954				lun_req->num_be_args,
2955				lun_req->be_args,
2956				lun_req->error_str,
2957				sizeof(lun_req->error_str));
2958			if (lun_req->kern_be_args == NULL) {
2959				lun_req->status = CTL_LUN_ERROR;
2960				break;
2961			}
2962		}
2963
2964		retval = backend->ioctl(dev, cmd, addr, flag, td);
2965
2966		if (lun_req->num_be_args > 0) {
2967			ctl_copyout_args(lun_req->num_be_args,
2968				      lun_req->kern_be_args);
2969			ctl_free_args(lun_req->num_be_args,
2970				      lun_req->kern_be_args);
2971		}
2972		break;
2973	}
2974	case CTL_LUN_LIST: {
2975		struct sbuf *sb;
2976		struct ctl_lun *lun;
2977		struct ctl_lun_list *list;
2978		struct ctl_option *opt;
2979
2980		list = (struct ctl_lun_list *)addr;
2981
2982		/*
2983		 * Allocate a fixed length sbuf here, based on the length
2984		 * of the user's buffer.  We could allocate an auto-extending
2985		 * buffer, and then tell the user how much larger our
2986		 * amount of data is than his buffer, but that presents
2987		 * some problems:
2988		 *
2989		 * 1.  The sbuf(9) routines use a blocking malloc, and so
2990		 *     we can't hold a lock while calling them with an
2991		 *     auto-extending buffer.
2992 		 *
2993		 * 2.  There is not currently a LUN reference counting
2994		 *     mechanism, outside of outstanding transactions on
2995		 *     the LUN's OOA queue.  So a LUN could go away on us
2996		 *     while we're getting the LUN number, backend-specific
2997		 *     information, etc.  Thus, given the way things
2998		 *     currently work, we need to hold the CTL lock while
2999		 *     grabbing LUN information.
3000		 *
3001		 * So, from the user's standpoint, the best thing to do is
3002		 * allocate what he thinks is a reasonable buffer length,
3003		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3004		 * double the buffer length and try again.  (And repeat
3005		 * that until he succeeds.)
3006		 */
3007		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3008		if (sb == NULL) {
3009			list->status = CTL_LUN_LIST_ERROR;
3010			snprintf(list->error_str, sizeof(list->error_str),
3011				 "Unable to allocate %d bytes for LUN list",
3012				 list->alloc_len);
3013			break;
3014		}
3015
3016		sbuf_printf(sb, "<ctllunlist>\n");
3017
3018		mtx_lock(&softc->ctl_lock);
3019		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3020			mtx_lock(&lun->lun_lock);
3021			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3022					     (uintmax_t)lun->lun);
3023
3024			/*
3025			 * Bail out as soon as we see that we've overfilled
3026			 * the buffer.
3027			 */
3028			if (retval != 0)
3029				break;
3030
3031			retval = sbuf_printf(sb, "\t<backend_type>%s"
3032					     "</backend_type>\n",
3033					     (lun->backend == NULL) ?  "none" :
3034					     lun->backend->name);
3035
3036			if (retval != 0)
3037				break;
3038
3039			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3040					     lun->be_lun->lun_type);
3041
3042			if (retval != 0)
3043				break;
3044
3045			if (lun->backend == NULL) {
3046				retval = sbuf_printf(sb, "</lun>\n");
3047				if (retval != 0)
3048					break;
3049				continue;
3050			}
3051
3052			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3053					     (lun->be_lun->maxlba > 0) ?
3054					     lun->be_lun->maxlba + 1 : 0);
3055
3056			if (retval != 0)
3057				break;
3058
3059			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3060					     lun->be_lun->blocksize);
3061
3062			if (retval != 0)
3063				break;
3064
3065			retval = sbuf_printf(sb, "\t<serial_number>");
3066
3067			if (retval != 0)
3068				break;
3069
3070			retval = ctl_sbuf_printf_esc(sb,
3071						     lun->be_lun->serial_num);
3072
3073			if (retval != 0)
3074				break;
3075
3076			retval = sbuf_printf(sb, "</serial_number>\n");
3077
3078			if (retval != 0)
3079				break;
3080
3081			retval = sbuf_printf(sb, "\t<device_id>");
3082
3083			if (retval != 0)
3084				break;
3085
3086			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3087
3088			if (retval != 0)
3089				break;
3090
3091			retval = sbuf_printf(sb, "</device_id>\n");
3092
3093			if (retval != 0)
3094				break;
3095
3096			if (lun->backend->lun_info != NULL) {
3097				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3098				if (retval != 0)
3099					break;
3100			}
3101			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3102				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3103				    opt->name, opt->value, opt->name);
3104				if (retval != 0)
3105					break;
3106			}
3107
3108			retval = sbuf_printf(sb, "</lun>\n");
3109
3110			if (retval != 0)
3111				break;
3112			mtx_unlock(&lun->lun_lock);
3113		}
3114		if (lun != NULL)
3115			mtx_unlock(&lun->lun_lock);
3116		mtx_unlock(&softc->ctl_lock);
3117
3118		if ((retval != 0)
3119		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3120			retval = 0;
3121			sbuf_delete(sb);
3122			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3123			snprintf(list->error_str, sizeof(list->error_str),
3124				 "Out of space, %d bytes is too small",
3125				 list->alloc_len);
3126			break;
3127		}
3128
3129		sbuf_finish(sb);
3130
3131		retval = copyout(sbuf_data(sb), list->lun_xml,
3132				 sbuf_len(sb) + 1);
3133
3134		list->fill_len = sbuf_len(sb) + 1;
3135		list->status = CTL_LUN_LIST_OK;
3136		sbuf_delete(sb);
3137		break;
3138	}
3139	case CTL_ISCSI: {
3140		struct ctl_iscsi *ci;
3141		struct ctl_frontend *fe;
3142
3143		ci = (struct ctl_iscsi *)addr;
3144
3145		fe = ctl_frontend_find("iscsi");
3146		if (fe == NULL) {
3147			ci->status = CTL_ISCSI_ERROR;
3148			snprintf(ci->error_str, sizeof(ci->error_str),
3149			    "Frontend \"iscsi\" not found.");
3150			break;
3151		}
3152
3153		retval = fe->ioctl(dev, cmd, addr, flag, td);
3154		break;
3155	}
3156	case CTL_PORT_REQ: {
3157		struct ctl_req *req;
3158		struct ctl_frontend *fe;
3159
3160		req = (struct ctl_req *)addr;
3161
3162		fe = ctl_frontend_find(req->driver);
3163		if (fe == NULL) {
3164			req->status = CTL_LUN_ERROR;
3165			snprintf(req->error_str, sizeof(req->error_str),
3166			    "Frontend \"%s\" not found.", req->driver);
3167			break;
3168		}
3169		if (req->num_args > 0) {
3170			req->kern_args = ctl_copyin_args(req->num_args,
3171			    req->args, req->error_str, sizeof(req->error_str));
3172			if (req->kern_args == NULL) {
3173				req->status = CTL_LUN_ERROR;
3174				break;
3175			}
3176		}
3177
3178		retval = fe->ioctl(dev, cmd, addr, flag, td);
3179
3180		if (req->num_args > 0) {
3181			ctl_copyout_args(req->num_args, req->kern_args);
3182			ctl_free_args(req->num_args, req->kern_args);
3183		}
3184		break;
3185	}
3186	case CTL_PORT_LIST: {
3187		struct sbuf *sb;
3188		struct ctl_port *port;
3189		struct ctl_lun_list *list;
3190		struct ctl_option *opt;
3191
3192		list = (struct ctl_lun_list *)addr;
3193
3194		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3195		if (sb == NULL) {
3196			list->status = CTL_LUN_LIST_ERROR;
3197			snprintf(list->error_str, sizeof(list->error_str),
3198				 "Unable to allocate %d bytes for LUN list",
3199				 list->alloc_len);
3200			break;
3201		}
3202
3203		sbuf_printf(sb, "<ctlportlist>\n");
3204
3205		mtx_lock(&softc->ctl_lock);
3206		STAILQ_FOREACH(port, &softc->port_list, links) {
3207			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3208					     (uintmax_t)port->targ_port);
3209
3210			/*
3211			 * Bail out as soon as we see that we've overfilled
3212			 * the buffer.
3213			 */
3214			if (retval != 0)
3215				break;
3216
3217			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3218			    "</frontend_type>\n", port->frontend->name);
3219			if (retval != 0)
3220				break;
3221
3222			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3223					     port->port_type);
3224			if (retval != 0)
3225				break;
3226
3227			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3228			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3229			if (retval != 0)
3230				break;
3231
3232			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3233			    port->port_name);
3234			if (retval != 0)
3235				break;
3236
3237			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3238			    port->physical_port);
3239			if (retval != 0)
3240				break;
3241
3242			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3243			    port->virtual_port);
3244			if (retval != 0)
3245				break;
3246
3247			retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3248			    (uintmax_t)port->wwnn);
3249			if (retval != 0)
3250				break;
3251
3252			retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3253			    (uintmax_t)port->wwpn);
3254			if (retval != 0)
3255				break;
3256
3257			if (port->port_info != NULL) {
3258				retval = port->port_info(port->onoff_arg, sb);
3259				if (retval != 0)
3260					break;
3261			}
3262			STAILQ_FOREACH(opt, &port->options, links) {
3263				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3264				    opt->name, opt->value, opt->name);
3265				if (retval != 0)
3266					break;
3267			}
3268
3269			retval = sbuf_printf(sb, "</targ_port>\n");
3270			if (retval != 0)
3271				break;
3272		}
3273		mtx_unlock(&softc->ctl_lock);
3274
3275		if ((retval != 0)
3276		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3277			retval = 0;
3278			sbuf_delete(sb);
3279			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3280			snprintf(list->error_str, sizeof(list->error_str),
3281				 "Out of space, %d bytes is too small",
3282				 list->alloc_len);
3283			break;
3284		}
3285
3286		sbuf_finish(sb);
3287
3288		retval = copyout(sbuf_data(sb), list->lun_xml,
3289				 sbuf_len(sb) + 1);
3290
3291		list->fill_len = sbuf_len(sb) + 1;
3292		list->status = CTL_LUN_LIST_OK;
3293		sbuf_delete(sb);
3294		break;
3295	}
3296	default: {
3297		/* XXX KDM should we fix this? */
3298#if 0
3299		struct ctl_backend_driver *backend;
3300		unsigned int type;
3301		int found;
3302
3303		found = 0;
3304
3305		/*
3306		 * We encode the backend type as the ioctl type for backend
3307		 * ioctls.  So parse it out here, and then search for a
3308		 * backend of this type.
3309		 */
3310		type = _IOC_TYPE(cmd);
3311
3312		STAILQ_FOREACH(backend, &softc->be_list, links) {
3313			if (backend->type == type) {
3314				found = 1;
3315				break;
3316			}
3317		}
3318		if (found == 0) {
3319			printf("ctl: unknown ioctl command %#lx or backend "
3320			       "%d\n", cmd, type);
3321			retval = EINVAL;
3322			break;
3323		}
3324		retval = backend->ioctl(dev, cmd, addr, flag, td);
3325#endif
3326		retval = ENOTTY;
3327		break;
3328	}
3329	}
3330	return (retval);
3331}
3332
3333uint32_t
3334ctl_get_initindex(struct ctl_nexus *nexus)
3335{
3336	if (nexus->targ_port < CTL_MAX_PORTS)
3337		return (nexus->initid.id +
3338			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3339	else
3340		return (nexus->initid.id +
3341		       ((nexus->targ_port - CTL_MAX_PORTS) *
3342			CTL_MAX_INIT_PER_PORT));
3343}
3344
3345uint32_t
3346ctl_get_resindex(struct ctl_nexus *nexus)
3347{
3348	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3349}
3350
3351uint32_t
3352ctl_port_idx(int port_num)
3353{
3354	if (port_num < CTL_MAX_PORTS)
3355		return(port_num);
3356	else
3357		return(port_num - CTL_MAX_PORTS);
3358}
3359
3360static uint32_t
3361ctl_map_lun(int port_num, uint32_t lun_id)
3362{
3363	struct ctl_port *port;
3364
3365	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3366	if (port == NULL)
3367		return (UINT32_MAX);
3368	if (port->lun_map == NULL)
3369		return (lun_id);
3370	return (port->lun_map(port->targ_lun_arg, lun_id));
3371}
3372
3373static uint32_t
3374ctl_map_lun_back(int port_num, uint32_t lun_id)
3375{
3376	struct ctl_port *port;
3377	uint32_t i;
3378
3379	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3380	if (port->lun_map == NULL)
3381		return (lun_id);
3382	for (i = 0; i < CTL_MAX_LUNS; i++) {
3383		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3384			return (i);
3385	}
3386	return (UINT32_MAX);
3387}
3388
3389/*
3390 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3391 * that are a power of 2.
3392 */
3393int
3394ctl_ffz(uint32_t *mask, uint32_t size)
3395{
3396	uint32_t num_chunks, num_pieces;
3397	int i, j;
3398
3399	num_chunks = (size >> 5);
3400	if (num_chunks == 0)
3401		num_chunks++;
3402	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3403
3404	for (i = 0; i < num_chunks; i++) {
3405		for (j = 0; j < num_pieces; j++) {
3406			if ((mask[i] & (1 << j)) == 0)
3407				return ((i << 5) + j);
3408		}
3409	}
3410
3411	return (-1);
3412}
3413
3414int
3415ctl_set_mask(uint32_t *mask, uint32_t bit)
3416{
3417	uint32_t chunk, piece;
3418
3419	chunk = bit >> 5;
3420	piece = bit % (sizeof(uint32_t) * 8);
3421
3422	if ((mask[chunk] & (1 << piece)) != 0)
3423		return (-1);
3424	else
3425		mask[chunk] |= (1 << piece);
3426
3427	return (0);
3428}
3429
3430int
3431ctl_clear_mask(uint32_t *mask, uint32_t bit)
3432{
3433	uint32_t chunk, piece;
3434
3435	chunk = bit >> 5;
3436	piece = bit % (sizeof(uint32_t) * 8);
3437
3438	if ((mask[chunk] & (1 << piece)) == 0)
3439		return (-1);
3440	else
3441		mask[chunk] &= ~(1 << piece);
3442
3443	return (0);
3444}
3445
3446int
3447ctl_is_set(uint32_t *mask, uint32_t bit)
3448{
3449	uint32_t chunk, piece;
3450
3451	chunk = bit >> 5;
3452	piece = bit % (sizeof(uint32_t) * 8);
3453
3454	if ((mask[chunk] & (1 << piece)) == 0)
3455		return (0);
3456	else
3457		return (1);
3458}
3459
3460#ifdef unused
3461/*
3462 * The bus, target and lun are optional, they can be filled in later.
3463 * can_wait is used to determine whether we can wait on the malloc or not.
3464 */
3465union ctl_io*
3466ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3467	      uint32_t targ_lun, int can_wait)
3468{
3469	union ctl_io *io;
3470
3471	if (can_wait)
3472		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3473	else
3474		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3475
3476	if (io != NULL) {
3477		io->io_hdr.io_type = io_type;
3478		io->io_hdr.targ_port = targ_port;
3479		/*
3480		 * XXX KDM this needs to change/go away.  We need to move
3481		 * to a preallocated pool of ctl_scsiio structures.
3482		 */
3483		io->io_hdr.nexus.targ_target.id = targ_target;
3484		io->io_hdr.nexus.targ_lun = targ_lun;
3485	}
3486
3487	return (io);
3488}
3489
3490void
3491ctl_kfree_io(union ctl_io *io)
3492{
3493	free(io, M_CTL);
3494}
3495#endif /* unused */
3496
3497/*
3498 * ctl_softc, pool_type, total_ctl_io are passed in.
3499 * npool is passed out.
3500 */
3501int
3502ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3503		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3504{
3505	uint32_t i;
3506	union ctl_io *cur_io, *next_io;
3507	struct ctl_io_pool *pool;
3508	int retval;
3509
3510	retval = 0;
3511
3512	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3513					    M_NOWAIT | M_ZERO);
3514	if (pool == NULL) {
3515		retval = ENOMEM;
3516		goto bailout;
3517	}
3518
3519	pool->type = pool_type;
3520	pool->ctl_softc = ctl_softc;
3521
3522	mtx_lock(&ctl_softc->pool_lock);
3523	pool->id = ctl_softc->cur_pool_id++;
3524	mtx_unlock(&ctl_softc->pool_lock);
3525
3526	pool->flags = CTL_POOL_FLAG_NONE;
3527	pool->refcount = 1;		/* Reference for validity. */
3528	STAILQ_INIT(&pool->free_queue);
3529
3530	/*
3531	 * XXX KDM other options here:
3532	 * - allocate a page at a time
3533	 * - allocate one big chunk of memory.
3534	 * Page allocation might work well, but would take a little more
3535	 * tracking.
3536	 */
3537	for (i = 0; i < total_ctl_io; i++) {
3538		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3539						M_NOWAIT);
3540		if (cur_io == NULL) {
3541			retval = ENOMEM;
3542			break;
3543		}
3544		cur_io->io_hdr.pool = pool;
3545		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3546		pool->total_ctl_io++;
3547		pool->free_ctl_io++;
3548	}
3549
3550	if (retval != 0) {
3551		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3552		     cur_io != NULL; cur_io = next_io) {
3553			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3554							      links);
3555			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3556				      ctl_io_hdr, links);
3557			free(cur_io, M_CTLIO);
3558		}
3559
3560		free(pool, M_CTL);
3561		goto bailout;
3562	}
3563	mtx_lock(&ctl_softc->pool_lock);
3564	ctl_softc->num_pools++;
3565	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3566	/*
3567	 * Increment our usage count if this is an external consumer, so we
3568	 * can't get unloaded until the external consumer (most likely a
3569	 * FETD) unloads and frees his pool.
3570	 *
3571	 * XXX KDM will this increment the caller's module use count, or
3572	 * mine?
3573	 */
3574#if 0
3575	if ((pool_type != CTL_POOL_EMERGENCY)
3576	 && (pool_type != CTL_POOL_INTERNAL)
3577	 && (pool_type != CTL_POOL_4OTHERSC))
3578		MOD_INC_USE_COUNT;
3579#endif
3580
3581	mtx_unlock(&ctl_softc->pool_lock);
3582
3583	*npool = pool;
3584
3585bailout:
3586
3587	return (retval);
3588}
3589
3590static int
3591ctl_pool_acquire(struct ctl_io_pool *pool)
3592{
3593
3594	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3595
3596	if (pool->flags & CTL_POOL_FLAG_INVALID)
3597		return (EINVAL);
3598
3599	pool->refcount++;
3600
3601	return (0);
3602}
3603
3604static void
3605ctl_pool_release(struct ctl_io_pool *pool)
3606{
3607	struct ctl_softc *ctl_softc = pool->ctl_softc;
3608	union ctl_io *io;
3609
3610	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3611
3612	if (--pool->refcount != 0)
3613		return;
3614
3615	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3616		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3617			      links);
3618		free(io, M_CTLIO);
3619	}
3620
3621	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3622	ctl_softc->num_pools--;
3623
3624	/*
3625	 * XXX KDM will this decrement the caller's usage count or mine?
3626	 */
3627#if 0
3628	if ((pool->type != CTL_POOL_EMERGENCY)
3629	 && (pool->type != CTL_POOL_INTERNAL)
3630	 && (pool->type != CTL_POOL_4OTHERSC))
3631		MOD_DEC_USE_COUNT;
3632#endif
3633
3634	free(pool, M_CTL);
3635}
3636
3637void
3638ctl_pool_free(struct ctl_io_pool *pool)
3639{
3640	struct ctl_softc *ctl_softc;
3641
3642	if (pool == NULL)
3643		return;
3644
3645	ctl_softc = pool->ctl_softc;
3646	mtx_lock(&ctl_softc->pool_lock);
3647	pool->flags |= CTL_POOL_FLAG_INVALID;
3648	ctl_pool_release(pool);
3649	mtx_unlock(&ctl_softc->pool_lock);
3650}
3651
3652/*
3653 * This routine does not block (except for spinlocks of course).
3654 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3655 * possible.
3656 */
3657union ctl_io *
3658ctl_alloc_io(void *pool_ref)
3659{
3660	union ctl_io *io;
3661	struct ctl_softc *ctl_softc;
3662	struct ctl_io_pool *pool, *npool;
3663	struct ctl_io_pool *emergency_pool;
3664
3665	pool = (struct ctl_io_pool *)pool_ref;
3666
3667	if (pool == NULL) {
3668		printf("%s: pool is NULL\n", __func__);
3669		return (NULL);
3670	}
3671
3672	emergency_pool = NULL;
3673
3674	ctl_softc = pool->ctl_softc;
3675
3676	mtx_lock(&ctl_softc->pool_lock);
3677	/*
3678	 * First, try to get the io structure from the user's pool.
3679	 */
3680	if (ctl_pool_acquire(pool) == 0) {
3681		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3682		if (io != NULL) {
3683			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3684			pool->total_allocated++;
3685			pool->free_ctl_io--;
3686			mtx_unlock(&ctl_softc->pool_lock);
3687			return (io);
3688		} else
3689			ctl_pool_release(pool);
3690	}
3691	/*
3692	 * If he doesn't have any io structures left, search for an
3693	 * emergency pool and grab one from there.
3694	 */
3695	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3696		if (npool->type != CTL_POOL_EMERGENCY)
3697			continue;
3698
3699		if (ctl_pool_acquire(npool) != 0)
3700			continue;
3701
3702		emergency_pool = npool;
3703
3704		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3705		if (io != NULL) {
3706			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3707			npool->total_allocated++;
3708			npool->free_ctl_io--;
3709			mtx_unlock(&ctl_softc->pool_lock);
3710			return (io);
3711		} else
3712			ctl_pool_release(npool);
3713	}
3714
3715	/* Drop the spinlock before we malloc */
3716	mtx_unlock(&ctl_softc->pool_lock);
3717
3718	/*
3719	 * The emergency pool (if it exists) didn't have one, so try an
3720	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3721	 */
3722	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3723	if (io != NULL) {
3724		/*
3725		 * If the emergency pool exists but is empty, add this
3726		 * ctl_io to its list when it gets freed.
3727		 */
3728		if (emergency_pool != NULL) {
3729			mtx_lock(&ctl_softc->pool_lock);
3730			if (ctl_pool_acquire(emergency_pool) == 0) {
3731				io->io_hdr.pool = emergency_pool;
3732				emergency_pool->total_ctl_io++;
3733				/*
3734				 * Need to bump this, otherwise
3735				 * total_allocated and total_freed won't
3736				 * match when we no longer have anything
3737				 * outstanding.
3738				 */
3739				emergency_pool->total_allocated++;
3740			}
3741			mtx_unlock(&ctl_softc->pool_lock);
3742		} else
3743			io->io_hdr.pool = NULL;
3744	}
3745
3746	return (io);
3747}
3748
3749void
3750ctl_free_io(union ctl_io *io)
3751{
3752	if (io == NULL)
3753		return;
3754
3755	/*
3756	 * If this ctl_io has a pool, return it to that pool.
3757	 */
3758	if (io->io_hdr.pool != NULL) {
3759		struct ctl_io_pool *pool;
3760
3761		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3762		mtx_lock(&pool->ctl_softc->pool_lock);
3763		io->io_hdr.io_type = 0xff;
3764		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3765		pool->total_freed++;
3766		pool->free_ctl_io++;
3767		ctl_pool_release(pool);
3768		mtx_unlock(&pool->ctl_softc->pool_lock);
3769	} else {
3770		/*
3771		 * Otherwise, just free it.  We probably malloced it and
3772		 * the emergency pool wasn't available.
3773		 */
3774		free(io, M_CTLIO);
3775	}
3776
3777}
3778
3779void
3780ctl_zero_io(union ctl_io *io)
3781{
3782	void *pool_ref;
3783
3784	if (io == NULL)
3785		return;
3786
3787	/*
3788	 * May need to preserve linked list pointers at some point too.
3789	 */
3790	pool_ref = io->io_hdr.pool;
3791
3792	memset(io, 0, sizeof(*io));
3793
3794	io->io_hdr.pool = pool_ref;
3795}
3796
3797/*
3798 * This routine is currently used for internal copies of ctl_ios that need
3799 * to persist for some reason after we've already returned status to the
3800 * FETD.  (Thus the flag set.)
3801 *
3802 * XXX XXX
3803 * Note that this makes a blind copy of all fields in the ctl_io, except
3804 * for the pool reference.  This includes any memory that has been
3805 * allocated!  That memory will no longer be valid after done has been
3806 * called, so this would be VERY DANGEROUS for command that actually does
3807 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3808 * start and stop commands, which don't transfer any data, so this is not a
3809 * problem.  If it is used for anything else, the caller would also need to
3810 * allocate data buffer space and this routine would need to be modified to
3811 * copy the data buffer(s) as well.
3812 */
3813void
3814ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3815{
3816	void *pool_ref;
3817
3818	if ((src == NULL)
3819	 || (dest == NULL))
3820		return;
3821
3822	/*
3823	 * May need to preserve linked list pointers at some point too.
3824	 */
3825	pool_ref = dest->io_hdr.pool;
3826
3827	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3828
3829	dest->io_hdr.pool = pool_ref;
3830	/*
3831	 * We need to know that this is an internal copy, and doesn't need
3832	 * to get passed back to the FETD that allocated it.
3833	 */
3834	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3835}
3836
3837#ifdef NEEDTOPORT
3838static void
3839ctl_update_power_subpage(struct copan_power_subpage *page)
3840{
3841	int num_luns, num_partitions, config_type;
3842	struct ctl_softc *softc;
3843	cs_BOOL_t aor_present, shelf_50pct_power;
3844	cs_raidset_personality_t rs_type;
3845	int max_active_luns;
3846
3847	softc = control_softc;
3848
3849	/* subtract out the processor LUN */
3850	num_luns = softc->num_luns - 1;
3851	/*
3852	 * Default to 7 LUNs active, which was the only number we allowed
3853	 * in the past.
3854	 */
3855	max_active_luns = 7;
3856
3857	num_partitions = config_GetRsPartitionInfo();
3858	config_type = config_GetConfigType();
3859	shelf_50pct_power = config_GetShelfPowerMode();
3860	aor_present = config_IsAorRsPresent();
3861
3862	rs_type = ddb_GetRsRaidType(1);
3863	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3864	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3865		EPRINT(0, "Unsupported RS type %d!", rs_type);
3866	}
3867
3868
3869	page->total_luns = num_luns;
3870
3871	switch (config_type) {
3872	case 40:
3873		/*
3874		 * In a 40 drive configuration, it doesn't matter what DC
3875		 * cards we have, whether we have AOR enabled or not,
3876		 * partitioning or not, or what type of RAIDset we have.
3877		 * In that scenario, we can power up every LUN we present
3878		 * to the user.
3879		 */
3880		max_active_luns = num_luns;
3881
3882		break;
3883	case 64:
3884		if (shelf_50pct_power == CS_FALSE) {
3885			/* 25% power */
3886			if (aor_present == CS_TRUE) {
3887				if (rs_type ==
3888				     CS_RAIDSET_PERSONALITY_RAID5) {
3889					max_active_luns = 7;
3890				} else if (rs_type ==
3891					 CS_RAIDSET_PERSONALITY_RAID1){
3892					max_active_luns = 14;
3893				} else {
3894					/* XXX KDM now what?? */
3895				}
3896			} else {
3897				if (rs_type ==
3898				     CS_RAIDSET_PERSONALITY_RAID5) {
3899					max_active_luns = 8;
3900				} else if (rs_type ==
3901					 CS_RAIDSET_PERSONALITY_RAID1){
3902					max_active_luns = 16;
3903				} else {
3904					/* XXX KDM now what?? */
3905				}
3906			}
3907		} else {
3908			/* 50% power */
3909			/*
3910			 * With 50% power in a 64 drive configuration, we
3911			 * can power all LUNs we present.
3912			 */
3913			max_active_luns = num_luns;
3914		}
3915		break;
3916	case 112:
3917		if (shelf_50pct_power == CS_FALSE) {
3918			/* 25% power */
3919			if (aor_present == CS_TRUE) {
3920				if (rs_type ==
3921				     CS_RAIDSET_PERSONALITY_RAID5) {
3922					max_active_luns = 7;
3923				} else if (rs_type ==
3924					 CS_RAIDSET_PERSONALITY_RAID1){
3925					max_active_luns = 14;
3926				} else {
3927					/* XXX KDM now what?? */
3928				}
3929			} else {
3930				if (rs_type ==
3931				     CS_RAIDSET_PERSONALITY_RAID5) {
3932					max_active_luns = 8;
3933				} else if (rs_type ==
3934					 CS_RAIDSET_PERSONALITY_RAID1){
3935					max_active_luns = 16;
3936				} else {
3937					/* XXX KDM now what?? */
3938				}
3939			}
3940		} else {
3941			/* 50% power */
3942			if (aor_present == CS_TRUE) {
3943				if (rs_type ==
3944				     CS_RAIDSET_PERSONALITY_RAID5) {
3945					max_active_luns = 14;
3946				} else if (rs_type ==
3947					 CS_RAIDSET_PERSONALITY_RAID1){
3948					/*
3949					 * We're assuming here that disk
3950					 * caching is enabled, and so we're
3951					 * able to power up half of each
3952					 * LUN, and cache all writes.
3953					 */
3954					max_active_luns = num_luns;
3955				} else {
3956					/* XXX KDM now what?? */
3957				}
3958			} else {
3959				if (rs_type ==
3960				     CS_RAIDSET_PERSONALITY_RAID5) {
3961					max_active_luns = 15;
3962				} else if (rs_type ==
3963					 CS_RAIDSET_PERSONALITY_RAID1){
3964					max_active_luns = 30;
3965				} else {
3966					/* XXX KDM now what?? */
3967				}
3968			}
3969		}
3970		break;
3971	default:
3972		/*
3973		 * In this case, we have an unknown configuration, so we
3974		 * just use the default from above.
3975		 */
3976		break;
3977	}
3978
3979	page->max_active_luns = max_active_luns;
3980#if 0
3981	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3982	       page->total_luns, page->max_active_luns);
3983#endif
3984}
3985#endif /* NEEDTOPORT */
3986
3987/*
3988 * This routine could be used in the future to load default and/or saved
3989 * mode page parameters for a particuar lun.
3990 */
3991static int
3992ctl_init_page_index(struct ctl_lun *lun)
3993{
3994	int i;
3995	struct ctl_page_index *page_index;
3996	struct ctl_softc *softc;
3997
3998	memcpy(&lun->mode_pages.index, page_index_template,
3999	       sizeof(page_index_template));
4000
4001	softc = lun->ctl_softc;
4002
4003	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4004
4005		page_index = &lun->mode_pages.index[i];
4006		/*
4007		 * If this is a disk-only mode page, there's no point in
4008		 * setting it up.  For some pages, we have to have some
4009		 * basic information about the disk in order to calculate the
4010		 * mode page data.
4011		 */
4012		if ((lun->be_lun->lun_type != T_DIRECT)
4013		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4014			continue;
4015
4016		switch (page_index->page_code & SMPH_PC_MASK) {
4017		case SMS_FORMAT_DEVICE_PAGE: {
4018			struct scsi_format_page *format_page;
4019
4020			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4021				panic("subpage is incorrect!");
4022
4023			/*
4024			 * Sectors per track are set above.  Bytes per
4025			 * sector need to be set here on a per-LUN basis.
4026			 */
4027			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4028			       &format_page_default,
4029			       sizeof(format_page_default));
4030			memcpy(&lun->mode_pages.format_page[
4031			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4032			       sizeof(format_page_changeable));
4033			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4034			       &format_page_default,
4035			       sizeof(format_page_default));
4036			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4037			       &format_page_default,
4038			       sizeof(format_page_default));
4039
4040			format_page = &lun->mode_pages.format_page[
4041				CTL_PAGE_CURRENT];
4042			scsi_ulto2b(lun->be_lun->blocksize,
4043				    format_page->bytes_per_sector);
4044
4045			format_page = &lun->mode_pages.format_page[
4046				CTL_PAGE_DEFAULT];
4047			scsi_ulto2b(lun->be_lun->blocksize,
4048				    format_page->bytes_per_sector);
4049
4050			format_page = &lun->mode_pages.format_page[
4051				CTL_PAGE_SAVED];
4052			scsi_ulto2b(lun->be_lun->blocksize,
4053				    format_page->bytes_per_sector);
4054
4055			page_index->page_data =
4056				(uint8_t *)lun->mode_pages.format_page;
4057			break;
4058		}
4059		case SMS_RIGID_DISK_PAGE: {
4060			struct scsi_rigid_disk_page *rigid_disk_page;
4061			uint32_t sectors_per_cylinder;
4062			uint64_t cylinders;
4063#ifndef	__XSCALE__
4064			int shift;
4065#endif /* !__XSCALE__ */
4066
4067			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4068				panic("invalid subpage value %d",
4069				      page_index->subpage);
4070
4071			/*
4072			 * Rotation rate and sectors per track are set
4073			 * above.  We calculate the cylinders here based on
4074			 * capacity.  Due to the number of heads and
4075			 * sectors per track we're using, smaller arrays
4076			 * may turn out to have 0 cylinders.  Linux and
4077			 * FreeBSD don't pay attention to these mode pages
4078			 * to figure out capacity, but Solaris does.  It
4079			 * seems to deal with 0 cylinders just fine, and
4080			 * works out a fake geometry based on the capacity.
4081			 */
4082			memcpy(&lun->mode_pages.rigid_disk_page[
4083			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4084			       sizeof(rigid_disk_page_default));
4085			memcpy(&lun->mode_pages.rigid_disk_page[
4086			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4087			       sizeof(rigid_disk_page_changeable));
4088			memcpy(&lun->mode_pages.rigid_disk_page[
4089			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4090			       sizeof(rigid_disk_page_default));
4091			memcpy(&lun->mode_pages.rigid_disk_page[
4092			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4093			       sizeof(rigid_disk_page_default));
4094
4095			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4096				CTL_DEFAULT_HEADS;
4097
4098			/*
4099			 * The divide method here will be more accurate,
4100			 * probably, but results in floating point being
4101			 * used in the kernel on i386 (__udivdi3()).  On the
4102			 * XScale, though, __udivdi3() is implemented in
4103			 * software.
4104			 *
4105			 * The shift method for cylinder calculation is
4106			 * accurate if sectors_per_cylinder is a power of
4107			 * 2.  Otherwise it might be slightly off -- you
4108			 * might have a bit of a truncation problem.
4109			 */
4110#ifdef	__XSCALE__
4111			cylinders = (lun->be_lun->maxlba + 1) /
4112				sectors_per_cylinder;
4113#else
4114			for (shift = 31; shift > 0; shift--) {
4115				if (sectors_per_cylinder & (1 << shift))
4116					break;
4117			}
4118			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4119#endif
4120
4121			/*
4122			 * We've basically got 3 bytes, or 24 bits for the
4123			 * cylinder size in the mode page.  If we're over,
4124			 * just round down to 2^24.
4125			 */
4126			if (cylinders > 0xffffff)
4127				cylinders = 0xffffff;
4128
4129			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4130				CTL_PAGE_CURRENT];
4131			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4132
4133			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4134				CTL_PAGE_DEFAULT];
4135			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4136
4137			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4138				CTL_PAGE_SAVED];
4139			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4140
4141			page_index->page_data =
4142				(uint8_t *)lun->mode_pages.rigid_disk_page;
4143			break;
4144		}
4145		case SMS_CACHING_PAGE: {
4146
4147			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4148				panic("invalid subpage value %d",
4149				      page_index->subpage);
4150			/*
4151			 * Defaults should be okay here, no calculations
4152			 * needed.
4153			 */
4154			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4155			       &caching_page_default,
4156			       sizeof(caching_page_default));
4157			memcpy(&lun->mode_pages.caching_page[
4158			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4159			       sizeof(caching_page_changeable));
4160			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4161			       &caching_page_default,
4162			       sizeof(caching_page_default));
4163			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4164			       &caching_page_default,
4165			       sizeof(caching_page_default));
4166			page_index->page_data =
4167				(uint8_t *)lun->mode_pages.caching_page;
4168			break;
4169		}
4170		case SMS_CONTROL_MODE_PAGE: {
4171
4172			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4173				panic("invalid subpage value %d",
4174				      page_index->subpage);
4175
4176			/*
4177			 * Defaults should be okay here, no calculations
4178			 * needed.
4179			 */
4180			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4181			       &control_page_default,
4182			       sizeof(control_page_default));
4183			memcpy(&lun->mode_pages.control_page[
4184			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4185			       sizeof(control_page_changeable));
4186			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4187			       &control_page_default,
4188			       sizeof(control_page_default));
4189			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4190			       &control_page_default,
4191			       sizeof(control_page_default));
4192			page_index->page_data =
4193				(uint8_t *)lun->mode_pages.control_page;
4194			break;
4195
4196		}
4197		case SMS_VENDOR_SPECIFIC_PAGE:{
4198			switch (page_index->subpage) {
4199			case PWR_SUBPAGE_CODE: {
4200				struct copan_power_subpage *current_page,
4201							   *saved_page;
4202
4203				memcpy(&lun->mode_pages.power_subpage[
4204				       CTL_PAGE_CURRENT],
4205				       &power_page_default,
4206				       sizeof(power_page_default));
4207				memcpy(&lun->mode_pages.power_subpage[
4208				       CTL_PAGE_CHANGEABLE],
4209				       &power_page_changeable,
4210				       sizeof(power_page_changeable));
4211				memcpy(&lun->mode_pages.power_subpage[
4212				       CTL_PAGE_DEFAULT],
4213				       &power_page_default,
4214				       sizeof(power_page_default));
4215				memcpy(&lun->mode_pages.power_subpage[
4216				       CTL_PAGE_SAVED],
4217				       &power_page_default,
4218				       sizeof(power_page_default));
4219				page_index->page_data =
4220				    (uint8_t *)lun->mode_pages.power_subpage;
4221
4222				current_page = (struct copan_power_subpage *)
4223					(page_index->page_data +
4224					 (page_index->page_len *
4225					  CTL_PAGE_CURRENT));
4226			        saved_page = (struct copan_power_subpage *)
4227				        (page_index->page_data +
4228					 (page_index->page_len *
4229					  CTL_PAGE_SAVED));
4230				break;
4231			}
4232			case APS_SUBPAGE_CODE: {
4233				struct copan_aps_subpage *current_page,
4234							 *saved_page;
4235
4236				// This gets set multiple times but
4237				// it should always be the same. It's
4238				// only done during init so who cares.
4239				index_to_aps_page = i;
4240
4241				memcpy(&lun->mode_pages.aps_subpage[
4242				       CTL_PAGE_CURRENT],
4243				       &aps_page_default,
4244				       sizeof(aps_page_default));
4245				memcpy(&lun->mode_pages.aps_subpage[
4246				       CTL_PAGE_CHANGEABLE],
4247				       &aps_page_changeable,
4248				       sizeof(aps_page_changeable));
4249				memcpy(&lun->mode_pages.aps_subpage[
4250				       CTL_PAGE_DEFAULT],
4251				       &aps_page_default,
4252				       sizeof(aps_page_default));
4253				memcpy(&lun->mode_pages.aps_subpage[
4254				       CTL_PAGE_SAVED],
4255				       &aps_page_default,
4256				       sizeof(aps_page_default));
4257				page_index->page_data =
4258					(uint8_t *)lun->mode_pages.aps_subpage;
4259
4260				current_page = (struct copan_aps_subpage *)
4261					(page_index->page_data +
4262					 (page_index->page_len *
4263					  CTL_PAGE_CURRENT));
4264				saved_page = (struct copan_aps_subpage *)
4265					(page_index->page_data +
4266					 (page_index->page_len *
4267					  CTL_PAGE_SAVED));
4268				break;
4269			}
4270			case DBGCNF_SUBPAGE_CODE: {
4271				struct copan_debugconf_subpage *current_page,
4272							       *saved_page;
4273
4274				memcpy(&lun->mode_pages.debugconf_subpage[
4275				       CTL_PAGE_CURRENT],
4276				       &debugconf_page_default,
4277				       sizeof(debugconf_page_default));
4278				memcpy(&lun->mode_pages.debugconf_subpage[
4279				       CTL_PAGE_CHANGEABLE],
4280				       &debugconf_page_changeable,
4281				       sizeof(debugconf_page_changeable));
4282				memcpy(&lun->mode_pages.debugconf_subpage[
4283				       CTL_PAGE_DEFAULT],
4284				       &debugconf_page_default,
4285				       sizeof(debugconf_page_default));
4286				memcpy(&lun->mode_pages.debugconf_subpage[
4287				       CTL_PAGE_SAVED],
4288				       &debugconf_page_default,
4289				       sizeof(debugconf_page_default));
4290				page_index->page_data =
4291					(uint8_t *)lun->mode_pages.debugconf_subpage;
4292
4293				current_page = (struct copan_debugconf_subpage *)
4294					(page_index->page_data +
4295					 (page_index->page_len *
4296					  CTL_PAGE_CURRENT));
4297				saved_page = (struct copan_debugconf_subpage *)
4298					(page_index->page_data +
4299					 (page_index->page_len *
4300					  CTL_PAGE_SAVED));
4301				break;
4302			}
4303			default:
4304				panic("invalid subpage value %d",
4305				      page_index->subpage);
4306				break;
4307			}
4308   			break;
4309		}
4310		default:
4311			panic("invalid page value %d",
4312			      page_index->page_code & SMPH_PC_MASK);
4313			break;
4314    	}
4315	}
4316
4317	return (CTL_RETVAL_COMPLETE);
4318}
4319
4320/*
4321 * LUN allocation.
4322 *
4323 * Requirements:
4324 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4325 *   wants us to allocate the LUN and he can block.
4326 * - ctl_softc is always set
4327 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4328 *
4329 * Returns 0 for success, non-zero (errno) for failure.
4330 */
4331static int
4332ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4333	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4334{
4335	struct ctl_lun *nlun, *lun;
4336	struct ctl_port *port;
4337	struct scsi_vpd_id_descriptor *desc;
4338	struct scsi_vpd_id_t10 *t10id;
4339	const char *scsiname, *vendor;
4340	int lun_number, i, lun_malloced;
4341	int devidlen, idlen1, idlen2 = 0, len;
4342
4343	if (be_lun == NULL)
4344		return (EINVAL);
4345
4346	/*
4347	 * We currently only support Direct Access or Processor LUN types.
4348	 */
4349	switch (be_lun->lun_type) {
4350	case T_DIRECT:
4351		break;
4352	case T_PROCESSOR:
4353		break;
4354	case T_SEQUENTIAL:
4355	case T_CHANGER:
4356	default:
4357		be_lun->lun_config_status(be_lun->be_lun,
4358					  CTL_LUN_CONFIG_FAILURE);
4359		break;
4360	}
4361	if (ctl_lun == NULL) {
4362		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4363		lun_malloced = 1;
4364	} else {
4365		lun_malloced = 0;
4366		lun = ctl_lun;
4367	}
4368
4369	memset(lun, 0, sizeof(*lun));
4370	if (lun_malloced)
4371		lun->flags = CTL_LUN_MALLOCED;
4372
4373	/* Generate LUN ID. */
4374	devidlen = max(CTL_DEVID_MIN_LEN,
4375	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4376	idlen1 = sizeof(*t10id) + devidlen;
4377	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4378	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4379	if (scsiname != NULL) {
4380		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4381		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4382	}
4383	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4384	    M_CTL, M_WAITOK | M_ZERO);
4385	lun->lun_devid->len = len;
4386	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4387	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4388	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4389	desc->length = idlen1;
4390	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4391	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4392	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4393		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4394	} else {
4395		strncpy(t10id->vendor, vendor,
4396		    min(sizeof(t10id->vendor), strlen(vendor)));
4397	}
4398	strncpy((char *)t10id->vendor_spec_id,
4399	    (char *)be_lun->device_id, devidlen);
4400	if (scsiname != NULL) {
4401		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4402		    desc->length);
4403		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4404		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4405		    SVPD_ID_TYPE_SCSI_NAME;
4406		desc->length = idlen2;
4407		strlcpy(desc->identifier, scsiname, idlen2);
4408	}
4409
4410	mtx_lock(&ctl_softc->ctl_lock);
4411	/*
4412	 * See if the caller requested a particular LUN number.  If so, see
4413	 * if it is available.  Otherwise, allocate the first available LUN.
4414	 */
4415	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4416		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4417		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4418			mtx_unlock(&ctl_softc->ctl_lock);
4419			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4420				printf("ctl: requested LUN ID %d is higher "
4421				       "than CTL_MAX_LUNS - 1 (%d)\n",
4422				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4423			} else {
4424				/*
4425				 * XXX KDM return an error, or just assign
4426				 * another LUN ID in this case??
4427				 */
4428				printf("ctl: requested LUN ID %d is already "
4429				       "in use\n", be_lun->req_lun_id);
4430			}
4431			if (lun->flags & CTL_LUN_MALLOCED)
4432				free(lun, M_CTL);
4433			be_lun->lun_config_status(be_lun->be_lun,
4434						  CTL_LUN_CONFIG_FAILURE);
4435			return (ENOSPC);
4436		}
4437		lun_number = be_lun->req_lun_id;
4438	} else {
4439		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4440		if (lun_number == -1) {
4441			mtx_unlock(&ctl_softc->ctl_lock);
4442			printf("ctl: can't allocate LUN on target %ju, out of "
4443			       "LUNs\n", (uintmax_t)target_id.id);
4444			if (lun->flags & CTL_LUN_MALLOCED)
4445				free(lun, M_CTL);
4446			be_lun->lun_config_status(be_lun->be_lun,
4447						  CTL_LUN_CONFIG_FAILURE);
4448			return (ENOSPC);
4449		}
4450	}
4451	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4452
4453	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4454	lun->target = target_id;
4455	lun->lun = lun_number;
4456	lun->be_lun = be_lun;
4457	/*
4458	 * The processor LUN is always enabled.  Disk LUNs come on line
4459	 * disabled, and must be enabled by the backend.
4460	 */
4461	lun->flags |= CTL_LUN_DISABLED;
4462	lun->backend = be_lun->be;
4463	be_lun->ctl_lun = lun;
4464	be_lun->lun_id = lun_number;
4465	atomic_add_int(&be_lun->be->num_luns, 1);
4466	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4467		lun->flags |= CTL_LUN_STOPPED;
4468
4469	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4470		lun->flags |= CTL_LUN_INOPERABLE;
4471
4472	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4473		lun->flags |= CTL_LUN_PRIMARY_SC;
4474
4475	lun->ctl_softc = ctl_softc;
4476	TAILQ_INIT(&lun->ooa_queue);
4477	TAILQ_INIT(&lun->blocked_queue);
4478	STAILQ_INIT(&lun->error_list);
4479
4480	/*
4481	 * Initialize the mode page index.
4482	 */
4483	ctl_init_page_index(lun);
4484
4485	/*
4486	 * Set the poweron UA for all initiators on this LUN only.
4487	 */
4488	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4489		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4490
4491	/*
4492	 * Now, before we insert this lun on the lun list, set the lun
4493	 * inventory changed UA for all other luns.
4494	 */
4495	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4496		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4497			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4498		}
4499	}
4500
4501	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4502
4503	ctl_softc->ctl_luns[lun_number] = lun;
4504
4505	ctl_softc->num_luns++;
4506
4507	/* Setup statistics gathering */
4508	lun->stats.device_type = be_lun->lun_type;
4509	lun->stats.lun_number = lun_number;
4510	if (lun->stats.device_type == T_DIRECT)
4511		lun->stats.blocksize = be_lun->blocksize;
4512	else
4513		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4514	for (i = 0;i < CTL_MAX_PORTS;i++)
4515		lun->stats.ports[i].targ_port = i;
4516
4517	mtx_unlock(&ctl_softc->ctl_lock);
4518
4519	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4520
4521	/*
4522	 * Run through each registered FETD and bring it online if it isn't
4523	 * already.  Enable the target ID if it hasn't been enabled, and
4524	 * enable this particular LUN.
4525	 */
4526	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4527		int retval;
4528
4529		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4530		if (retval != 0) {
4531			printf("ctl_alloc_lun: FETD %s port %d returned error "
4532			       "%d for lun_enable on target %ju lun %d\n",
4533			       port->port_name, port->targ_port, retval,
4534			       (uintmax_t)target_id.id, lun_number);
4535		} else
4536			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4537	}
4538	return (0);
4539}
4540
4541/*
4542 * Delete a LUN.
4543 * Assumptions:
4544 * - LUN has already been marked invalid and any pending I/O has been taken
4545 *   care of.
4546 */
4547static int
4548ctl_free_lun(struct ctl_lun *lun)
4549{
4550	struct ctl_softc *softc;
4551#if 0
4552	struct ctl_port *port;
4553#endif
4554	struct ctl_lun *nlun;
4555	int i;
4556
4557	softc = lun->ctl_softc;
4558
4559	mtx_assert(&softc->ctl_lock, MA_OWNED);
4560
4561	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4562
4563	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4564
4565	softc->ctl_luns[lun->lun] = NULL;
4566
4567	if (!TAILQ_EMPTY(&lun->ooa_queue))
4568		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4569
4570	softc->num_luns--;
4571
4572	/*
4573	 * XXX KDM this scheme only works for a single target/multiple LUN
4574	 * setup.  It needs to be revamped for a multiple target scheme.
4575	 *
4576	 * XXX KDM this results in port->lun_disable() getting called twice,
4577	 * once when ctl_disable_lun() is called, and a second time here.
4578	 * We really need to re-think the LUN disable semantics.  There
4579	 * should probably be several steps/levels to LUN removal:
4580	 *  - disable
4581	 *  - invalidate
4582	 *  - free
4583 	 *
4584	 * Right now we only have a disable method when communicating to
4585	 * the front end ports, at least for individual LUNs.
4586	 */
4587#if 0
4588	STAILQ_FOREACH(port, &softc->port_list, links) {
4589		int retval;
4590
4591		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4592					 lun->lun);
4593		if (retval != 0) {
4594			printf("ctl_free_lun: FETD %s port %d returned error "
4595			       "%d for lun_disable on target %ju lun %jd\n",
4596			       port->port_name, port->targ_port, retval,
4597			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4598		}
4599
4600		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4601			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4602
4603			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4604			if (retval != 0) {
4605				printf("ctl_free_lun: FETD %s port %d "
4606				       "returned error %d for targ_disable on "
4607				       "target %ju\n", port->port_name,
4608				       port->targ_port, retval,
4609				       (uintmax_t)lun->target.id);
4610			} else
4611				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4612
4613			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4614				continue;
4615
4616#if 0
4617			port->port_offline(port->onoff_arg);
4618			port->status &= ~CTL_PORT_STATUS_ONLINE;
4619#endif
4620		}
4621	}
4622#endif
4623
4624	/*
4625	 * Tell the backend to free resources, if this LUN has a backend.
4626	 */
4627	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4628	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4629
4630	mtx_destroy(&lun->lun_lock);
4631	free(lun->lun_devid, M_CTL);
4632	if (lun->flags & CTL_LUN_MALLOCED)
4633		free(lun, M_CTL);
4634
4635	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4636		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4637			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4638		}
4639	}
4640
4641	return (0);
4642}
4643
4644static void
4645ctl_create_lun(struct ctl_be_lun *be_lun)
4646{
4647	struct ctl_softc *ctl_softc;
4648
4649	ctl_softc = control_softc;
4650
4651	/*
4652	 * ctl_alloc_lun() should handle all potential failure cases.
4653	 */
4654	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4655}
4656
4657int
4658ctl_add_lun(struct ctl_be_lun *be_lun)
4659{
4660	struct ctl_softc *ctl_softc = control_softc;
4661
4662	mtx_lock(&ctl_softc->ctl_lock);
4663	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4664	mtx_unlock(&ctl_softc->ctl_lock);
4665	wakeup(&ctl_softc->pending_lun_queue);
4666
4667	return (0);
4668}
4669
4670int
4671ctl_enable_lun(struct ctl_be_lun *be_lun)
4672{
4673	struct ctl_softc *ctl_softc;
4674	struct ctl_port *port, *nport;
4675	struct ctl_lun *lun;
4676	int retval;
4677
4678	ctl_softc = control_softc;
4679
4680	lun = (struct ctl_lun *)be_lun->ctl_lun;
4681
4682	mtx_lock(&ctl_softc->ctl_lock);
4683	mtx_lock(&lun->lun_lock);
4684	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4685		/*
4686		 * eh?  Why did we get called if the LUN is already
4687		 * enabled?
4688		 */
4689		mtx_unlock(&lun->lun_lock);
4690		mtx_unlock(&ctl_softc->ctl_lock);
4691		return (0);
4692	}
4693	lun->flags &= ~CTL_LUN_DISABLED;
4694	mtx_unlock(&lun->lun_lock);
4695
4696	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4697		nport = STAILQ_NEXT(port, links);
4698
4699		/*
4700		 * Drop the lock while we call the FETD's enable routine.
4701		 * This can lead to a callback into CTL (at least in the
4702		 * case of the internal initiator frontend.
4703		 */
4704		mtx_unlock(&ctl_softc->ctl_lock);
4705		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4706		mtx_lock(&ctl_softc->ctl_lock);
4707		if (retval != 0) {
4708			printf("%s: FETD %s port %d returned error "
4709			       "%d for lun_enable on target %ju lun %jd\n",
4710			       __func__, port->port_name, port->targ_port, retval,
4711			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4712		}
4713#if 0
4714		 else {
4715            /* NOTE:  TODO:  why does lun enable affect port status? */
4716			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4717		}
4718#endif
4719	}
4720
4721	mtx_unlock(&ctl_softc->ctl_lock);
4722
4723	return (0);
4724}
4725
4726int
4727ctl_disable_lun(struct ctl_be_lun *be_lun)
4728{
4729	struct ctl_softc *ctl_softc;
4730	struct ctl_port *port;
4731	struct ctl_lun *lun;
4732	int retval;
4733
4734	ctl_softc = control_softc;
4735
4736	lun = (struct ctl_lun *)be_lun->ctl_lun;
4737
4738	mtx_lock(&ctl_softc->ctl_lock);
4739	mtx_lock(&lun->lun_lock);
4740	if (lun->flags & CTL_LUN_DISABLED) {
4741		mtx_unlock(&lun->lun_lock);
4742		mtx_unlock(&ctl_softc->ctl_lock);
4743		return (0);
4744	}
4745	lun->flags |= CTL_LUN_DISABLED;
4746	mtx_unlock(&lun->lun_lock);
4747
4748	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4749		mtx_unlock(&ctl_softc->ctl_lock);
4750		/*
4751		 * Drop the lock before we call the frontend's disable
4752		 * routine, to avoid lock order reversals.
4753		 *
4754		 * XXX KDM what happens if the frontend list changes while
4755		 * we're traversing it?  It's unlikely, but should be handled.
4756		 */
4757		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4758					 lun->lun);
4759		mtx_lock(&ctl_softc->ctl_lock);
4760		if (retval != 0) {
4761			printf("ctl_alloc_lun: FETD %s port %d returned error "
4762			       "%d for lun_disable on target %ju lun %jd\n",
4763			       port->port_name, port->targ_port, retval,
4764			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4765		}
4766	}
4767
4768	mtx_unlock(&ctl_softc->ctl_lock);
4769
4770	return (0);
4771}
4772
4773int
4774ctl_start_lun(struct ctl_be_lun *be_lun)
4775{
4776	struct ctl_softc *ctl_softc;
4777	struct ctl_lun *lun;
4778
4779	ctl_softc = control_softc;
4780
4781	lun = (struct ctl_lun *)be_lun->ctl_lun;
4782
4783	mtx_lock(&lun->lun_lock);
4784	lun->flags &= ~CTL_LUN_STOPPED;
4785	mtx_unlock(&lun->lun_lock);
4786
4787	return (0);
4788}
4789
4790int
4791ctl_stop_lun(struct ctl_be_lun *be_lun)
4792{
4793	struct ctl_softc *ctl_softc;
4794	struct ctl_lun *lun;
4795
4796	ctl_softc = control_softc;
4797
4798	lun = (struct ctl_lun *)be_lun->ctl_lun;
4799
4800	mtx_lock(&lun->lun_lock);
4801	lun->flags |= CTL_LUN_STOPPED;
4802	mtx_unlock(&lun->lun_lock);
4803
4804	return (0);
4805}
4806
4807int
4808ctl_lun_offline(struct ctl_be_lun *be_lun)
4809{
4810	struct ctl_softc *ctl_softc;
4811	struct ctl_lun *lun;
4812
4813	ctl_softc = control_softc;
4814
4815	lun = (struct ctl_lun *)be_lun->ctl_lun;
4816
4817	mtx_lock(&lun->lun_lock);
4818	lun->flags |= CTL_LUN_OFFLINE;
4819	mtx_unlock(&lun->lun_lock);
4820
4821	return (0);
4822}
4823
4824int
4825ctl_lun_online(struct ctl_be_lun *be_lun)
4826{
4827	struct ctl_softc *ctl_softc;
4828	struct ctl_lun *lun;
4829
4830	ctl_softc = control_softc;
4831
4832	lun = (struct ctl_lun *)be_lun->ctl_lun;
4833
4834	mtx_lock(&lun->lun_lock);
4835	lun->flags &= ~CTL_LUN_OFFLINE;
4836	mtx_unlock(&lun->lun_lock);
4837
4838	return (0);
4839}
4840
4841int
4842ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4843{
4844	struct ctl_softc *ctl_softc;
4845	struct ctl_lun *lun;
4846
4847	ctl_softc = control_softc;
4848
4849	lun = (struct ctl_lun *)be_lun->ctl_lun;
4850
4851	mtx_lock(&lun->lun_lock);
4852
4853	/*
4854	 * The LUN needs to be disabled before it can be marked invalid.
4855	 */
4856	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4857		mtx_unlock(&lun->lun_lock);
4858		return (-1);
4859	}
4860	/*
4861	 * Mark the LUN invalid.
4862	 */
4863	lun->flags |= CTL_LUN_INVALID;
4864
4865	/*
4866	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4867	 * If we have something in the OOA queue, we'll free it when the
4868	 * last I/O completes.
4869	 */
4870	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4871		mtx_unlock(&lun->lun_lock);
4872		mtx_lock(&ctl_softc->ctl_lock);
4873		ctl_free_lun(lun);
4874		mtx_unlock(&ctl_softc->ctl_lock);
4875	} else
4876		mtx_unlock(&lun->lun_lock);
4877
4878	return (0);
4879}
4880
4881int
4882ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4883{
4884	struct ctl_softc *ctl_softc;
4885	struct ctl_lun *lun;
4886
4887	ctl_softc = control_softc;
4888	lun = (struct ctl_lun *)be_lun->ctl_lun;
4889
4890	mtx_lock(&lun->lun_lock);
4891	lun->flags |= CTL_LUN_INOPERABLE;
4892	mtx_unlock(&lun->lun_lock);
4893
4894	return (0);
4895}
4896
4897int
4898ctl_lun_operable(struct ctl_be_lun *be_lun)
4899{
4900	struct ctl_softc *ctl_softc;
4901	struct ctl_lun *lun;
4902
4903	ctl_softc = control_softc;
4904	lun = (struct ctl_lun *)be_lun->ctl_lun;
4905
4906	mtx_lock(&lun->lun_lock);
4907	lun->flags &= ~CTL_LUN_INOPERABLE;
4908	mtx_unlock(&lun->lun_lock);
4909
4910	return (0);
4911}
4912
4913int
4914ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4915		   int lock)
4916{
4917	struct ctl_softc *softc;
4918	struct ctl_lun *lun;
4919	struct copan_aps_subpage *current_sp;
4920	struct ctl_page_index *page_index;
4921	int i;
4922
4923	softc = control_softc;
4924
4925	mtx_lock(&softc->ctl_lock);
4926
4927	lun = (struct ctl_lun *)be_lun->ctl_lun;
4928	mtx_lock(&lun->lun_lock);
4929
4930	page_index = NULL;
4931	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4932		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4933		     APS_PAGE_CODE)
4934			continue;
4935
4936		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4937			continue;
4938		page_index = &lun->mode_pages.index[i];
4939	}
4940
4941	if (page_index == NULL) {
4942		mtx_unlock(&lun->lun_lock);
4943		mtx_unlock(&softc->ctl_lock);
4944		printf("%s: APS subpage not found for lun %ju!\n", __func__,
4945		       (uintmax_t)lun->lun);
4946		return (1);
4947	}
4948#if 0
4949	if ((softc->aps_locked_lun != 0)
4950	 && (softc->aps_locked_lun != lun->lun)) {
4951		printf("%s: attempt to lock LUN %llu when %llu is already "
4952		       "locked\n");
4953		mtx_unlock(&lun->lun_lock);
4954		mtx_unlock(&softc->ctl_lock);
4955		return (1);
4956	}
4957#endif
4958
4959	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4960		(page_index->page_len * CTL_PAGE_CURRENT));
4961
4962	if (lock != 0) {
4963		current_sp->lock_active = APS_LOCK_ACTIVE;
4964		softc->aps_locked_lun = lun->lun;
4965	} else {
4966		current_sp->lock_active = 0;
4967		softc->aps_locked_lun = 0;
4968	}
4969
4970
4971	/*
4972	 * If we're in HA mode, try to send the lock message to the other
4973	 * side.
4974	 */
4975	if (ctl_is_single == 0) {
4976		int isc_retval;
4977		union ctl_ha_msg lock_msg;
4978
4979		lock_msg.hdr.nexus = *nexus;
4980		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4981		if (lock != 0)
4982			lock_msg.aps.lock_flag = 1;
4983		else
4984			lock_msg.aps.lock_flag = 0;
4985		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4986					 sizeof(lock_msg), 0);
4987		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4988			printf("%s: APS (lock=%d) error returned from "
4989			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4990			mtx_unlock(&lun->lun_lock);
4991			mtx_unlock(&softc->ctl_lock);
4992			return (1);
4993		}
4994	}
4995
4996	mtx_unlock(&lun->lun_lock);
4997	mtx_unlock(&softc->ctl_lock);
4998
4999	return (0);
5000}
5001
5002void
5003ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5004{
5005	struct ctl_lun *lun;
5006	struct ctl_softc *softc;
5007	int i;
5008
5009	softc = control_softc;
5010
5011	lun = (struct ctl_lun *)be_lun->ctl_lun;
5012
5013	mtx_lock(&lun->lun_lock);
5014
5015	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5016		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
5017
5018	mtx_unlock(&lun->lun_lock);
5019}
5020
5021/*
5022 * Backend "memory move is complete" callback for requests that never
5023 * make it down to say RAIDCore's configuration code.
5024 */
5025int
5026ctl_config_move_done(union ctl_io *io)
5027{
5028	int retval;
5029
5030	retval = CTL_RETVAL_COMPLETE;
5031
5032
5033	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5034	/*
5035	 * XXX KDM this shouldn't happen, but what if it does?
5036	 */
5037	if (io->io_hdr.io_type != CTL_IO_SCSI)
5038		panic("I/O type isn't CTL_IO_SCSI!");
5039
5040	if ((io->io_hdr.port_status == 0)
5041	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5042	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5043		io->io_hdr.status = CTL_SUCCESS;
5044	else if ((io->io_hdr.port_status != 0)
5045	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5046	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5047		/*
5048		 * For hardware error sense keys, the sense key
5049		 * specific value is defined to be a retry count,
5050		 * but we use it to pass back an internal FETD
5051		 * error code.  XXX KDM  Hopefully the FETD is only
5052		 * using 16 bits for an error code, since that's
5053		 * all the space we have in the sks field.
5054		 */
5055		ctl_set_internal_failure(&io->scsiio,
5056					 /*sks_valid*/ 1,
5057					 /*retry_count*/
5058					 io->io_hdr.port_status);
5059		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5060			free(io->scsiio.kern_data_ptr, M_CTL);
5061		ctl_done(io);
5062		goto bailout;
5063	}
5064
5065	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5066	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5067	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5068		/*
5069		 * XXX KDM just assuming a single pointer here, and not a
5070		 * S/G list.  If we start using S/G lists for config data,
5071		 * we'll need to know how to clean them up here as well.
5072		 */
5073		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5074			free(io->scsiio.kern_data_ptr, M_CTL);
5075		/* Hopefully the user has already set the status... */
5076		ctl_done(io);
5077	} else {
5078		/*
5079		 * XXX KDM now we need to continue data movement.  Some
5080		 * options:
5081		 * - call ctl_scsiio() again?  We don't do this for data
5082		 *   writes, because for those at least we know ahead of
5083		 *   time where the write will go and how long it is.  For
5084		 *   config writes, though, that information is largely
5085		 *   contained within the write itself, thus we need to
5086		 *   parse out the data again.
5087		 *
5088		 * - Call some other function once the data is in?
5089		 */
5090
5091		/*
5092		 * XXX KDM call ctl_scsiio() again for now, and check flag
5093		 * bits to see whether we're allocated or not.
5094		 */
5095		retval = ctl_scsiio(&io->scsiio);
5096	}
5097bailout:
5098	return (retval);
5099}
5100
5101/*
5102 * This gets called by a backend driver when it is done with a
5103 * data_submit method.
5104 */
5105void
5106ctl_data_submit_done(union ctl_io *io)
5107{
5108	/*
5109	 * If the IO_CONT flag is set, we need to call the supplied
5110	 * function to continue processing the I/O, instead of completing
5111	 * the I/O just yet.
5112	 *
5113	 * If there is an error, though, we don't want to keep processing.
5114	 * Instead, just send status back to the initiator.
5115	 */
5116	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5117	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5118	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5119	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5120		io->scsiio.io_cont(io);
5121		return;
5122	}
5123	ctl_done(io);
5124}
5125
5126/*
5127 * This gets called by a backend driver when it is done with a
5128 * configuration write.
5129 */
5130void
5131ctl_config_write_done(union ctl_io *io)
5132{
5133	/*
5134	 * If the IO_CONT flag is set, we need to call the supplied
5135	 * function to continue processing the I/O, instead of completing
5136	 * the I/O just yet.
5137	 *
5138	 * If there is an error, though, we don't want to keep processing.
5139	 * Instead, just send status back to the initiator.
5140	 */
5141	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5142	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5143	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5144		io->scsiio.io_cont(io);
5145		return;
5146	}
5147	/*
5148	 * Since a configuration write can be done for commands that actually
5149	 * have data allocated, like write buffer, and commands that have
5150	 * no data, like start/stop unit, we need to check here.
5151	 */
5152	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5153		free(io->scsiio.kern_data_ptr, M_CTL);
5154	ctl_done(io);
5155}
5156
5157/*
5158 * SCSI release command.
5159 */
5160int
5161ctl_scsi_release(struct ctl_scsiio *ctsio)
5162{
5163	int length, longid, thirdparty_id, resv_id;
5164	struct ctl_softc *ctl_softc;
5165	struct ctl_lun *lun;
5166
5167	length = 0;
5168	resv_id = 0;
5169
5170	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5171
5172	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5173	ctl_softc = control_softc;
5174
5175	switch (ctsio->cdb[0]) {
5176	case RELEASE_10: {
5177		struct scsi_release_10 *cdb;
5178
5179		cdb = (struct scsi_release_10 *)ctsio->cdb;
5180
5181		if (cdb->byte2 & SR10_LONGID)
5182			longid = 1;
5183		else
5184			thirdparty_id = cdb->thirdparty_id;
5185
5186		resv_id = cdb->resv_id;
5187		length = scsi_2btoul(cdb->length);
5188		break;
5189	}
5190	}
5191
5192
5193	/*
5194	 * XXX KDM right now, we only support LUN reservation.  We don't
5195	 * support 3rd party reservations, or extent reservations, which
5196	 * might actually need the parameter list.  If we've gotten this
5197	 * far, we've got a LUN reservation.  Anything else got kicked out
5198	 * above.  So, according to SPC, ignore the length.
5199	 */
5200	length = 0;
5201
5202	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5203	 && (length > 0)) {
5204		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5205		ctsio->kern_data_len = length;
5206		ctsio->kern_total_len = length;
5207		ctsio->kern_data_resid = 0;
5208		ctsio->kern_rel_offset = 0;
5209		ctsio->kern_sg_entries = 0;
5210		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5211		ctsio->be_move_done = ctl_config_move_done;
5212		ctl_datamove((union ctl_io *)ctsio);
5213
5214		return (CTL_RETVAL_COMPLETE);
5215	}
5216
5217	if (length > 0)
5218		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5219
5220	mtx_lock(&lun->lun_lock);
5221
5222	/*
5223	 * According to SPC, it is not an error for an intiator to attempt
5224	 * to release a reservation on a LUN that isn't reserved, or that
5225	 * is reserved by another initiator.  The reservation can only be
5226	 * released, though, by the initiator who made it or by one of
5227	 * several reset type events.
5228	 */
5229	if (lun->flags & CTL_LUN_RESERVED) {
5230		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5231		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5232		 && (ctsio->io_hdr.nexus.targ_target.id ==
5233		     lun->rsv_nexus.targ_target.id)) {
5234			lun->flags &= ~CTL_LUN_RESERVED;
5235		}
5236	}
5237
5238	mtx_unlock(&lun->lun_lock);
5239
5240	ctsio->scsi_status = SCSI_STATUS_OK;
5241	ctsio->io_hdr.status = CTL_SUCCESS;
5242
5243	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5244		free(ctsio->kern_data_ptr, M_CTL);
5245		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5246	}
5247
5248	ctl_done((union ctl_io *)ctsio);
5249	return (CTL_RETVAL_COMPLETE);
5250}
5251
5252int
5253ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5254{
5255	int extent, thirdparty, longid;
5256	int resv_id, length;
5257	uint64_t thirdparty_id;
5258	struct ctl_softc *ctl_softc;
5259	struct ctl_lun *lun;
5260
5261	extent = 0;
5262	thirdparty = 0;
5263	longid = 0;
5264	resv_id = 0;
5265	length = 0;
5266	thirdparty_id = 0;
5267
5268	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5269
5270	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5271	ctl_softc = control_softc;
5272
5273	switch (ctsio->cdb[0]) {
5274	case RESERVE_10: {
5275		struct scsi_reserve_10 *cdb;
5276
5277		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5278
5279		if (cdb->byte2 & SR10_LONGID)
5280			longid = 1;
5281		else
5282			thirdparty_id = cdb->thirdparty_id;
5283
5284		resv_id = cdb->resv_id;
5285		length = scsi_2btoul(cdb->length);
5286		break;
5287	}
5288	}
5289
5290	/*
5291	 * XXX KDM right now, we only support LUN reservation.  We don't
5292	 * support 3rd party reservations, or extent reservations, which
5293	 * might actually need the parameter list.  If we've gotten this
5294	 * far, we've got a LUN reservation.  Anything else got kicked out
5295	 * above.  So, according to SPC, ignore the length.
5296	 */
5297	length = 0;
5298
5299	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5300	 && (length > 0)) {
5301		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5302		ctsio->kern_data_len = length;
5303		ctsio->kern_total_len = length;
5304		ctsio->kern_data_resid = 0;
5305		ctsio->kern_rel_offset = 0;
5306		ctsio->kern_sg_entries = 0;
5307		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5308		ctsio->be_move_done = ctl_config_move_done;
5309		ctl_datamove((union ctl_io *)ctsio);
5310
5311		return (CTL_RETVAL_COMPLETE);
5312	}
5313
5314	if (length > 0)
5315		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5316
5317	mtx_lock(&lun->lun_lock);
5318	if (lun->flags & CTL_LUN_RESERVED) {
5319		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5320		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5321		 || (ctsio->io_hdr.nexus.targ_target.id !=
5322		     lun->rsv_nexus.targ_target.id)) {
5323			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5324			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5325			goto bailout;
5326		}
5327	}
5328
5329	lun->flags |= CTL_LUN_RESERVED;
5330	lun->rsv_nexus = ctsio->io_hdr.nexus;
5331
5332	ctsio->scsi_status = SCSI_STATUS_OK;
5333	ctsio->io_hdr.status = CTL_SUCCESS;
5334
5335bailout:
5336	mtx_unlock(&lun->lun_lock);
5337
5338	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5339		free(ctsio->kern_data_ptr, M_CTL);
5340		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5341	}
5342
5343	ctl_done((union ctl_io *)ctsio);
5344	return (CTL_RETVAL_COMPLETE);
5345}
5346
5347int
5348ctl_start_stop(struct ctl_scsiio *ctsio)
5349{
5350	struct scsi_start_stop_unit *cdb;
5351	struct ctl_lun *lun;
5352	struct ctl_softc *ctl_softc;
5353	int retval;
5354
5355	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5356
5357	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5358	ctl_softc = control_softc;
5359	retval = 0;
5360
5361	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5362
5363	/*
5364	 * XXX KDM
5365	 * We don't support the immediate bit on a stop unit.  In order to
5366	 * do that, we would need to code up a way to know that a stop is
5367	 * pending, and hold off any new commands until it completes, one
5368	 * way or another.  Then we could accept or reject those commands
5369	 * depending on its status.  We would almost need to do the reverse
5370	 * of what we do below for an immediate start -- return the copy of
5371	 * the ctl_io to the FETD with status to send to the host (and to
5372	 * free the copy!) and then free the original I/O once the stop
5373	 * actually completes.  That way, the OOA queue mechanism can work
5374	 * to block commands that shouldn't proceed.  Another alternative
5375	 * would be to put the copy in the queue in place of the original,
5376	 * and return the original back to the caller.  That could be
5377	 * slightly safer..
5378	 */
5379	if ((cdb->byte2 & SSS_IMMED)
5380	 && ((cdb->how & SSS_START) == 0)) {
5381		ctl_set_invalid_field(ctsio,
5382				      /*sks_valid*/ 1,
5383				      /*command*/ 1,
5384				      /*field*/ 1,
5385				      /*bit_valid*/ 1,
5386				      /*bit*/ 0);
5387		ctl_done((union ctl_io *)ctsio);
5388		return (CTL_RETVAL_COMPLETE);
5389	}
5390
5391	if ((lun->flags & CTL_LUN_PR_RESERVED)
5392	 && ((cdb->how & SSS_START)==0)) {
5393		uint32_t residx;
5394
5395		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5396		if (!lun->per_res[residx].registered
5397		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5398
5399			ctl_set_reservation_conflict(ctsio);
5400			ctl_done((union ctl_io *)ctsio);
5401			return (CTL_RETVAL_COMPLETE);
5402		}
5403	}
5404
5405	/*
5406	 * If there is no backend on this device, we can't start or stop
5407	 * it.  In theory we shouldn't get any start/stop commands in the
5408	 * first place at this level if the LUN doesn't have a backend.
5409	 * That should get stopped by the command decode code.
5410	 */
5411	if (lun->backend == NULL) {
5412		ctl_set_invalid_opcode(ctsio);
5413		ctl_done((union ctl_io *)ctsio);
5414		return (CTL_RETVAL_COMPLETE);
5415	}
5416
5417	/*
5418	 * XXX KDM Copan-specific offline behavior.
5419	 * Figure out a reasonable way to port this?
5420	 */
5421#ifdef NEEDTOPORT
5422	mtx_lock(&lun->lun_lock);
5423
5424	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5425	 && (lun->flags & CTL_LUN_OFFLINE)) {
5426		/*
5427		 * If the LUN is offline, and the on/offline bit isn't set,
5428		 * reject the start or stop.  Otherwise, let it through.
5429		 */
5430		mtx_unlock(&lun->lun_lock);
5431		ctl_set_lun_not_ready(ctsio);
5432		ctl_done((union ctl_io *)ctsio);
5433	} else {
5434		mtx_unlock(&lun->lun_lock);
5435#endif /* NEEDTOPORT */
5436		/*
5437		 * This could be a start or a stop when we're online,
5438		 * or a stop/offline or start/online.  A start or stop when
5439		 * we're offline is covered in the case above.
5440		 */
5441		/*
5442		 * In the non-immediate case, we send the request to
5443		 * the backend and return status to the user when
5444		 * it is done.
5445		 *
5446		 * In the immediate case, we allocate a new ctl_io
5447		 * to hold a copy of the request, and send that to
5448		 * the backend.  We then set good status on the
5449		 * user's request and return it immediately.
5450		 */
5451		if (cdb->byte2 & SSS_IMMED) {
5452			union ctl_io *new_io;
5453
5454			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5455			if (new_io == NULL) {
5456				ctl_set_busy(ctsio);
5457				ctl_done((union ctl_io *)ctsio);
5458			} else {
5459				ctl_copy_io((union ctl_io *)ctsio,
5460					    new_io);
5461				retval = lun->backend->config_write(new_io);
5462				ctl_set_success(ctsio);
5463				ctl_done((union ctl_io *)ctsio);
5464			}
5465		} else {
5466			retval = lun->backend->config_write(
5467				(union ctl_io *)ctsio);
5468		}
5469#ifdef NEEDTOPORT
5470	}
5471#endif
5472	return (retval);
5473}
5474
5475/*
5476 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5477 * we don't really do anything with the LBA and length fields if the user
5478 * passes them in.  Instead we'll just flush out the cache for the entire
5479 * LUN.
5480 */
5481int
5482ctl_sync_cache(struct ctl_scsiio *ctsio)
5483{
5484	struct ctl_lun *lun;
5485	struct ctl_softc *ctl_softc;
5486	uint64_t starting_lba;
5487	uint32_t block_count;
5488	int retval;
5489
5490	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5491
5492	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5493	ctl_softc = control_softc;
5494	retval = 0;
5495
5496	switch (ctsio->cdb[0]) {
5497	case SYNCHRONIZE_CACHE: {
5498		struct scsi_sync_cache *cdb;
5499		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5500
5501		starting_lba = scsi_4btoul(cdb->begin_lba);
5502		block_count = scsi_2btoul(cdb->lb_count);
5503		break;
5504	}
5505	case SYNCHRONIZE_CACHE_16: {
5506		struct scsi_sync_cache_16 *cdb;
5507		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5508
5509		starting_lba = scsi_8btou64(cdb->begin_lba);
5510		block_count = scsi_4btoul(cdb->lb_count);
5511		break;
5512	}
5513	default:
5514		ctl_set_invalid_opcode(ctsio);
5515		ctl_done((union ctl_io *)ctsio);
5516		goto bailout;
5517		break; /* NOTREACHED */
5518	}
5519
5520	/*
5521	 * We check the LBA and length, but don't do anything with them.
5522	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5523	 * get flushed.  This check will just help satisfy anyone who wants
5524	 * to see an error for an out of range LBA.
5525	 */
5526	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5527		ctl_set_lba_out_of_range(ctsio);
5528		ctl_done((union ctl_io *)ctsio);
5529		goto bailout;
5530	}
5531
5532	/*
5533	 * If this LUN has no backend, we can't flush the cache anyway.
5534	 */
5535	if (lun->backend == NULL) {
5536		ctl_set_invalid_opcode(ctsio);
5537		ctl_done((union ctl_io *)ctsio);
5538		goto bailout;
5539	}
5540
5541	/*
5542	 * Check to see whether we're configured to send the SYNCHRONIZE
5543	 * CACHE command directly to the back end.
5544	 */
5545	mtx_lock(&lun->lun_lock);
5546	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5547	 && (++(lun->sync_count) >= lun->sync_interval)) {
5548		lun->sync_count = 0;
5549		mtx_unlock(&lun->lun_lock);
5550		retval = lun->backend->config_write((union ctl_io *)ctsio);
5551	} else {
5552		mtx_unlock(&lun->lun_lock);
5553		ctl_set_success(ctsio);
5554		ctl_done((union ctl_io *)ctsio);
5555	}
5556
5557bailout:
5558
5559	return (retval);
5560}
5561
5562int
5563ctl_format(struct ctl_scsiio *ctsio)
5564{
5565	struct scsi_format *cdb;
5566	struct ctl_lun *lun;
5567	struct ctl_softc *ctl_softc;
5568	int length, defect_list_len;
5569
5570	CTL_DEBUG_PRINT(("ctl_format\n"));
5571
5572	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5573	ctl_softc = control_softc;
5574
5575	cdb = (struct scsi_format *)ctsio->cdb;
5576
5577	length = 0;
5578	if (cdb->byte2 & SF_FMTDATA) {
5579		if (cdb->byte2 & SF_LONGLIST)
5580			length = sizeof(struct scsi_format_header_long);
5581		else
5582			length = sizeof(struct scsi_format_header_short);
5583	}
5584
5585	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5586	 && (length > 0)) {
5587		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5588		ctsio->kern_data_len = length;
5589		ctsio->kern_total_len = length;
5590		ctsio->kern_data_resid = 0;
5591		ctsio->kern_rel_offset = 0;
5592		ctsio->kern_sg_entries = 0;
5593		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5594		ctsio->be_move_done = ctl_config_move_done;
5595		ctl_datamove((union ctl_io *)ctsio);
5596
5597		return (CTL_RETVAL_COMPLETE);
5598	}
5599
5600	defect_list_len = 0;
5601
5602	if (cdb->byte2 & SF_FMTDATA) {
5603		if (cdb->byte2 & SF_LONGLIST) {
5604			struct scsi_format_header_long *header;
5605
5606			header = (struct scsi_format_header_long *)
5607				ctsio->kern_data_ptr;
5608
5609			defect_list_len = scsi_4btoul(header->defect_list_len);
5610			if (defect_list_len != 0) {
5611				ctl_set_invalid_field(ctsio,
5612						      /*sks_valid*/ 1,
5613						      /*command*/ 0,
5614						      /*field*/ 2,
5615						      /*bit_valid*/ 0,
5616						      /*bit*/ 0);
5617				goto bailout;
5618			}
5619		} else {
5620			struct scsi_format_header_short *header;
5621
5622			header = (struct scsi_format_header_short *)
5623				ctsio->kern_data_ptr;
5624
5625			defect_list_len = scsi_2btoul(header->defect_list_len);
5626			if (defect_list_len != 0) {
5627				ctl_set_invalid_field(ctsio,
5628						      /*sks_valid*/ 1,
5629						      /*command*/ 0,
5630						      /*field*/ 2,
5631						      /*bit_valid*/ 0,
5632						      /*bit*/ 0);
5633				goto bailout;
5634			}
5635		}
5636	}
5637
5638	/*
5639	 * The format command will clear out the "Medium format corrupted"
5640	 * status if set by the configuration code.  That status is really
5641	 * just a way to notify the host that we have lost the media, and
5642	 * get them to issue a command that will basically make them think
5643	 * they're blowing away the media.
5644	 */
5645	mtx_lock(&lun->lun_lock);
5646	lun->flags &= ~CTL_LUN_INOPERABLE;
5647	mtx_unlock(&lun->lun_lock);
5648
5649	ctsio->scsi_status = SCSI_STATUS_OK;
5650	ctsio->io_hdr.status = CTL_SUCCESS;
5651bailout:
5652
5653	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5654		free(ctsio->kern_data_ptr, M_CTL);
5655		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5656	}
5657
5658	ctl_done((union ctl_io *)ctsio);
5659	return (CTL_RETVAL_COMPLETE);
5660}
5661
5662int
5663ctl_read_buffer(struct ctl_scsiio *ctsio)
5664{
5665	struct scsi_read_buffer *cdb;
5666	struct ctl_lun *lun;
5667	int buffer_offset, len;
5668	static uint8_t descr[4];
5669	static uint8_t echo_descr[4] = { 0 };
5670
5671	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5672
5673	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5674	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5675
5676	if (lun->flags & CTL_LUN_PR_RESERVED) {
5677		uint32_t residx;
5678
5679		/*
5680		 * XXX KDM need a lock here.
5681		 */
5682		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5683		if ((lun->res_type == SPR_TYPE_EX_AC
5684		  && residx != lun->pr_res_idx)
5685		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5686		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5687		  && !lun->per_res[residx].registered)) {
5688			ctl_set_reservation_conflict(ctsio);
5689			ctl_done((union ctl_io *)ctsio);
5690			return (CTL_RETVAL_COMPLETE);
5691	        }
5692	}
5693
5694	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5695	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5696	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5697		ctl_set_invalid_field(ctsio,
5698				      /*sks_valid*/ 1,
5699				      /*command*/ 1,
5700				      /*field*/ 1,
5701				      /*bit_valid*/ 1,
5702				      /*bit*/ 4);
5703		ctl_done((union ctl_io *)ctsio);
5704		return (CTL_RETVAL_COMPLETE);
5705	}
5706
5707	len = scsi_3btoul(cdb->length);
5708	buffer_offset = scsi_3btoul(cdb->offset);
5709
5710	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5711		ctl_set_invalid_field(ctsio,
5712				      /*sks_valid*/ 1,
5713				      /*command*/ 1,
5714				      /*field*/ 6,
5715				      /*bit_valid*/ 0,
5716				      /*bit*/ 0);
5717		ctl_done((union ctl_io *)ctsio);
5718		return (CTL_RETVAL_COMPLETE);
5719	}
5720
5721	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5722		descr[0] = 0;
5723		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5724		ctsio->kern_data_ptr = descr;
5725		len = min(len, sizeof(descr));
5726	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5727		ctsio->kern_data_ptr = echo_descr;
5728		len = min(len, sizeof(echo_descr));
5729	} else
5730		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5731	ctsio->kern_data_len = len;
5732	ctsio->kern_total_len = len;
5733	ctsio->kern_data_resid = 0;
5734	ctsio->kern_rel_offset = 0;
5735	ctsio->kern_sg_entries = 0;
5736	ctsio->be_move_done = ctl_config_move_done;
5737	ctl_datamove((union ctl_io *)ctsio);
5738
5739	return (CTL_RETVAL_COMPLETE);
5740}
5741
5742int
5743ctl_write_buffer(struct ctl_scsiio *ctsio)
5744{
5745	struct scsi_write_buffer *cdb;
5746	struct ctl_lun *lun;
5747	int buffer_offset, len;
5748
5749	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5750
5751	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5752	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5753
5754	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5755		ctl_set_invalid_field(ctsio,
5756				      /*sks_valid*/ 1,
5757				      /*command*/ 1,
5758				      /*field*/ 1,
5759				      /*bit_valid*/ 1,
5760				      /*bit*/ 4);
5761		ctl_done((union ctl_io *)ctsio);
5762		return (CTL_RETVAL_COMPLETE);
5763	}
5764
5765	len = scsi_3btoul(cdb->length);
5766	buffer_offset = scsi_3btoul(cdb->offset);
5767
5768	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5769		ctl_set_invalid_field(ctsio,
5770				      /*sks_valid*/ 1,
5771				      /*command*/ 1,
5772				      /*field*/ 6,
5773				      /*bit_valid*/ 0,
5774				      /*bit*/ 0);
5775		ctl_done((union ctl_io *)ctsio);
5776		return (CTL_RETVAL_COMPLETE);
5777	}
5778
5779	/*
5780	 * If we've got a kernel request that hasn't been malloced yet,
5781	 * malloc it and tell the caller the data buffer is here.
5782	 */
5783	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5784		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5785		ctsio->kern_data_len = len;
5786		ctsio->kern_total_len = len;
5787		ctsio->kern_data_resid = 0;
5788		ctsio->kern_rel_offset = 0;
5789		ctsio->kern_sg_entries = 0;
5790		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5791		ctsio->be_move_done = ctl_config_move_done;
5792		ctl_datamove((union ctl_io *)ctsio);
5793
5794		return (CTL_RETVAL_COMPLETE);
5795	}
5796
5797	ctl_done((union ctl_io *)ctsio);
5798
5799	return (CTL_RETVAL_COMPLETE);
5800}
5801
5802int
5803ctl_write_same(struct ctl_scsiio *ctsio)
5804{
5805	struct ctl_lun *lun;
5806	struct ctl_lba_len_flags *lbalen;
5807	uint64_t lba;
5808	uint32_t num_blocks;
5809	int len, retval;
5810	uint8_t byte2;
5811
5812	retval = CTL_RETVAL_COMPLETE;
5813
5814	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5815
5816	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5817
5818	switch (ctsio->cdb[0]) {
5819	case WRITE_SAME_10: {
5820		struct scsi_write_same_10 *cdb;
5821
5822		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5823
5824		lba = scsi_4btoul(cdb->addr);
5825		num_blocks = scsi_2btoul(cdb->length);
5826		byte2 = cdb->byte2;
5827		break;
5828	}
5829	case WRITE_SAME_16: {
5830		struct scsi_write_same_16 *cdb;
5831
5832		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5833
5834		lba = scsi_8btou64(cdb->addr);
5835		num_blocks = scsi_4btoul(cdb->length);
5836		byte2 = cdb->byte2;
5837		break;
5838	}
5839	default:
5840		/*
5841		 * We got a command we don't support.  This shouldn't
5842		 * happen, commands should be filtered out above us.
5843		 */
5844		ctl_set_invalid_opcode(ctsio);
5845		ctl_done((union ctl_io *)ctsio);
5846
5847		return (CTL_RETVAL_COMPLETE);
5848		break; /* NOTREACHED */
5849	}
5850
5851	/*
5852	 * The first check is to make sure we're in bounds, the second
5853	 * check is to catch wrap-around problems.  If the lba + num blocks
5854	 * is less than the lba, then we've wrapped around and the block
5855	 * range is invalid anyway.
5856	 */
5857	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5858	 || ((lba + num_blocks) < lba)) {
5859		ctl_set_lba_out_of_range(ctsio);
5860		ctl_done((union ctl_io *)ctsio);
5861		return (CTL_RETVAL_COMPLETE);
5862	}
5863
5864	/* Zero number of blocks means "to the last logical block" */
5865	if (num_blocks == 0) {
5866		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5867			ctl_set_invalid_field(ctsio,
5868					      /*sks_valid*/ 0,
5869					      /*command*/ 1,
5870					      /*field*/ 0,
5871					      /*bit_valid*/ 0,
5872					      /*bit*/ 0);
5873			ctl_done((union ctl_io *)ctsio);
5874			return (CTL_RETVAL_COMPLETE);
5875		}
5876		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5877	}
5878
5879	len = lun->be_lun->blocksize;
5880
5881	/*
5882	 * If we've got a kernel request that hasn't been malloced yet,
5883	 * malloc it and tell the caller the data buffer is here.
5884	 */
5885	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5886		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5887		ctsio->kern_data_len = len;
5888		ctsio->kern_total_len = len;
5889		ctsio->kern_data_resid = 0;
5890		ctsio->kern_rel_offset = 0;
5891		ctsio->kern_sg_entries = 0;
5892		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5893		ctsio->be_move_done = ctl_config_move_done;
5894		ctl_datamove((union ctl_io *)ctsio);
5895
5896		return (CTL_RETVAL_COMPLETE);
5897	}
5898
5899	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5900	lbalen->lba = lba;
5901	lbalen->len = num_blocks;
5902	lbalen->flags = byte2;
5903	retval = lun->backend->config_write((union ctl_io *)ctsio);
5904
5905	return (retval);
5906}
5907
5908int
5909ctl_unmap(struct ctl_scsiio *ctsio)
5910{
5911	struct ctl_lun *lun;
5912	struct scsi_unmap *cdb;
5913	struct ctl_ptr_len_flags *ptrlen;
5914	struct scsi_unmap_header *hdr;
5915	struct scsi_unmap_desc *buf, *end;
5916	uint64_t lba;
5917	uint32_t num_blocks;
5918	int len, retval;
5919	uint8_t byte2;
5920
5921	retval = CTL_RETVAL_COMPLETE;
5922
5923	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5924
5925	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5926	cdb = (struct scsi_unmap *)ctsio->cdb;
5927
5928	len = scsi_2btoul(cdb->length);
5929	byte2 = cdb->byte2;
5930
5931	/*
5932	 * If we've got a kernel request that hasn't been malloced yet,
5933	 * malloc it and tell the caller the data buffer is here.
5934	 */
5935	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5936		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5937		ctsio->kern_data_len = len;
5938		ctsio->kern_total_len = len;
5939		ctsio->kern_data_resid = 0;
5940		ctsio->kern_rel_offset = 0;
5941		ctsio->kern_sg_entries = 0;
5942		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5943		ctsio->be_move_done = ctl_config_move_done;
5944		ctl_datamove((union ctl_io *)ctsio);
5945
5946		return (CTL_RETVAL_COMPLETE);
5947	}
5948
5949	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5950	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5951	if (len < sizeof (*hdr) ||
5952	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5953	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5954	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5955		ctl_set_invalid_field(ctsio,
5956				      /*sks_valid*/ 0,
5957				      /*command*/ 0,
5958				      /*field*/ 0,
5959				      /*bit_valid*/ 0,
5960				      /*bit*/ 0);
5961		ctl_done((union ctl_io *)ctsio);
5962		return (CTL_RETVAL_COMPLETE);
5963	}
5964	len = scsi_2btoul(hdr->desc_length);
5965	buf = (struct scsi_unmap_desc *)(hdr + 1);
5966	end = buf + len / sizeof(*buf);
5967
5968	ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5969	ptrlen->ptr = (void *)buf;
5970	ptrlen->len = len;
5971	ptrlen->flags = byte2;
5972
5973	for (; buf < end; buf++) {
5974		lba = scsi_8btou64(buf->lba);
5975		num_blocks = scsi_4btoul(buf->length);
5976		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5977		 || ((lba + num_blocks) < lba)) {
5978			ctl_set_lba_out_of_range(ctsio);
5979			ctl_done((union ctl_io *)ctsio);
5980			return (CTL_RETVAL_COMPLETE);
5981		}
5982	}
5983
5984	retval = lun->backend->config_write((union ctl_io *)ctsio);
5985
5986	return (retval);
5987}
5988
5989/*
5990 * Note that this function currently doesn't actually do anything inside
5991 * CTL to enforce things if the DQue bit is turned on.
5992 *
5993 * Also note that this function can't be used in the default case, because
5994 * the DQue bit isn't set in the changeable mask for the control mode page
5995 * anyway.  This is just here as an example for how to implement a page
5996 * handler, and a placeholder in case we want to allow the user to turn
5997 * tagged queueing on and off.
5998 *
5999 * The D_SENSE bit handling is functional, however, and will turn
6000 * descriptor sense on and off for a given LUN.
6001 */
6002int
6003ctl_control_page_handler(struct ctl_scsiio *ctsio,
6004			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6005{
6006	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6007	struct ctl_lun *lun;
6008	struct ctl_softc *softc;
6009	int set_ua;
6010	uint32_t initidx;
6011
6012	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6013	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6014	set_ua = 0;
6015
6016	user_cp = (struct scsi_control_page *)page_ptr;
6017	current_cp = (struct scsi_control_page *)
6018		(page_index->page_data + (page_index->page_len *
6019		CTL_PAGE_CURRENT));
6020	saved_cp = (struct scsi_control_page *)
6021		(page_index->page_data + (page_index->page_len *
6022		CTL_PAGE_SAVED));
6023
6024	softc = control_softc;
6025
6026	mtx_lock(&lun->lun_lock);
6027	if (((current_cp->rlec & SCP_DSENSE) == 0)
6028	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6029		/*
6030		 * Descriptor sense is currently turned off and the user
6031		 * wants to turn it on.
6032		 */
6033		current_cp->rlec |= SCP_DSENSE;
6034		saved_cp->rlec |= SCP_DSENSE;
6035		lun->flags |= CTL_LUN_SENSE_DESC;
6036		set_ua = 1;
6037	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6038		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6039		/*
6040		 * Descriptor sense is currently turned on, and the user
6041		 * wants to turn it off.
6042		 */
6043		current_cp->rlec &= ~SCP_DSENSE;
6044		saved_cp->rlec &= ~SCP_DSENSE;
6045		lun->flags &= ~CTL_LUN_SENSE_DESC;
6046		set_ua = 1;
6047	}
6048	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
6049		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6050#ifdef NEEDTOPORT
6051			csevent_log(CSC_CTL | CSC_SHELF_SW |
6052				    CTL_UNTAG_TO_UNTAG,
6053				    csevent_LogType_Trace,
6054				    csevent_Severity_Information,
6055				    csevent_AlertLevel_Green,
6056				    csevent_FRU_Firmware,
6057				    csevent_FRU_Unknown,
6058				    "Received untagged to untagged transition");
6059#endif /* NEEDTOPORT */
6060		} else {
6061#ifdef NEEDTOPORT
6062			csevent_log(CSC_CTL | CSC_SHELF_SW |
6063				    CTL_UNTAG_TO_TAG,
6064				    csevent_LogType_ConfigChange,
6065				    csevent_Severity_Information,
6066				    csevent_AlertLevel_Green,
6067				    csevent_FRU_Firmware,
6068				    csevent_FRU_Unknown,
6069				    "Received untagged to tagged "
6070				    "queueing transition");
6071#endif /* NEEDTOPORT */
6072
6073			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6074			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6075			set_ua = 1;
6076		}
6077	} else {
6078		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6079#ifdef NEEDTOPORT
6080			csevent_log(CSC_CTL | CSC_SHELF_SW |
6081				    CTL_TAG_TO_UNTAG,
6082				    csevent_LogType_ConfigChange,
6083				    csevent_Severity_Warning,
6084				    csevent_AlertLevel_Yellow,
6085				    csevent_FRU_Firmware,
6086				    csevent_FRU_Unknown,
6087				    "Received tagged queueing to untagged "
6088				    "transition");
6089#endif /* NEEDTOPORT */
6090
6091			current_cp->queue_flags |= SCP_QUEUE_DQUE;
6092			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
6093			set_ua = 1;
6094		} else {
6095#ifdef NEEDTOPORT
6096			csevent_log(CSC_CTL | CSC_SHELF_SW |
6097				    CTL_TAG_TO_TAG,
6098				    csevent_LogType_Trace,
6099				    csevent_Severity_Information,
6100				    csevent_AlertLevel_Green,
6101				    csevent_FRU_Firmware,
6102				    csevent_FRU_Unknown,
6103				    "Received tagged queueing to tagged "
6104				    "queueing transition");
6105#endif /* NEEDTOPORT */
6106		}
6107	}
6108	if (set_ua != 0) {
6109		int i;
6110		/*
6111		 * Let other initiators know that the mode
6112		 * parameters for this LUN have changed.
6113		 */
6114		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6115			if (i == initidx)
6116				continue;
6117
6118			lun->pending_sense[i].ua_pending |=
6119				CTL_UA_MODE_CHANGE;
6120		}
6121	}
6122	mtx_unlock(&lun->lun_lock);
6123
6124	return (0);
6125}
6126
6127int
6128ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6129		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6130{
6131	return (0);
6132}
6133
6134int
6135ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6136			   struct ctl_page_index *page_index, int pc)
6137{
6138	struct copan_power_subpage *page;
6139
6140	page = (struct copan_power_subpage *)page_index->page_data +
6141		(page_index->page_len * pc);
6142
6143	switch (pc) {
6144	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6145		/*
6146		 * We don't update the changable bits for this page.
6147		 */
6148		break;
6149	case SMS_PAGE_CTRL_CURRENT >> 6:
6150	case SMS_PAGE_CTRL_DEFAULT >> 6:
6151	case SMS_PAGE_CTRL_SAVED >> 6:
6152#ifdef NEEDTOPORT
6153		ctl_update_power_subpage(page);
6154#endif
6155		break;
6156	default:
6157#ifdef NEEDTOPORT
6158		EPRINT(0, "Invalid PC %d!!", pc);
6159#endif
6160		break;
6161	}
6162	return (0);
6163}
6164
6165
6166int
6167ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6168		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6169{
6170	struct copan_aps_subpage *user_sp;
6171	struct copan_aps_subpage *current_sp;
6172	union ctl_modepage_info *modepage_info;
6173	struct ctl_softc *softc;
6174	struct ctl_lun *lun;
6175	int retval;
6176
6177	retval = CTL_RETVAL_COMPLETE;
6178	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6179		     (page_index->page_len * CTL_PAGE_CURRENT));
6180	softc = control_softc;
6181	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6182
6183	user_sp = (struct copan_aps_subpage *)page_ptr;
6184
6185	modepage_info = (union ctl_modepage_info *)
6186		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6187
6188	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6189	modepage_info->header.subpage = page_index->subpage;
6190	modepage_info->aps.lock_active = user_sp->lock_active;
6191
6192	mtx_lock(&softc->ctl_lock);
6193
6194	/*
6195	 * If there is a request to lock the LUN and another LUN is locked
6196	 * this is an error. If the requested LUN is already locked ignore
6197	 * the request. If no LUN is locked attempt to lock it.
6198	 * if there is a request to unlock the LUN and the LUN is currently
6199	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6200	 * if another LUN is locked or no LUN is locked.
6201	 */
6202	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6203		if (softc->aps_locked_lun == lun->lun) {
6204			/*
6205			 * This LUN is already locked, so we're done.
6206			 */
6207			retval = CTL_RETVAL_COMPLETE;
6208		} else if (softc->aps_locked_lun == 0) {
6209			/*
6210			 * No one has the lock, pass the request to the
6211			 * backend.
6212			 */
6213			retval = lun->backend->config_write(
6214				(union ctl_io *)ctsio);
6215		} else {
6216			/*
6217			 * Someone else has the lock, throw out the request.
6218			 */
6219			ctl_set_already_locked(ctsio);
6220			free(ctsio->kern_data_ptr, M_CTL);
6221			ctl_done((union ctl_io *)ctsio);
6222
6223			/*
6224			 * Set the return value so that ctl_do_mode_select()
6225			 * won't try to complete the command.  We already
6226			 * completed it here.
6227			 */
6228			retval = CTL_RETVAL_ERROR;
6229		}
6230	} else if (softc->aps_locked_lun == lun->lun) {
6231		/*
6232		 * This LUN is locked, so pass the unlock request to the
6233		 * backend.
6234		 */
6235		retval = lun->backend->config_write((union ctl_io *)ctsio);
6236	}
6237	mtx_unlock(&softc->ctl_lock);
6238
6239	return (retval);
6240}
6241
6242int
6243ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6244				struct ctl_page_index *page_index,
6245				uint8_t *page_ptr)
6246{
6247	uint8_t *c;
6248	int i;
6249
6250	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6251	ctl_time_io_secs =
6252		(c[0] << 8) |
6253		(c[1] << 0) |
6254		0;
6255	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6256	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6257	printf("page data:");
6258	for (i=0; i<8; i++)
6259		printf(" %.2x",page_ptr[i]);
6260	printf("\n");
6261	return (0);
6262}
6263
6264int
6265ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6266			       struct ctl_page_index *page_index,
6267			       int pc)
6268{
6269	struct copan_debugconf_subpage *page;
6270
6271	page = (struct copan_debugconf_subpage *)page_index->page_data +
6272		(page_index->page_len * pc);
6273
6274	switch (pc) {
6275	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6276	case SMS_PAGE_CTRL_DEFAULT >> 6:
6277	case SMS_PAGE_CTRL_SAVED >> 6:
6278		/*
6279		 * We don't update the changable or default bits for this page.
6280		 */
6281		break;
6282	case SMS_PAGE_CTRL_CURRENT >> 6:
6283		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6284		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6285		break;
6286	default:
6287#ifdef NEEDTOPORT
6288		EPRINT(0, "Invalid PC %d!!", pc);
6289#endif /* NEEDTOPORT */
6290		break;
6291	}
6292	return (0);
6293}
6294
6295
6296static int
6297ctl_do_mode_select(union ctl_io *io)
6298{
6299	struct scsi_mode_page_header *page_header;
6300	struct ctl_page_index *page_index;
6301	struct ctl_scsiio *ctsio;
6302	int control_dev, page_len;
6303	int page_len_offset, page_len_size;
6304	union ctl_modepage_info *modepage_info;
6305	struct ctl_lun *lun;
6306	int *len_left, *len_used;
6307	int retval, i;
6308
6309	ctsio = &io->scsiio;
6310	page_index = NULL;
6311	page_len = 0;
6312	retval = CTL_RETVAL_COMPLETE;
6313
6314	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6315
6316	if (lun->be_lun->lun_type != T_DIRECT)
6317		control_dev = 1;
6318	else
6319		control_dev = 0;
6320
6321	modepage_info = (union ctl_modepage_info *)
6322		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6323	len_left = &modepage_info->header.len_left;
6324	len_used = &modepage_info->header.len_used;
6325
6326do_next_page:
6327
6328	page_header = (struct scsi_mode_page_header *)
6329		(ctsio->kern_data_ptr + *len_used);
6330
6331	if (*len_left == 0) {
6332		free(ctsio->kern_data_ptr, M_CTL);
6333		ctl_set_success(ctsio);
6334		ctl_done((union ctl_io *)ctsio);
6335		return (CTL_RETVAL_COMPLETE);
6336	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6337
6338		free(ctsio->kern_data_ptr, M_CTL);
6339		ctl_set_param_len_error(ctsio);
6340		ctl_done((union ctl_io *)ctsio);
6341		return (CTL_RETVAL_COMPLETE);
6342
6343	} else if ((page_header->page_code & SMPH_SPF)
6344		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6345
6346		free(ctsio->kern_data_ptr, M_CTL);
6347		ctl_set_param_len_error(ctsio);
6348		ctl_done((union ctl_io *)ctsio);
6349		return (CTL_RETVAL_COMPLETE);
6350	}
6351
6352
6353	/*
6354	 * XXX KDM should we do something with the block descriptor?
6355	 */
6356	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6357
6358		if ((control_dev != 0)
6359		 && (lun->mode_pages.index[i].page_flags &
6360		     CTL_PAGE_FLAG_DISK_ONLY))
6361			continue;
6362
6363		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6364		    (page_header->page_code & SMPH_PC_MASK))
6365			continue;
6366
6367		/*
6368		 * If neither page has a subpage code, then we've got a
6369		 * match.
6370		 */
6371		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6372		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6373			page_index = &lun->mode_pages.index[i];
6374			page_len = page_header->page_length;
6375			break;
6376		}
6377
6378		/*
6379		 * If both pages have subpages, then the subpage numbers
6380		 * have to match.
6381		 */
6382		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6383		  && (page_header->page_code & SMPH_SPF)) {
6384			struct scsi_mode_page_header_sp *sph;
6385
6386			sph = (struct scsi_mode_page_header_sp *)page_header;
6387
6388			if (lun->mode_pages.index[i].subpage ==
6389			    sph->subpage) {
6390				page_index = &lun->mode_pages.index[i];
6391				page_len = scsi_2btoul(sph->page_length);
6392				break;
6393			}
6394		}
6395	}
6396
6397	/*
6398	 * If we couldn't find the page, or if we don't have a mode select
6399	 * handler for it, send back an error to the user.
6400	 */
6401	if ((page_index == NULL)
6402	 || (page_index->select_handler == NULL)) {
6403		ctl_set_invalid_field(ctsio,
6404				      /*sks_valid*/ 1,
6405				      /*command*/ 0,
6406				      /*field*/ *len_used,
6407				      /*bit_valid*/ 0,
6408				      /*bit*/ 0);
6409		free(ctsio->kern_data_ptr, M_CTL);
6410		ctl_done((union ctl_io *)ctsio);
6411		return (CTL_RETVAL_COMPLETE);
6412	}
6413
6414	if (page_index->page_code & SMPH_SPF) {
6415		page_len_offset = 2;
6416		page_len_size = 2;
6417	} else {
6418		page_len_size = 1;
6419		page_len_offset = 1;
6420	}
6421
6422	/*
6423	 * If the length the initiator gives us isn't the one we specify in
6424	 * the mode page header, or if they didn't specify enough data in
6425	 * the CDB to avoid truncating this page, kick out the request.
6426	 */
6427	if ((page_len != (page_index->page_len - page_len_offset -
6428			  page_len_size))
6429	 || (*len_left < page_index->page_len)) {
6430
6431
6432		ctl_set_invalid_field(ctsio,
6433				      /*sks_valid*/ 1,
6434				      /*command*/ 0,
6435				      /*field*/ *len_used + page_len_offset,
6436				      /*bit_valid*/ 0,
6437				      /*bit*/ 0);
6438		free(ctsio->kern_data_ptr, M_CTL);
6439		ctl_done((union ctl_io *)ctsio);
6440		return (CTL_RETVAL_COMPLETE);
6441	}
6442
6443	/*
6444	 * Run through the mode page, checking to make sure that the bits
6445	 * the user changed are actually legal for him to change.
6446	 */
6447	for (i = 0; i < page_index->page_len; i++) {
6448		uint8_t *user_byte, *change_mask, *current_byte;
6449		int bad_bit;
6450		int j;
6451
6452		user_byte = (uint8_t *)page_header + i;
6453		change_mask = page_index->page_data +
6454			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6455		current_byte = page_index->page_data +
6456			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6457
6458		/*
6459		 * Check to see whether the user set any bits in this byte
6460		 * that he is not allowed to set.
6461		 */
6462		if ((*user_byte & ~(*change_mask)) ==
6463		    (*current_byte & ~(*change_mask)))
6464			continue;
6465
6466		/*
6467		 * Go through bit by bit to determine which one is illegal.
6468		 */
6469		bad_bit = 0;
6470		for (j = 7; j >= 0; j--) {
6471			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6472			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6473				bad_bit = i;
6474				break;
6475			}
6476		}
6477		ctl_set_invalid_field(ctsio,
6478				      /*sks_valid*/ 1,
6479				      /*command*/ 0,
6480				      /*field*/ *len_used + i,
6481				      /*bit_valid*/ 1,
6482				      /*bit*/ bad_bit);
6483		free(ctsio->kern_data_ptr, M_CTL);
6484		ctl_done((union ctl_io *)ctsio);
6485		return (CTL_RETVAL_COMPLETE);
6486	}
6487
6488	/*
6489	 * Decrement these before we call the page handler, since we may
6490	 * end up getting called back one way or another before the handler
6491	 * returns to this context.
6492	 */
6493	*len_left -= page_index->page_len;
6494	*len_used += page_index->page_len;
6495
6496	retval = page_index->select_handler(ctsio, page_index,
6497					    (uint8_t *)page_header);
6498
6499	/*
6500	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6501	 * wait until this queued command completes to finish processing
6502	 * the mode page.  If it returns anything other than
6503	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6504	 * already set the sense information, freed the data pointer, and
6505	 * completed the io for us.
6506	 */
6507	if (retval != CTL_RETVAL_COMPLETE)
6508		goto bailout_no_done;
6509
6510	/*
6511	 * If the initiator sent us more than one page, parse the next one.
6512	 */
6513	if (*len_left > 0)
6514		goto do_next_page;
6515
6516	ctl_set_success(ctsio);
6517	free(ctsio->kern_data_ptr, M_CTL);
6518	ctl_done((union ctl_io *)ctsio);
6519
6520bailout_no_done:
6521
6522	return (CTL_RETVAL_COMPLETE);
6523
6524}
6525
6526int
6527ctl_mode_select(struct ctl_scsiio *ctsio)
6528{
6529	int param_len, pf, sp;
6530	int header_size, bd_len;
6531	int len_left, len_used;
6532	struct ctl_page_index *page_index;
6533	struct ctl_lun *lun;
6534	int control_dev, page_len;
6535	union ctl_modepage_info *modepage_info;
6536	int retval;
6537
6538	pf = 0;
6539	sp = 0;
6540	page_len = 0;
6541	len_used = 0;
6542	len_left = 0;
6543	retval = 0;
6544	bd_len = 0;
6545	page_index = NULL;
6546
6547	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6548
6549	if (lun->be_lun->lun_type != T_DIRECT)
6550		control_dev = 1;
6551	else
6552		control_dev = 0;
6553
6554	switch (ctsio->cdb[0]) {
6555	case MODE_SELECT_6: {
6556		struct scsi_mode_select_6 *cdb;
6557
6558		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6559
6560		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6561		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6562
6563		param_len = cdb->length;
6564		header_size = sizeof(struct scsi_mode_header_6);
6565		break;
6566	}
6567	case MODE_SELECT_10: {
6568		struct scsi_mode_select_10 *cdb;
6569
6570		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6571
6572		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6573		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6574
6575		param_len = scsi_2btoul(cdb->length);
6576		header_size = sizeof(struct scsi_mode_header_10);
6577		break;
6578	}
6579	default:
6580		ctl_set_invalid_opcode(ctsio);
6581		ctl_done((union ctl_io *)ctsio);
6582		return (CTL_RETVAL_COMPLETE);
6583		break; /* NOTREACHED */
6584	}
6585
6586	/*
6587	 * From SPC-3:
6588	 * "A parameter list length of zero indicates that the Data-Out Buffer
6589	 * shall be empty. This condition shall not be considered as an error."
6590	 */
6591	if (param_len == 0) {
6592		ctl_set_success(ctsio);
6593		ctl_done((union ctl_io *)ctsio);
6594		return (CTL_RETVAL_COMPLETE);
6595	}
6596
6597	/*
6598	 * Since we'll hit this the first time through, prior to
6599	 * allocation, we don't need to free a data buffer here.
6600	 */
6601	if (param_len < header_size) {
6602		ctl_set_param_len_error(ctsio);
6603		ctl_done((union ctl_io *)ctsio);
6604		return (CTL_RETVAL_COMPLETE);
6605	}
6606
6607	/*
6608	 * Allocate the data buffer and grab the user's data.  In theory,
6609	 * we shouldn't have to sanity check the parameter list length here
6610	 * because the maximum size is 64K.  We should be able to malloc
6611	 * that much without too many problems.
6612	 */
6613	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6614		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6615		ctsio->kern_data_len = param_len;
6616		ctsio->kern_total_len = param_len;
6617		ctsio->kern_data_resid = 0;
6618		ctsio->kern_rel_offset = 0;
6619		ctsio->kern_sg_entries = 0;
6620		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6621		ctsio->be_move_done = ctl_config_move_done;
6622		ctl_datamove((union ctl_io *)ctsio);
6623
6624		return (CTL_RETVAL_COMPLETE);
6625	}
6626
6627	switch (ctsio->cdb[0]) {
6628	case MODE_SELECT_6: {
6629		struct scsi_mode_header_6 *mh6;
6630
6631		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6632		bd_len = mh6->blk_desc_len;
6633		break;
6634	}
6635	case MODE_SELECT_10: {
6636		struct scsi_mode_header_10 *mh10;
6637
6638		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6639		bd_len = scsi_2btoul(mh10->blk_desc_len);
6640		break;
6641	}
6642	default:
6643		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6644		break;
6645	}
6646
6647	if (param_len < (header_size + bd_len)) {
6648		free(ctsio->kern_data_ptr, M_CTL);
6649		ctl_set_param_len_error(ctsio);
6650		ctl_done((union ctl_io *)ctsio);
6651		return (CTL_RETVAL_COMPLETE);
6652	}
6653
6654	/*
6655	 * Set the IO_CONT flag, so that if this I/O gets passed to
6656	 * ctl_config_write_done(), it'll get passed back to
6657	 * ctl_do_mode_select() for further processing, or completion if
6658	 * we're all done.
6659	 */
6660	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6661	ctsio->io_cont = ctl_do_mode_select;
6662
6663	modepage_info = (union ctl_modepage_info *)
6664		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6665
6666	memset(modepage_info, 0, sizeof(*modepage_info));
6667
6668	len_left = param_len - header_size - bd_len;
6669	len_used = header_size + bd_len;
6670
6671	modepage_info->header.len_left = len_left;
6672	modepage_info->header.len_used = len_used;
6673
6674	return (ctl_do_mode_select((union ctl_io *)ctsio));
6675}
6676
6677int
6678ctl_mode_sense(struct ctl_scsiio *ctsio)
6679{
6680	struct ctl_lun *lun;
6681	int pc, page_code, dbd, llba, subpage;
6682	int alloc_len, page_len, header_len, total_len;
6683	struct scsi_mode_block_descr *block_desc;
6684	struct ctl_page_index *page_index;
6685	int control_dev;
6686
6687	dbd = 0;
6688	llba = 0;
6689	block_desc = NULL;
6690	page_index = NULL;
6691
6692	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6693
6694	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6695
6696	if (lun->be_lun->lun_type != T_DIRECT)
6697		control_dev = 1;
6698	else
6699		control_dev = 0;
6700
6701	if (lun->flags & CTL_LUN_PR_RESERVED) {
6702		uint32_t residx;
6703
6704		/*
6705		 * XXX KDM need a lock here.
6706		 */
6707		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6708		if ((lun->res_type == SPR_TYPE_EX_AC
6709		  && residx != lun->pr_res_idx)
6710		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6711		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6712		  && !lun->per_res[residx].registered)) {
6713			ctl_set_reservation_conflict(ctsio);
6714			ctl_done((union ctl_io *)ctsio);
6715			return (CTL_RETVAL_COMPLETE);
6716		}
6717	}
6718
6719	switch (ctsio->cdb[0]) {
6720	case MODE_SENSE_6: {
6721		struct scsi_mode_sense_6 *cdb;
6722
6723		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6724
6725		header_len = sizeof(struct scsi_mode_hdr_6);
6726		if (cdb->byte2 & SMS_DBD)
6727			dbd = 1;
6728		else
6729			header_len += sizeof(struct scsi_mode_block_descr);
6730
6731		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6732		page_code = cdb->page & SMS_PAGE_CODE;
6733		subpage = cdb->subpage;
6734		alloc_len = cdb->length;
6735		break;
6736	}
6737	case MODE_SENSE_10: {
6738		struct scsi_mode_sense_10 *cdb;
6739
6740		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6741
6742		header_len = sizeof(struct scsi_mode_hdr_10);
6743
6744		if (cdb->byte2 & SMS_DBD)
6745			dbd = 1;
6746		else
6747			header_len += sizeof(struct scsi_mode_block_descr);
6748		if (cdb->byte2 & SMS10_LLBAA)
6749			llba = 1;
6750		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6751		page_code = cdb->page & SMS_PAGE_CODE;
6752		subpage = cdb->subpage;
6753		alloc_len = scsi_2btoul(cdb->length);
6754		break;
6755	}
6756	default:
6757		ctl_set_invalid_opcode(ctsio);
6758		ctl_done((union ctl_io *)ctsio);
6759		return (CTL_RETVAL_COMPLETE);
6760		break; /* NOTREACHED */
6761	}
6762
6763	/*
6764	 * We have to make a first pass through to calculate the size of
6765	 * the pages that match the user's query.  Then we allocate enough
6766	 * memory to hold it, and actually copy the data into the buffer.
6767	 */
6768	switch (page_code) {
6769	case SMS_ALL_PAGES_PAGE: {
6770		int i;
6771
6772		page_len = 0;
6773
6774		/*
6775		 * At the moment, values other than 0 and 0xff here are
6776		 * reserved according to SPC-3.
6777		 */
6778		if ((subpage != SMS_SUBPAGE_PAGE_0)
6779		 && (subpage != SMS_SUBPAGE_ALL)) {
6780			ctl_set_invalid_field(ctsio,
6781					      /*sks_valid*/ 1,
6782					      /*command*/ 1,
6783					      /*field*/ 3,
6784					      /*bit_valid*/ 0,
6785					      /*bit*/ 0);
6786			ctl_done((union ctl_io *)ctsio);
6787			return (CTL_RETVAL_COMPLETE);
6788		}
6789
6790		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6791			if ((control_dev != 0)
6792			 && (lun->mode_pages.index[i].page_flags &
6793			     CTL_PAGE_FLAG_DISK_ONLY))
6794				continue;
6795
6796			/*
6797			 * We don't use this subpage if the user didn't
6798			 * request all subpages.
6799			 */
6800			if ((lun->mode_pages.index[i].subpage != 0)
6801			 && (subpage == SMS_SUBPAGE_PAGE_0))
6802				continue;
6803
6804#if 0
6805			printf("found page %#x len %d\n",
6806			       lun->mode_pages.index[i].page_code &
6807			       SMPH_PC_MASK,
6808			       lun->mode_pages.index[i].page_len);
6809#endif
6810			page_len += lun->mode_pages.index[i].page_len;
6811		}
6812		break;
6813	}
6814	default: {
6815		int i;
6816
6817		page_len = 0;
6818
6819		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6820			/* Look for the right page code */
6821			if ((lun->mode_pages.index[i].page_code &
6822			     SMPH_PC_MASK) != page_code)
6823				continue;
6824
6825			/* Look for the right subpage or the subpage wildcard*/
6826			if ((lun->mode_pages.index[i].subpage != subpage)
6827			 && (subpage != SMS_SUBPAGE_ALL))
6828				continue;
6829
6830			/* Make sure the page is supported for this dev type */
6831			if ((control_dev != 0)
6832			 && (lun->mode_pages.index[i].page_flags &
6833			     CTL_PAGE_FLAG_DISK_ONLY))
6834				continue;
6835
6836#if 0
6837			printf("found page %#x len %d\n",
6838			       lun->mode_pages.index[i].page_code &
6839			       SMPH_PC_MASK,
6840			       lun->mode_pages.index[i].page_len);
6841#endif
6842
6843			page_len += lun->mode_pages.index[i].page_len;
6844		}
6845
6846		if (page_len == 0) {
6847			ctl_set_invalid_field(ctsio,
6848					      /*sks_valid*/ 1,
6849					      /*command*/ 1,
6850					      /*field*/ 2,
6851					      /*bit_valid*/ 1,
6852					      /*bit*/ 5);
6853			ctl_done((union ctl_io *)ctsio);
6854			return (CTL_RETVAL_COMPLETE);
6855		}
6856		break;
6857	}
6858	}
6859
6860	total_len = header_len + page_len;
6861#if 0
6862	printf("header_len = %d, page_len = %d, total_len = %d\n",
6863	       header_len, page_len, total_len);
6864#endif
6865
6866	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6867	ctsio->kern_sg_entries = 0;
6868	ctsio->kern_data_resid = 0;
6869	ctsio->kern_rel_offset = 0;
6870	if (total_len < alloc_len) {
6871		ctsio->residual = alloc_len - total_len;
6872		ctsio->kern_data_len = total_len;
6873		ctsio->kern_total_len = total_len;
6874	} else {
6875		ctsio->residual = 0;
6876		ctsio->kern_data_len = alloc_len;
6877		ctsio->kern_total_len = alloc_len;
6878	}
6879
6880	switch (ctsio->cdb[0]) {
6881	case MODE_SENSE_6: {
6882		struct scsi_mode_hdr_6 *header;
6883
6884		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6885
6886		header->datalen = ctl_min(total_len - 1, 254);
6887
6888		if (dbd)
6889			header->block_descr_len = 0;
6890		else
6891			header->block_descr_len =
6892				sizeof(struct scsi_mode_block_descr);
6893		block_desc = (struct scsi_mode_block_descr *)&header[1];
6894		break;
6895	}
6896	case MODE_SENSE_10: {
6897		struct scsi_mode_hdr_10 *header;
6898		int datalen;
6899
6900		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6901
6902		datalen = ctl_min(total_len - 2, 65533);
6903		scsi_ulto2b(datalen, header->datalen);
6904		if (dbd)
6905			scsi_ulto2b(0, header->block_descr_len);
6906		else
6907			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6908				    header->block_descr_len);
6909		block_desc = (struct scsi_mode_block_descr *)&header[1];
6910		break;
6911	}
6912	default:
6913		panic("invalid CDB type %#x", ctsio->cdb[0]);
6914		break; /* NOTREACHED */
6915	}
6916
6917	/*
6918	 * If we've got a disk, use its blocksize in the block
6919	 * descriptor.  Otherwise, just set it to 0.
6920	 */
6921	if (dbd == 0) {
6922		if (control_dev != 0)
6923			scsi_ulto3b(lun->be_lun->blocksize,
6924				    block_desc->block_len);
6925		else
6926			scsi_ulto3b(0, block_desc->block_len);
6927	}
6928
6929	switch (page_code) {
6930	case SMS_ALL_PAGES_PAGE: {
6931		int i, data_used;
6932
6933		data_used = header_len;
6934		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6935			struct ctl_page_index *page_index;
6936
6937			page_index = &lun->mode_pages.index[i];
6938
6939			if ((control_dev != 0)
6940			 && (page_index->page_flags &
6941			    CTL_PAGE_FLAG_DISK_ONLY))
6942				continue;
6943
6944			/*
6945			 * We don't use this subpage if the user didn't
6946			 * request all subpages.  We already checked (above)
6947			 * to make sure the user only specified a subpage
6948			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6949			 */
6950			if ((page_index->subpage != 0)
6951			 && (subpage == SMS_SUBPAGE_PAGE_0))
6952				continue;
6953
6954			/*
6955			 * Call the handler, if it exists, to update the
6956			 * page to the latest values.
6957			 */
6958			if (page_index->sense_handler != NULL)
6959				page_index->sense_handler(ctsio, page_index,pc);
6960
6961			memcpy(ctsio->kern_data_ptr + data_used,
6962			       page_index->page_data +
6963			       (page_index->page_len * pc),
6964			       page_index->page_len);
6965			data_used += page_index->page_len;
6966		}
6967		break;
6968	}
6969	default: {
6970		int i, data_used;
6971
6972		data_used = header_len;
6973
6974		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6975			struct ctl_page_index *page_index;
6976
6977			page_index = &lun->mode_pages.index[i];
6978
6979			/* Look for the right page code */
6980			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6981				continue;
6982
6983			/* Look for the right subpage or the subpage wildcard*/
6984			if ((page_index->subpage != subpage)
6985			 && (subpage != SMS_SUBPAGE_ALL))
6986				continue;
6987
6988			/* Make sure the page is supported for this dev type */
6989			if ((control_dev != 0)
6990			 && (page_index->page_flags &
6991			     CTL_PAGE_FLAG_DISK_ONLY))
6992				continue;
6993
6994			/*
6995			 * Call the handler, if it exists, to update the
6996			 * page to the latest values.
6997			 */
6998			if (page_index->sense_handler != NULL)
6999				page_index->sense_handler(ctsio, page_index,pc);
7000
7001			memcpy(ctsio->kern_data_ptr + data_used,
7002			       page_index->page_data +
7003			       (page_index->page_len * pc),
7004			       page_index->page_len);
7005			data_used += page_index->page_len;
7006		}
7007		break;
7008	}
7009	}
7010
7011	ctsio->scsi_status = SCSI_STATUS_OK;
7012
7013	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7014	ctsio->be_move_done = ctl_config_move_done;
7015	ctl_datamove((union ctl_io *)ctsio);
7016
7017	return (CTL_RETVAL_COMPLETE);
7018}
7019
7020int
7021ctl_read_capacity(struct ctl_scsiio *ctsio)
7022{
7023	struct scsi_read_capacity *cdb;
7024	struct scsi_read_capacity_data *data;
7025	struct ctl_lun *lun;
7026	uint32_t lba;
7027
7028	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7029
7030	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7031
7032	lba = scsi_4btoul(cdb->addr);
7033	if (((cdb->pmi & SRC_PMI) == 0)
7034	 && (lba != 0)) {
7035		ctl_set_invalid_field(/*ctsio*/ ctsio,
7036				      /*sks_valid*/ 1,
7037				      /*command*/ 1,
7038				      /*field*/ 2,
7039				      /*bit_valid*/ 0,
7040				      /*bit*/ 0);
7041		ctl_done((union ctl_io *)ctsio);
7042		return (CTL_RETVAL_COMPLETE);
7043	}
7044
7045	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7046
7047	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7048	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7049	ctsio->residual = 0;
7050	ctsio->kern_data_len = sizeof(*data);
7051	ctsio->kern_total_len = sizeof(*data);
7052	ctsio->kern_data_resid = 0;
7053	ctsio->kern_rel_offset = 0;
7054	ctsio->kern_sg_entries = 0;
7055
7056	/*
7057	 * If the maximum LBA is greater than 0xfffffffe, the user must
7058	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7059	 * serivce action set.
7060	 */
7061	if (lun->be_lun->maxlba > 0xfffffffe)
7062		scsi_ulto4b(0xffffffff, data->addr);
7063	else
7064		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7065
7066	/*
7067	 * XXX KDM this may not be 512 bytes...
7068	 */
7069	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7070
7071	ctsio->scsi_status = SCSI_STATUS_OK;
7072
7073	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7074	ctsio->be_move_done = ctl_config_move_done;
7075	ctl_datamove((union ctl_io *)ctsio);
7076
7077	return (CTL_RETVAL_COMPLETE);
7078}
7079
7080int
7081ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7082{
7083	struct scsi_read_capacity_16 *cdb;
7084	struct scsi_read_capacity_data_long *data;
7085	struct ctl_lun *lun;
7086	uint64_t lba;
7087	uint32_t alloc_len;
7088
7089	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7090
7091	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7092
7093	alloc_len = scsi_4btoul(cdb->alloc_len);
7094	lba = scsi_8btou64(cdb->addr);
7095
7096	if ((cdb->reladr & SRC16_PMI)
7097	 && (lba != 0)) {
7098		ctl_set_invalid_field(/*ctsio*/ ctsio,
7099				      /*sks_valid*/ 1,
7100				      /*command*/ 1,
7101				      /*field*/ 2,
7102				      /*bit_valid*/ 0,
7103				      /*bit*/ 0);
7104		ctl_done((union ctl_io *)ctsio);
7105		return (CTL_RETVAL_COMPLETE);
7106	}
7107
7108	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7109
7110	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7111	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7112
7113	if (sizeof(*data) < alloc_len) {
7114		ctsio->residual = alloc_len - sizeof(*data);
7115		ctsio->kern_data_len = sizeof(*data);
7116		ctsio->kern_total_len = sizeof(*data);
7117	} else {
7118		ctsio->residual = 0;
7119		ctsio->kern_data_len = alloc_len;
7120		ctsio->kern_total_len = alloc_len;
7121	}
7122	ctsio->kern_data_resid = 0;
7123	ctsio->kern_rel_offset = 0;
7124	ctsio->kern_sg_entries = 0;
7125
7126	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7127	/* XXX KDM this may not be 512 bytes... */
7128	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7129	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7130	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7131	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7132		data->lalba_lbp[0] |= SRC16_LBPME;
7133
7134	ctsio->scsi_status = SCSI_STATUS_OK;
7135
7136	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7137	ctsio->be_move_done = ctl_config_move_done;
7138	ctl_datamove((union ctl_io *)ctsio);
7139
7140	return (CTL_RETVAL_COMPLETE);
7141}
7142
7143int
7144ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7145{
7146	struct scsi_maintenance_in *cdb;
7147	int retval;
7148	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7149	int num_target_port_groups, num_target_ports, single;
7150	struct ctl_lun *lun;
7151	struct ctl_softc *softc;
7152	struct ctl_port *port;
7153	struct scsi_target_group_data *rtg_ptr;
7154	struct scsi_target_group_data_extended *rtg_ext_ptr;
7155	struct scsi_target_port_group_descriptor *tpg_desc;
7156
7157	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7158
7159	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7160	softc = control_softc;
7161	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7162
7163	retval = CTL_RETVAL_COMPLETE;
7164
7165	switch (cdb->byte2 & STG_PDF_MASK) {
7166	case STG_PDF_LENGTH:
7167		ext = 0;
7168		break;
7169	case STG_PDF_EXTENDED:
7170		ext = 1;
7171		break;
7172	default:
7173		ctl_set_invalid_field(/*ctsio*/ ctsio,
7174				      /*sks_valid*/ 1,
7175				      /*command*/ 1,
7176				      /*field*/ 2,
7177				      /*bit_valid*/ 1,
7178				      /*bit*/ 5);
7179		ctl_done((union ctl_io *)ctsio);
7180		return(retval);
7181	}
7182
7183	single = ctl_is_single;
7184	if (single)
7185		num_target_port_groups = 1;
7186	else
7187		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7188	num_target_ports = 0;
7189	mtx_lock(&softc->ctl_lock);
7190	STAILQ_FOREACH(port, &softc->port_list, links) {
7191		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7192			continue;
7193		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7194			continue;
7195		num_target_ports++;
7196	}
7197	mtx_unlock(&softc->ctl_lock);
7198
7199	if (ext)
7200		total_len = sizeof(struct scsi_target_group_data_extended);
7201	else
7202		total_len = sizeof(struct scsi_target_group_data);
7203	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7204		num_target_port_groups +
7205	    sizeof(struct scsi_target_port_descriptor) *
7206		num_target_ports * num_target_port_groups;
7207
7208	alloc_len = scsi_4btoul(cdb->length);
7209
7210	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7211
7212	ctsio->kern_sg_entries = 0;
7213
7214	if (total_len < alloc_len) {
7215		ctsio->residual = alloc_len - total_len;
7216		ctsio->kern_data_len = total_len;
7217		ctsio->kern_total_len = total_len;
7218	} else {
7219		ctsio->residual = 0;
7220		ctsio->kern_data_len = alloc_len;
7221		ctsio->kern_total_len = alloc_len;
7222	}
7223	ctsio->kern_data_resid = 0;
7224	ctsio->kern_rel_offset = 0;
7225
7226	if (ext) {
7227		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7228		    ctsio->kern_data_ptr;
7229		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7230		rtg_ext_ptr->format_type = 0x10;
7231		rtg_ext_ptr->implicit_transition_time = 0;
7232		tpg_desc = &rtg_ext_ptr->groups[0];
7233	} else {
7234		rtg_ptr = (struct scsi_target_group_data *)
7235		    ctsio->kern_data_ptr;
7236		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7237		tpg_desc = &rtg_ptr->groups[0];
7238	}
7239
7240	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7241	mtx_lock(&softc->ctl_lock);
7242	for (g = 0; g < num_target_port_groups; g++) {
7243		if (g == pg)
7244			tpg_desc->pref_state = TPG_PRIMARY |
7245			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7246		else
7247			tpg_desc->pref_state =
7248			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7249		tpg_desc->support = TPG_AO_SUP;
7250		if (!single)
7251			tpg_desc->support |= TPG_AN_SUP;
7252		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7253		tpg_desc->status = TPG_IMPLICIT;
7254		pc = 0;
7255		STAILQ_FOREACH(port, &softc->port_list, links) {
7256			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7257				continue;
7258			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7259			    CTL_MAX_LUNS)
7260				continue;
7261			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7262			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7263			    relative_target_port_identifier);
7264			pc++;
7265		}
7266		tpg_desc->target_port_count = pc;
7267		tpg_desc = (struct scsi_target_port_group_descriptor *)
7268		    &tpg_desc->descriptors[pc];
7269	}
7270	mtx_unlock(&softc->ctl_lock);
7271
7272	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7273	ctsio->be_move_done = ctl_config_move_done;
7274
7275	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7276			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7277			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7278			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7279			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7280
7281	ctl_datamove((union ctl_io *)ctsio);
7282	return(retval);
7283}
7284
7285int
7286ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7287{
7288	struct ctl_lun *lun;
7289	struct scsi_report_supported_opcodes *cdb;
7290	const struct ctl_cmd_entry *entry, *sentry;
7291	struct scsi_report_supported_opcodes_all *all;
7292	struct scsi_report_supported_opcodes_descr *descr;
7293	struct scsi_report_supported_opcodes_one *one;
7294	int retval;
7295	int alloc_len, total_len;
7296	int opcode, service_action, i, j, num;
7297
7298	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7299
7300	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7301	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7302
7303	retval = CTL_RETVAL_COMPLETE;
7304
7305	opcode = cdb->requested_opcode;
7306	service_action = scsi_2btoul(cdb->requested_service_action);
7307	switch (cdb->options & RSO_OPTIONS_MASK) {
7308	case RSO_OPTIONS_ALL:
7309		num = 0;
7310		for (i = 0; i < 256; i++) {
7311			entry = &ctl_cmd_table[i];
7312			if (entry->flags & CTL_CMD_FLAG_SA5) {
7313				for (j = 0; j < 32; j++) {
7314					sentry = &((const struct ctl_cmd_entry *)
7315					    entry->execute)[j];
7316					if (ctl_cmd_applicable(
7317					    lun->be_lun->lun_type, sentry))
7318						num++;
7319				}
7320			} else {
7321				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7322				    entry))
7323					num++;
7324			}
7325		}
7326		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7327		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7328		break;
7329	case RSO_OPTIONS_OC:
7330		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7331			ctl_set_invalid_field(/*ctsio*/ ctsio,
7332					      /*sks_valid*/ 1,
7333					      /*command*/ 1,
7334					      /*field*/ 2,
7335					      /*bit_valid*/ 1,
7336					      /*bit*/ 2);
7337			ctl_done((union ctl_io *)ctsio);
7338			return (CTL_RETVAL_COMPLETE);
7339		}
7340		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7341		break;
7342	case RSO_OPTIONS_OC_SA:
7343		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7344		    service_action >= 32) {
7345			ctl_set_invalid_field(/*ctsio*/ ctsio,
7346					      /*sks_valid*/ 1,
7347					      /*command*/ 1,
7348					      /*field*/ 2,
7349					      /*bit_valid*/ 1,
7350					      /*bit*/ 2);
7351			ctl_done((union ctl_io *)ctsio);
7352			return (CTL_RETVAL_COMPLETE);
7353		}
7354		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7355		break;
7356	default:
7357		ctl_set_invalid_field(/*ctsio*/ ctsio,
7358				      /*sks_valid*/ 1,
7359				      /*command*/ 1,
7360				      /*field*/ 2,
7361				      /*bit_valid*/ 1,
7362				      /*bit*/ 2);
7363		ctl_done((union ctl_io *)ctsio);
7364		return (CTL_RETVAL_COMPLETE);
7365	}
7366
7367	alloc_len = scsi_4btoul(cdb->length);
7368
7369	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7370
7371	ctsio->kern_sg_entries = 0;
7372
7373	if (total_len < alloc_len) {
7374		ctsio->residual = alloc_len - total_len;
7375		ctsio->kern_data_len = total_len;
7376		ctsio->kern_total_len = total_len;
7377	} else {
7378		ctsio->residual = 0;
7379		ctsio->kern_data_len = alloc_len;
7380		ctsio->kern_total_len = alloc_len;
7381	}
7382	ctsio->kern_data_resid = 0;
7383	ctsio->kern_rel_offset = 0;
7384
7385	switch (cdb->options & RSO_OPTIONS_MASK) {
7386	case RSO_OPTIONS_ALL:
7387		all = (struct scsi_report_supported_opcodes_all *)
7388		    ctsio->kern_data_ptr;
7389		num = 0;
7390		for (i = 0; i < 256; i++) {
7391			entry = &ctl_cmd_table[i];
7392			if (entry->flags & CTL_CMD_FLAG_SA5) {
7393				for (j = 0; j < 32; j++) {
7394					sentry = &((const struct ctl_cmd_entry *)
7395					    entry->execute)[j];
7396					if (!ctl_cmd_applicable(
7397					    lun->be_lun->lun_type, sentry))
7398						continue;
7399					descr = &all->descr[num++];
7400					descr->opcode = i;
7401					scsi_ulto2b(j, descr->service_action);
7402					descr->flags = RSO_SERVACTV;
7403					scsi_ulto2b(sentry->length,
7404					    descr->cdb_length);
7405				}
7406			} else {
7407				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7408				    entry))
7409					continue;
7410				descr = &all->descr[num++];
7411				descr->opcode = i;
7412				scsi_ulto2b(0, descr->service_action);
7413				descr->flags = 0;
7414				scsi_ulto2b(entry->length, descr->cdb_length);
7415			}
7416		}
7417		scsi_ulto4b(
7418		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7419		    all->length);
7420		break;
7421	case RSO_OPTIONS_OC:
7422		one = (struct scsi_report_supported_opcodes_one *)
7423		    ctsio->kern_data_ptr;
7424		entry = &ctl_cmd_table[opcode];
7425		goto fill_one;
7426	case RSO_OPTIONS_OC_SA:
7427		one = (struct scsi_report_supported_opcodes_one *)
7428		    ctsio->kern_data_ptr;
7429		entry = &ctl_cmd_table[opcode];
7430		entry = &((const struct ctl_cmd_entry *)
7431		    entry->execute)[service_action];
7432fill_one:
7433		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7434			one->support = 3;
7435			scsi_ulto2b(entry->length, one->cdb_length);
7436			one->cdb_usage[0] = opcode;
7437			memcpy(&one->cdb_usage[1], entry->usage,
7438			    entry->length - 1);
7439		} else
7440			one->support = 1;
7441		break;
7442	}
7443
7444	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7445	ctsio->be_move_done = ctl_config_move_done;
7446
7447	ctl_datamove((union ctl_io *)ctsio);
7448	return(retval);
7449}
7450
7451int
7452ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7453{
7454	struct ctl_lun *lun;
7455	struct scsi_report_supported_tmf *cdb;
7456	struct scsi_report_supported_tmf_data *data;
7457	int retval;
7458	int alloc_len, total_len;
7459
7460	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7461
7462	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7463	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7464
7465	retval = CTL_RETVAL_COMPLETE;
7466
7467	total_len = sizeof(struct scsi_report_supported_tmf_data);
7468	alloc_len = scsi_4btoul(cdb->length);
7469
7470	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7471
7472	ctsio->kern_sg_entries = 0;
7473
7474	if (total_len < alloc_len) {
7475		ctsio->residual = alloc_len - total_len;
7476		ctsio->kern_data_len = total_len;
7477		ctsio->kern_total_len = total_len;
7478	} else {
7479		ctsio->residual = 0;
7480		ctsio->kern_data_len = alloc_len;
7481		ctsio->kern_total_len = alloc_len;
7482	}
7483	ctsio->kern_data_resid = 0;
7484	ctsio->kern_rel_offset = 0;
7485
7486	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7487	data->byte1 |= RST_ATS | RST_LURS | RST_TRS;
7488
7489	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7490	ctsio->be_move_done = ctl_config_move_done;
7491
7492	ctl_datamove((union ctl_io *)ctsio);
7493	return (retval);
7494}
7495
7496int
7497ctl_report_timestamp(struct ctl_scsiio *ctsio)
7498{
7499	struct ctl_lun *lun;
7500	struct scsi_report_timestamp *cdb;
7501	struct scsi_report_timestamp_data *data;
7502	struct timeval tv;
7503	int64_t timestamp;
7504	int retval;
7505	int alloc_len, total_len;
7506
7507	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7508
7509	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7510	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7511
7512	retval = CTL_RETVAL_COMPLETE;
7513
7514	total_len = sizeof(struct scsi_report_timestamp_data);
7515	alloc_len = scsi_4btoul(cdb->length);
7516
7517	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7518
7519	ctsio->kern_sg_entries = 0;
7520
7521	if (total_len < alloc_len) {
7522		ctsio->residual = alloc_len - total_len;
7523		ctsio->kern_data_len = total_len;
7524		ctsio->kern_total_len = total_len;
7525	} else {
7526		ctsio->residual = 0;
7527		ctsio->kern_data_len = alloc_len;
7528		ctsio->kern_total_len = alloc_len;
7529	}
7530	ctsio->kern_data_resid = 0;
7531	ctsio->kern_rel_offset = 0;
7532
7533	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7534	scsi_ulto2b(sizeof(*data) - 2, data->length);
7535	data->origin = RTS_ORIG_OUTSIDE;
7536	getmicrotime(&tv);
7537	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7538	scsi_ulto4b(timestamp >> 16, data->timestamp);
7539	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7540
7541	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7542	ctsio->be_move_done = ctl_config_move_done;
7543
7544	ctl_datamove((union ctl_io *)ctsio);
7545	return (retval);
7546}
7547
7548int
7549ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7550{
7551	struct scsi_per_res_in *cdb;
7552	int alloc_len, total_len = 0;
7553	/* struct scsi_per_res_in_rsrv in_data; */
7554	struct ctl_lun *lun;
7555	struct ctl_softc *softc;
7556
7557	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7558
7559	softc = control_softc;
7560
7561	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7562
7563	alloc_len = scsi_2btoul(cdb->length);
7564
7565	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7566
7567retry:
7568	mtx_lock(&lun->lun_lock);
7569	switch (cdb->action) {
7570	case SPRI_RK: /* read keys */
7571		total_len = sizeof(struct scsi_per_res_in_keys) +
7572			lun->pr_key_count *
7573			sizeof(struct scsi_per_res_key);
7574		break;
7575	case SPRI_RR: /* read reservation */
7576		if (lun->flags & CTL_LUN_PR_RESERVED)
7577			total_len = sizeof(struct scsi_per_res_in_rsrv);
7578		else
7579			total_len = sizeof(struct scsi_per_res_in_header);
7580		break;
7581	case SPRI_RC: /* report capabilities */
7582		total_len = sizeof(struct scsi_per_res_cap);
7583		break;
7584	default:
7585		panic("Invalid PR type %x", cdb->action);
7586	}
7587	mtx_unlock(&lun->lun_lock);
7588
7589	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7590
7591	if (total_len < alloc_len) {
7592		ctsio->residual = alloc_len - total_len;
7593		ctsio->kern_data_len = total_len;
7594		ctsio->kern_total_len = total_len;
7595	} else {
7596		ctsio->residual = 0;
7597		ctsio->kern_data_len = alloc_len;
7598		ctsio->kern_total_len = alloc_len;
7599	}
7600
7601	ctsio->kern_data_resid = 0;
7602	ctsio->kern_rel_offset = 0;
7603	ctsio->kern_sg_entries = 0;
7604
7605	mtx_lock(&lun->lun_lock);
7606	switch (cdb->action) {
7607	case SPRI_RK: { // read keys
7608        struct scsi_per_res_in_keys *res_keys;
7609		int i, key_count;
7610
7611		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7612
7613		/*
7614		 * We had to drop the lock to allocate our buffer, which
7615		 * leaves time for someone to come in with another
7616		 * persistent reservation.  (That is unlikely, though,
7617		 * since this should be the only persistent reservation
7618		 * command active right now.)
7619		 */
7620		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7621		    (lun->pr_key_count *
7622		     sizeof(struct scsi_per_res_key)))){
7623			mtx_unlock(&lun->lun_lock);
7624			free(ctsio->kern_data_ptr, M_CTL);
7625			printf("%s: reservation length changed, retrying\n",
7626			       __func__);
7627			goto retry;
7628		}
7629
7630		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7631
7632		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7633			     lun->pr_key_count, res_keys->header.length);
7634
7635		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7636			if (!lun->per_res[i].registered)
7637				continue;
7638
7639			/*
7640			 * We used lun->pr_key_count to calculate the
7641			 * size to allocate.  If it turns out the number of
7642			 * initiators with the registered flag set is
7643			 * larger than that (i.e. they haven't been kept in
7644			 * sync), we've got a problem.
7645			 */
7646			if (key_count >= lun->pr_key_count) {
7647#ifdef NEEDTOPORT
7648				csevent_log(CSC_CTL | CSC_SHELF_SW |
7649					    CTL_PR_ERROR,
7650					    csevent_LogType_Fault,
7651					    csevent_AlertLevel_Yellow,
7652					    csevent_FRU_ShelfController,
7653					    csevent_FRU_Firmware,
7654				        csevent_FRU_Unknown,
7655					    "registered keys %d >= key "
7656					    "count %d", key_count,
7657					    lun->pr_key_count);
7658#endif
7659				key_count++;
7660				continue;
7661			}
7662			memcpy(res_keys->keys[key_count].key,
7663			       lun->per_res[i].res_key.key,
7664			       ctl_min(sizeof(res_keys->keys[key_count].key),
7665			       sizeof(lun->per_res[i].res_key)));
7666			key_count++;
7667		}
7668		break;
7669	}
7670	case SPRI_RR: { // read reservation
7671		struct scsi_per_res_in_rsrv *res;
7672		int tmp_len, header_only;
7673
7674		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7675
7676		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7677
7678		if (lun->flags & CTL_LUN_PR_RESERVED)
7679		{
7680			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7681			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7682				    res->header.length);
7683			header_only = 0;
7684		} else {
7685			tmp_len = sizeof(struct scsi_per_res_in_header);
7686			scsi_ulto4b(0, res->header.length);
7687			header_only = 1;
7688		}
7689
7690		/*
7691		 * We had to drop the lock to allocate our buffer, which
7692		 * leaves time for someone to come in with another
7693		 * persistent reservation.  (That is unlikely, though,
7694		 * since this should be the only persistent reservation
7695		 * command active right now.)
7696		 */
7697		if (tmp_len != total_len) {
7698			mtx_unlock(&lun->lun_lock);
7699			free(ctsio->kern_data_ptr, M_CTL);
7700			printf("%s: reservation status changed, retrying\n",
7701			       __func__);
7702			goto retry;
7703		}
7704
7705		/*
7706		 * No reservation held, so we're done.
7707		 */
7708		if (header_only != 0)
7709			break;
7710
7711		/*
7712		 * If the registration is an All Registrants type, the key
7713		 * is 0, since it doesn't really matter.
7714		 */
7715		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7716			memcpy(res->data.reservation,
7717			       &lun->per_res[lun->pr_res_idx].res_key,
7718			       sizeof(struct scsi_per_res_key));
7719		}
7720		res->data.scopetype = lun->res_type;
7721		break;
7722	}
7723	case SPRI_RC:     //report capabilities
7724	{
7725		struct scsi_per_res_cap *res_cap;
7726		uint16_t type_mask;
7727
7728		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7729		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7730		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_3;
7731		type_mask = SPRI_TM_WR_EX_AR |
7732			    SPRI_TM_EX_AC_RO |
7733			    SPRI_TM_WR_EX_RO |
7734			    SPRI_TM_EX_AC |
7735			    SPRI_TM_WR_EX |
7736			    SPRI_TM_EX_AC_AR;
7737		scsi_ulto2b(type_mask, res_cap->type_mask);
7738		break;
7739	}
7740	case SPRI_RS: //read full status
7741	default:
7742		/*
7743		 * This is a bug, because we just checked for this above,
7744		 * and should have returned an error.
7745		 */
7746		panic("Invalid PR type %x", cdb->action);
7747		break; /* NOTREACHED */
7748	}
7749	mtx_unlock(&lun->lun_lock);
7750
7751	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7752	ctsio->be_move_done = ctl_config_move_done;
7753
7754	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7755			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7756			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7757			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7758			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7759
7760	ctl_datamove((union ctl_io *)ctsio);
7761
7762	return (CTL_RETVAL_COMPLETE);
7763}
7764
7765/*
7766 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7767 * it should return.
7768 */
7769static int
7770ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7771		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7772		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7773		struct scsi_per_res_out_parms* param)
7774{
7775	union ctl_ha_msg persis_io;
7776	int retval, i;
7777	int isc_retval;
7778
7779	retval = 0;
7780
7781	mtx_lock(&lun->lun_lock);
7782	if (sa_res_key == 0) {
7783		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7784			/* validate scope and type */
7785			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7786			     SPR_LU_SCOPE) {
7787				mtx_unlock(&lun->lun_lock);
7788				ctl_set_invalid_field(/*ctsio*/ ctsio,
7789						      /*sks_valid*/ 1,
7790						      /*command*/ 1,
7791						      /*field*/ 2,
7792						      /*bit_valid*/ 1,
7793						      /*bit*/ 4);
7794				ctl_done((union ctl_io *)ctsio);
7795				return (1);
7796			}
7797
7798		        if (type>8 || type==2 || type==4 || type==0) {
7799				mtx_unlock(&lun->lun_lock);
7800				ctl_set_invalid_field(/*ctsio*/ ctsio,
7801       	           				      /*sks_valid*/ 1,
7802						      /*command*/ 1,
7803						      /*field*/ 2,
7804						      /*bit_valid*/ 1,
7805						      /*bit*/ 0);
7806				ctl_done((union ctl_io *)ctsio);
7807				return (1);
7808		        }
7809
7810			/* temporarily unregister this nexus */
7811			lun->per_res[residx].registered = 0;
7812
7813			/*
7814			 * Unregister everybody else and build UA for
7815			 * them
7816			 */
7817			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7818				if (lun->per_res[i].registered == 0)
7819					continue;
7820
7821				if (!persis_offset
7822				 && i <CTL_MAX_INITIATORS)
7823					lun->pending_sense[i].ua_pending |=
7824						CTL_UA_REG_PREEMPT;
7825				else if (persis_offset
7826				      && i >= persis_offset)
7827					lun->pending_sense[i-persis_offset
7828						].ua_pending |=
7829						CTL_UA_REG_PREEMPT;
7830				lun->per_res[i].registered = 0;
7831				memset(&lun->per_res[i].res_key, 0,
7832				       sizeof(struct scsi_per_res_key));
7833			}
7834			lun->per_res[residx].registered = 1;
7835			lun->pr_key_count = 1;
7836			lun->res_type = type;
7837			if (lun->res_type != SPR_TYPE_WR_EX_AR
7838			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7839				lun->pr_res_idx = residx;
7840
7841			/* send msg to other side */
7842			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7843			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7844			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7845			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7846			persis_io.pr.pr_info.res_type = type;
7847			memcpy(persis_io.pr.pr_info.sa_res_key,
7848			       param->serv_act_res_key,
7849			       sizeof(param->serv_act_res_key));
7850			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7851			     &persis_io, sizeof(persis_io), 0)) >
7852			     CTL_HA_STATUS_SUCCESS) {
7853				printf("CTL:Persis Out error returned "
7854				       "from ctl_ha_msg_send %d\n",
7855				       isc_retval);
7856			}
7857		} else {
7858			/* not all registrants */
7859			mtx_unlock(&lun->lun_lock);
7860			free(ctsio->kern_data_ptr, M_CTL);
7861			ctl_set_invalid_field(ctsio,
7862					      /*sks_valid*/ 1,
7863					      /*command*/ 0,
7864					      /*field*/ 8,
7865					      /*bit_valid*/ 0,
7866					      /*bit*/ 0);
7867			ctl_done((union ctl_io *)ctsio);
7868			return (1);
7869		}
7870	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7871		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7872		int found = 0;
7873
7874		if (res_key == sa_res_key) {
7875			/* special case */
7876			/*
7877			 * The spec implies this is not good but doesn't
7878			 * say what to do. There are two choices either
7879			 * generate a res conflict or check condition
7880			 * with illegal field in parameter data. Since
7881			 * that is what is done when the sa_res_key is
7882			 * zero I'll take that approach since this has
7883			 * to do with the sa_res_key.
7884			 */
7885			mtx_unlock(&lun->lun_lock);
7886			free(ctsio->kern_data_ptr, M_CTL);
7887			ctl_set_invalid_field(ctsio,
7888					      /*sks_valid*/ 1,
7889					      /*command*/ 0,
7890					      /*field*/ 8,
7891					      /*bit_valid*/ 0,
7892					      /*bit*/ 0);
7893			ctl_done((union ctl_io *)ctsio);
7894			return (1);
7895		}
7896
7897		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7898			if (lun->per_res[i].registered
7899			 && memcmp(param->serv_act_res_key,
7900			    lun->per_res[i].res_key.key,
7901			    sizeof(struct scsi_per_res_key)) != 0)
7902				continue;
7903
7904			found = 1;
7905			lun->per_res[i].registered = 0;
7906			memset(&lun->per_res[i].res_key, 0,
7907			       sizeof(struct scsi_per_res_key));
7908			lun->pr_key_count--;
7909
7910			if (!persis_offset
7911			 && i < CTL_MAX_INITIATORS)
7912				lun->pending_sense[i].ua_pending |=
7913					CTL_UA_REG_PREEMPT;
7914			else if (persis_offset
7915			      && i >= persis_offset)
7916				lun->pending_sense[i-persis_offset].ua_pending|=
7917					CTL_UA_REG_PREEMPT;
7918		}
7919		if (!found) {
7920			mtx_unlock(&lun->lun_lock);
7921			free(ctsio->kern_data_ptr, M_CTL);
7922			ctl_set_reservation_conflict(ctsio);
7923			ctl_done((union ctl_io *)ctsio);
7924			return (CTL_RETVAL_COMPLETE);
7925		}
7926		/* send msg to other side */
7927		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7928		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7929		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7930		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7931		persis_io.pr.pr_info.res_type = type;
7932		memcpy(persis_io.pr.pr_info.sa_res_key,
7933		       param->serv_act_res_key,
7934		       sizeof(param->serv_act_res_key));
7935		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7936		     &persis_io, sizeof(persis_io), 0)) >
7937		     CTL_HA_STATUS_SUCCESS) {
7938			printf("CTL:Persis Out error returned from "
7939			       "ctl_ha_msg_send %d\n", isc_retval);
7940		}
7941	} else {
7942		/* Reserved but not all registrants */
7943		/* sa_res_key is res holder */
7944		if (memcmp(param->serv_act_res_key,
7945                   lun->per_res[lun->pr_res_idx].res_key.key,
7946                   sizeof(struct scsi_per_res_key)) == 0) {
7947			/* validate scope and type */
7948			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7949			     SPR_LU_SCOPE) {
7950				mtx_unlock(&lun->lun_lock);
7951				ctl_set_invalid_field(/*ctsio*/ ctsio,
7952						      /*sks_valid*/ 1,
7953						      /*command*/ 1,
7954						      /*field*/ 2,
7955						      /*bit_valid*/ 1,
7956						      /*bit*/ 4);
7957				ctl_done((union ctl_io *)ctsio);
7958				return (1);
7959			}
7960
7961			if (type>8 || type==2 || type==4 || type==0) {
7962				mtx_unlock(&lun->lun_lock);
7963				ctl_set_invalid_field(/*ctsio*/ ctsio,
7964						      /*sks_valid*/ 1,
7965						      /*command*/ 1,
7966						      /*field*/ 2,
7967						      /*bit_valid*/ 1,
7968						      /*bit*/ 0);
7969				ctl_done((union ctl_io *)ctsio);
7970				return (1);
7971			}
7972
7973			/*
7974			 * Do the following:
7975			 * if sa_res_key != res_key remove all
7976			 * registrants w/sa_res_key and generate UA
7977			 * for these registrants(Registrations
7978			 * Preempted) if it wasn't an exclusive
7979			 * reservation generate UA(Reservations
7980			 * Preempted) for all other registered nexuses
7981			 * if the type has changed. Establish the new
7982			 * reservation and holder. If res_key and
7983			 * sa_res_key are the same do the above
7984			 * except don't unregister the res holder.
7985			 */
7986
7987			/*
7988			 * Temporarily unregister so it won't get
7989			 * removed or UA generated
7990			 */
7991			lun->per_res[residx].registered = 0;
7992			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7993				if (lun->per_res[i].registered == 0)
7994					continue;
7995
7996				if (memcmp(param->serv_act_res_key,
7997				    lun->per_res[i].res_key.key,
7998				    sizeof(struct scsi_per_res_key)) == 0) {
7999					lun->per_res[i].registered = 0;
8000					memset(&lun->per_res[i].res_key,
8001					       0,
8002					       sizeof(struct scsi_per_res_key));
8003					lun->pr_key_count--;
8004
8005					if (!persis_offset
8006					 && i < CTL_MAX_INITIATORS)
8007						lun->pending_sense[i
8008							].ua_pending |=
8009							CTL_UA_REG_PREEMPT;
8010					else if (persis_offset
8011					      && i >= persis_offset)
8012						lun->pending_sense[
8013						  i-persis_offset].ua_pending |=
8014						  CTL_UA_REG_PREEMPT;
8015				} else if (type != lun->res_type
8016					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8017					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8018						if (!persis_offset
8019						 && i < CTL_MAX_INITIATORS)
8020							lun->pending_sense[i
8021							].ua_pending |=
8022							CTL_UA_RES_RELEASE;
8023						else if (persis_offset
8024						      && i >= persis_offset)
8025							lun->pending_sense[
8026							i-persis_offset
8027							].ua_pending |=
8028							CTL_UA_RES_RELEASE;
8029				}
8030			}
8031			lun->per_res[residx].registered = 1;
8032			lun->res_type = type;
8033			if (lun->res_type != SPR_TYPE_WR_EX_AR
8034			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8035				lun->pr_res_idx = residx;
8036			else
8037				lun->pr_res_idx =
8038					CTL_PR_ALL_REGISTRANTS;
8039
8040			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8041			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8042			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8043			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8044			persis_io.pr.pr_info.res_type = type;
8045			memcpy(persis_io.pr.pr_info.sa_res_key,
8046			       param->serv_act_res_key,
8047			       sizeof(param->serv_act_res_key));
8048			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8049			     &persis_io, sizeof(persis_io), 0)) >
8050			     CTL_HA_STATUS_SUCCESS) {
8051				printf("CTL:Persis Out error returned "
8052				       "from ctl_ha_msg_send %d\n",
8053				       isc_retval);
8054			}
8055		} else {
8056			/*
8057			 * sa_res_key is not the res holder just
8058			 * remove registrants
8059			 */
8060			int found=0;
8061
8062			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8063				if (memcmp(param->serv_act_res_key,
8064				    lun->per_res[i].res_key.key,
8065				    sizeof(struct scsi_per_res_key)) != 0)
8066					continue;
8067
8068				found = 1;
8069				lun->per_res[i].registered = 0;
8070				memset(&lun->per_res[i].res_key, 0,
8071				       sizeof(struct scsi_per_res_key));
8072				lun->pr_key_count--;
8073
8074				if (!persis_offset
8075				 && i < CTL_MAX_INITIATORS)
8076					lun->pending_sense[i].ua_pending |=
8077						CTL_UA_REG_PREEMPT;
8078				else if (persis_offset
8079				      && i >= persis_offset)
8080					lun->pending_sense[
8081						i-persis_offset].ua_pending |=
8082						CTL_UA_REG_PREEMPT;
8083			}
8084
8085			if (!found) {
8086				mtx_unlock(&lun->lun_lock);
8087				free(ctsio->kern_data_ptr, M_CTL);
8088				ctl_set_reservation_conflict(ctsio);
8089				ctl_done((union ctl_io *)ctsio);
8090		        	return (1);
8091			}
8092			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8093			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8094			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8095			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8096			persis_io.pr.pr_info.res_type = type;
8097			memcpy(persis_io.pr.pr_info.sa_res_key,
8098			       param->serv_act_res_key,
8099			       sizeof(param->serv_act_res_key));
8100			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8101			     &persis_io, sizeof(persis_io), 0)) >
8102			     CTL_HA_STATUS_SUCCESS) {
8103				printf("CTL:Persis Out error returned "
8104				       "from ctl_ha_msg_send %d\n",
8105				isc_retval);
8106			}
8107		}
8108	}
8109
8110	lun->PRGeneration++;
8111	mtx_unlock(&lun->lun_lock);
8112
8113	return (retval);
8114}
8115
8116static void
8117ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8118{
8119	int i;
8120
8121	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8122	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8123	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8124		   msg->pr.pr_info.sa_res_key,
8125		   sizeof(struct scsi_per_res_key)) != 0) {
8126		uint64_t sa_res_key;
8127		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8128
8129		if (sa_res_key == 0) {
8130			/* temporarily unregister this nexus */
8131			lun->per_res[msg->pr.pr_info.residx].registered = 0;
8132
8133			/*
8134			 * Unregister everybody else and build UA for
8135			 * them
8136			 */
8137			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8138				if (lun->per_res[i].registered == 0)
8139					continue;
8140
8141				if (!persis_offset
8142				 && i < CTL_MAX_INITIATORS)
8143					lun->pending_sense[i].ua_pending |=
8144						CTL_UA_REG_PREEMPT;
8145				else if (persis_offset && i >= persis_offset)
8146					lun->pending_sense[i -
8147						persis_offset].ua_pending |=
8148						CTL_UA_REG_PREEMPT;
8149				lun->per_res[i].registered = 0;
8150				memset(&lun->per_res[i].res_key, 0,
8151				       sizeof(struct scsi_per_res_key));
8152			}
8153
8154			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8155			lun->pr_key_count = 1;
8156			lun->res_type = msg->pr.pr_info.res_type;
8157			if (lun->res_type != SPR_TYPE_WR_EX_AR
8158			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8159				lun->pr_res_idx = msg->pr.pr_info.residx;
8160		} else {
8161		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8162				if (memcmp(msg->pr.pr_info.sa_res_key,
8163		                   lun->per_res[i].res_key.key,
8164		                   sizeof(struct scsi_per_res_key)) != 0)
8165					continue;
8166
8167				lun->per_res[i].registered = 0;
8168				memset(&lun->per_res[i].res_key, 0,
8169				       sizeof(struct scsi_per_res_key));
8170				lun->pr_key_count--;
8171
8172				if (!persis_offset
8173				 && i < persis_offset)
8174					lun->pending_sense[i].ua_pending |=
8175						CTL_UA_REG_PREEMPT;
8176				else if (persis_offset
8177				      && i >= persis_offset)
8178					lun->pending_sense[i -
8179						persis_offset].ua_pending |=
8180						CTL_UA_REG_PREEMPT;
8181			}
8182		}
8183	} else {
8184		/*
8185		 * Temporarily unregister so it won't get removed
8186		 * or UA generated
8187		 */
8188		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8189		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8190			if (lun->per_res[i].registered == 0)
8191				continue;
8192
8193			if (memcmp(msg->pr.pr_info.sa_res_key,
8194	                   lun->per_res[i].res_key.key,
8195	                   sizeof(struct scsi_per_res_key)) == 0) {
8196				lun->per_res[i].registered = 0;
8197				memset(&lun->per_res[i].res_key, 0,
8198				       sizeof(struct scsi_per_res_key));
8199				lun->pr_key_count--;
8200				if (!persis_offset
8201				 && i < CTL_MAX_INITIATORS)
8202					lun->pending_sense[i].ua_pending |=
8203						CTL_UA_REG_PREEMPT;
8204				else if (persis_offset
8205				      && i >= persis_offset)
8206					lun->pending_sense[i -
8207						persis_offset].ua_pending |=
8208						CTL_UA_REG_PREEMPT;
8209			} else if (msg->pr.pr_info.res_type != lun->res_type
8210				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8211				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8212					if (!persis_offset
8213					 && i < persis_offset)
8214						lun->pending_sense[i
8215							].ua_pending |=
8216							CTL_UA_RES_RELEASE;
8217					else if (persis_offset
8218					      && i >= persis_offset)
8219					lun->pending_sense[i -
8220						persis_offset].ua_pending |=
8221						CTL_UA_RES_RELEASE;
8222			}
8223		}
8224		lun->per_res[msg->pr.pr_info.residx].registered = 1;
8225		lun->res_type = msg->pr.pr_info.res_type;
8226		if (lun->res_type != SPR_TYPE_WR_EX_AR
8227		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8228			lun->pr_res_idx = msg->pr.pr_info.residx;
8229		else
8230			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8231	}
8232	lun->PRGeneration++;
8233
8234}
8235
8236
8237int
8238ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8239{
8240	int retval;
8241	int isc_retval;
8242	u_int32_t param_len;
8243	struct scsi_per_res_out *cdb;
8244	struct ctl_lun *lun;
8245	struct scsi_per_res_out_parms* param;
8246	struct ctl_softc *softc;
8247	uint32_t residx;
8248	uint64_t res_key, sa_res_key;
8249	uint8_t type;
8250	union ctl_ha_msg persis_io;
8251	int    i;
8252
8253	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8254
8255	retval = CTL_RETVAL_COMPLETE;
8256
8257	softc = control_softc;
8258
8259	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8260	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8261
8262	/*
8263	 * We only support whole-LUN scope.  The scope & type are ignored for
8264	 * register, register and ignore existing key and clear.
8265	 * We sometimes ignore scope and type on preempts too!!
8266	 * Verify reservation type here as well.
8267	 */
8268	type = cdb->scope_type & SPR_TYPE_MASK;
8269	if ((cdb->action == SPRO_RESERVE)
8270	 || (cdb->action == SPRO_RELEASE)) {
8271		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8272			ctl_set_invalid_field(/*ctsio*/ ctsio,
8273					      /*sks_valid*/ 1,
8274					      /*command*/ 1,
8275					      /*field*/ 2,
8276					      /*bit_valid*/ 1,
8277					      /*bit*/ 4);
8278			ctl_done((union ctl_io *)ctsio);
8279			return (CTL_RETVAL_COMPLETE);
8280		}
8281
8282		if (type>8 || type==2 || type==4 || type==0) {
8283			ctl_set_invalid_field(/*ctsio*/ ctsio,
8284					      /*sks_valid*/ 1,
8285					      /*command*/ 1,
8286					      /*field*/ 2,
8287					      /*bit_valid*/ 1,
8288					      /*bit*/ 0);
8289			ctl_done((union ctl_io *)ctsio);
8290			return (CTL_RETVAL_COMPLETE);
8291		}
8292	}
8293
8294	param_len = scsi_4btoul(cdb->length);
8295
8296	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8297		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8298		ctsio->kern_data_len = param_len;
8299		ctsio->kern_total_len = param_len;
8300		ctsio->kern_data_resid = 0;
8301		ctsio->kern_rel_offset = 0;
8302		ctsio->kern_sg_entries = 0;
8303		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8304		ctsio->be_move_done = ctl_config_move_done;
8305		ctl_datamove((union ctl_io *)ctsio);
8306
8307		return (CTL_RETVAL_COMPLETE);
8308	}
8309
8310	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8311
8312	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8313	res_key = scsi_8btou64(param->res_key.key);
8314	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8315
8316	/*
8317	 * Validate the reservation key here except for SPRO_REG_IGNO
8318	 * This must be done for all other service actions
8319	 */
8320	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8321		mtx_lock(&lun->lun_lock);
8322		if (lun->per_res[residx].registered) {
8323		    if (memcmp(param->res_key.key,
8324			       lun->per_res[residx].res_key.key,
8325			       ctl_min(sizeof(param->res_key),
8326			       sizeof(lun->per_res[residx].res_key))) != 0) {
8327				/*
8328				 * The current key passed in doesn't match
8329				 * the one the initiator previously
8330				 * registered.
8331				 */
8332				mtx_unlock(&lun->lun_lock);
8333				free(ctsio->kern_data_ptr, M_CTL);
8334				ctl_set_reservation_conflict(ctsio);
8335				ctl_done((union ctl_io *)ctsio);
8336				return (CTL_RETVAL_COMPLETE);
8337			}
8338		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8339			/*
8340			 * We are not registered
8341			 */
8342			mtx_unlock(&lun->lun_lock);
8343			free(ctsio->kern_data_ptr, M_CTL);
8344			ctl_set_reservation_conflict(ctsio);
8345			ctl_done((union ctl_io *)ctsio);
8346			return (CTL_RETVAL_COMPLETE);
8347		} else if (res_key != 0) {
8348			/*
8349			 * We are not registered and trying to register but
8350			 * the register key isn't zero.
8351			 */
8352			mtx_unlock(&lun->lun_lock);
8353			free(ctsio->kern_data_ptr, M_CTL);
8354			ctl_set_reservation_conflict(ctsio);
8355			ctl_done((union ctl_io *)ctsio);
8356			return (CTL_RETVAL_COMPLETE);
8357		}
8358		mtx_unlock(&lun->lun_lock);
8359	}
8360
8361	switch (cdb->action & SPRO_ACTION_MASK) {
8362	case SPRO_REGISTER:
8363	case SPRO_REG_IGNO: {
8364
8365#if 0
8366		printf("Registration received\n");
8367#endif
8368
8369		/*
8370		 * We don't support any of these options, as we report in
8371		 * the read capabilities request (see
8372		 * ctl_persistent_reserve_in(), above).
8373		 */
8374		if ((param->flags & SPR_SPEC_I_PT)
8375		 || (param->flags & SPR_ALL_TG_PT)
8376		 || (param->flags & SPR_APTPL)) {
8377			int bit_ptr;
8378
8379			if (param->flags & SPR_APTPL)
8380				bit_ptr = 0;
8381			else if (param->flags & SPR_ALL_TG_PT)
8382				bit_ptr = 2;
8383			else /* SPR_SPEC_I_PT */
8384				bit_ptr = 3;
8385
8386			free(ctsio->kern_data_ptr, M_CTL);
8387			ctl_set_invalid_field(ctsio,
8388					      /*sks_valid*/ 1,
8389					      /*command*/ 0,
8390					      /*field*/ 20,
8391					      /*bit_valid*/ 1,
8392					      /*bit*/ bit_ptr);
8393			ctl_done((union ctl_io *)ctsio);
8394			return (CTL_RETVAL_COMPLETE);
8395		}
8396
8397		mtx_lock(&lun->lun_lock);
8398
8399		/*
8400		 * The initiator wants to clear the
8401		 * key/unregister.
8402		 */
8403		if (sa_res_key == 0) {
8404			if ((res_key == 0
8405			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8406			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8407			  && !lun->per_res[residx].registered)) {
8408				mtx_unlock(&lun->lun_lock);
8409				goto done;
8410			}
8411
8412			lun->per_res[residx].registered = 0;
8413			memset(&lun->per_res[residx].res_key,
8414			       0, sizeof(lun->per_res[residx].res_key));
8415			lun->pr_key_count--;
8416
8417			if (residx == lun->pr_res_idx) {
8418				lun->flags &= ~CTL_LUN_PR_RESERVED;
8419				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8420
8421				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8422				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8423				 && lun->pr_key_count) {
8424					/*
8425					 * If the reservation is a registrants
8426					 * only type we need to generate a UA
8427					 * for other registered inits.  The
8428					 * sense code should be RESERVATIONS
8429					 * RELEASED
8430					 */
8431
8432					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8433						if (lun->per_res[
8434						    i+persis_offset].registered
8435						    == 0)
8436							continue;
8437						lun->pending_sense[i
8438							].ua_pending |=
8439							CTL_UA_RES_RELEASE;
8440					}
8441				}
8442				lun->res_type = 0;
8443			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8444				if (lun->pr_key_count==0) {
8445					lun->flags &= ~CTL_LUN_PR_RESERVED;
8446					lun->res_type = 0;
8447					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8448				}
8449			}
8450			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8451			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8452			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8453			persis_io.pr.pr_info.residx = residx;
8454			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8455			     &persis_io, sizeof(persis_io), 0 )) >
8456			     CTL_HA_STATUS_SUCCESS) {
8457				printf("CTL:Persis Out error returned from "
8458				       "ctl_ha_msg_send %d\n", isc_retval);
8459			}
8460		} else /* sa_res_key != 0 */ {
8461
8462			/*
8463			 * If we aren't registered currently then increment
8464			 * the key count and set the registered flag.
8465			 */
8466			if (!lun->per_res[residx].registered) {
8467				lun->pr_key_count++;
8468				lun->per_res[residx].registered = 1;
8469			}
8470
8471			memcpy(&lun->per_res[residx].res_key,
8472			       param->serv_act_res_key,
8473			       ctl_min(sizeof(param->serv_act_res_key),
8474			       sizeof(lun->per_res[residx].res_key)));
8475
8476			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8477			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8478			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8479			persis_io.pr.pr_info.residx = residx;
8480			memcpy(persis_io.pr.pr_info.sa_res_key,
8481			       param->serv_act_res_key,
8482			       sizeof(param->serv_act_res_key));
8483			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8484			     &persis_io, sizeof(persis_io), 0)) >
8485			     CTL_HA_STATUS_SUCCESS) {
8486				printf("CTL:Persis Out error returned from "
8487				       "ctl_ha_msg_send %d\n", isc_retval);
8488			}
8489		}
8490		lun->PRGeneration++;
8491		mtx_unlock(&lun->lun_lock);
8492
8493		break;
8494	}
8495	case SPRO_RESERVE:
8496#if 0
8497                printf("Reserve executed type %d\n", type);
8498#endif
8499		mtx_lock(&lun->lun_lock);
8500		if (lun->flags & CTL_LUN_PR_RESERVED) {
8501			/*
8502			 * if this isn't the reservation holder and it's
8503			 * not a "all registrants" type or if the type is
8504			 * different then we have a conflict
8505			 */
8506			if ((lun->pr_res_idx != residx
8507			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8508			 || lun->res_type != type) {
8509				mtx_unlock(&lun->lun_lock);
8510				free(ctsio->kern_data_ptr, M_CTL);
8511				ctl_set_reservation_conflict(ctsio);
8512				ctl_done((union ctl_io *)ctsio);
8513				return (CTL_RETVAL_COMPLETE);
8514			}
8515			mtx_unlock(&lun->lun_lock);
8516		} else /* create a reservation */ {
8517			/*
8518			 * If it's not an "all registrants" type record
8519			 * reservation holder
8520			 */
8521			if (type != SPR_TYPE_WR_EX_AR
8522			 && type != SPR_TYPE_EX_AC_AR)
8523				lun->pr_res_idx = residx; /* Res holder */
8524			else
8525				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8526
8527			lun->flags |= CTL_LUN_PR_RESERVED;
8528			lun->res_type = type;
8529
8530			mtx_unlock(&lun->lun_lock);
8531
8532			/* send msg to other side */
8533			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8534			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8535			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8536			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8537			persis_io.pr.pr_info.res_type = type;
8538			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8539			     &persis_io, sizeof(persis_io), 0)) >
8540			     CTL_HA_STATUS_SUCCESS) {
8541				printf("CTL:Persis Out error returned from "
8542				       "ctl_ha_msg_send %d\n", isc_retval);
8543			}
8544		}
8545		break;
8546
8547	case SPRO_RELEASE:
8548		mtx_lock(&lun->lun_lock);
8549		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8550			/* No reservation exists return good status */
8551			mtx_unlock(&lun->lun_lock);
8552			goto done;
8553		}
8554		/*
8555		 * Is this nexus a reservation holder?
8556		 */
8557		if (lun->pr_res_idx != residx
8558		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8559			/*
8560			 * not a res holder return good status but
8561			 * do nothing
8562			 */
8563			mtx_unlock(&lun->lun_lock);
8564			goto done;
8565		}
8566
8567		if (lun->res_type != type) {
8568			mtx_unlock(&lun->lun_lock);
8569			free(ctsio->kern_data_ptr, M_CTL);
8570			ctl_set_illegal_pr_release(ctsio);
8571			ctl_done((union ctl_io *)ctsio);
8572			return (CTL_RETVAL_COMPLETE);
8573		}
8574
8575		/* okay to release */
8576		lun->flags &= ~CTL_LUN_PR_RESERVED;
8577		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8578		lun->res_type = 0;
8579
8580		/*
8581		 * if this isn't an exclusive access
8582		 * res generate UA for all other
8583		 * registrants.
8584		 */
8585		if (type != SPR_TYPE_EX_AC
8586		 && type != SPR_TYPE_WR_EX) {
8587			/*
8588			 * temporarily unregister so we don't generate UA
8589			 */
8590			lun->per_res[residx].registered = 0;
8591
8592			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8593				if (lun->per_res[i+persis_offset].registered
8594				    == 0)
8595					continue;
8596				lun->pending_sense[i].ua_pending |=
8597					CTL_UA_RES_RELEASE;
8598			}
8599
8600			lun->per_res[residx].registered = 1;
8601		}
8602		mtx_unlock(&lun->lun_lock);
8603		/* Send msg to other side */
8604		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8605		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8606		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8607		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8608		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8609			printf("CTL:Persis Out error returned from "
8610			       "ctl_ha_msg_send %d\n", isc_retval);
8611		}
8612		break;
8613
8614	case SPRO_CLEAR:
8615		/* send msg to other side */
8616
8617		mtx_lock(&lun->lun_lock);
8618		lun->flags &= ~CTL_LUN_PR_RESERVED;
8619		lun->res_type = 0;
8620		lun->pr_key_count = 0;
8621		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8622
8623
8624		memset(&lun->per_res[residx].res_key,
8625		       0, sizeof(lun->per_res[residx].res_key));
8626		lun->per_res[residx].registered = 0;
8627
8628		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8629			if (lun->per_res[i].registered) {
8630				if (!persis_offset && i < CTL_MAX_INITIATORS)
8631					lun->pending_sense[i].ua_pending |=
8632						CTL_UA_RES_PREEMPT;
8633				else if (persis_offset && i >= persis_offset)
8634					lun->pending_sense[i-persis_offset
8635					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8636
8637				memset(&lun->per_res[i].res_key,
8638				       0, sizeof(struct scsi_per_res_key));
8639				lun->per_res[i].registered = 0;
8640			}
8641		lun->PRGeneration++;
8642		mtx_unlock(&lun->lun_lock);
8643		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8644		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8645		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8646		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8647		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8648			printf("CTL:Persis Out error returned from "
8649			       "ctl_ha_msg_send %d\n", isc_retval);
8650		}
8651		break;
8652
8653	case SPRO_PREEMPT: {
8654		int nretval;
8655
8656		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8657					  residx, ctsio, cdb, param);
8658		if (nretval != 0)
8659			return (CTL_RETVAL_COMPLETE);
8660		break;
8661	}
8662	default:
8663		panic("Invalid PR type %x", cdb->action);
8664	}
8665
8666done:
8667	free(ctsio->kern_data_ptr, M_CTL);
8668	ctl_set_success(ctsio);
8669	ctl_done((union ctl_io *)ctsio);
8670
8671	return (retval);
8672}
8673
8674/*
8675 * This routine is for handling a message from the other SC pertaining to
8676 * persistent reserve out. All the error checking will have been done
8677 * so only perorming the action need be done here to keep the two
8678 * in sync.
8679 */
8680static void
8681ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8682{
8683	struct ctl_lun *lun;
8684	struct ctl_softc *softc;
8685	int i;
8686	uint32_t targ_lun;
8687
8688	softc = control_softc;
8689
8690	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8691	lun = softc->ctl_luns[targ_lun];
8692	mtx_lock(&lun->lun_lock);
8693	switch(msg->pr.pr_info.action) {
8694	case CTL_PR_REG_KEY:
8695		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8696			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8697			lun->pr_key_count++;
8698		}
8699		lun->PRGeneration++;
8700		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8701		       msg->pr.pr_info.sa_res_key,
8702		       sizeof(struct scsi_per_res_key));
8703		break;
8704
8705	case CTL_PR_UNREG_KEY:
8706		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8707		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8708		       0, sizeof(struct scsi_per_res_key));
8709		lun->pr_key_count--;
8710
8711		/* XXX Need to see if the reservation has been released */
8712		/* if so do we need to generate UA? */
8713		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8714			lun->flags &= ~CTL_LUN_PR_RESERVED;
8715			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8716
8717			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8718			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8719			 && lun->pr_key_count) {
8720				/*
8721				 * If the reservation is a registrants
8722				 * only type we need to generate a UA
8723				 * for other registered inits.  The
8724				 * sense code should be RESERVATIONS
8725				 * RELEASED
8726				 */
8727
8728				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8729					if (lun->per_res[i+
8730					    persis_offset].registered == 0)
8731						continue;
8732
8733					lun->pending_sense[i
8734						].ua_pending |=
8735						CTL_UA_RES_RELEASE;
8736				}
8737			}
8738			lun->res_type = 0;
8739		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8740			if (lun->pr_key_count==0) {
8741				lun->flags &= ~CTL_LUN_PR_RESERVED;
8742				lun->res_type = 0;
8743				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8744			}
8745		}
8746		lun->PRGeneration++;
8747		break;
8748
8749	case CTL_PR_RESERVE:
8750		lun->flags |= CTL_LUN_PR_RESERVED;
8751		lun->res_type = msg->pr.pr_info.res_type;
8752		lun->pr_res_idx = msg->pr.pr_info.residx;
8753
8754		break;
8755
8756	case CTL_PR_RELEASE:
8757		/*
8758		 * if this isn't an exclusive access res generate UA for all
8759		 * other registrants.
8760		 */
8761		if (lun->res_type != SPR_TYPE_EX_AC
8762		 && lun->res_type != SPR_TYPE_WR_EX) {
8763			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8764				if (lun->per_res[i+persis_offset].registered)
8765					lun->pending_sense[i].ua_pending |=
8766						CTL_UA_RES_RELEASE;
8767		}
8768
8769		lun->flags &= ~CTL_LUN_PR_RESERVED;
8770		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8771		lun->res_type = 0;
8772		break;
8773
8774	case CTL_PR_PREEMPT:
8775		ctl_pro_preempt_other(lun, msg);
8776		break;
8777	case CTL_PR_CLEAR:
8778		lun->flags &= ~CTL_LUN_PR_RESERVED;
8779		lun->res_type = 0;
8780		lun->pr_key_count = 0;
8781		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8782
8783		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8784			if (lun->per_res[i].registered == 0)
8785				continue;
8786			if (!persis_offset
8787			 && i < CTL_MAX_INITIATORS)
8788				lun->pending_sense[i].ua_pending |=
8789					CTL_UA_RES_PREEMPT;
8790			else if (persis_offset
8791			      && i >= persis_offset)
8792   				lun->pending_sense[i-persis_offset].ua_pending|=
8793					CTL_UA_RES_PREEMPT;
8794			memset(&lun->per_res[i].res_key, 0,
8795			       sizeof(struct scsi_per_res_key));
8796			lun->per_res[i].registered = 0;
8797		}
8798		lun->PRGeneration++;
8799		break;
8800	}
8801
8802	mtx_unlock(&lun->lun_lock);
8803}
8804
8805int
8806ctl_read_write(struct ctl_scsiio *ctsio)
8807{
8808	struct ctl_lun *lun;
8809	struct ctl_lba_len_flags *lbalen;
8810	uint64_t lba;
8811	uint32_t num_blocks;
8812	int fua, dpo;
8813	int retval;
8814	int isread;
8815
8816	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8817
8818	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8819
8820	fua = 0;
8821	dpo = 0;
8822
8823	retval = CTL_RETVAL_COMPLETE;
8824
8825	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8826	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8827	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8828		uint32_t residx;
8829
8830		/*
8831		 * XXX KDM need a lock here.
8832		 */
8833		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8834		if ((lun->res_type == SPR_TYPE_EX_AC
8835		  && residx != lun->pr_res_idx)
8836		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8837		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8838		  && !lun->per_res[residx].registered)) {
8839			ctl_set_reservation_conflict(ctsio);
8840			ctl_done((union ctl_io *)ctsio);
8841			return (CTL_RETVAL_COMPLETE);
8842	        }
8843	}
8844
8845	switch (ctsio->cdb[0]) {
8846	case READ_6:
8847	case WRITE_6: {
8848		struct scsi_rw_6 *cdb;
8849
8850		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8851
8852		lba = scsi_3btoul(cdb->addr);
8853		/* only 5 bits are valid in the most significant address byte */
8854		lba &= 0x1fffff;
8855		num_blocks = cdb->length;
8856		/*
8857		 * This is correct according to SBC-2.
8858		 */
8859		if (num_blocks == 0)
8860			num_blocks = 256;
8861		break;
8862	}
8863	case READ_10:
8864	case WRITE_10: {
8865		struct scsi_rw_10 *cdb;
8866
8867		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8868
8869		if (cdb->byte2 & SRW10_FUA)
8870			fua = 1;
8871		if (cdb->byte2 & SRW10_DPO)
8872			dpo = 1;
8873
8874		lba = scsi_4btoul(cdb->addr);
8875		num_blocks = scsi_2btoul(cdb->length);
8876		break;
8877	}
8878	case WRITE_VERIFY_10: {
8879		struct scsi_write_verify_10 *cdb;
8880
8881		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8882
8883		/*
8884		 * XXX KDM we should do actual write verify support at some
8885		 * point.  This is obviously fake, we're just translating
8886		 * things to a write.  So we don't even bother checking the
8887		 * BYTCHK field, since we don't do any verification.  If
8888		 * the user asks for it, we'll just pretend we did it.
8889		 */
8890		if (cdb->byte2 & SWV_DPO)
8891			dpo = 1;
8892
8893		lba = scsi_4btoul(cdb->addr);
8894		num_blocks = scsi_2btoul(cdb->length);
8895		break;
8896	}
8897	case READ_12:
8898	case WRITE_12: {
8899		struct scsi_rw_12 *cdb;
8900
8901		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8902
8903		if (cdb->byte2 & SRW12_FUA)
8904			fua = 1;
8905		if (cdb->byte2 & SRW12_DPO)
8906			dpo = 1;
8907		lba = scsi_4btoul(cdb->addr);
8908		num_blocks = scsi_4btoul(cdb->length);
8909		break;
8910	}
8911	case WRITE_VERIFY_12: {
8912		struct scsi_write_verify_12 *cdb;
8913
8914		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8915
8916		if (cdb->byte2 & SWV_DPO)
8917			dpo = 1;
8918
8919		lba = scsi_4btoul(cdb->addr);
8920		num_blocks = scsi_4btoul(cdb->length);
8921
8922		break;
8923	}
8924	case READ_16:
8925	case WRITE_16: {
8926		struct scsi_rw_16 *cdb;
8927
8928		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8929
8930		if (cdb->byte2 & SRW12_FUA)
8931			fua = 1;
8932		if (cdb->byte2 & SRW12_DPO)
8933			dpo = 1;
8934
8935		lba = scsi_8btou64(cdb->addr);
8936		num_blocks = scsi_4btoul(cdb->length);
8937		break;
8938	}
8939	case WRITE_VERIFY_16: {
8940		struct scsi_write_verify_16 *cdb;
8941
8942		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8943
8944		if (cdb->byte2 & SWV_DPO)
8945			dpo = 1;
8946
8947		lba = scsi_8btou64(cdb->addr);
8948		num_blocks = scsi_4btoul(cdb->length);
8949		break;
8950	}
8951	default:
8952		/*
8953		 * We got a command we don't support.  This shouldn't
8954		 * happen, commands should be filtered out above us.
8955		 */
8956		ctl_set_invalid_opcode(ctsio);
8957		ctl_done((union ctl_io *)ctsio);
8958
8959		return (CTL_RETVAL_COMPLETE);
8960		break; /* NOTREACHED */
8961	}
8962
8963	/*
8964	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8965	 * interesting for us, but if RAIDCore is in write-back mode,
8966	 * getting it to do write-through for a particular transaction may
8967	 * not be possible.
8968	 */
8969
8970	/*
8971	 * The first check is to make sure we're in bounds, the second
8972	 * check is to catch wrap-around problems.  If the lba + num blocks
8973	 * is less than the lba, then we've wrapped around and the block
8974	 * range is invalid anyway.
8975	 */
8976	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8977	 || ((lba + num_blocks) < lba)) {
8978		ctl_set_lba_out_of_range(ctsio);
8979		ctl_done((union ctl_io *)ctsio);
8980		return (CTL_RETVAL_COMPLETE);
8981	}
8982
8983	/*
8984	 * According to SBC-3, a transfer length of 0 is not an error.
8985	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8986	 * translates to 256 blocks for those commands.
8987	 */
8988	if (num_blocks == 0) {
8989		ctl_set_success(ctsio);
8990		ctl_done((union ctl_io *)ctsio);
8991		return (CTL_RETVAL_COMPLETE);
8992	}
8993
8994	lbalen = (struct ctl_lba_len_flags *)
8995	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8996	lbalen->lba = lba;
8997	lbalen->len = num_blocks;
8998	lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
8999
9000	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9001	ctsio->kern_rel_offset = 0;
9002
9003	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9004
9005	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9006
9007	return (retval);
9008}
9009
9010static int
9011ctl_cnw_cont(union ctl_io *io)
9012{
9013	struct ctl_scsiio *ctsio;
9014	struct ctl_lun *lun;
9015	struct ctl_lba_len_flags *lbalen;
9016	int retval;
9017
9018	ctsio = &io->scsiio;
9019	ctsio->io_hdr.status = CTL_STATUS_NONE;
9020	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9021	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9022	lbalen = (struct ctl_lba_len_flags *)
9023	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9024	lbalen->flags = CTL_LLF_WRITE;
9025
9026	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9027	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9028	return (retval);
9029}
9030
9031int
9032ctl_cnw(struct ctl_scsiio *ctsio)
9033{
9034	struct ctl_lun *lun;
9035	struct ctl_lba_len_flags *lbalen;
9036	uint64_t lba;
9037	uint32_t num_blocks;
9038	int fua, dpo;
9039	int retval;
9040
9041	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9042
9043	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9044
9045	fua = 0;
9046	dpo = 0;
9047
9048	retval = CTL_RETVAL_COMPLETE;
9049
9050	switch (ctsio->cdb[0]) {
9051	case COMPARE_AND_WRITE: {
9052		struct scsi_compare_and_write *cdb;
9053
9054		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9055
9056		if (cdb->byte2 & SRW10_FUA)
9057			fua = 1;
9058		if (cdb->byte2 & SRW10_DPO)
9059			dpo = 1;
9060		lba = scsi_8btou64(cdb->addr);
9061		num_blocks = cdb->length;
9062		break;
9063	}
9064	default:
9065		/*
9066		 * We got a command we don't support.  This shouldn't
9067		 * happen, commands should be filtered out above us.
9068		 */
9069		ctl_set_invalid_opcode(ctsio);
9070		ctl_done((union ctl_io *)ctsio);
9071
9072		return (CTL_RETVAL_COMPLETE);
9073		break; /* NOTREACHED */
9074	}
9075
9076	/*
9077	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
9078	 * interesting for us, but if RAIDCore is in write-back mode,
9079	 * getting it to do write-through for a particular transaction may
9080	 * not be possible.
9081	 */
9082
9083	/*
9084	 * The first check is to make sure we're in bounds, the second
9085	 * check is to catch wrap-around problems.  If the lba + num blocks
9086	 * is less than the lba, then we've wrapped around and the block
9087	 * range is invalid anyway.
9088	 */
9089	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9090	 || ((lba + num_blocks) < lba)) {
9091		ctl_set_lba_out_of_range(ctsio);
9092		ctl_done((union ctl_io *)ctsio);
9093		return (CTL_RETVAL_COMPLETE);
9094	}
9095
9096	/*
9097	 * According to SBC-3, a transfer length of 0 is not an error.
9098	 */
9099	if (num_blocks == 0) {
9100		ctl_set_success(ctsio);
9101		ctl_done((union ctl_io *)ctsio);
9102		return (CTL_RETVAL_COMPLETE);
9103	}
9104
9105	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9106	ctsio->kern_rel_offset = 0;
9107
9108	/*
9109	 * Set the IO_CONT flag, so that if this I/O gets passed to
9110	 * ctl_data_submit_done(), it'll get passed back to
9111	 * ctl_ctl_cnw_cont() for further processing.
9112	 */
9113	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9114	ctsio->io_cont = ctl_cnw_cont;
9115
9116	lbalen = (struct ctl_lba_len_flags *)
9117	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9118	lbalen->lba = lba;
9119	lbalen->len = num_blocks;
9120	lbalen->flags = CTL_LLF_COMPARE;
9121
9122	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9123	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9124	return (retval);
9125}
9126
9127int
9128ctl_verify(struct ctl_scsiio *ctsio)
9129{
9130	struct ctl_lun *lun;
9131	struct ctl_lba_len_flags *lbalen;
9132	uint64_t lba;
9133	uint32_t num_blocks;
9134	int bytchk, dpo;
9135	int retval;
9136
9137	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9138
9139	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9140
9141	bytchk = 0;
9142	dpo = 0;
9143	retval = CTL_RETVAL_COMPLETE;
9144
9145	switch (ctsio->cdb[0]) {
9146	case VERIFY_10: {
9147		struct scsi_verify_10 *cdb;
9148
9149		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9150		if (cdb->byte2 & SVFY_BYTCHK)
9151			bytchk = 1;
9152		if (cdb->byte2 & SVFY_DPO)
9153			dpo = 1;
9154		lba = scsi_4btoul(cdb->addr);
9155		num_blocks = scsi_2btoul(cdb->length);
9156		break;
9157	}
9158	case VERIFY_12: {
9159		struct scsi_verify_12 *cdb;
9160
9161		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9162		if (cdb->byte2 & SVFY_BYTCHK)
9163			bytchk = 1;
9164		if (cdb->byte2 & SVFY_DPO)
9165			dpo = 1;
9166		lba = scsi_4btoul(cdb->addr);
9167		num_blocks = scsi_4btoul(cdb->length);
9168		break;
9169	}
9170	case VERIFY_16: {
9171		struct scsi_rw_16 *cdb;
9172
9173		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9174		if (cdb->byte2 & SVFY_BYTCHK)
9175			bytchk = 1;
9176		if (cdb->byte2 & SVFY_DPO)
9177			dpo = 1;
9178		lba = scsi_8btou64(cdb->addr);
9179		num_blocks = scsi_4btoul(cdb->length);
9180		break;
9181	}
9182	default:
9183		/*
9184		 * We got a command we don't support.  This shouldn't
9185		 * happen, commands should be filtered out above us.
9186		 */
9187		ctl_set_invalid_opcode(ctsio);
9188		ctl_done((union ctl_io *)ctsio);
9189		return (CTL_RETVAL_COMPLETE);
9190	}
9191
9192	/*
9193	 * The first check is to make sure we're in bounds, the second
9194	 * check is to catch wrap-around problems.  If the lba + num blocks
9195	 * is less than the lba, then we've wrapped around and the block
9196	 * range is invalid anyway.
9197	 */
9198	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9199	 || ((lba + num_blocks) < lba)) {
9200		ctl_set_lba_out_of_range(ctsio);
9201		ctl_done((union ctl_io *)ctsio);
9202		return (CTL_RETVAL_COMPLETE);
9203	}
9204
9205	/*
9206	 * According to SBC-3, a transfer length of 0 is not an error.
9207	 */
9208	if (num_blocks == 0) {
9209		ctl_set_success(ctsio);
9210		ctl_done((union ctl_io *)ctsio);
9211		return (CTL_RETVAL_COMPLETE);
9212	}
9213
9214	lbalen = (struct ctl_lba_len_flags *)
9215	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9216	lbalen->lba = lba;
9217	lbalen->len = num_blocks;
9218	if (bytchk) {
9219		lbalen->flags = CTL_LLF_COMPARE;
9220		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9221	} else {
9222		lbalen->flags = CTL_LLF_VERIFY;
9223		ctsio->kern_total_len = 0;
9224	}
9225	ctsio->kern_rel_offset = 0;
9226
9227	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9228	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9229	return (retval);
9230}
9231
9232int
9233ctl_report_luns(struct ctl_scsiio *ctsio)
9234{
9235	struct scsi_report_luns *cdb;
9236	struct scsi_report_luns_data *lun_data;
9237	struct ctl_lun *lun, *request_lun;
9238	int num_luns, retval;
9239	uint32_t alloc_len, lun_datalen;
9240	int num_filled, well_known;
9241	uint32_t initidx, targ_lun_id, lun_id;
9242
9243	retval = CTL_RETVAL_COMPLETE;
9244	well_known = 0;
9245
9246	cdb = (struct scsi_report_luns *)ctsio->cdb;
9247
9248	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9249
9250	mtx_lock(&control_softc->ctl_lock);
9251	num_luns = control_softc->num_luns;
9252	mtx_unlock(&control_softc->ctl_lock);
9253
9254	switch (cdb->select_report) {
9255	case RPL_REPORT_DEFAULT:
9256	case RPL_REPORT_ALL:
9257		break;
9258	case RPL_REPORT_WELLKNOWN:
9259		well_known = 1;
9260		num_luns = 0;
9261		break;
9262	default:
9263		ctl_set_invalid_field(ctsio,
9264				      /*sks_valid*/ 1,
9265				      /*command*/ 1,
9266				      /*field*/ 2,
9267				      /*bit_valid*/ 0,
9268				      /*bit*/ 0);
9269		ctl_done((union ctl_io *)ctsio);
9270		return (retval);
9271		break; /* NOTREACHED */
9272	}
9273
9274	alloc_len = scsi_4btoul(cdb->length);
9275	/*
9276	 * The initiator has to allocate at least 16 bytes for this request,
9277	 * so he can at least get the header and the first LUN.  Otherwise
9278	 * we reject the request (per SPC-3 rev 14, section 6.21).
9279	 */
9280	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9281	    sizeof(struct scsi_report_luns_lundata))) {
9282		ctl_set_invalid_field(ctsio,
9283				      /*sks_valid*/ 1,
9284				      /*command*/ 1,
9285				      /*field*/ 6,
9286				      /*bit_valid*/ 0,
9287				      /*bit*/ 0);
9288		ctl_done((union ctl_io *)ctsio);
9289		return (retval);
9290	}
9291
9292	request_lun = (struct ctl_lun *)
9293		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9294
9295	lun_datalen = sizeof(*lun_data) +
9296		(num_luns * sizeof(struct scsi_report_luns_lundata));
9297
9298	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9299	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9300	ctsio->kern_sg_entries = 0;
9301
9302	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9303
9304	mtx_lock(&control_softc->ctl_lock);
9305	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9306		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9307		if (lun_id >= CTL_MAX_LUNS)
9308			continue;
9309		lun = control_softc->ctl_luns[lun_id];
9310		if (lun == NULL)
9311			continue;
9312
9313		if (targ_lun_id <= 0xff) {
9314			/*
9315			 * Peripheral addressing method, bus number 0.
9316			 */
9317			lun_data->luns[num_filled].lundata[0] =
9318				RPL_LUNDATA_ATYP_PERIPH;
9319			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9320			num_filled++;
9321		} else if (targ_lun_id <= 0x3fff) {
9322			/*
9323			 * Flat addressing method.
9324			 */
9325			lun_data->luns[num_filled].lundata[0] =
9326				RPL_LUNDATA_ATYP_FLAT |
9327				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9328#ifdef OLDCTLHEADERS
9329				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9330				(targ_lun_id & SRLD_BUS_LUN_MASK);
9331#endif
9332			lun_data->luns[num_filled].lundata[1] =
9333#ifdef OLDCTLHEADERS
9334				targ_lun_id >> SRLD_BUS_LUN_BITS;
9335#endif
9336				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9337			num_filled++;
9338		} else {
9339			printf("ctl_report_luns: bogus LUN number %jd, "
9340			       "skipping\n", (intmax_t)targ_lun_id);
9341		}
9342		/*
9343		 * According to SPC-3, rev 14 section 6.21:
9344		 *
9345		 * "The execution of a REPORT LUNS command to any valid and
9346		 * installed logical unit shall clear the REPORTED LUNS DATA
9347		 * HAS CHANGED unit attention condition for all logical
9348		 * units of that target with respect to the requesting
9349		 * initiator. A valid and installed logical unit is one
9350		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9351		 * INQUIRY data (see 6.4.2)."
9352		 *
9353		 * If request_lun is NULL, the LUN this report luns command
9354		 * was issued to is either disabled or doesn't exist. In that
9355		 * case, we shouldn't clear any pending lun change unit
9356		 * attention.
9357		 */
9358		if (request_lun != NULL) {
9359			mtx_lock(&lun->lun_lock);
9360			lun->pending_sense[initidx].ua_pending &=
9361				~CTL_UA_LUN_CHANGE;
9362			mtx_unlock(&lun->lun_lock);
9363		}
9364	}
9365	mtx_unlock(&control_softc->ctl_lock);
9366
9367	/*
9368	 * It's quite possible that we've returned fewer LUNs than we allocated
9369	 * space for.  Trim it.
9370	 */
9371	lun_datalen = sizeof(*lun_data) +
9372		(num_filled * sizeof(struct scsi_report_luns_lundata));
9373
9374	if (lun_datalen < alloc_len) {
9375		ctsio->residual = alloc_len - lun_datalen;
9376		ctsio->kern_data_len = lun_datalen;
9377		ctsio->kern_total_len = lun_datalen;
9378	} else {
9379		ctsio->residual = 0;
9380		ctsio->kern_data_len = alloc_len;
9381		ctsio->kern_total_len = alloc_len;
9382	}
9383	ctsio->kern_data_resid = 0;
9384	ctsio->kern_rel_offset = 0;
9385	ctsio->kern_sg_entries = 0;
9386
9387	/*
9388	 * We set this to the actual data length, regardless of how much
9389	 * space we actually have to return results.  If the user looks at
9390	 * this value, he'll know whether or not he allocated enough space
9391	 * and reissue the command if necessary.  We don't support well
9392	 * known logical units, so if the user asks for that, return none.
9393	 */
9394	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9395
9396	/*
9397	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9398	 * this request.
9399	 */
9400	ctsio->scsi_status = SCSI_STATUS_OK;
9401
9402	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9403	ctsio->be_move_done = ctl_config_move_done;
9404	ctl_datamove((union ctl_io *)ctsio);
9405
9406	return (retval);
9407}
9408
9409int
9410ctl_request_sense(struct ctl_scsiio *ctsio)
9411{
9412	struct scsi_request_sense *cdb;
9413	struct scsi_sense_data *sense_ptr;
9414	struct ctl_lun *lun;
9415	uint32_t initidx;
9416	int have_error;
9417	scsi_sense_data_type sense_format;
9418
9419	cdb = (struct scsi_request_sense *)ctsio->cdb;
9420
9421	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9422
9423	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9424
9425	/*
9426	 * Determine which sense format the user wants.
9427	 */
9428	if (cdb->byte2 & SRS_DESC)
9429		sense_format = SSD_TYPE_DESC;
9430	else
9431		sense_format = SSD_TYPE_FIXED;
9432
9433	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9434	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9435	ctsio->kern_sg_entries = 0;
9436
9437	/*
9438	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9439	 * larger than the largest allowed value for the length field in the
9440	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9441	 */
9442	ctsio->residual = 0;
9443	ctsio->kern_data_len = cdb->length;
9444	ctsio->kern_total_len = cdb->length;
9445
9446	ctsio->kern_data_resid = 0;
9447	ctsio->kern_rel_offset = 0;
9448	ctsio->kern_sg_entries = 0;
9449
9450	/*
9451	 * If we don't have a LUN, we don't have any pending sense.
9452	 */
9453	if (lun == NULL)
9454		goto no_sense;
9455
9456	have_error = 0;
9457	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9458	/*
9459	 * Check for pending sense, and then for pending unit attentions.
9460	 * Pending sense gets returned first, then pending unit attentions.
9461	 */
9462	mtx_lock(&lun->lun_lock);
9463	if (ctl_is_set(lun->have_ca, initidx)) {
9464		scsi_sense_data_type stored_format;
9465
9466		/*
9467		 * Check to see which sense format was used for the stored
9468		 * sense data.
9469		 */
9470		stored_format = scsi_sense_type(
9471		    &lun->pending_sense[initidx].sense);
9472
9473		/*
9474		 * If the user requested a different sense format than the
9475		 * one we stored, then we need to convert it to the other
9476		 * format.  If we're going from descriptor to fixed format
9477		 * sense data, we may lose things in translation, depending
9478		 * on what options were used.
9479		 *
9480		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9481		 * for some reason we'll just copy it out as-is.
9482		 */
9483		if ((stored_format == SSD_TYPE_FIXED)
9484		 && (sense_format == SSD_TYPE_DESC))
9485			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9486			    &lun->pending_sense[initidx].sense,
9487			    (struct scsi_sense_data_desc *)sense_ptr);
9488		else if ((stored_format == SSD_TYPE_DESC)
9489		      && (sense_format == SSD_TYPE_FIXED))
9490			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9491			    &lun->pending_sense[initidx].sense,
9492			    (struct scsi_sense_data_fixed *)sense_ptr);
9493		else
9494			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
9495			       ctl_min(sizeof(*sense_ptr),
9496			       sizeof(lun->pending_sense[initidx].sense)));
9497
9498		ctl_clear_mask(lun->have_ca, initidx);
9499		have_error = 1;
9500	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
9501		ctl_ua_type ua_type;
9502
9503		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
9504				       sense_ptr, sense_format);
9505		if (ua_type != CTL_UA_NONE) {
9506			have_error = 1;
9507			/* We're reporting this UA, so clear it */
9508			lun->pending_sense[initidx].ua_pending &= ~ua_type;
9509		}
9510	}
9511	mtx_unlock(&lun->lun_lock);
9512
9513	/*
9514	 * We already have a pending error, return it.
9515	 */
9516	if (have_error != 0) {
9517		/*
9518		 * We report the SCSI status as OK, since the status of the
9519		 * request sense command itself is OK.
9520		 */
9521		ctsio->scsi_status = SCSI_STATUS_OK;
9522
9523		/*
9524		 * We report 0 for the sense length, because we aren't doing
9525		 * autosense in this case.  We're reporting sense as
9526		 * parameter data.
9527		 */
9528		ctsio->sense_len = 0;
9529		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9530		ctsio->be_move_done = ctl_config_move_done;
9531		ctl_datamove((union ctl_io *)ctsio);
9532
9533		return (CTL_RETVAL_COMPLETE);
9534	}
9535
9536no_sense:
9537
9538	/*
9539	 * No sense information to report, so we report that everything is
9540	 * okay.
9541	 */
9542	ctl_set_sense_data(sense_ptr,
9543			   lun,
9544			   sense_format,
9545			   /*current_error*/ 1,
9546			   /*sense_key*/ SSD_KEY_NO_SENSE,
9547			   /*asc*/ 0x00,
9548			   /*ascq*/ 0x00,
9549			   SSD_ELEM_NONE);
9550
9551	ctsio->scsi_status = SCSI_STATUS_OK;
9552
9553	/*
9554	 * We report 0 for the sense length, because we aren't doing
9555	 * autosense in this case.  We're reporting sense as parameter data.
9556	 */
9557	ctsio->sense_len = 0;
9558	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9559	ctsio->be_move_done = ctl_config_move_done;
9560	ctl_datamove((union ctl_io *)ctsio);
9561
9562	return (CTL_RETVAL_COMPLETE);
9563}
9564
9565int
9566ctl_tur(struct ctl_scsiio *ctsio)
9567{
9568	struct ctl_lun *lun;
9569
9570	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9571
9572	CTL_DEBUG_PRINT(("ctl_tur\n"));
9573
9574	if (lun == NULL)
9575		return (EINVAL);
9576
9577	ctsio->scsi_status = SCSI_STATUS_OK;
9578	ctsio->io_hdr.status = CTL_SUCCESS;
9579
9580	ctl_done((union ctl_io *)ctsio);
9581
9582	return (CTL_RETVAL_COMPLETE);
9583}
9584
9585#ifdef notyet
9586static int
9587ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9588{
9589
9590}
9591#endif
9592
9593static int
9594ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9595{
9596	struct scsi_vpd_supported_pages *pages;
9597	int sup_page_size;
9598	struct ctl_lun *lun;
9599
9600	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9601
9602	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9603	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9604	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9605	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9606	ctsio->kern_sg_entries = 0;
9607
9608	if (sup_page_size < alloc_len) {
9609		ctsio->residual = alloc_len - sup_page_size;
9610		ctsio->kern_data_len = sup_page_size;
9611		ctsio->kern_total_len = sup_page_size;
9612	} else {
9613		ctsio->residual = 0;
9614		ctsio->kern_data_len = alloc_len;
9615		ctsio->kern_total_len = alloc_len;
9616	}
9617	ctsio->kern_data_resid = 0;
9618	ctsio->kern_rel_offset = 0;
9619	ctsio->kern_sg_entries = 0;
9620
9621	/*
9622	 * The control device is always connected.  The disk device, on the
9623	 * other hand, may not be online all the time.  Need to change this
9624	 * to figure out whether the disk device is actually online or not.
9625	 */
9626	if (lun != NULL)
9627		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9628				lun->be_lun->lun_type;
9629	else
9630		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9631
9632	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9633	/* Supported VPD pages */
9634	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9635	/* Serial Number */
9636	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9637	/* Device Identification */
9638	pages->page_list[2] = SVPD_DEVICE_ID;
9639	/* SCSI Ports */
9640	pages->page_list[3] = SVPD_SCSI_PORTS;
9641	/* Block limits */
9642	pages->page_list[4] = SVPD_BLOCK_LIMITS;
9643	/* Logical Block Provisioning */
9644	pages->page_list[5] = SVPD_LBP;
9645
9646	ctsio->scsi_status = SCSI_STATUS_OK;
9647
9648	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9649	ctsio->be_move_done = ctl_config_move_done;
9650	ctl_datamove((union ctl_io *)ctsio);
9651
9652	return (CTL_RETVAL_COMPLETE);
9653}
9654
9655static int
9656ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9657{
9658	struct scsi_vpd_unit_serial_number *sn_ptr;
9659	struct ctl_lun *lun;
9660
9661	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9662
9663	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9664	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9665	ctsio->kern_sg_entries = 0;
9666
9667	if (sizeof(*sn_ptr) < alloc_len) {
9668		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9669		ctsio->kern_data_len = sizeof(*sn_ptr);
9670		ctsio->kern_total_len = sizeof(*sn_ptr);
9671	} else {
9672		ctsio->residual = 0;
9673		ctsio->kern_data_len = alloc_len;
9674		ctsio->kern_total_len = alloc_len;
9675	}
9676	ctsio->kern_data_resid = 0;
9677	ctsio->kern_rel_offset = 0;
9678	ctsio->kern_sg_entries = 0;
9679
9680	/*
9681	 * The control device is always connected.  The disk device, on the
9682	 * other hand, may not be online all the time.  Need to change this
9683	 * to figure out whether the disk device is actually online or not.
9684	 */
9685	if (lun != NULL)
9686		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9687				  lun->be_lun->lun_type;
9688	else
9689		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9690
9691	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9692	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9693	/*
9694	 * If we don't have a LUN, we just leave the serial number as
9695	 * all spaces.
9696	 */
9697	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9698	if (lun != NULL) {
9699		strncpy((char *)sn_ptr->serial_num,
9700			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9701	}
9702	ctsio->scsi_status = SCSI_STATUS_OK;
9703
9704	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9705	ctsio->be_move_done = ctl_config_move_done;
9706	ctl_datamove((union ctl_io *)ctsio);
9707
9708	return (CTL_RETVAL_COMPLETE);
9709}
9710
9711
9712static int
9713ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9714{
9715	struct scsi_vpd_device_id *devid_ptr;
9716	struct scsi_vpd_id_descriptor *desc;
9717	struct ctl_softc *ctl_softc;
9718	struct ctl_lun *lun;
9719	struct ctl_port *port;
9720	int data_len;
9721	uint8_t proto;
9722
9723	ctl_softc = control_softc;
9724
9725	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9726	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9727
9728	data_len = sizeof(struct scsi_vpd_device_id) +
9729	    sizeof(struct scsi_vpd_id_descriptor) +
9730		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9731	    sizeof(struct scsi_vpd_id_descriptor) +
9732		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9733	if (lun && lun->lun_devid)
9734		data_len += lun->lun_devid->len;
9735	if (port->port_devid)
9736		data_len += port->port_devid->len;
9737	if (port->target_devid)
9738		data_len += port->target_devid->len;
9739
9740	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9741	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9742	ctsio->kern_sg_entries = 0;
9743
9744	if (data_len < alloc_len) {
9745		ctsio->residual = alloc_len - data_len;
9746		ctsio->kern_data_len = data_len;
9747		ctsio->kern_total_len = data_len;
9748	} else {
9749		ctsio->residual = 0;
9750		ctsio->kern_data_len = alloc_len;
9751		ctsio->kern_total_len = alloc_len;
9752	}
9753	ctsio->kern_data_resid = 0;
9754	ctsio->kern_rel_offset = 0;
9755	ctsio->kern_sg_entries = 0;
9756
9757	/*
9758	 * The control device is always connected.  The disk device, on the
9759	 * other hand, may not be online all the time.
9760	 */
9761	if (lun != NULL)
9762		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9763				     lun->be_lun->lun_type;
9764	else
9765		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9766	devid_ptr->page_code = SVPD_DEVICE_ID;
9767	scsi_ulto2b(data_len - 4, devid_ptr->length);
9768
9769	if (port->port_type == CTL_PORT_FC)
9770		proto = SCSI_PROTO_FC << 4;
9771	else if (port->port_type == CTL_PORT_ISCSI)
9772		proto = SCSI_PROTO_ISCSI << 4;
9773	else
9774		proto = SCSI_PROTO_SPI << 4;
9775	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9776
9777	/*
9778	 * We're using a LUN association here.  i.e., this device ID is a
9779	 * per-LUN identifier.
9780	 */
9781	if (lun && lun->lun_devid) {
9782		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9783		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9784		    lun->lun_devid->len);
9785	}
9786
9787	/*
9788	 * This is for the WWPN which is a port association.
9789	 */
9790	if (port->port_devid) {
9791		memcpy(desc, port->port_devid->data, port->port_devid->len);
9792		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9793		    port->port_devid->len);
9794	}
9795
9796	/*
9797	 * This is for the Relative Target Port(type 4h) identifier
9798	 */
9799	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9800	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9801	    SVPD_ID_TYPE_RELTARG;
9802	desc->length = 4;
9803	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9804	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9805	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9806
9807	/*
9808	 * This is for the Target Port Group(type 5h) identifier
9809	 */
9810	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9811	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9812	    SVPD_ID_TYPE_TPORTGRP;
9813	desc->length = 4;
9814	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9815	    &desc->identifier[2]);
9816	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9817	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9818
9819	/*
9820	 * This is for the Target identifier
9821	 */
9822	if (port->target_devid) {
9823		memcpy(desc, port->target_devid->data, port->target_devid->len);
9824	}
9825
9826	ctsio->scsi_status = SCSI_STATUS_OK;
9827	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9828	ctsio->be_move_done = ctl_config_move_done;
9829	ctl_datamove((union ctl_io *)ctsio);
9830
9831	return (CTL_RETVAL_COMPLETE);
9832}
9833
9834static int
9835ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9836{
9837	struct ctl_softc *softc = control_softc;
9838	struct scsi_vpd_scsi_ports *sp;
9839	struct scsi_vpd_port_designation *pd;
9840	struct scsi_vpd_port_designation_cont *pdc;
9841	struct ctl_lun *lun;
9842	struct ctl_port *port;
9843	int data_len, num_target_ports, id_len, g, pg, p;
9844	int num_target_port_groups, single;
9845
9846	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9847
9848	single = ctl_is_single;
9849	if (single)
9850		num_target_port_groups = 1;
9851	else
9852		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
9853	num_target_ports = 0;
9854	id_len = 0;
9855	mtx_lock(&softc->ctl_lock);
9856	STAILQ_FOREACH(port, &softc->port_list, links) {
9857		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9858			continue;
9859		if (ctl_map_lun_back(port->targ_port, lun->lun) >=
9860		    CTL_MAX_LUNS)
9861			continue;
9862		num_target_ports++;
9863		if (port->port_devid)
9864			id_len += port->port_devid->len;
9865	}
9866	mtx_unlock(&softc->ctl_lock);
9867
9868	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
9869	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9870	     sizeof(struct scsi_vpd_port_designation_cont)) + id_len;
9871	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9872	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9873	ctsio->kern_sg_entries = 0;
9874
9875	if (data_len < alloc_len) {
9876		ctsio->residual = alloc_len - data_len;
9877		ctsio->kern_data_len = data_len;
9878		ctsio->kern_total_len = data_len;
9879	} else {
9880		ctsio->residual = 0;
9881		ctsio->kern_data_len = alloc_len;
9882		ctsio->kern_total_len = alloc_len;
9883	}
9884	ctsio->kern_data_resid = 0;
9885	ctsio->kern_rel_offset = 0;
9886	ctsio->kern_sg_entries = 0;
9887
9888	/*
9889	 * The control device is always connected.  The disk device, on the
9890	 * other hand, may not be online all the time.  Need to change this
9891	 * to figure out whether the disk device is actually online or not.
9892	 */
9893	if (lun != NULL)
9894		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9895				  lun->be_lun->lun_type;
9896	else
9897		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9898
9899	sp->page_code = SVPD_SCSI_PORTS;
9900	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9901	    sp->page_length);
9902	pd = &sp->design[0];
9903
9904	mtx_lock(&softc->ctl_lock);
9905	if (softc->flags & CTL_FLAG_MASTER_SHELF)
9906		pg = 0;
9907	else
9908		pg = 1;
9909	for (g = 0; g < num_target_port_groups; g++) {
9910		STAILQ_FOREACH(port, &softc->port_list, links) {
9911			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9912				continue;
9913			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
9914			    CTL_MAX_LUNS)
9915				continue;
9916			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
9917			scsi_ulto2b(p, pd->relative_port_id);
9918			scsi_ulto2b(0, pd->initiator_transportid_length);
9919			pdc = (struct scsi_vpd_port_designation_cont *)
9920			    &pd->initiator_transportid[0];
9921			if (port->port_devid && g == pg) {
9922				id_len = port->port_devid->len;
9923				scsi_ulto2b(port->port_devid->len,
9924				    pdc->target_port_descriptors_length);
9925				memcpy(pdc->target_port_descriptors,
9926				    port->port_devid->data, port->port_devid->len);
9927			} else {
9928				id_len = 0;
9929				scsi_ulto2b(0, pdc->target_port_descriptors_length);
9930			}
9931			pd = (struct scsi_vpd_port_designation *)
9932			    ((uint8_t *)pdc->target_port_descriptors + id_len);
9933		}
9934	}
9935	mtx_unlock(&softc->ctl_lock);
9936
9937	ctsio->scsi_status = SCSI_STATUS_OK;
9938	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9939	ctsio->be_move_done = ctl_config_move_done;
9940	ctl_datamove((union ctl_io *)ctsio);
9941
9942	return (CTL_RETVAL_COMPLETE);
9943}
9944
9945static int
9946ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9947{
9948	struct scsi_vpd_block_limits *bl_ptr;
9949	struct ctl_lun *lun;
9950	int bs;
9951
9952	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9953	bs = lun->be_lun->blocksize;
9954
9955	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9956	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9957	ctsio->kern_sg_entries = 0;
9958
9959	if (sizeof(*bl_ptr) < alloc_len) {
9960		ctsio->residual = alloc_len - sizeof(*bl_ptr);
9961		ctsio->kern_data_len = sizeof(*bl_ptr);
9962		ctsio->kern_total_len = sizeof(*bl_ptr);
9963	} else {
9964		ctsio->residual = 0;
9965		ctsio->kern_data_len = alloc_len;
9966		ctsio->kern_total_len = alloc_len;
9967	}
9968	ctsio->kern_data_resid = 0;
9969	ctsio->kern_rel_offset = 0;
9970	ctsio->kern_sg_entries = 0;
9971
9972	/*
9973	 * The control device is always connected.  The disk device, on the
9974	 * other hand, may not be online all the time.  Need to change this
9975	 * to figure out whether the disk device is actually online or not.
9976	 */
9977	if (lun != NULL)
9978		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9979				  lun->be_lun->lun_type;
9980	else
9981		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9982
9983	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9984	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
9985	bl_ptr->max_cmp_write_len = 0xff;
9986	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9987	scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
9988	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9989		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9990		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9991	}
9992	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9993
9994	ctsio->scsi_status = SCSI_STATUS_OK;
9995	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9996	ctsio->be_move_done = ctl_config_move_done;
9997	ctl_datamove((union ctl_io *)ctsio);
9998
9999	return (CTL_RETVAL_COMPLETE);
10000}
10001
10002static int
10003ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10004{
10005	struct scsi_vpd_logical_block_prov *lbp_ptr;
10006	struct ctl_lun *lun;
10007	int bs;
10008
10009	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10010	bs = lun->be_lun->blocksize;
10011
10012	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10013	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10014	ctsio->kern_sg_entries = 0;
10015
10016	if (sizeof(*lbp_ptr) < alloc_len) {
10017		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10018		ctsio->kern_data_len = sizeof(*lbp_ptr);
10019		ctsio->kern_total_len = sizeof(*lbp_ptr);
10020	} else {
10021		ctsio->residual = 0;
10022		ctsio->kern_data_len = alloc_len;
10023		ctsio->kern_total_len = alloc_len;
10024	}
10025	ctsio->kern_data_resid = 0;
10026	ctsio->kern_rel_offset = 0;
10027	ctsio->kern_sg_entries = 0;
10028
10029	/*
10030	 * The control device is always connected.  The disk device, on the
10031	 * other hand, may not be online all the time.  Need to change this
10032	 * to figure out whether the disk device is actually online or not.
10033	 */
10034	if (lun != NULL)
10035		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10036				  lun->be_lun->lun_type;
10037	else
10038		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10039
10040	lbp_ptr->page_code = SVPD_LBP;
10041	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
10042		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10;
10043
10044	ctsio->scsi_status = SCSI_STATUS_OK;
10045	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10046	ctsio->be_move_done = ctl_config_move_done;
10047	ctl_datamove((union ctl_io *)ctsio);
10048
10049	return (CTL_RETVAL_COMPLETE);
10050}
10051
10052static int
10053ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10054{
10055	struct scsi_inquiry *cdb;
10056	struct ctl_lun *lun;
10057	int alloc_len, retval;
10058
10059	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10060	cdb = (struct scsi_inquiry *)ctsio->cdb;
10061
10062	retval = CTL_RETVAL_COMPLETE;
10063
10064	alloc_len = scsi_2btoul(cdb->length);
10065
10066	switch (cdb->page_code) {
10067	case SVPD_SUPPORTED_PAGES:
10068		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10069		break;
10070	case SVPD_UNIT_SERIAL_NUMBER:
10071		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10072		break;
10073	case SVPD_DEVICE_ID:
10074		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10075		break;
10076	case SVPD_SCSI_PORTS:
10077		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10078		break;
10079	case SVPD_BLOCK_LIMITS:
10080		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10081		break;
10082	case SVPD_LBP:
10083		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10084		break;
10085	default:
10086		ctl_set_invalid_field(ctsio,
10087				      /*sks_valid*/ 1,
10088				      /*command*/ 1,
10089				      /*field*/ 2,
10090				      /*bit_valid*/ 0,
10091				      /*bit*/ 0);
10092		ctl_done((union ctl_io *)ctsio);
10093		retval = CTL_RETVAL_COMPLETE;
10094		break;
10095	}
10096
10097	return (retval);
10098}
10099
10100static int
10101ctl_inquiry_std(struct ctl_scsiio *ctsio)
10102{
10103	struct scsi_inquiry_data *inq_ptr;
10104	struct scsi_inquiry *cdb;
10105	struct ctl_softc *ctl_softc;
10106	struct ctl_lun *lun;
10107	char *val;
10108	uint32_t alloc_len;
10109	int is_fc;
10110
10111	ctl_softc = control_softc;
10112
10113	/*
10114	 * Figure out whether we're talking to a Fibre Channel port or not.
10115	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10116	 * SCSI front ends.
10117	 */
10118	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
10119	    CTL_PORT_FC)
10120		is_fc = 0;
10121	else
10122		is_fc = 1;
10123
10124	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10125	cdb = (struct scsi_inquiry *)ctsio->cdb;
10126	alloc_len = scsi_2btoul(cdb->length);
10127
10128	/*
10129	 * We malloc the full inquiry data size here and fill it
10130	 * in.  If the user only asks for less, we'll give him
10131	 * that much.
10132	 */
10133	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
10134	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10135	ctsio->kern_sg_entries = 0;
10136	ctsio->kern_data_resid = 0;
10137	ctsio->kern_rel_offset = 0;
10138
10139	if (sizeof(*inq_ptr) < alloc_len) {
10140		ctsio->residual = alloc_len - sizeof(*inq_ptr);
10141		ctsio->kern_data_len = sizeof(*inq_ptr);
10142		ctsio->kern_total_len = sizeof(*inq_ptr);
10143	} else {
10144		ctsio->residual = 0;
10145		ctsio->kern_data_len = alloc_len;
10146		ctsio->kern_total_len = alloc_len;
10147	}
10148
10149	/*
10150	 * If we have a LUN configured, report it as connected.  Otherwise,
10151	 * report that it is offline or no device is supported, depending
10152	 * on the value of inquiry_pq_no_lun.
10153	 *
10154	 * According to the spec (SPC-4 r34), the peripheral qualifier
10155	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10156	 *
10157	 * "A peripheral device having the specified peripheral device type
10158	 * is not connected to this logical unit. However, the device
10159	 * server is capable of supporting the specified peripheral device
10160	 * type on this logical unit."
10161	 *
10162	 * According to the same spec, the peripheral qualifier
10163	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10164	 *
10165	 * "The device server is not capable of supporting a peripheral
10166	 * device on this logical unit. For this peripheral qualifier the
10167	 * peripheral device type shall be set to 1Fh. All other peripheral
10168	 * device type values are reserved for this peripheral qualifier."
10169	 *
10170	 * Given the text, it would seem that we probably want to report that
10171	 * the LUN is offline here.  There is no LUN connected, but we can
10172	 * support a LUN at the given LUN number.
10173	 *
10174	 * In the real world, though, it sounds like things are a little
10175	 * different:
10176	 *
10177	 * - Linux, when presented with a LUN with the offline peripheral
10178	 *   qualifier, will create an sg driver instance for it.  So when
10179	 *   you attach it to CTL, you wind up with a ton of sg driver
10180	 *   instances.  (One for every LUN that Linux bothered to probe.)
10181	 *   Linux does this despite the fact that it issues a REPORT LUNs
10182	 *   to LUN 0 to get the inventory of supported LUNs.
10183	 *
10184	 * - There is other anecdotal evidence (from Emulex folks) about
10185	 *   arrays that use the offline peripheral qualifier for LUNs that
10186	 *   are on the "passive" path in an active/passive array.
10187	 *
10188	 * So the solution is provide a hopefully reasonable default
10189	 * (return bad/no LUN) and allow the user to change the behavior
10190	 * with a tunable/sysctl variable.
10191	 */
10192	if (lun != NULL)
10193		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10194				  lun->be_lun->lun_type;
10195	else if (ctl_softc->inquiry_pq_no_lun == 0)
10196		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10197	else
10198		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10199
10200	/* RMB in byte 2 is 0 */
10201	inq_ptr->version = SCSI_REV_SPC3;
10202
10203	/*
10204	 * According to SAM-3, even if a device only supports a single
10205	 * level of LUN addressing, it should still set the HISUP bit:
10206	 *
10207	 * 4.9.1 Logical unit numbers overview
10208	 *
10209	 * All logical unit number formats described in this standard are
10210	 * hierarchical in structure even when only a single level in that
10211	 * hierarchy is used. The HISUP bit shall be set to one in the
10212	 * standard INQUIRY data (see SPC-2) when any logical unit number
10213	 * format described in this standard is used.  Non-hierarchical
10214	 * formats are outside the scope of this standard.
10215	 *
10216	 * Therefore we set the HiSup bit here.
10217	 *
10218	 * The reponse format is 2, per SPC-3.
10219	 */
10220	inq_ptr->response_format = SID_HiSup | 2;
10221
10222	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
10223	CTL_DEBUG_PRINT(("additional_length = %d\n",
10224			 inq_ptr->additional_length));
10225
10226	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
10227	/* 16 bit addressing */
10228	if (is_fc == 0)
10229		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10230	/* XXX set the SID_MultiP bit here if we're actually going to
10231	   respond on multiple ports */
10232	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10233
10234	/* 16 bit data bus, synchronous transfers */
10235	/* XXX these flags don't apply for FC */
10236	if (is_fc == 0)
10237		inq_ptr->flags = SID_WBus16 | SID_Sync;
10238	/*
10239	 * XXX KDM do we want to support tagged queueing on the control
10240	 * device at all?
10241	 */
10242	if ((lun == NULL)
10243	 || (lun->be_lun->lun_type != T_PROCESSOR))
10244		inq_ptr->flags |= SID_CmdQue;
10245	/*
10246	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10247	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10248	 * name and 4 bytes for the revision.
10249	 */
10250	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10251	    "vendor")) == NULL) {
10252		strcpy(inq_ptr->vendor, CTL_VENDOR);
10253	} else {
10254		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10255		strncpy(inq_ptr->vendor, val,
10256		    min(sizeof(inq_ptr->vendor), strlen(val)));
10257	}
10258	if (lun == NULL) {
10259		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10260	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10261		switch (lun->be_lun->lun_type) {
10262		case T_DIRECT:
10263			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10264			break;
10265		case T_PROCESSOR:
10266			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
10267			break;
10268		default:
10269			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
10270			break;
10271		}
10272	} else {
10273		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10274		strncpy(inq_ptr->product, val,
10275		    min(sizeof(inq_ptr->product), strlen(val)));
10276	}
10277
10278	/*
10279	 * XXX make this a macro somewhere so it automatically gets
10280	 * incremented when we make changes.
10281	 */
10282	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10283	    "revision")) == NULL) {
10284		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10285	} else {
10286		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10287		strncpy(inq_ptr->revision, val,
10288		    min(sizeof(inq_ptr->revision), strlen(val)));
10289	}
10290
10291	/*
10292	 * For parallel SCSI, we support double transition and single
10293	 * transition clocking.  We also support QAS (Quick Arbitration
10294	 * and Selection) and Information Unit transfers on both the
10295	 * control and array devices.
10296	 */
10297	if (is_fc == 0)
10298		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10299				    SID_SPI_IUS;
10300
10301	/* SAM-3 */
10302	scsi_ulto2b(0x0060, inq_ptr->version1);
10303	/* SPC-3 (no version claimed) XXX should we claim a version? */
10304	scsi_ulto2b(0x0300, inq_ptr->version2);
10305	if (is_fc) {
10306		/* FCP-2 ANSI INCITS.350:2003 */
10307		scsi_ulto2b(0x0917, inq_ptr->version3);
10308	} else {
10309		/* SPI-4 ANSI INCITS.362:200x */
10310		scsi_ulto2b(0x0B56, inq_ptr->version3);
10311	}
10312
10313	if (lun == NULL) {
10314		/* SBC-2 (no version claimed) XXX should we claim a version? */
10315		scsi_ulto2b(0x0320, inq_ptr->version4);
10316	} else {
10317		switch (lun->be_lun->lun_type) {
10318		case T_DIRECT:
10319			/*
10320			 * SBC-2 (no version claimed) XXX should we claim a
10321			 * version?
10322			 */
10323			scsi_ulto2b(0x0320, inq_ptr->version4);
10324			break;
10325		case T_PROCESSOR:
10326		default:
10327			break;
10328		}
10329	}
10330
10331	ctsio->scsi_status = SCSI_STATUS_OK;
10332	if (ctsio->kern_data_len > 0) {
10333		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10334		ctsio->be_move_done = ctl_config_move_done;
10335		ctl_datamove((union ctl_io *)ctsio);
10336	} else {
10337		ctsio->io_hdr.status = CTL_SUCCESS;
10338		ctl_done((union ctl_io *)ctsio);
10339	}
10340
10341	return (CTL_RETVAL_COMPLETE);
10342}
10343
10344int
10345ctl_inquiry(struct ctl_scsiio *ctsio)
10346{
10347	struct scsi_inquiry *cdb;
10348	int retval;
10349
10350	cdb = (struct scsi_inquiry *)ctsio->cdb;
10351
10352	retval = 0;
10353
10354	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10355
10356	/*
10357	 * Right now, we don't support the CmdDt inquiry information.
10358	 * This would be nice to support in the future.  When we do
10359	 * support it, we should change this test so that it checks to make
10360	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10361	 */
10362#ifdef notyet
10363	if (((cdb->byte2 & SI_EVPD)
10364	 && (cdb->byte2 & SI_CMDDT)))
10365#endif
10366	if (cdb->byte2 & SI_CMDDT) {
10367		/*
10368		 * Point to the SI_CMDDT bit.  We might change this
10369		 * when we support SI_CMDDT, but since both bits would be
10370		 * "wrong", this should probably just stay as-is then.
10371		 */
10372		ctl_set_invalid_field(ctsio,
10373				      /*sks_valid*/ 1,
10374				      /*command*/ 1,
10375				      /*field*/ 1,
10376				      /*bit_valid*/ 1,
10377				      /*bit*/ 1);
10378		ctl_done((union ctl_io *)ctsio);
10379		return (CTL_RETVAL_COMPLETE);
10380	}
10381	if (cdb->byte2 & SI_EVPD)
10382		retval = ctl_inquiry_evpd(ctsio);
10383#ifdef notyet
10384	else if (cdb->byte2 & SI_CMDDT)
10385		retval = ctl_inquiry_cmddt(ctsio);
10386#endif
10387	else
10388		retval = ctl_inquiry_std(ctsio);
10389
10390	return (retval);
10391}
10392
10393/*
10394 * For known CDB types, parse the LBA and length.
10395 */
10396static int
10397ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10398{
10399	if (io->io_hdr.io_type != CTL_IO_SCSI)
10400		return (1);
10401
10402	switch (io->scsiio.cdb[0]) {
10403	case COMPARE_AND_WRITE: {
10404		struct scsi_compare_and_write *cdb;
10405
10406		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10407
10408		*lba = scsi_8btou64(cdb->addr);
10409		*len = cdb->length;
10410		break;
10411	}
10412	case READ_6:
10413	case WRITE_6: {
10414		struct scsi_rw_6 *cdb;
10415
10416		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10417
10418		*lba = scsi_3btoul(cdb->addr);
10419		/* only 5 bits are valid in the most significant address byte */
10420		*lba &= 0x1fffff;
10421		*len = cdb->length;
10422		break;
10423	}
10424	case READ_10:
10425	case WRITE_10: {
10426		struct scsi_rw_10 *cdb;
10427
10428		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10429
10430		*lba = scsi_4btoul(cdb->addr);
10431		*len = scsi_2btoul(cdb->length);
10432		break;
10433	}
10434	case WRITE_VERIFY_10: {
10435		struct scsi_write_verify_10 *cdb;
10436
10437		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10438
10439		*lba = scsi_4btoul(cdb->addr);
10440		*len = scsi_2btoul(cdb->length);
10441		break;
10442	}
10443	case READ_12:
10444	case WRITE_12: {
10445		struct scsi_rw_12 *cdb;
10446
10447		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10448
10449		*lba = scsi_4btoul(cdb->addr);
10450		*len = scsi_4btoul(cdb->length);
10451		break;
10452	}
10453	case WRITE_VERIFY_12: {
10454		struct scsi_write_verify_12 *cdb;
10455
10456		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10457
10458		*lba = scsi_4btoul(cdb->addr);
10459		*len = scsi_4btoul(cdb->length);
10460		break;
10461	}
10462	case READ_16:
10463	case WRITE_16: {
10464		struct scsi_rw_16 *cdb;
10465
10466		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10467
10468		*lba = scsi_8btou64(cdb->addr);
10469		*len = scsi_4btoul(cdb->length);
10470		break;
10471	}
10472	case WRITE_VERIFY_16: {
10473		struct scsi_write_verify_16 *cdb;
10474
10475		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10476
10477
10478		*lba = scsi_8btou64(cdb->addr);
10479		*len = scsi_4btoul(cdb->length);
10480		break;
10481	}
10482	case WRITE_SAME_10: {
10483		struct scsi_write_same_10 *cdb;
10484
10485		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10486
10487		*lba = scsi_4btoul(cdb->addr);
10488		*len = scsi_2btoul(cdb->length);
10489		break;
10490	}
10491	case WRITE_SAME_16: {
10492		struct scsi_write_same_16 *cdb;
10493
10494		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10495
10496		*lba = scsi_8btou64(cdb->addr);
10497		*len = scsi_4btoul(cdb->length);
10498		break;
10499	}
10500	case VERIFY_10: {
10501		struct scsi_verify_10 *cdb;
10502
10503		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10504
10505		*lba = scsi_4btoul(cdb->addr);
10506		*len = scsi_2btoul(cdb->length);
10507		break;
10508	}
10509	case VERIFY_12: {
10510		struct scsi_verify_12 *cdb;
10511
10512		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10513
10514		*lba = scsi_4btoul(cdb->addr);
10515		*len = scsi_4btoul(cdb->length);
10516		break;
10517	}
10518	case VERIFY_16: {
10519		struct scsi_verify_16 *cdb;
10520
10521		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10522
10523		*lba = scsi_8btou64(cdb->addr);
10524		*len = scsi_4btoul(cdb->length);
10525		break;
10526	}
10527	default:
10528		return (1);
10529		break; /* NOTREACHED */
10530	}
10531
10532	return (0);
10533}
10534
10535static ctl_action
10536ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10537{
10538	uint64_t endlba1, endlba2;
10539
10540	endlba1 = lba1 + len1 - 1;
10541	endlba2 = lba2 + len2 - 1;
10542
10543	if ((endlba1 < lba2)
10544	 || (endlba2 < lba1))
10545		return (CTL_ACTION_PASS);
10546	else
10547		return (CTL_ACTION_BLOCK);
10548}
10549
10550static ctl_action
10551ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10552{
10553	uint64_t lba1, lba2;
10554	uint32_t len1, len2;
10555	int retval;
10556
10557	retval = ctl_get_lba_len(io1, &lba1, &len1);
10558	if (retval != 0)
10559		return (CTL_ACTION_ERROR);
10560
10561	retval = ctl_get_lba_len(io2, &lba2, &len2);
10562	if (retval != 0)
10563		return (CTL_ACTION_ERROR);
10564
10565	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10566}
10567
10568static ctl_action
10569ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10570{
10571	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10572	ctl_serialize_action *serialize_row;
10573
10574	/*
10575	 * The initiator attempted multiple untagged commands at the same
10576	 * time.  Can't do that.
10577	 */
10578	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10579	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10580	 && ((pending_io->io_hdr.nexus.targ_port ==
10581	      ooa_io->io_hdr.nexus.targ_port)
10582	  && (pending_io->io_hdr.nexus.initid.id ==
10583	      ooa_io->io_hdr.nexus.initid.id))
10584	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10585		return (CTL_ACTION_OVERLAP);
10586
10587	/*
10588	 * The initiator attempted to send multiple tagged commands with
10589	 * the same ID.  (It's fine if different initiators have the same
10590	 * tag ID.)
10591	 *
10592	 * Even if all of those conditions are true, we don't kill the I/O
10593	 * if the command ahead of us has been aborted.  We won't end up
10594	 * sending it to the FETD, and it's perfectly legal to resend a
10595	 * command with the same tag number as long as the previous
10596	 * instance of this tag number has been aborted somehow.
10597	 */
10598	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10599	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10600	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10601	 && ((pending_io->io_hdr.nexus.targ_port ==
10602	      ooa_io->io_hdr.nexus.targ_port)
10603	  && (pending_io->io_hdr.nexus.initid.id ==
10604	      ooa_io->io_hdr.nexus.initid.id))
10605	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10606		return (CTL_ACTION_OVERLAP_TAG);
10607
10608	/*
10609	 * If we get a head of queue tag, SAM-3 says that we should
10610	 * immediately execute it.
10611	 *
10612	 * What happens if this command would normally block for some other
10613	 * reason?  e.g. a request sense with a head of queue tag
10614	 * immediately after a write.  Normally that would block, but this
10615	 * will result in its getting executed immediately...
10616	 *
10617	 * We currently return "pass" instead of "skip", so we'll end up
10618	 * going through the rest of the queue to check for overlapped tags.
10619	 *
10620	 * XXX KDM check for other types of blockage first??
10621	 */
10622	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10623		return (CTL_ACTION_PASS);
10624
10625	/*
10626	 * Ordered tags have to block until all items ahead of them
10627	 * have completed.  If we get called with an ordered tag, we always
10628	 * block, if something else is ahead of us in the queue.
10629	 */
10630	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10631		return (CTL_ACTION_BLOCK);
10632
10633	/*
10634	 * Simple tags get blocked until all head of queue and ordered tags
10635	 * ahead of them have completed.  I'm lumping untagged commands in
10636	 * with simple tags here.  XXX KDM is that the right thing to do?
10637	 */
10638	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10639	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10640	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10641	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10642		return (CTL_ACTION_BLOCK);
10643
10644	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio);
10645	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio);
10646
10647	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10648
10649	switch (serialize_row[pending_entry->seridx]) {
10650	case CTL_SER_BLOCK:
10651		return (CTL_ACTION_BLOCK);
10652		break; /* NOTREACHED */
10653	case CTL_SER_EXTENT:
10654		return (ctl_extent_check(pending_io, ooa_io));
10655		break; /* NOTREACHED */
10656	case CTL_SER_PASS:
10657		return (CTL_ACTION_PASS);
10658		break; /* NOTREACHED */
10659	case CTL_SER_SKIP:
10660		return (CTL_ACTION_SKIP);
10661		break;
10662	default:
10663		panic("invalid serialization value %d",
10664		      serialize_row[pending_entry->seridx]);
10665		break; /* NOTREACHED */
10666	}
10667
10668	return (CTL_ACTION_ERROR);
10669}
10670
10671/*
10672 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10673 * Assumptions:
10674 * - pending_io is generally either incoming, or on the blocked queue
10675 * - starting I/O is the I/O we want to start the check with.
10676 */
10677static ctl_action
10678ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10679	      union ctl_io *starting_io)
10680{
10681	union ctl_io *ooa_io;
10682	ctl_action action;
10683
10684	mtx_assert(&lun->lun_lock, MA_OWNED);
10685
10686	/*
10687	 * Run back along the OOA queue, starting with the current
10688	 * blocked I/O and going through every I/O before it on the
10689	 * queue.  If starting_io is NULL, we'll just end up returning
10690	 * CTL_ACTION_PASS.
10691	 */
10692	for (ooa_io = starting_io; ooa_io != NULL;
10693	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10694	     ooa_links)){
10695
10696		/*
10697		 * This routine just checks to see whether
10698		 * cur_blocked is blocked by ooa_io, which is ahead
10699		 * of it in the queue.  It doesn't queue/dequeue
10700		 * cur_blocked.
10701		 */
10702		action = ctl_check_for_blockage(pending_io, ooa_io);
10703		switch (action) {
10704		case CTL_ACTION_BLOCK:
10705		case CTL_ACTION_OVERLAP:
10706		case CTL_ACTION_OVERLAP_TAG:
10707		case CTL_ACTION_SKIP:
10708		case CTL_ACTION_ERROR:
10709			return (action);
10710			break; /* NOTREACHED */
10711		case CTL_ACTION_PASS:
10712			break;
10713		default:
10714			panic("invalid action %d", action);
10715			break;  /* NOTREACHED */
10716		}
10717	}
10718
10719	return (CTL_ACTION_PASS);
10720}
10721
10722/*
10723 * Assumptions:
10724 * - An I/O has just completed, and has been removed from the per-LUN OOA
10725 *   queue, so some items on the blocked queue may now be unblocked.
10726 */
10727static int
10728ctl_check_blocked(struct ctl_lun *lun)
10729{
10730	union ctl_io *cur_blocked, *next_blocked;
10731
10732	mtx_assert(&lun->lun_lock, MA_OWNED);
10733
10734	/*
10735	 * Run forward from the head of the blocked queue, checking each
10736	 * entry against the I/Os prior to it on the OOA queue to see if
10737	 * there is still any blockage.
10738	 *
10739	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10740	 * with our removing a variable on it while it is traversing the
10741	 * list.
10742	 */
10743	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10744	     cur_blocked != NULL; cur_blocked = next_blocked) {
10745		union ctl_io *prev_ooa;
10746		ctl_action action;
10747
10748		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10749							  blocked_links);
10750
10751		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10752						      ctl_ooaq, ooa_links);
10753
10754		/*
10755		 * If cur_blocked happens to be the first item in the OOA
10756		 * queue now, prev_ooa will be NULL, and the action
10757		 * returned will just be CTL_ACTION_PASS.
10758		 */
10759		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10760
10761		switch (action) {
10762		case CTL_ACTION_BLOCK:
10763			/* Nothing to do here, still blocked */
10764			break;
10765		case CTL_ACTION_OVERLAP:
10766		case CTL_ACTION_OVERLAP_TAG:
10767			/*
10768			 * This shouldn't happen!  In theory we've already
10769			 * checked this command for overlap...
10770			 */
10771			break;
10772		case CTL_ACTION_PASS:
10773		case CTL_ACTION_SKIP: {
10774			struct ctl_softc *softc;
10775			const struct ctl_cmd_entry *entry;
10776			uint32_t initidx;
10777			int isc_retval;
10778
10779			/*
10780			 * The skip case shouldn't happen, this transaction
10781			 * should have never made it onto the blocked queue.
10782			 */
10783			/*
10784			 * This I/O is no longer blocked, we can remove it
10785			 * from the blocked queue.  Since this is a TAILQ
10786			 * (doubly linked list), we can do O(1) removals
10787			 * from any place on the list.
10788			 */
10789			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10790				     blocked_links);
10791			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10792
10793			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10794				/*
10795				 * Need to send IO back to original side to
10796				 * run
10797				 */
10798				union ctl_ha_msg msg_info;
10799
10800				msg_info.hdr.original_sc =
10801					cur_blocked->io_hdr.original_sc;
10802				msg_info.hdr.serializing_sc = cur_blocked;
10803				msg_info.hdr.msg_type = CTL_MSG_R2R;
10804				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10805				     &msg_info, sizeof(msg_info), 0)) >
10806				     CTL_HA_STATUS_SUCCESS) {
10807					printf("CTL:Check Blocked error from "
10808					       "ctl_ha_msg_send %d\n",
10809					       isc_retval);
10810				}
10811				break;
10812			}
10813			entry = ctl_get_cmd_entry(&cur_blocked->scsiio);
10814			softc = control_softc;
10815
10816			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
10817
10818			/*
10819			 * Check this I/O for LUN state changes that may
10820			 * have happened while this command was blocked.
10821			 * The LUN state may have been changed by a command
10822			 * ahead of us in the queue, so we need to re-check
10823			 * for any states that can be caused by SCSI
10824			 * commands.
10825			 */
10826			if (ctl_scsiio_lun_check(softc, lun, entry,
10827						 &cur_blocked->scsiio) == 0) {
10828				cur_blocked->io_hdr.flags |=
10829				                      CTL_FLAG_IS_WAS_ON_RTR;
10830				ctl_enqueue_rtr(cur_blocked);
10831			} else
10832				ctl_done(cur_blocked);
10833			break;
10834		}
10835		default:
10836			/*
10837			 * This probably shouldn't happen -- we shouldn't
10838			 * get CTL_ACTION_ERROR, or anything else.
10839			 */
10840			break;
10841		}
10842	}
10843
10844	return (CTL_RETVAL_COMPLETE);
10845}
10846
10847/*
10848 * This routine (with one exception) checks LUN flags that can be set by
10849 * commands ahead of us in the OOA queue.  These flags have to be checked
10850 * when a command initially comes in, and when we pull a command off the
10851 * blocked queue and are preparing to execute it.  The reason we have to
10852 * check these flags for commands on the blocked queue is that the LUN
10853 * state may have been changed by a command ahead of us while we're on the
10854 * blocked queue.
10855 *
10856 * Ordering is somewhat important with these checks, so please pay
10857 * careful attention to the placement of any new checks.
10858 */
10859static int
10860ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
10861    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10862{
10863	int retval;
10864
10865	retval = 0;
10866
10867	mtx_assert(&lun->lun_lock, MA_OWNED);
10868
10869	/*
10870	 * If this shelf is a secondary shelf controller, we have to reject
10871	 * any media access commands.
10872	 */
10873#if 0
10874	/* No longer needed for HA */
10875	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10876	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10877		ctl_set_lun_standby(ctsio);
10878		retval = 1;
10879		goto bailout;
10880	}
10881#endif
10882
10883	/*
10884	 * Check for a reservation conflict.  If this command isn't allowed
10885	 * even on reserved LUNs, and if this initiator isn't the one who
10886	 * reserved us, reject the command with a reservation conflict.
10887	 */
10888	if ((lun->flags & CTL_LUN_RESERVED)
10889	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10890		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10891		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10892		 || (ctsio->io_hdr.nexus.targ_target.id !=
10893		     lun->rsv_nexus.targ_target.id)) {
10894			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10895			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10896			retval = 1;
10897			goto bailout;
10898		}
10899	}
10900
10901	if ( (lun->flags & CTL_LUN_PR_RESERVED)
10902	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10903		uint32_t residx;
10904
10905		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10906		/*
10907		 * if we aren't registered or it's a res holder type
10908		 * reservation and this isn't the res holder then set a
10909		 * conflict.
10910		 * NOTE: Commands which might be allowed on write exclusive
10911		 * type reservations are checked in the particular command
10912		 * for a conflict. Read and SSU are the only ones.
10913		 */
10914		if (!lun->per_res[residx].registered
10915		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10916			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10917			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10918			retval = 1;
10919			goto bailout;
10920		}
10921
10922	}
10923
10924	if ((lun->flags & CTL_LUN_OFFLINE)
10925	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10926		ctl_set_lun_not_ready(ctsio);
10927		retval = 1;
10928		goto bailout;
10929	}
10930
10931	/*
10932	 * If the LUN is stopped, see if this particular command is allowed
10933	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10934	 */
10935	if ((lun->flags & CTL_LUN_STOPPED)
10936	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10937		/* "Logical unit not ready, initializing cmd. required" */
10938		ctl_set_lun_stopped(ctsio);
10939		retval = 1;
10940		goto bailout;
10941	}
10942
10943	if ((lun->flags & CTL_LUN_INOPERABLE)
10944	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10945		/* "Medium format corrupted" */
10946		ctl_set_medium_format_corrupted(ctsio);
10947		retval = 1;
10948		goto bailout;
10949	}
10950
10951bailout:
10952	return (retval);
10953
10954}
10955
10956static void
10957ctl_failover_io(union ctl_io *io, int have_lock)
10958{
10959	ctl_set_busy(&io->scsiio);
10960	ctl_done(io);
10961}
10962
10963static void
10964ctl_failover(void)
10965{
10966	struct ctl_lun *lun;
10967	struct ctl_softc *ctl_softc;
10968	union ctl_io *next_io, *pending_io;
10969	union ctl_io *io;
10970	int lun_idx;
10971	int i;
10972
10973	ctl_softc = control_softc;
10974
10975	mtx_lock(&ctl_softc->ctl_lock);
10976	/*
10977	 * Remove any cmds from the other SC from the rtr queue.  These
10978	 * will obviously only be for LUNs for which we're the primary.
10979	 * We can't send status or get/send data for these commands.
10980	 * Since they haven't been executed yet, we can just remove them.
10981	 * We'll either abort them or delete them below, depending on
10982	 * which HA mode we're in.
10983	 */
10984#ifdef notyet
10985	mtx_lock(&ctl_softc->queue_lock);
10986	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10987	     io != NULL; io = next_io) {
10988		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10989		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10990			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10991				      ctl_io_hdr, links);
10992	}
10993	mtx_unlock(&ctl_softc->queue_lock);
10994#endif
10995
10996	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10997		lun = ctl_softc->ctl_luns[lun_idx];
10998		if (lun==NULL)
10999			continue;
11000
11001		/*
11002		 * Processor LUNs are primary on both sides.
11003		 * XXX will this always be true?
11004		 */
11005		if (lun->be_lun->lun_type == T_PROCESSOR)
11006			continue;
11007
11008		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11009		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11010			printf("FAILOVER: primary lun %d\n", lun_idx);
11011		        /*
11012			 * Remove all commands from the other SC. First from the
11013			 * blocked queue then from the ooa queue. Once we have
11014			 * removed them. Call ctl_check_blocked to see if there
11015			 * is anything that can run.
11016			 */
11017			for (io = (union ctl_io *)TAILQ_FIRST(
11018			     &lun->blocked_queue); io != NULL; io = next_io) {
11019
11020		        	next_io = (union ctl_io *)TAILQ_NEXT(
11021				    &io->io_hdr, blocked_links);
11022
11023				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11024					TAILQ_REMOVE(&lun->blocked_queue,
11025						     &io->io_hdr,blocked_links);
11026					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11027					TAILQ_REMOVE(&lun->ooa_queue,
11028						     &io->io_hdr, ooa_links);
11029
11030					ctl_free_io(io);
11031				}
11032			}
11033
11034			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11035	     		     io != NULL; io = next_io) {
11036
11037		        	next_io = (union ctl_io *)TAILQ_NEXT(
11038				    &io->io_hdr, ooa_links);
11039
11040				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11041
11042					TAILQ_REMOVE(&lun->ooa_queue,
11043						&io->io_hdr,
11044					     	ooa_links);
11045
11046					ctl_free_io(io);
11047				}
11048			}
11049			ctl_check_blocked(lun);
11050		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11051			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11052
11053			printf("FAILOVER: primary lun %d\n", lun_idx);
11054			/*
11055			 * Abort all commands from the other SC.  We can't
11056			 * send status back for them now.  These should get
11057			 * cleaned up when they are completed or come out
11058			 * for a datamove operation.
11059			 */
11060			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11061	     		     io != NULL; io = next_io) {
11062		        	next_io = (union ctl_io *)TAILQ_NEXT(
11063					&io->io_hdr, ooa_links);
11064
11065				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11066					io->io_hdr.flags |= CTL_FLAG_ABORT;
11067			}
11068		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11069			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11070
11071			printf("FAILOVER: secondary lun %d\n", lun_idx);
11072
11073			lun->flags |= CTL_LUN_PRIMARY_SC;
11074
11075			/*
11076			 * We send all I/O that was sent to this controller
11077			 * and redirected to the other side back with
11078			 * busy status, and have the initiator retry it.
11079			 * Figuring out how much data has been transferred,
11080			 * etc. and picking up where we left off would be
11081			 * very tricky.
11082			 *
11083			 * XXX KDM need to remove I/O from the blocked
11084			 * queue as well!
11085			 */
11086			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11087			     &lun->ooa_queue); pending_io != NULL;
11088			     pending_io = next_io) {
11089
11090				next_io =  (union ctl_io *)TAILQ_NEXT(
11091					&pending_io->io_hdr, ooa_links);
11092
11093				pending_io->io_hdr.flags &=
11094					~CTL_FLAG_SENT_2OTHER_SC;
11095
11096				if (pending_io->io_hdr.flags &
11097				    CTL_FLAG_IO_ACTIVE) {
11098					pending_io->io_hdr.flags |=
11099						CTL_FLAG_FAILOVER;
11100				} else {
11101					ctl_set_busy(&pending_io->scsiio);
11102					ctl_done(pending_io);
11103				}
11104			}
11105
11106			/*
11107			 * Build Unit Attention
11108			 */
11109			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11110				lun->pending_sense[i].ua_pending |=
11111				                     CTL_UA_ASYM_ACC_CHANGE;
11112			}
11113		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11114			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11115			printf("FAILOVER: secondary lun %d\n", lun_idx);
11116			/*
11117			 * if the first io on the OOA is not on the RtR queue
11118			 * add it.
11119			 */
11120			lun->flags |= CTL_LUN_PRIMARY_SC;
11121
11122			pending_io = (union ctl_io *)TAILQ_FIRST(
11123			    &lun->ooa_queue);
11124			if (pending_io==NULL) {
11125				printf("Nothing on OOA queue\n");
11126				continue;
11127			}
11128
11129			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11130			if ((pending_io->io_hdr.flags &
11131			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11132				pending_io->io_hdr.flags |=
11133				    CTL_FLAG_IS_WAS_ON_RTR;
11134				ctl_enqueue_rtr(pending_io);
11135			}
11136#if 0
11137			else
11138			{
11139				printf("Tag 0x%04x is running\n",
11140				      pending_io->scsiio.tag_num);
11141			}
11142#endif
11143
11144			next_io = (union ctl_io *)TAILQ_NEXT(
11145			    &pending_io->io_hdr, ooa_links);
11146			for (pending_io=next_io; pending_io != NULL;
11147			     pending_io = next_io) {
11148				pending_io->io_hdr.flags &=
11149				    ~CTL_FLAG_SENT_2OTHER_SC;
11150				next_io = (union ctl_io *)TAILQ_NEXT(
11151					&pending_io->io_hdr, ooa_links);
11152				if (pending_io->io_hdr.flags &
11153				    CTL_FLAG_IS_WAS_ON_RTR) {
11154#if 0
11155				        printf("Tag 0x%04x is running\n",
11156				      		pending_io->scsiio.tag_num);
11157#endif
11158					continue;
11159				}
11160
11161				switch (ctl_check_ooa(lun, pending_io,
11162			            (union ctl_io *)TAILQ_PREV(
11163				    &pending_io->io_hdr, ctl_ooaq,
11164				    ooa_links))) {
11165
11166				case CTL_ACTION_BLOCK:
11167					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11168							  &pending_io->io_hdr,
11169							  blocked_links);
11170					pending_io->io_hdr.flags |=
11171					    CTL_FLAG_BLOCKED;
11172					break;
11173				case CTL_ACTION_PASS:
11174				case CTL_ACTION_SKIP:
11175					pending_io->io_hdr.flags |=
11176					    CTL_FLAG_IS_WAS_ON_RTR;
11177					ctl_enqueue_rtr(pending_io);
11178					break;
11179				case CTL_ACTION_OVERLAP:
11180					ctl_set_overlapped_cmd(
11181					    (struct ctl_scsiio *)pending_io);
11182					ctl_done(pending_io);
11183					break;
11184				case CTL_ACTION_OVERLAP_TAG:
11185					ctl_set_overlapped_tag(
11186					    (struct ctl_scsiio *)pending_io,
11187					    pending_io->scsiio.tag_num & 0xff);
11188					ctl_done(pending_io);
11189					break;
11190				case CTL_ACTION_ERROR:
11191				default:
11192					ctl_set_internal_failure(
11193						(struct ctl_scsiio *)pending_io,
11194						0,  // sks_valid
11195						0); //retry count
11196					ctl_done(pending_io);
11197					break;
11198				}
11199			}
11200
11201			/*
11202			 * Build Unit Attention
11203			 */
11204			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11205				lun->pending_sense[i].ua_pending |=
11206				                     CTL_UA_ASYM_ACC_CHANGE;
11207			}
11208		} else {
11209			panic("Unhandled HA mode failover, LUN flags = %#x, "
11210			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11211		}
11212	}
11213	ctl_pause_rtr = 0;
11214	mtx_unlock(&ctl_softc->ctl_lock);
11215}
11216
11217static int
11218ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11219{
11220	struct ctl_lun *lun;
11221	const struct ctl_cmd_entry *entry;
11222	uint32_t initidx, targ_lun;
11223	int retval;
11224
11225	retval = 0;
11226
11227	lun = NULL;
11228
11229	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11230	if ((targ_lun < CTL_MAX_LUNS)
11231	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11232		lun = ctl_softc->ctl_luns[targ_lun];
11233		/*
11234		 * If the LUN is invalid, pretend that it doesn't exist.
11235		 * It will go away as soon as all pending I/O has been
11236		 * completed.
11237		 */
11238		if (lun->flags & CTL_LUN_DISABLED) {
11239			lun = NULL;
11240		} else {
11241			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11242			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11243				lun->be_lun;
11244			if (lun->be_lun->lun_type == T_PROCESSOR) {
11245				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11246			}
11247
11248			/*
11249			 * Every I/O goes into the OOA queue for a
11250			 * particular LUN, and stays there until completion.
11251			 */
11252			mtx_lock(&lun->lun_lock);
11253			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11254			    ooa_links);
11255		}
11256	} else {
11257		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11258		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11259	}
11260
11261	/* Get command entry and return error if it is unsuppotyed. */
11262	entry = ctl_validate_command(ctsio);
11263	if (entry == NULL) {
11264		if (lun)
11265			mtx_unlock(&lun->lun_lock);
11266		return (retval);
11267	}
11268
11269	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11270	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11271
11272	/*
11273	 * Check to see whether we can send this command to LUNs that don't
11274	 * exist.  This should pretty much only be the case for inquiry
11275	 * and request sense.  Further checks, below, really require having
11276	 * a LUN, so we can't really check the command anymore.  Just put
11277	 * it on the rtr queue.
11278	 */
11279	if (lun == NULL) {
11280		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11281			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11282			ctl_enqueue_rtr((union ctl_io *)ctsio);
11283			return (retval);
11284		}
11285
11286		ctl_set_unsupported_lun(ctsio);
11287		ctl_done((union ctl_io *)ctsio);
11288		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11289		return (retval);
11290	} else {
11291		/*
11292		 * Make sure we support this particular command on this LUN.
11293		 * e.g., we don't support writes to the control LUN.
11294		 */
11295		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11296			mtx_unlock(&lun->lun_lock);
11297			ctl_set_invalid_opcode(ctsio);
11298			ctl_done((union ctl_io *)ctsio);
11299			return (retval);
11300		}
11301	}
11302
11303	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11304
11305	/*
11306	 * If we've got a request sense, it'll clear the contingent
11307	 * allegiance condition.  Otherwise, if we have a CA condition for
11308	 * this initiator, clear it, because it sent down a command other
11309	 * than request sense.
11310	 */
11311	if ((ctsio->cdb[0] != REQUEST_SENSE)
11312	 && (ctl_is_set(lun->have_ca, initidx)))
11313		ctl_clear_mask(lun->have_ca, initidx);
11314
11315	/*
11316	 * If the command has this flag set, it handles its own unit
11317	 * attention reporting, we shouldn't do anything.  Otherwise we
11318	 * check for any pending unit attentions, and send them back to the
11319	 * initiator.  We only do this when a command initially comes in,
11320	 * not when we pull it off the blocked queue.
11321	 *
11322	 * According to SAM-3, section 5.3.2, the order that things get
11323	 * presented back to the host is basically unit attentions caused
11324	 * by some sort of reset event, busy status, reservation conflicts
11325	 * or task set full, and finally any other status.
11326	 *
11327	 * One issue here is that some of the unit attentions we report
11328	 * don't fall into the "reset" category (e.g. "reported luns data
11329	 * has changed").  So reporting it here, before the reservation
11330	 * check, may be technically wrong.  I guess the only thing to do
11331	 * would be to check for and report the reset events here, and then
11332	 * check for the other unit attention types after we check for a
11333	 * reservation conflict.
11334	 *
11335	 * XXX KDM need to fix this
11336	 */
11337	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11338		ctl_ua_type ua_type;
11339
11340		ua_type = lun->pending_sense[initidx].ua_pending;
11341		if (ua_type != CTL_UA_NONE) {
11342			scsi_sense_data_type sense_format;
11343
11344			if (lun != NULL)
11345				sense_format = (lun->flags &
11346				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11347				    SSD_TYPE_FIXED;
11348			else
11349				sense_format = SSD_TYPE_FIXED;
11350
11351			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11352					       sense_format);
11353			if (ua_type != CTL_UA_NONE) {
11354				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11355				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11356						       CTL_AUTOSENSE;
11357				ctsio->sense_len = SSD_FULL_SIZE;
11358				lun->pending_sense[initidx].ua_pending &=
11359					~ua_type;
11360				mtx_unlock(&lun->lun_lock);
11361				ctl_done((union ctl_io *)ctsio);
11362				return (retval);
11363			}
11364		}
11365	}
11366
11367
11368	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11369		mtx_unlock(&lun->lun_lock);
11370		ctl_done((union ctl_io *)ctsio);
11371		return (retval);
11372	}
11373
11374	/*
11375	 * XXX CHD this is where we want to send IO to other side if
11376	 * this LUN is secondary on this SC. We will need to make a copy
11377	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11378	 * the copy we send as FROM_OTHER.
11379	 * We also need to stuff the address of the original IO so we can
11380	 * find it easily. Something similar will need be done on the other
11381	 * side so when we are done we can find the copy.
11382	 */
11383	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11384		union ctl_ha_msg msg_info;
11385		int isc_retval;
11386
11387		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11388
11389		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11390		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11391#if 0
11392		printf("1. ctsio %p\n", ctsio);
11393#endif
11394		msg_info.hdr.serializing_sc = NULL;
11395		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11396		msg_info.scsi.tag_num = ctsio->tag_num;
11397		msg_info.scsi.tag_type = ctsio->tag_type;
11398		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11399
11400		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11401
11402		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11403		    (void *)&msg_info, sizeof(msg_info), 0)) >
11404		    CTL_HA_STATUS_SUCCESS) {
11405			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11406			       isc_retval);
11407			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11408		} else {
11409#if 0
11410			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11411#endif
11412		}
11413
11414		/*
11415		 * XXX KDM this I/O is off the incoming queue, but hasn't
11416		 * been inserted on any other queue.  We may need to come
11417		 * up with a holding queue while we wait for serialization
11418		 * so that we have an idea of what we're waiting for from
11419		 * the other side.
11420		 */
11421		mtx_unlock(&lun->lun_lock);
11422		return (retval);
11423	}
11424
11425	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11426			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11427			      ctl_ooaq, ooa_links))) {
11428	case CTL_ACTION_BLOCK:
11429		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11430		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11431				  blocked_links);
11432		mtx_unlock(&lun->lun_lock);
11433		return (retval);
11434	case CTL_ACTION_PASS:
11435	case CTL_ACTION_SKIP:
11436		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11437		mtx_unlock(&lun->lun_lock);
11438		ctl_enqueue_rtr((union ctl_io *)ctsio);
11439		break;
11440	case CTL_ACTION_OVERLAP:
11441		mtx_unlock(&lun->lun_lock);
11442		ctl_set_overlapped_cmd(ctsio);
11443		ctl_done((union ctl_io *)ctsio);
11444		break;
11445	case CTL_ACTION_OVERLAP_TAG:
11446		mtx_unlock(&lun->lun_lock);
11447		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11448		ctl_done((union ctl_io *)ctsio);
11449		break;
11450	case CTL_ACTION_ERROR:
11451	default:
11452		mtx_unlock(&lun->lun_lock);
11453		ctl_set_internal_failure(ctsio,
11454					 /*sks_valid*/ 0,
11455					 /*retry_count*/ 0);
11456		ctl_done((union ctl_io *)ctsio);
11457		break;
11458	}
11459	return (retval);
11460}
11461
11462const struct ctl_cmd_entry *
11463ctl_get_cmd_entry(struct ctl_scsiio *ctsio)
11464{
11465	const struct ctl_cmd_entry *entry;
11466	int service_action;
11467
11468	entry = &ctl_cmd_table[ctsio->cdb[0]];
11469	if (entry->flags & CTL_CMD_FLAG_SA5) {
11470		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11471		entry = &((const struct ctl_cmd_entry *)
11472		    entry->execute)[service_action];
11473	}
11474	return (entry);
11475}
11476
11477const struct ctl_cmd_entry *
11478ctl_validate_command(struct ctl_scsiio *ctsio)
11479{
11480	const struct ctl_cmd_entry *entry;
11481	int i;
11482	uint8_t diff;
11483
11484	entry = ctl_get_cmd_entry(ctsio);
11485	if (entry->execute == NULL) {
11486		ctl_set_invalid_opcode(ctsio);
11487		ctl_done((union ctl_io *)ctsio);
11488		return (NULL);
11489	}
11490	KASSERT(entry->length > 0,
11491	    ("Not defined length for command 0x%02x/0x%02x",
11492	     ctsio->cdb[0], ctsio->cdb[1]));
11493	for (i = 1; i < entry->length; i++) {
11494		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11495		if (diff == 0)
11496			continue;
11497		ctl_set_invalid_field(ctsio,
11498				      /*sks_valid*/ 1,
11499				      /*command*/ 1,
11500				      /*field*/ i,
11501				      /*bit_valid*/ 1,
11502				      /*bit*/ fls(diff) - 1);
11503		ctl_done((union ctl_io *)ctsio);
11504		return (NULL);
11505	}
11506	return (entry);
11507}
11508
11509static int
11510ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11511{
11512
11513	switch (lun_type) {
11514	case T_PROCESSOR:
11515		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11516		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11517			return (0);
11518		break;
11519	case T_DIRECT:
11520		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11521		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11522			return (0);
11523		break;
11524	default:
11525		return (0);
11526	}
11527	return (1);
11528}
11529
11530static int
11531ctl_scsiio(struct ctl_scsiio *ctsio)
11532{
11533	int retval;
11534	const struct ctl_cmd_entry *entry;
11535
11536	retval = CTL_RETVAL_COMPLETE;
11537
11538	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11539
11540	entry = ctl_get_cmd_entry(ctsio);
11541
11542	/*
11543	 * If this I/O has been aborted, just send it straight to
11544	 * ctl_done() without executing it.
11545	 */
11546	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11547		ctl_done((union ctl_io *)ctsio);
11548		goto bailout;
11549	}
11550
11551	/*
11552	 * All the checks should have been handled by ctl_scsiio_precheck().
11553	 * We should be clear now to just execute the I/O.
11554	 */
11555	retval = entry->execute(ctsio);
11556
11557bailout:
11558	return (retval);
11559}
11560
11561/*
11562 * Since we only implement one target right now, a bus reset simply resets
11563 * our single target.
11564 */
11565static int
11566ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11567{
11568	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11569}
11570
11571static int
11572ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11573		 ctl_ua_type ua_type)
11574{
11575	struct ctl_lun *lun;
11576	int retval;
11577
11578	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11579		union ctl_ha_msg msg_info;
11580
11581		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11582		msg_info.hdr.nexus = io->io_hdr.nexus;
11583		if (ua_type==CTL_UA_TARG_RESET)
11584			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11585		else
11586			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11587		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11588		msg_info.hdr.original_sc = NULL;
11589		msg_info.hdr.serializing_sc = NULL;
11590		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11591		    (void *)&msg_info, sizeof(msg_info), 0)) {
11592		}
11593	}
11594	retval = 0;
11595
11596	mtx_lock(&ctl_softc->ctl_lock);
11597	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11598		retval += ctl_lun_reset(lun, io, ua_type);
11599	mtx_unlock(&ctl_softc->ctl_lock);
11600
11601	return (retval);
11602}
11603
11604/*
11605 * The LUN should always be set.  The I/O is optional, and is used to
11606 * distinguish between I/Os sent by this initiator, and by other
11607 * initiators.  We set unit attention for initiators other than this one.
11608 * SAM-3 is vague on this point.  It does say that a unit attention should
11609 * be established for other initiators when a LUN is reset (see section
11610 * 5.7.3), but it doesn't specifically say that the unit attention should
11611 * be established for this particular initiator when a LUN is reset.  Here
11612 * is the relevant text, from SAM-3 rev 8:
11613 *
11614 * 5.7.2 When a SCSI initiator port aborts its own tasks
11615 *
11616 * When a SCSI initiator port causes its own task(s) to be aborted, no
11617 * notification that the task(s) have been aborted shall be returned to
11618 * the SCSI initiator port other than the completion response for the
11619 * command or task management function action that caused the task(s) to
11620 * be aborted and notification(s) associated with related effects of the
11621 * action (e.g., a reset unit attention condition).
11622 *
11623 * XXX KDM for now, we're setting unit attention for all initiators.
11624 */
11625static int
11626ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11627{
11628	union ctl_io *xio;
11629#if 0
11630	uint32_t initindex;
11631#endif
11632	int i;
11633
11634	mtx_lock(&lun->lun_lock);
11635	/*
11636	 * Run through the OOA queue and abort each I/O.
11637	 */
11638#if 0
11639	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11640#endif
11641	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11642	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11643		xio->io_hdr.flags |= CTL_FLAG_ABORT;
11644	}
11645
11646	/*
11647	 * This version sets unit attention for every
11648	 */
11649#if 0
11650	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11651	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11652		if (initindex == i)
11653			continue;
11654		lun->pending_sense[i].ua_pending |= ua_type;
11655	}
11656#endif
11657
11658	/*
11659	 * A reset (any kind, really) clears reservations established with
11660	 * RESERVE/RELEASE.  It does not clear reservations established
11661	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11662	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11663	 * reservations made with the RESERVE/RELEASE commands, because
11664	 * those commands are obsolete in SPC-3.
11665	 */
11666	lun->flags &= ~CTL_LUN_RESERVED;
11667
11668	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11669		ctl_clear_mask(lun->have_ca, i);
11670		lun->pending_sense[i].ua_pending |= ua_type;
11671	}
11672	mtx_unlock(&lun->lun_lock);
11673
11674	return (0);
11675}
11676
11677static int
11678ctl_abort_task(union ctl_io *io)
11679{
11680	union ctl_io *xio;
11681	struct ctl_lun *lun;
11682	struct ctl_softc *ctl_softc;
11683#if 0
11684	struct sbuf sb;
11685	char printbuf[128];
11686#endif
11687	int found;
11688	uint32_t targ_lun;
11689
11690	ctl_softc = control_softc;
11691	found = 0;
11692
11693	/*
11694	 * Look up the LUN.
11695	 */
11696	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11697	mtx_lock(&ctl_softc->ctl_lock);
11698	if ((targ_lun < CTL_MAX_LUNS)
11699	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11700		lun = ctl_softc->ctl_luns[targ_lun];
11701	else {
11702		mtx_unlock(&ctl_softc->ctl_lock);
11703		goto bailout;
11704	}
11705
11706#if 0
11707	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11708	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11709#endif
11710
11711	mtx_lock(&lun->lun_lock);
11712	mtx_unlock(&ctl_softc->ctl_lock);
11713	/*
11714	 * Run through the OOA queue and attempt to find the given I/O.
11715	 * The target port, initiator ID, tag type and tag number have to
11716	 * match the values that we got from the initiator.  If we have an
11717	 * untagged command to abort, simply abort the first untagged command
11718	 * we come to.  We only allow one untagged command at a time of course.
11719	 */
11720#if 0
11721	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11722#endif
11723	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11724	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11725#if 0
11726		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11727
11728		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11729			    lun->lun, xio->scsiio.tag_num,
11730			    xio->scsiio.tag_type,
11731			    (xio->io_hdr.blocked_links.tqe_prev
11732			    == NULL) ? "" : " BLOCKED",
11733			    (xio->io_hdr.flags &
11734			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11735			    (xio->io_hdr.flags &
11736			    CTL_FLAG_ABORT) ? " ABORT" : "",
11737			    (xio->io_hdr.flags &
11738			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11739		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11740		sbuf_finish(&sb);
11741		printf("%s\n", sbuf_data(&sb));
11742#endif
11743
11744		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
11745		 && (xio->io_hdr.nexus.initid.id ==
11746		     io->io_hdr.nexus.initid.id)) {
11747			/*
11748			 * If the abort says that the task is untagged, the
11749			 * task in the queue must be untagged.  Otherwise,
11750			 * we just check to see whether the tag numbers
11751			 * match.  This is because the QLogic firmware
11752			 * doesn't pass back the tag type in an abort
11753			 * request.
11754			 */
11755#if 0
11756			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11757			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11758			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
11759#endif
11760			/*
11761			 * XXX KDM we've got problems with FC, because it
11762			 * doesn't send down a tag type with aborts.  So we
11763			 * can only really go by the tag number...
11764			 * This may cause problems with parallel SCSI.
11765			 * Need to figure that out!!
11766			 */
11767			if (xio->scsiio.tag_num == io->taskio.tag_num) {
11768				xio->io_hdr.flags |= CTL_FLAG_ABORT;
11769				found = 1;
11770				if ((io->io_hdr.flags &
11771				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
11772				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11773					union ctl_ha_msg msg_info;
11774
11775					io->io_hdr.flags |=
11776					                CTL_FLAG_SENT_2OTHER_SC;
11777					msg_info.hdr.nexus = io->io_hdr.nexus;
11778					msg_info.task.task_action =
11779						CTL_TASK_ABORT_TASK;
11780					msg_info.task.tag_num =
11781						io->taskio.tag_num;
11782					msg_info.task.tag_type =
11783						io->taskio.tag_type;
11784					msg_info.hdr.msg_type =
11785						CTL_MSG_MANAGE_TASKS;
11786					msg_info.hdr.original_sc = NULL;
11787					msg_info.hdr.serializing_sc = NULL;
11788#if 0
11789					printf("Sent Abort to other side\n");
11790#endif
11791					if (CTL_HA_STATUS_SUCCESS !=
11792					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11793		    				(void *)&msg_info,
11794						sizeof(msg_info), 0)) {
11795					}
11796				}
11797#if 0
11798				printf("ctl_abort_task: found I/O to abort\n");
11799#endif
11800				break;
11801			}
11802		}
11803	}
11804	mtx_unlock(&lun->lun_lock);
11805
11806bailout:
11807
11808	if (found == 0) {
11809		/*
11810		 * This isn't really an error.  It's entirely possible for
11811		 * the abort and command completion to cross on the wire.
11812		 * This is more of an informative/diagnostic error.
11813		 */
11814#if 0
11815		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11816		       "%d:%d:%d:%d tag %d type %d\n",
11817		       io->io_hdr.nexus.initid.id,
11818		       io->io_hdr.nexus.targ_port,
11819		       io->io_hdr.nexus.targ_target.id,
11820		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11821		       io->taskio.tag_type);
11822#endif
11823		return (1);
11824	} else
11825		return (0);
11826}
11827
11828static void
11829ctl_run_task(union ctl_io *io)
11830{
11831	struct ctl_softc *ctl_softc;
11832	int retval;
11833	const char *task_desc;
11834
11835	CTL_DEBUG_PRINT(("ctl_run_task\n"));
11836
11837	ctl_softc = control_softc;
11838	retval = 0;
11839
11840	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
11841	    ("ctl_run_task: Unextected io_type %d\n",
11842	     io->io_hdr.io_type));
11843
11844	task_desc = ctl_scsi_task_string(&io->taskio);
11845	if (task_desc != NULL) {
11846#ifdef NEEDTOPORT
11847		csevent_log(CSC_CTL | CSC_SHELF_SW |
11848			    CTL_TASK_REPORT,
11849			    csevent_LogType_Trace,
11850			    csevent_Severity_Information,
11851			    csevent_AlertLevel_Green,
11852			    csevent_FRU_Firmware,
11853			    csevent_FRU_Unknown,
11854			    "CTL: received task: %s",task_desc);
11855#endif
11856	} else {
11857#ifdef NEEDTOPORT
11858		csevent_log(CSC_CTL | CSC_SHELF_SW |
11859			    CTL_TASK_REPORT,
11860			    csevent_LogType_Trace,
11861			    csevent_Severity_Information,
11862			    csevent_AlertLevel_Green,
11863			    csevent_FRU_Firmware,
11864			    csevent_FRU_Unknown,
11865			    "CTL: received unknown task "
11866			    "type: %d (%#x)",
11867			    io->taskio.task_action,
11868			    io->taskio.task_action);
11869#endif
11870	}
11871	switch (io->taskio.task_action) {
11872	case CTL_TASK_ABORT_TASK:
11873		retval = ctl_abort_task(io);
11874		break;
11875	case CTL_TASK_ABORT_TASK_SET:
11876		break;
11877	case CTL_TASK_CLEAR_ACA:
11878		break;
11879	case CTL_TASK_CLEAR_TASK_SET:
11880		break;
11881	case CTL_TASK_LUN_RESET: {
11882		struct ctl_lun *lun;
11883		uint32_t targ_lun;
11884		int retval;
11885
11886		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11887		mtx_lock(&ctl_softc->ctl_lock);
11888		if ((targ_lun < CTL_MAX_LUNS)
11889		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11890			lun = ctl_softc->ctl_luns[targ_lun];
11891		else {
11892			mtx_unlock(&ctl_softc->ctl_lock);
11893			retval = 1;
11894			break;
11895		}
11896
11897		if (!(io->io_hdr.flags &
11898		    CTL_FLAG_FROM_OTHER_SC)) {
11899			union ctl_ha_msg msg_info;
11900
11901			io->io_hdr.flags |=
11902				CTL_FLAG_SENT_2OTHER_SC;
11903			msg_info.hdr.msg_type =
11904				CTL_MSG_MANAGE_TASKS;
11905			msg_info.hdr.nexus = io->io_hdr.nexus;
11906			msg_info.task.task_action =
11907				CTL_TASK_LUN_RESET;
11908			msg_info.hdr.original_sc = NULL;
11909			msg_info.hdr.serializing_sc = NULL;
11910			if (CTL_HA_STATUS_SUCCESS !=
11911			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11912			    (void *)&msg_info,
11913			    sizeof(msg_info), 0)) {
11914			}
11915		}
11916
11917		retval = ctl_lun_reset(lun, io,
11918				       CTL_UA_LUN_RESET);
11919		mtx_unlock(&ctl_softc->ctl_lock);
11920		break;
11921	}
11922	case CTL_TASK_TARGET_RESET:
11923		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
11924		break;
11925	case CTL_TASK_BUS_RESET:
11926		retval = ctl_bus_reset(ctl_softc, io);
11927		break;
11928	case CTL_TASK_PORT_LOGIN:
11929		break;
11930	case CTL_TASK_PORT_LOGOUT:
11931		break;
11932	default:
11933		printf("ctl_run_task: got unknown task management event %d\n",
11934		       io->taskio.task_action);
11935		break;
11936	}
11937	if (retval == 0)
11938		io->io_hdr.status = CTL_SUCCESS;
11939	else
11940		io->io_hdr.status = CTL_ERROR;
11941
11942	/*
11943	 * This will queue this I/O to the done queue, but the
11944	 * work thread won't be able to process it until we
11945	 * return and the lock is released.
11946	 */
11947	ctl_done(io);
11948}
11949
11950/*
11951 * For HA operation.  Handle commands that come in from the other
11952 * controller.
11953 */
11954static void
11955ctl_handle_isc(union ctl_io *io)
11956{
11957	int free_io;
11958	struct ctl_lun *lun;
11959	struct ctl_softc *ctl_softc;
11960	uint32_t targ_lun;
11961
11962	ctl_softc = control_softc;
11963
11964	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11965	lun = ctl_softc->ctl_luns[targ_lun];
11966
11967	switch (io->io_hdr.msg_type) {
11968	case CTL_MSG_SERIALIZE:
11969		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
11970		break;
11971	case CTL_MSG_R2R: {
11972		const struct ctl_cmd_entry *entry;
11973
11974		/*
11975		 * This is only used in SER_ONLY mode.
11976		 */
11977		free_io = 0;
11978		entry = ctl_get_cmd_entry(&io->scsiio);
11979		mtx_lock(&lun->lun_lock);
11980		if (ctl_scsiio_lun_check(ctl_softc, lun,
11981		    entry, (struct ctl_scsiio *)io) != 0) {
11982			mtx_unlock(&lun->lun_lock);
11983			ctl_done(io);
11984			break;
11985		}
11986		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11987		mtx_unlock(&lun->lun_lock);
11988		ctl_enqueue_rtr(io);
11989		break;
11990	}
11991	case CTL_MSG_FINISH_IO:
11992		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11993			free_io = 0;
11994			ctl_done(io);
11995		} else {
11996			free_io = 1;
11997			mtx_lock(&lun->lun_lock);
11998			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11999				     ooa_links);
12000			ctl_check_blocked(lun);
12001			mtx_unlock(&lun->lun_lock);
12002		}
12003		break;
12004	case CTL_MSG_PERS_ACTION:
12005		ctl_hndl_per_res_out_on_other_sc(
12006			(union ctl_ha_msg *)&io->presio.pr_msg);
12007		free_io = 1;
12008		break;
12009	case CTL_MSG_BAD_JUJU:
12010		free_io = 0;
12011		ctl_done(io);
12012		break;
12013	case CTL_MSG_DATAMOVE:
12014		/* Only used in XFER mode */
12015		free_io = 0;
12016		ctl_datamove_remote(io);
12017		break;
12018	case CTL_MSG_DATAMOVE_DONE:
12019		/* Only used in XFER mode */
12020		free_io = 0;
12021		io->scsiio.be_move_done(io);
12022		break;
12023	default:
12024		free_io = 1;
12025		printf("%s: Invalid message type %d\n",
12026		       __func__, io->io_hdr.msg_type);
12027		break;
12028	}
12029	if (free_io)
12030		ctl_free_io(io);
12031
12032}
12033
12034
12035/*
12036 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12037 * there is no match.
12038 */
12039static ctl_lun_error_pattern
12040ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12041{
12042	const struct ctl_cmd_entry *entry;
12043	ctl_lun_error_pattern filtered_pattern, pattern;
12044
12045	pattern = desc->error_pattern;
12046
12047	/*
12048	 * XXX KDM we need more data passed into this function to match a
12049	 * custom pattern, and we actually need to implement custom pattern
12050	 * matching.
12051	 */
12052	if (pattern & CTL_LUN_PAT_CMD)
12053		return (CTL_LUN_PAT_CMD);
12054
12055	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12056		return (CTL_LUN_PAT_ANY);
12057
12058	entry = ctl_get_cmd_entry(ctsio);
12059
12060	filtered_pattern = entry->pattern & pattern;
12061
12062	/*
12063	 * If the user requested specific flags in the pattern (e.g.
12064	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12065	 * flags.
12066	 *
12067	 * If the user did not specify any flags, it doesn't matter whether
12068	 * or not the command supports the flags.
12069	 */
12070	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12071	     (pattern & ~CTL_LUN_PAT_MASK))
12072		return (CTL_LUN_PAT_NONE);
12073
12074	/*
12075	 * If the user asked for a range check, see if the requested LBA
12076	 * range overlaps with this command's LBA range.
12077	 */
12078	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12079		uint64_t lba1;
12080		uint32_t len1;
12081		ctl_action action;
12082		int retval;
12083
12084		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12085		if (retval != 0)
12086			return (CTL_LUN_PAT_NONE);
12087
12088		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12089					      desc->lba_range.len);
12090		/*
12091		 * A "pass" means that the LBA ranges don't overlap, so
12092		 * this doesn't match the user's range criteria.
12093		 */
12094		if (action == CTL_ACTION_PASS)
12095			return (CTL_LUN_PAT_NONE);
12096	}
12097
12098	return (filtered_pattern);
12099}
12100
12101static void
12102ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12103{
12104	struct ctl_error_desc *desc, *desc2;
12105
12106	mtx_assert(&lun->lun_lock, MA_OWNED);
12107
12108	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12109		ctl_lun_error_pattern pattern;
12110		/*
12111		 * Check to see whether this particular command matches
12112		 * the pattern in the descriptor.
12113		 */
12114		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12115		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12116			continue;
12117
12118		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12119		case CTL_LUN_INJ_ABORTED:
12120			ctl_set_aborted(&io->scsiio);
12121			break;
12122		case CTL_LUN_INJ_MEDIUM_ERR:
12123			ctl_set_medium_error(&io->scsiio);
12124			break;
12125		case CTL_LUN_INJ_UA:
12126			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12127			 * OCCURRED */
12128			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12129			break;
12130		case CTL_LUN_INJ_CUSTOM:
12131			/*
12132			 * We're assuming the user knows what he is doing.
12133			 * Just copy the sense information without doing
12134			 * checks.
12135			 */
12136			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12137			      ctl_min(sizeof(desc->custom_sense),
12138				      sizeof(io->scsiio.sense_data)));
12139			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12140			io->scsiio.sense_len = SSD_FULL_SIZE;
12141			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12142			break;
12143		case CTL_LUN_INJ_NONE:
12144		default:
12145			/*
12146			 * If this is an error injection type we don't know
12147			 * about, clear the continuous flag (if it is set)
12148			 * so it will get deleted below.
12149			 */
12150			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12151			break;
12152		}
12153		/*
12154		 * By default, each error injection action is a one-shot
12155		 */
12156		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12157			continue;
12158
12159		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12160
12161		free(desc, M_CTL);
12162	}
12163}
12164
12165#ifdef CTL_IO_DELAY
12166static void
12167ctl_datamove_timer_wakeup(void *arg)
12168{
12169	union ctl_io *io;
12170
12171	io = (union ctl_io *)arg;
12172
12173	ctl_datamove(io);
12174}
12175#endif /* CTL_IO_DELAY */
12176
12177void
12178ctl_datamove(union ctl_io *io)
12179{
12180	void (*fe_datamove)(union ctl_io *io);
12181
12182	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12183
12184	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12185
12186#ifdef CTL_TIME_IO
12187	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12188		char str[256];
12189		char path_str[64];
12190		struct sbuf sb;
12191
12192		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12193		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12194
12195		sbuf_cat(&sb, path_str);
12196		switch (io->io_hdr.io_type) {
12197		case CTL_IO_SCSI:
12198			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12199			sbuf_printf(&sb, "\n");
12200			sbuf_cat(&sb, path_str);
12201			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12202				    io->scsiio.tag_num, io->scsiio.tag_type);
12203			break;
12204		case CTL_IO_TASK:
12205			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12206				    "Tag Type: %d\n", io->taskio.task_action,
12207				    io->taskio.tag_num, io->taskio.tag_type);
12208			break;
12209		default:
12210			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12211			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12212			break;
12213		}
12214		sbuf_cat(&sb, path_str);
12215		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12216			    (intmax_t)time_uptime - io->io_hdr.start_time);
12217		sbuf_finish(&sb);
12218		printf("%s", sbuf_data(&sb));
12219	}
12220#endif /* CTL_TIME_IO */
12221
12222#ifdef CTL_IO_DELAY
12223	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12224		struct ctl_lun *lun;
12225
12226		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12227
12228		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12229	} else {
12230		struct ctl_lun *lun;
12231
12232		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12233		if ((lun != NULL)
12234		 && (lun->delay_info.datamove_delay > 0)) {
12235			struct callout *callout;
12236
12237			callout = (struct callout *)&io->io_hdr.timer_bytes;
12238			callout_init(callout, /*mpsafe*/ 1);
12239			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12240			callout_reset(callout,
12241				      lun->delay_info.datamove_delay * hz,
12242				      ctl_datamove_timer_wakeup, io);
12243			if (lun->delay_info.datamove_type ==
12244			    CTL_DELAY_TYPE_ONESHOT)
12245				lun->delay_info.datamove_delay = 0;
12246			return;
12247		}
12248	}
12249#endif
12250
12251	/*
12252	 * This command has been aborted.  Set the port status, so we fail
12253	 * the data move.
12254	 */
12255	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12256		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12257		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12258		       io->io_hdr.nexus.targ_port,
12259		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12260		       io->io_hdr.nexus.targ_lun);
12261		io->io_hdr.status = CTL_CMD_ABORTED;
12262		io->io_hdr.port_status = 31337;
12263		/*
12264		 * Note that the backend, in this case, will get the
12265		 * callback in its context.  In other cases it may get
12266		 * called in the frontend's interrupt thread context.
12267		 */
12268		io->scsiio.be_move_done(io);
12269		return;
12270	}
12271
12272	/*
12273	 * If we're in XFER mode and this I/O is from the other shelf
12274	 * controller, we need to send the DMA to the other side to
12275	 * actually transfer the data to/from the host.  In serialize only
12276	 * mode the transfer happens below CTL and ctl_datamove() is only
12277	 * called on the machine that originally received the I/O.
12278	 */
12279	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12280	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12281		union ctl_ha_msg msg;
12282		uint32_t sg_entries_sent;
12283		int do_sg_copy;
12284		int i;
12285
12286		memset(&msg, 0, sizeof(msg));
12287		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12288		msg.hdr.original_sc = io->io_hdr.original_sc;
12289		msg.hdr.serializing_sc = io;
12290		msg.hdr.nexus = io->io_hdr.nexus;
12291		msg.dt.flags = io->io_hdr.flags;
12292		/*
12293		 * We convert everything into a S/G list here.  We can't
12294		 * pass by reference, only by value between controllers.
12295		 * So we can't pass a pointer to the S/G list, only as many
12296		 * S/G entries as we can fit in here.  If it's possible for
12297		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12298		 * then we need to break this up into multiple transfers.
12299		 */
12300		if (io->scsiio.kern_sg_entries == 0) {
12301			msg.dt.kern_sg_entries = 1;
12302			/*
12303			 * If this is in cached memory, flush the cache
12304			 * before we send the DMA request to the other
12305			 * controller.  We want to do this in either the
12306			 * read or the write case.  The read case is
12307			 * straightforward.  In the write case, we want to
12308			 * make sure nothing is in the local cache that
12309			 * could overwrite the DMAed data.
12310			 */
12311			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12312				/*
12313				 * XXX KDM use bus_dmamap_sync() here.
12314				 */
12315			}
12316
12317			/*
12318			 * Convert to a physical address if this is a
12319			 * virtual address.
12320			 */
12321			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12322				msg.dt.sg_list[0].addr =
12323					io->scsiio.kern_data_ptr;
12324			} else {
12325				/*
12326				 * XXX KDM use busdma here!
12327				 */
12328#if 0
12329				msg.dt.sg_list[0].addr = (void *)
12330					vtophys(io->scsiio.kern_data_ptr);
12331#endif
12332			}
12333
12334			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12335			do_sg_copy = 0;
12336		} else {
12337			struct ctl_sg_entry *sgl;
12338
12339			do_sg_copy = 1;
12340			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12341			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12342			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12343				/*
12344				 * XXX KDM use bus_dmamap_sync() here.
12345				 */
12346			}
12347		}
12348
12349		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12350		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12351		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12352		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12353		msg.dt.sg_sequence = 0;
12354
12355		/*
12356		 * Loop until we've sent all of the S/G entries.  On the
12357		 * other end, we'll recompose these S/G entries into one
12358		 * contiguous list before passing it to the
12359		 */
12360		for (sg_entries_sent = 0; sg_entries_sent <
12361		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12362			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12363				sizeof(msg.dt.sg_list[0])),
12364				msg.dt.kern_sg_entries - sg_entries_sent);
12365
12366			if (do_sg_copy != 0) {
12367				struct ctl_sg_entry *sgl;
12368				int j;
12369
12370				sgl = (struct ctl_sg_entry *)
12371					io->scsiio.kern_data_ptr;
12372				/*
12373				 * If this is in cached memory, flush the cache
12374				 * before we send the DMA request to the other
12375				 * controller.  We want to do this in either
12376				 * the * read or the write case.  The read
12377				 * case is straightforward.  In the write
12378				 * case, we want to make sure nothing is
12379				 * in the local cache that could overwrite
12380				 * the DMAed data.
12381				 */
12382
12383				for (i = sg_entries_sent, j = 0;
12384				     i < msg.dt.cur_sg_entries; i++, j++) {
12385					if ((io->io_hdr.flags &
12386					     CTL_FLAG_NO_DATASYNC) == 0) {
12387						/*
12388						 * XXX KDM use bus_dmamap_sync()
12389						 */
12390					}
12391					if ((io->io_hdr.flags &
12392					     CTL_FLAG_BUS_ADDR) == 0) {
12393						/*
12394						 * XXX KDM use busdma.
12395						 */
12396#if 0
12397						msg.dt.sg_list[j].addr =(void *)
12398						       vtophys(sgl[i].addr);
12399#endif
12400					} else {
12401						msg.dt.sg_list[j].addr =
12402							sgl[i].addr;
12403					}
12404					msg.dt.sg_list[j].len = sgl[i].len;
12405				}
12406			}
12407
12408			sg_entries_sent += msg.dt.cur_sg_entries;
12409			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12410				msg.dt.sg_last = 1;
12411			else
12412				msg.dt.sg_last = 0;
12413
12414			/*
12415			 * XXX KDM drop and reacquire the lock here?
12416			 */
12417			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12418			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12419				/*
12420				 * XXX do something here.
12421				 */
12422			}
12423
12424			msg.dt.sent_sg_entries = sg_entries_sent;
12425		}
12426		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12427		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12428			ctl_failover_io(io, /*have_lock*/ 0);
12429
12430	} else {
12431
12432		/*
12433		 * Lookup the fe_datamove() function for this particular
12434		 * front end.
12435		 */
12436		fe_datamove =
12437		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12438
12439		fe_datamove(io);
12440	}
12441}
12442
12443static void
12444ctl_send_datamove_done(union ctl_io *io, int have_lock)
12445{
12446	union ctl_ha_msg msg;
12447	int isc_status;
12448
12449	memset(&msg, 0, sizeof(msg));
12450
12451	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12452	msg.hdr.original_sc = io;
12453	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12454	msg.hdr.nexus = io->io_hdr.nexus;
12455	msg.hdr.status = io->io_hdr.status;
12456	msg.scsi.tag_num = io->scsiio.tag_num;
12457	msg.scsi.tag_type = io->scsiio.tag_type;
12458	msg.scsi.scsi_status = io->scsiio.scsi_status;
12459	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12460	       sizeof(io->scsiio.sense_data));
12461	msg.scsi.sense_len = io->scsiio.sense_len;
12462	msg.scsi.sense_residual = io->scsiio.sense_residual;
12463	msg.scsi.fetd_status = io->io_hdr.port_status;
12464	msg.scsi.residual = io->scsiio.residual;
12465	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12466
12467	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12468		ctl_failover_io(io, /*have_lock*/ have_lock);
12469		return;
12470	}
12471
12472	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12473	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12474		/* XXX do something if this fails */
12475	}
12476
12477}
12478
12479/*
12480 * The DMA to the remote side is done, now we need to tell the other side
12481 * we're done so it can continue with its data movement.
12482 */
12483static void
12484ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12485{
12486	union ctl_io *io;
12487
12488	io = rq->context;
12489
12490	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12491		printf("%s: ISC DMA write failed with error %d", __func__,
12492		       rq->ret);
12493		ctl_set_internal_failure(&io->scsiio,
12494					 /*sks_valid*/ 1,
12495					 /*retry_count*/ rq->ret);
12496	}
12497
12498	ctl_dt_req_free(rq);
12499
12500	/*
12501	 * In this case, we had to malloc the memory locally.  Free it.
12502	 */
12503	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12504		int i;
12505		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12506			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12507	}
12508	/*
12509	 * The data is in local and remote memory, so now we need to send
12510	 * status (good or back) back to the other side.
12511	 */
12512	ctl_send_datamove_done(io, /*have_lock*/ 0);
12513}
12514
12515/*
12516 * We've moved the data from the host/controller into local memory.  Now we
12517 * need to push it over to the remote controller's memory.
12518 */
12519static int
12520ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12521{
12522	int retval;
12523
12524	retval = 0;
12525
12526	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12527					  ctl_datamove_remote_write_cb);
12528
12529	return (retval);
12530}
12531
12532static void
12533ctl_datamove_remote_write(union ctl_io *io)
12534{
12535	int retval;
12536	void (*fe_datamove)(union ctl_io *io);
12537
12538	/*
12539	 * - Get the data from the host/HBA into local memory.
12540	 * - DMA memory from the local controller to the remote controller.
12541	 * - Send status back to the remote controller.
12542	 */
12543
12544	retval = ctl_datamove_remote_sgl_setup(io);
12545	if (retval != 0)
12546		return;
12547
12548	/* Switch the pointer over so the FETD knows what to do */
12549	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12550
12551	/*
12552	 * Use a custom move done callback, since we need to send completion
12553	 * back to the other controller, not to the backend on this side.
12554	 */
12555	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12556
12557	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12558
12559	fe_datamove(io);
12560
12561	return;
12562
12563}
12564
12565static int
12566ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12567{
12568#if 0
12569	char str[256];
12570	char path_str[64];
12571	struct sbuf sb;
12572#endif
12573
12574	/*
12575	 * In this case, we had to malloc the memory locally.  Free it.
12576	 */
12577	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12578		int i;
12579		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12580			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12581	}
12582
12583#if 0
12584	scsi_path_string(io, path_str, sizeof(path_str));
12585	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12586	sbuf_cat(&sb, path_str);
12587	scsi_command_string(&io->scsiio, NULL, &sb);
12588	sbuf_printf(&sb, "\n");
12589	sbuf_cat(&sb, path_str);
12590	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12591		    io->scsiio.tag_num, io->scsiio.tag_type);
12592	sbuf_cat(&sb, path_str);
12593	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12594		    io->io_hdr.flags, io->io_hdr.status);
12595	sbuf_finish(&sb);
12596	printk("%s", sbuf_data(&sb));
12597#endif
12598
12599
12600	/*
12601	 * The read is done, now we need to send status (good or bad) back
12602	 * to the other side.
12603	 */
12604	ctl_send_datamove_done(io, /*have_lock*/ 0);
12605
12606	return (0);
12607}
12608
12609static void
12610ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12611{
12612	union ctl_io *io;
12613	void (*fe_datamove)(union ctl_io *io);
12614
12615	io = rq->context;
12616
12617	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12618		printf("%s: ISC DMA read failed with error %d", __func__,
12619		       rq->ret);
12620		ctl_set_internal_failure(&io->scsiio,
12621					 /*sks_valid*/ 1,
12622					 /*retry_count*/ rq->ret);
12623	}
12624
12625	ctl_dt_req_free(rq);
12626
12627	/* Switch the pointer over so the FETD knows what to do */
12628	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12629
12630	/*
12631	 * Use a custom move done callback, since we need to send completion
12632	 * back to the other controller, not to the backend on this side.
12633	 */
12634	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12635
12636	/* XXX KDM add checks like the ones in ctl_datamove? */
12637
12638	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12639
12640	fe_datamove(io);
12641}
12642
12643static int
12644ctl_datamove_remote_sgl_setup(union ctl_io *io)
12645{
12646	struct ctl_sg_entry *local_sglist, *remote_sglist;
12647	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12648	struct ctl_softc *softc;
12649	int retval;
12650	int i;
12651
12652	retval = 0;
12653	softc = control_softc;
12654
12655	local_sglist = io->io_hdr.local_sglist;
12656	local_dma_sglist = io->io_hdr.local_dma_sglist;
12657	remote_sglist = io->io_hdr.remote_sglist;
12658	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12659
12660	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12661		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12662			local_sglist[i].len = remote_sglist[i].len;
12663
12664			/*
12665			 * XXX Detect the situation where the RS-level I/O
12666			 * redirector on the other side has already read the
12667			 * data off of the AOR RS on this side, and
12668			 * transferred it to remote (mirror) memory on the
12669			 * other side.  Since we already have the data in
12670			 * memory here, we just need to use it.
12671			 *
12672			 * XXX KDM this can probably be removed once we
12673			 * get the cache device code in and take the
12674			 * current AOR implementation out.
12675			 */
12676#ifdef NEEDTOPORT
12677			if ((remote_sglist[i].addr >=
12678			     (void *)vtophys(softc->mirr->addr))
12679			 && (remote_sglist[i].addr <
12680			     ((void *)vtophys(softc->mirr->addr) +
12681			     CacheMirrorOffset))) {
12682				local_sglist[i].addr = remote_sglist[i].addr -
12683					CacheMirrorOffset;
12684				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12685				     CTL_FLAG_DATA_IN)
12686					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12687			} else {
12688				local_sglist[i].addr = remote_sglist[i].addr +
12689					CacheMirrorOffset;
12690			}
12691#endif
12692#if 0
12693			printf("%s: local %p, remote %p, len %d\n",
12694			       __func__, local_sglist[i].addr,
12695			       remote_sglist[i].addr, local_sglist[i].len);
12696#endif
12697		}
12698	} else {
12699		uint32_t len_to_go;
12700
12701		/*
12702		 * In this case, we don't have automatically allocated
12703		 * memory for this I/O on this controller.  This typically
12704		 * happens with internal CTL I/O -- e.g. inquiry, mode
12705		 * sense, etc.  Anything coming from RAIDCore will have
12706		 * a mirror area available.
12707		 */
12708		len_to_go = io->scsiio.kern_data_len;
12709
12710		/*
12711		 * Clear the no datasync flag, we have to use malloced
12712		 * buffers.
12713		 */
12714		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12715
12716		/*
12717		 * The difficult thing here is that the size of the various
12718		 * S/G segments may be different than the size from the
12719		 * remote controller.  That'll make it harder when DMAing
12720		 * the data back to the other side.
12721		 */
12722		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12723		     sizeof(io->io_hdr.remote_sglist[0])) &&
12724		     (len_to_go > 0); i++) {
12725			local_sglist[i].len = ctl_min(len_to_go, 131072);
12726			CTL_SIZE_8B(local_dma_sglist[i].len,
12727				    local_sglist[i].len);
12728			local_sglist[i].addr =
12729				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
12730
12731			local_dma_sglist[i].addr = local_sglist[i].addr;
12732
12733			if (local_sglist[i].addr == NULL) {
12734				int j;
12735
12736				printf("malloc failed for %zd bytes!",
12737				       local_dma_sglist[i].len);
12738				for (j = 0; j < i; j++) {
12739					free(local_sglist[j].addr, M_CTL);
12740				}
12741				ctl_set_internal_failure(&io->scsiio,
12742							 /*sks_valid*/ 1,
12743							 /*retry_count*/ 4857);
12744				retval = 1;
12745				goto bailout_error;
12746
12747			}
12748			/* XXX KDM do we need a sync here? */
12749
12750			len_to_go -= local_sglist[i].len;
12751		}
12752		/*
12753		 * Reset the number of S/G entries accordingly.  The
12754		 * original number of S/G entries is available in
12755		 * rem_sg_entries.
12756		 */
12757		io->scsiio.kern_sg_entries = i;
12758
12759#if 0
12760		printf("%s: kern_sg_entries = %d\n", __func__,
12761		       io->scsiio.kern_sg_entries);
12762		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12763			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
12764			       local_sglist[i].addr, local_sglist[i].len,
12765			       local_dma_sglist[i].len);
12766#endif
12767	}
12768
12769
12770	return (retval);
12771
12772bailout_error:
12773
12774	ctl_send_datamove_done(io, /*have_lock*/ 0);
12775
12776	return (retval);
12777}
12778
12779static int
12780ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12781			 ctl_ha_dt_cb callback)
12782{
12783	struct ctl_ha_dt_req *rq;
12784	struct ctl_sg_entry *remote_sglist, *local_sglist;
12785	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
12786	uint32_t local_used, remote_used, total_used;
12787	int retval;
12788	int i, j;
12789
12790	retval = 0;
12791
12792	rq = ctl_dt_req_alloc();
12793
12794	/*
12795	 * If we failed to allocate the request, and if the DMA didn't fail
12796	 * anyway, set busy status.  This is just a resource allocation
12797	 * failure.
12798	 */
12799	if ((rq == NULL)
12800	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12801		ctl_set_busy(&io->scsiio);
12802
12803	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12804
12805		if (rq != NULL)
12806			ctl_dt_req_free(rq);
12807
12808		/*
12809		 * The data move failed.  We need to return status back
12810		 * to the other controller.  No point in trying to DMA
12811		 * data to the remote controller.
12812		 */
12813
12814		ctl_send_datamove_done(io, /*have_lock*/ 0);
12815
12816		retval = 1;
12817
12818		goto bailout;
12819	}
12820
12821	local_sglist = io->io_hdr.local_sglist;
12822	local_dma_sglist = io->io_hdr.local_dma_sglist;
12823	remote_sglist = io->io_hdr.remote_sglist;
12824	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12825	local_used = 0;
12826	remote_used = 0;
12827	total_used = 0;
12828
12829	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12830		rq->ret = CTL_HA_STATUS_SUCCESS;
12831		rq->context = io;
12832		callback(rq);
12833		goto bailout;
12834	}
12835
12836	/*
12837	 * Pull/push the data over the wire from/to the other controller.
12838	 * This takes into account the possibility that the local and
12839	 * remote sglists may not be identical in terms of the size of
12840	 * the elements and the number of elements.
12841	 *
12842	 * One fundamental assumption here is that the length allocated for
12843	 * both the local and remote sglists is identical.  Otherwise, we've
12844	 * essentially got a coding error of some sort.
12845	 */
12846	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12847		int isc_ret;
12848		uint32_t cur_len, dma_length;
12849		uint8_t *tmp_ptr;
12850
12851		rq->id = CTL_HA_DATA_CTL;
12852		rq->command = command;
12853		rq->context = io;
12854
12855		/*
12856		 * Both pointers should be aligned.  But it is possible
12857		 * that the allocation length is not.  They should both
12858		 * also have enough slack left over at the end, though,
12859		 * to round up to the next 8 byte boundary.
12860		 */
12861		cur_len = ctl_min(local_sglist[i].len - local_used,
12862				  remote_sglist[j].len - remote_used);
12863
12864		/*
12865		 * In this case, we have a size issue and need to decrease
12866		 * the size, except in the case where we actually have less
12867		 * than 8 bytes left.  In that case, we need to increase
12868		 * the DMA length to get the last bit.
12869		 */
12870		if ((cur_len & 0x7) != 0) {
12871			if (cur_len > 0x7) {
12872				cur_len = cur_len - (cur_len & 0x7);
12873				dma_length = cur_len;
12874			} else {
12875				CTL_SIZE_8B(dma_length, cur_len);
12876			}
12877
12878		} else
12879			dma_length = cur_len;
12880
12881		/*
12882		 * If we had to allocate memory for this I/O, instead of using
12883		 * the non-cached mirror memory, we'll need to flush the cache
12884		 * before trying to DMA to the other controller.
12885		 *
12886		 * We could end up doing this multiple times for the same
12887		 * segment if we have a larger local segment than remote
12888		 * segment.  That shouldn't be an issue.
12889		 */
12890		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12891			/*
12892			 * XXX KDM use bus_dmamap_sync() here.
12893			 */
12894		}
12895
12896		rq->size = dma_length;
12897
12898		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12899		tmp_ptr += local_used;
12900
12901		/* Use physical addresses when talking to ISC hardware */
12902		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12903			/* XXX KDM use busdma */
12904#if 0
12905			rq->local = vtophys(tmp_ptr);
12906#endif
12907		} else
12908			rq->local = tmp_ptr;
12909
12910		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12911		tmp_ptr += remote_used;
12912		rq->remote = tmp_ptr;
12913
12914		rq->callback = NULL;
12915
12916		local_used += cur_len;
12917		if (local_used >= local_sglist[i].len) {
12918			i++;
12919			local_used = 0;
12920		}
12921
12922		remote_used += cur_len;
12923		if (remote_used >= remote_sglist[j].len) {
12924			j++;
12925			remote_used = 0;
12926		}
12927		total_used += cur_len;
12928
12929		if (total_used >= io->scsiio.kern_data_len)
12930			rq->callback = callback;
12931
12932		if ((rq->size & 0x7) != 0) {
12933			printf("%s: warning: size %d is not on 8b boundary\n",
12934			       __func__, rq->size);
12935		}
12936		if (((uintptr_t)rq->local & 0x7) != 0) {
12937			printf("%s: warning: local %p not on 8b boundary\n",
12938			       __func__, rq->local);
12939		}
12940		if (((uintptr_t)rq->remote & 0x7) != 0) {
12941			printf("%s: warning: remote %p not on 8b boundary\n",
12942			       __func__, rq->local);
12943		}
12944#if 0
12945		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12946		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12947		       rq->local, rq->remote, rq->size);
12948#endif
12949
12950		isc_ret = ctl_dt_single(rq);
12951		if (isc_ret == CTL_HA_STATUS_WAIT)
12952			continue;
12953
12954		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12955			rq->ret = CTL_HA_STATUS_SUCCESS;
12956		} else {
12957			rq->ret = isc_ret;
12958		}
12959		callback(rq);
12960		goto bailout;
12961	}
12962
12963bailout:
12964	return (retval);
12965
12966}
12967
12968static void
12969ctl_datamove_remote_read(union ctl_io *io)
12970{
12971	int retval;
12972	int i;
12973
12974	/*
12975	 * This will send an error to the other controller in the case of a
12976	 * failure.
12977	 */
12978	retval = ctl_datamove_remote_sgl_setup(io);
12979	if (retval != 0)
12980		return;
12981
12982	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12983					  ctl_datamove_remote_read_cb);
12984	if ((retval != 0)
12985	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12986		/*
12987		 * Make sure we free memory if there was an error..  The
12988		 * ctl_datamove_remote_xfer() function will send the
12989		 * datamove done message, or call the callback with an
12990		 * error if there is a problem.
12991		 */
12992		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12993			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12994	}
12995
12996	return;
12997}
12998
12999/*
13000 * Process a datamove request from the other controller.  This is used for
13001 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13002 * first.  Once that is complete, the data gets DMAed into the remote
13003 * controller's memory.  For reads, we DMA from the remote controller's
13004 * memory into our memory first, and then move it out to the FETD.
13005 */
13006static void
13007ctl_datamove_remote(union ctl_io *io)
13008{
13009	struct ctl_softc *softc;
13010
13011	softc = control_softc;
13012
13013	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13014
13015	/*
13016	 * Note that we look for an aborted I/O here, but don't do some of
13017	 * the other checks that ctl_datamove() normally does.  We don't
13018	 * need to run the task queue, because this I/O is on the ISC
13019	 * queue, which is executed by the work thread after the task queue.
13020	 * We don't need to run the datamove delay code, since that should
13021	 * have been done if need be on the other controller.
13022	 */
13023	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13024
13025		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13026		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13027		       io->io_hdr.nexus.targ_port,
13028		       io->io_hdr.nexus.targ_target.id,
13029		       io->io_hdr.nexus.targ_lun);
13030		io->io_hdr.status = CTL_CMD_ABORTED;
13031		io->io_hdr.port_status = 31338;
13032
13033		ctl_send_datamove_done(io, /*have_lock*/ 0);
13034
13035		return;
13036	}
13037
13038	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13039		ctl_datamove_remote_write(io);
13040	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13041		ctl_datamove_remote_read(io);
13042	} else {
13043		union ctl_ha_msg msg;
13044		struct scsi_sense_data *sense;
13045		uint8_t sks[3];
13046		int retry_count;
13047
13048		memset(&msg, 0, sizeof(msg));
13049
13050		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13051		msg.hdr.status = CTL_SCSI_ERROR;
13052		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13053
13054		retry_count = 4243;
13055
13056		sense = &msg.scsi.sense_data;
13057		sks[0] = SSD_SCS_VALID;
13058		sks[1] = (retry_count >> 8) & 0xff;
13059		sks[2] = retry_count & 0xff;
13060
13061		/* "Internal target failure" */
13062		scsi_set_sense_data(sense,
13063				    /*sense_format*/ SSD_TYPE_NONE,
13064				    /*current_error*/ 1,
13065				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13066				    /*asc*/ 0x44,
13067				    /*ascq*/ 0x00,
13068				    /*type*/ SSD_ELEM_SKS,
13069				    /*size*/ sizeof(sks),
13070				    /*data*/ sks,
13071				    SSD_ELEM_NONE);
13072
13073		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13074		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13075			ctl_failover_io(io, /*have_lock*/ 1);
13076			return;
13077		}
13078
13079		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13080		    CTL_HA_STATUS_SUCCESS) {
13081			/* XXX KDM what to do if this fails? */
13082		}
13083		return;
13084	}
13085
13086}
13087
13088static int
13089ctl_process_done(union ctl_io *io)
13090{
13091	struct ctl_lun *lun;
13092	struct ctl_softc *ctl_softc;
13093	void (*fe_done)(union ctl_io *io);
13094	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13095
13096	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13097
13098	fe_done =
13099	    control_softc->ctl_ports[targ_port]->fe_done;
13100
13101#ifdef CTL_TIME_IO
13102	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13103		char str[256];
13104		char path_str[64];
13105		struct sbuf sb;
13106
13107		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13108		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13109
13110		sbuf_cat(&sb, path_str);
13111		switch (io->io_hdr.io_type) {
13112		case CTL_IO_SCSI:
13113			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13114			sbuf_printf(&sb, "\n");
13115			sbuf_cat(&sb, path_str);
13116			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13117				    io->scsiio.tag_num, io->scsiio.tag_type);
13118			break;
13119		case CTL_IO_TASK:
13120			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13121				    "Tag Type: %d\n", io->taskio.task_action,
13122				    io->taskio.tag_num, io->taskio.tag_type);
13123			break;
13124		default:
13125			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13126			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13127			break;
13128		}
13129		sbuf_cat(&sb, path_str);
13130		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13131			    (intmax_t)time_uptime - io->io_hdr.start_time);
13132		sbuf_finish(&sb);
13133		printf("%s", sbuf_data(&sb));
13134	}
13135#endif /* CTL_TIME_IO */
13136
13137	switch (io->io_hdr.io_type) {
13138	case CTL_IO_SCSI:
13139		break;
13140	case CTL_IO_TASK:
13141		if (bootverbose || verbose > 0)
13142			ctl_io_error_print(io, NULL);
13143		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13144			ctl_free_io(io);
13145		else
13146			fe_done(io);
13147		return (CTL_RETVAL_COMPLETE);
13148		break;
13149	default:
13150		printf("ctl_process_done: invalid io type %d\n",
13151		       io->io_hdr.io_type);
13152		panic("ctl_process_done: invalid io type %d\n",
13153		      io->io_hdr.io_type);
13154		break; /* NOTREACHED */
13155	}
13156
13157	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13158	if (lun == NULL) {
13159		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13160				 io->io_hdr.nexus.targ_mapped_lun));
13161		fe_done(io);
13162		goto bailout;
13163	}
13164	ctl_softc = lun->ctl_softc;
13165
13166	mtx_lock(&lun->lun_lock);
13167
13168	/*
13169	 * Check to see if we have any errors to inject here.  We only
13170	 * inject errors for commands that don't already have errors set.
13171	 */
13172	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13173	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13174		ctl_inject_error(lun, io);
13175
13176	/*
13177	 * XXX KDM how do we treat commands that aren't completed
13178	 * successfully?
13179	 *
13180	 * XXX KDM should we also track I/O latency?
13181	 */
13182	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13183	    io->io_hdr.io_type == CTL_IO_SCSI) {
13184#ifdef CTL_TIME_IO
13185		struct bintime cur_bt;
13186#endif
13187		int type;
13188
13189		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13190		    CTL_FLAG_DATA_IN)
13191			type = CTL_STATS_READ;
13192		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13193		    CTL_FLAG_DATA_OUT)
13194			type = CTL_STATS_WRITE;
13195		else
13196			type = CTL_STATS_NO_IO;
13197
13198		lun->stats.ports[targ_port].bytes[type] +=
13199		    io->scsiio.kern_total_len;
13200		lun->stats.ports[targ_port].operations[type]++;
13201#ifdef CTL_TIME_IO
13202		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13203		   &io->io_hdr.dma_bt);
13204		lun->stats.ports[targ_port].num_dmas[type] +=
13205		    io->io_hdr.num_dmas;
13206		getbintime(&cur_bt);
13207		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13208		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13209#endif
13210	}
13211
13212	/*
13213	 * Remove this from the OOA queue.
13214	 */
13215	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13216
13217	/*
13218	 * Run through the blocked queue on this LUN and see if anything
13219	 * has become unblocked, now that this transaction is done.
13220	 */
13221	ctl_check_blocked(lun);
13222
13223	/*
13224	 * If the LUN has been invalidated, free it if there is nothing
13225	 * left on its OOA queue.
13226	 */
13227	if ((lun->flags & CTL_LUN_INVALID)
13228	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13229		mtx_unlock(&lun->lun_lock);
13230		mtx_lock(&ctl_softc->ctl_lock);
13231		ctl_free_lun(lun);
13232		mtx_unlock(&ctl_softc->ctl_lock);
13233	} else
13234		mtx_unlock(&lun->lun_lock);
13235
13236	/*
13237	 * If this command has been aborted, make sure we set the status
13238	 * properly.  The FETD is responsible for freeing the I/O and doing
13239	 * whatever it needs to do to clean up its state.
13240	 */
13241	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13242		io->io_hdr.status = CTL_CMD_ABORTED;
13243
13244	/*
13245	 * We print out status for every task management command.  For SCSI
13246	 * commands, we filter out any unit attention errors; they happen
13247	 * on every boot, and would clutter up the log.  Note:  task
13248	 * management commands aren't printed here, they are printed above,
13249	 * since they should never even make it down here.
13250	 */
13251	switch (io->io_hdr.io_type) {
13252	case CTL_IO_SCSI: {
13253		int error_code, sense_key, asc, ascq;
13254
13255		sense_key = 0;
13256
13257		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13258		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13259			/*
13260			 * Since this is just for printing, no need to
13261			 * show errors here.
13262			 */
13263			scsi_extract_sense_len(&io->scsiio.sense_data,
13264					       io->scsiio.sense_len,
13265					       &error_code,
13266					       &sense_key,
13267					       &asc,
13268					       &ascq,
13269					       /*show_errors*/ 0);
13270		}
13271
13272		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13273		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13274		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13275		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13276
13277			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13278				ctl_softc->skipped_prints++;
13279			} else {
13280				uint32_t skipped_prints;
13281
13282				skipped_prints = ctl_softc->skipped_prints;
13283
13284				ctl_softc->skipped_prints = 0;
13285				ctl_softc->last_print_jiffies = time_uptime;
13286
13287				if (skipped_prints > 0) {
13288#ifdef NEEDTOPORT
13289					csevent_log(CSC_CTL | CSC_SHELF_SW |
13290					    CTL_ERROR_REPORT,
13291					    csevent_LogType_Trace,
13292					    csevent_Severity_Information,
13293					    csevent_AlertLevel_Green,
13294					    csevent_FRU_Firmware,
13295					    csevent_FRU_Unknown,
13296					    "High CTL error volume, %d prints "
13297					    "skipped", skipped_prints);
13298#endif
13299				}
13300				if (bootverbose || verbose > 0)
13301					ctl_io_error_print(io, NULL);
13302			}
13303		}
13304		break;
13305	}
13306	case CTL_IO_TASK:
13307		if (bootverbose || verbose > 0)
13308			ctl_io_error_print(io, NULL);
13309		break;
13310	default:
13311		break;
13312	}
13313
13314	/*
13315	 * Tell the FETD or the other shelf controller we're done with this
13316	 * command.  Note that only SCSI commands get to this point.  Task
13317	 * management commands are completed above.
13318	 *
13319	 * We only send status to the other controller if we're in XFER
13320	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13321	 * received the I/O (from CTL's perspective), and so the status is
13322	 * generated there.
13323	 *
13324	 * XXX KDM if we hold the lock here, we could cause a deadlock
13325	 * if the frontend comes back in in this context to queue
13326	 * something.
13327	 */
13328	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13329	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13330		union ctl_ha_msg msg;
13331
13332		memset(&msg, 0, sizeof(msg));
13333		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13334		msg.hdr.original_sc = io->io_hdr.original_sc;
13335		msg.hdr.nexus = io->io_hdr.nexus;
13336		msg.hdr.status = io->io_hdr.status;
13337		msg.scsi.scsi_status = io->scsiio.scsi_status;
13338		msg.scsi.tag_num = io->scsiio.tag_num;
13339		msg.scsi.tag_type = io->scsiio.tag_type;
13340		msg.scsi.sense_len = io->scsiio.sense_len;
13341		msg.scsi.sense_residual = io->scsiio.sense_residual;
13342		msg.scsi.residual = io->scsiio.residual;
13343		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13344		       sizeof(io->scsiio.sense_data));
13345		/*
13346		 * We copy this whether or not this is an I/O-related
13347		 * command.  Otherwise, we'd have to go and check to see
13348		 * whether it's a read/write command, and it really isn't
13349		 * worth it.
13350		 */
13351		memcpy(&msg.scsi.lbalen,
13352		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13353		       sizeof(msg.scsi.lbalen));
13354
13355		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13356				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13357			/* XXX do something here */
13358		}
13359
13360		ctl_free_io(io);
13361	} else
13362		fe_done(io);
13363
13364bailout:
13365
13366	return (CTL_RETVAL_COMPLETE);
13367}
13368
13369/*
13370 * Front end should call this if it doesn't do autosense.  When the request
13371 * sense comes back in from the initiator, we'll dequeue this and send it.
13372 */
13373int
13374ctl_queue_sense(union ctl_io *io)
13375{
13376	struct ctl_lun *lun;
13377	struct ctl_softc *ctl_softc;
13378	uint32_t initidx, targ_lun;
13379
13380	ctl_softc = control_softc;
13381
13382	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13383
13384	/*
13385	 * LUN lookup will likely move to the ctl_work_thread() once we
13386	 * have our new queueing infrastructure (that doesn't put things on
13387	 * a per-LUN queue initially).  That is so that we can handle
13388	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13389	 * can't deal with that right now.
13390	 */
13391	mtx_lock(&ctl_softc->ctl_lock);
13392
13393	/*
13394	 * If we don't have a LUN for this, just toss the sense
13395	 * information.
13396	 */
13397	targ_lun = io->io_hdr.nexus.targ_lun;
13398	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13399	if ((targ_lun < CTL_MAX_LUNS)
13400	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13401		lun = ctl_softc->ctl_luns[targ_lun];
13402	else
13403		goto bailout;
13404
13405	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13406
13407	mtx_lock(&lun->lun_lock);
13408	/*
13409	 * Already have CA set for this LUN...toss the sense information.
13410	 */
13411	if (ctl_is_set(lun->have_ca, initidx)) {
13412		mtx_unlock(&lun->lun_lock);
13413		goto bailout;
13414	}
13415
13416	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
13417	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
13418	       sizeof(io->scsiio.sense_data)));
13419	ctl_set_mask(lun->have_ca, initidx);
13420	mtx_unlock(&lun->lun_lock);
13421
13422bailout:
13423	mtx_unlock(&ctl_softc->ctl_lock);
13424
13425	ctl_free_io(io);
13426
13427	return (CTL_RETVAL_COMPLETE);
13428}
13429
13430/*
13431 * Primary command inlet from frontend ports.  All SCSI and task I/O
13432 * requests must go through this function.
13433 */
13434int
13435ctl_queue(union ctl_io *io)
13436{
13437	struct ctl_softc *ctl_softc;
13438
13439	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13440
13441	ctl_softc = control_softc;
13442
13443#ifdef CTL_TIME_IO
13444	io->io_hdr.start_time = time_uptime;
13445	getbintime(&io->io_hdr.start_bt);
13446#endif /* CTL_TIME_IO */
13447
13448	/* Map FE-specific LUN ID into global one. */
13449	io->io_hdr.nexus.targ_mapped_lun =
13450	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13451
13452	switch (io->io_hdr.io_type) {
13453	case CTL_IO_SCSI:
13454	case CTL_IO_TASK:
13455		ctl_enqueue_incoming(io);
13456		break;
13457	default:
13458		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13459		return (EINVAL);
13460	}
13461
13462	return (CTL_RETVAL_COMPLETE);
13463}
13464
13465#ifdef CTL_IO_DELAY
13466static void
13467ctl_done_timer_wakeup(void *arg)
13468{
13469	union ctl_io *io;
13470
13471	io = (union ctl_io *)arg;
13472	ctl_done(io);
13473}
13474#endif /* CTL_IO_DELAY */
13475
13476void
13477ctl_done(union ctl_io *io)
13478{
13479	struct ctl_softc *ctl_softc;
13480
13481	ctl_softc = control_softc;
13482
13483	/*
13484	 * Enable this to catch duplicate completion issues.
13485	 */
13486#if 0
13487	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13488		printf("%s: type %d msg %d cdb %x iptl: "
13489		       "%d:%d:%d:%d tag 0x%04x "
13490		       "flag %#x status %x\n",
13491			__func__,
13492			io->io_hdr.io_type,
13493			io->io_hdr.msg_type,
13494			io->scsiio.cdb[0],
13495			io->io_hdr.nexus.initid.id,
13496			io->io_hdr.nexus.targ_port,
13497			io->io_hdr.nexus.targ_target.id,
13498			io->io_hdr.nexus.targ_lun,
13499			(io->io_hdr.io_type ==
13500			CTL_IO_TASK) ?
13501			io->taskio.tag_num :
13502			io->scsiio.tag_num,
13503		        io->io_hdr.flags,
13504			io->io_hdr.status);
13505	} else
13506		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13507#endif
13508
13509	/*
13510	 * This is an internal copy of an I/O, and should not go through
13511	 * the normal done processing logic.
13512	 */
13513	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13514		return;
13515
13516	/*
13517	 * We need to send a msg to the serializing shelf to finish the IO
13518	 * as well.  We don't send a finish message to the other shelf if
13519	 * this is a task management command.  Task management commands
13520	 * aren't serialized in the OOA queue, but rather just executed on
13521	 * both shelf controllers for commands that originated on that
13522	 * controller.
13523	 */
13524	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13525	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13526		union ctl_ha_msg msg_io;
13527
13528		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13529		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13530		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13531		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13532		}
13533		/* continue on to finish IO */
13534	}
13535#ifdef CTL_IO_DELAY
13536	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13537		struct ctl_lun *lun;
13538
13539		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13540
13541		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13542	} else {
13543		struct ctl_lun *lun;
13544
13545		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13546
13547		if ((lun != NULL)
13548		 && (lun->delay_info.done_delay > 0)) {
13549			struct callout *callout;
13550
13551			callout = (struct callout *)&io->io_hdr.timer_bytes;
13552			callout_init(callout, /*mpsafe*/ 1);
13553			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13554			callout_reset(callout,
13555				      lun->delay_info.done_delay * hz,
13556				      ctl_done_timer_wakeup, io);
13557			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13558				lun->delay_info.done_delay = 0;
13559			return;
13560		}
13561	}
13562#endif /* CTL_IO_DELAY */
13563
13564	ctl_enqueue_done(io);
13565}
13566
13567int
13568ctl_isc(struct ctl_scsiio *ctsio)
13569{
13570	struct ctl_lun *lun;
13571	int retval;
13572
13573	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13574
13575	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13576
13577	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13578
13579	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13580
13581	return (retval);
13582}
13583
13584
13585static void
13586ctl_work_thread(void *arg)
13587{
13588	struct ctl_thread *thr = (struct ctl_thread *)arg;
13589	struct ctl_softc *softc = thr->ctl_softc;
13590	union ctl_io *io;
13591	int retval;
13592
13593	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13594
13595	for (;;) {
13596		retval = 0;
13597
13598		/*
13599		 * We handle the queues in this order:
13600		 * - ISC
13601		 * - done queue (to free up resources, unblock other commands)
13602		 * - RtR queue
13603		 * - incoming queue
13604		 *
13605		 * If those queues are empty, we break out of the loop and
13606		 * go to sleep.
13607		 */
13608		mtx_lock(&thr->queue_lock);
13609		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13610		if (io != NULL) {
13611			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13612			mtx_unlock(&thr->queue_lock);
13613			ctl_handle_isc(io);
13614			continue;
13615		}
13616		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13617		if (io != NULL) {
13618			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13619			/* clear any blocked commands, call fe_done */
13620			mtx_unlock(&thr->queue_lock);
13621			retval = ctl_process_done(io);
13622			continue;
13623		}
13624		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13625		if (io != NULL) {
13626			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13627			mtx_unlock(&thr->queue_lock);
13628			if (io->io_hdr.io_type == CTL_IO_TASK)
13629				ctl_run_task(io);
13630			else
13631				ctl_scsiio_precheck(softc, &io->scsiio);
13632			continue;
13633		}
13634		if (!ctl_pause_rtr) {
13635			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13636			if (io != NULL) {
13637				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13638				mtx_unlock(&thr->queue_lock);
13639				retval = ctl_scsiio(&io->scsiio);
13640				if (retval != CTL_RETVAL_COMPLETE)
13641					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13642				continue;
13643			}
13644		}
13645
13646		/* Sleep until we have something to do. */
13647		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13648	}
13649}
13650
13651static void
13652ctl_lun_thread(void *arg)
13653{
13654	struct ctl_softc *softc = (struct ctl_softc *)arg;
13655	struct ctl_be_lun *be_lun;
13656	int retval;
13657
13658	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13659
13660	for (;;) {
13661		retval = 0;
13662		mtx_lock(&softc->ctl_lock);
13663		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13664		if (be_lun != NULL) {
13665			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13666			mtx_unlock(&softc->ctl_lock);
13667			ctl_create_lun(be_lun);
13668			continue;
13669		}
13670
13671		/* Sleep until we have something to do. */
13672		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13673		    PDROP | PRIBIO, "-", 0);
13674	}
13675}
13676
13677static void
13678ctl_enqueue_incoming(union ctl_io *io)
13679{
13680	struct ctl_softc *softc = control_softc;
13681	struct ctl_thread *thr;
13682	u_int idx;
13683
13684	idx = (io->io_hdr.nexus.targ_port * 127 +
13685	       io->io_hdr.nexus.initid.id) % worker_threads;
13686	thr = &softc->threads[idx];
13687	mtx_lock(&thr->queue_lock);
13688	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13689	mtx_unlock(&thr->queue_lock);
13690	wakeup(thr);
13691}
13692
13693static void
13694ctl_enqueue_rtr(union ctl_io *io)
13695{
13696	struct ctl_softc *softc = control_softc;
13697	struct ctl_thread *thr;
13698
13699	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13700	mtx_lock(&thr->queue_lock);
13701	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13702	mtx_unlock(&thr->queue_lock);
13703	wakeup(thr);
13704}
13705
13706static void
13707ctl_enqueue_done(union ctl_io *io)
13708{
13709	struct ctl_softc *softc = control_softc;
13710	struct ctl_thread *thr;
13711
13712	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13713	mtx_lock(&thr->queue_lock);
13714	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13715	mtx_unlock(&thr->queue_lock);
13716	wakeup(thr);
13717}
13718
13719static void
13720ctl_enqueue_isc(union ctl_io *io)
13721{
13722	struct ctl_softc *softc = control_softc;
13723	struct ctl_thread *thr;
13724
13725	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13726	mtx_lock(&thr->queue_lock);
13727	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13728	mtx_unlock(&thr->queue_lock);
13729	wakeup(thr);
13730}
13731
13732/* Initialization and failover */
13733
13734void
13735ctl_init_isc_msg(void)
13736{
13737	printf("CTL: Still calling this thing\n");
13738}
13739
13740/*
13741 * Init component
13742 * 	Initializes component into configuration defined by bootMode
13743 *	(see hasc-sv.c)
13744 *  	returns hasc_Status:
13745 * 		OK
13746 *		ERROR - fatal error
13747 */
13748static ctl_ha_comp_status
13749ctl_isc_init(struct ctl_ha_component *c)
13750{
13751	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13752
13753	c->status = ret;
13754	return ret;
13755}
13756
13757/* Start component
13758 * 	Starts component in state requested. If component starts successfully,
13759 *	it must set its own state to the requestrd state
13760 *	When requested state is HASC_STATE_HA, the component may refine it
13761 * 	by adding _SLAVE or _MASTER flags.
13762 *	Currently allowed state transitions are:
13763 *	UNKNOWN->HA		- initial startup
13764 *	UNKNOWN->SINGLE - initial startup when no parter detected
13765 *	HA->SINGLE		- failover
13766 * returns ctl_ha_comp_status:
13767 * 		OK	- component successfully started in requested state
13768 *		FAILED  - could not start the requested state, failover may
13769 * 			  be possible
13770 *		ERROR	- fatal error detected, no future startup possible
13771 */
13772static ctl_ha_comp_status
13773ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13774{
13775	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13776
13777	printf("%s: go\n", __func__);
13778
13779	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13780	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13781		ctl_is_single = 0;
13782		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13783		    != CTL_HA_STATUS_SUCCESS) {
13784			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13785			ret = CTL_HA_COMP_STATUS_ERROR;
13786		}
13787	} else if (CTL_HA_STATE_IS_HA(c->state)
13788		&& CTL_HA_STATE_IS_SINGLE(state)){
13789		// HA->SINGLE transition
13790	        ctl_failover();
13791		ctl_is_single = 1;
13792	} else {
13793		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13794		       c->state, state);
13795		ret = CTL_HA_COMP_STATUS_ERROR;
13796	}
13797	if (CTL_HA_STATE_IS_SINGLE(state))
13798		ctl_is_single = 1;
13799
13800	c->state = state;
13801	c->status = ret;
13802	return ret;
13803}
13804
13805/*
13806 * Quiesce component
13807 * The component must clear any error conditions (set status to OK) and
13808 * prepare itself to another Start call
13809 * returns ctl_ha_comp_status:
13810 * 	OK
13811 *	ERROR
13812 */
13813static ctl_ha_comp_status
13814ctl_isc_quiesce(struct ctl_ha_component *c)
13815{
13816	int ret = CTL_HA_COMP_STATUS_OK;
13817
13818	ctl_pause_rtr = 1;
13819	c->status = ret;
13820	return ret;
13821}
13822
13823struct ctl_ha_component ctl_ha_component_ctlisc =
13824{
13825	.name = "CTL ISC",
13826	.state = CTL_HA_STATE_UNKNOWN,
13827	.init = ctl_isc_init,
13828	.start = ctl_isc_start,
13829	.quiesce = ctl_isc_quiesce
13830};
13831
13832/*
13833 *  vim: ts=8
13834 */
13835